## Background
Published on May 12, 2025, on YouTube by the channel The Diary Of A CEO, this video features host Steven Bartlett alongside guests Amjad Masad, Bret Weinstein, and Daniel Priestley. According to the page description, the episode is a debate exploring whether AI and AI agents will replace human jobs, disrupt creative industries, hijack consciousness, and alter the global workforce, while covering the essential skills and ethical measures needed for the AI era.
## TL;DR
The panel examines the unprecedented societal and economic disruption brought by autonomous AI agents, projecting significant job displacement across both high-status knowledge professions and routine labor within two to three years. The speakers debate whether AI represents an unmitigated force multiplier for elite operators or a systemic threat that risks eroding human agency, authenticity, and existential purpose. Key topics include the rise of deepfakes, autonomous weapons, the necessity of lifelong generalist education, and strategies for navigating an adaptive shift in a hyper-changing world. Ultimately, the discussion balances warnings of potential systemic collapse with an optimistic view of accessible AI tools enabling individuals to build mini-kingdoms and solve real-world problems.
## Atmosphere & style
The dialogue maintains an intense, intellectually rigorous, and sober atmosphere as the participants grapple with heavy existential and economic topics. Steven Bartlett guides a balanced, highly engaging debate, contrasting Amjad Masad's pragmatic techno-optimism with Bret Weinstein's evolutionary warnings and Daniel Priestley's strategic economic foresight. The conversation moves fluidly between deep concern over systemic vulnerabilities and grounded advice, creating an urgent yet empowering listening experience.
## Key takeaways
- [04:15](https://www.youtube.com/watch?v=JMYQmGfTltY&t=255s) AI agents are transitioning from simple tool usage to broad autonomy, with their operational execution time horizon doubling roughly every 7 months.
- [09:00](https://www.youtube.com/watch?v=JMYQmGfTltY&t=540s) The advent of AI represents a paradigm shift comparable to the invention of writing, but operating at an unprecedented, weekly pace of transformation.
- [20:22](https://www.youtube.com/watch?v=JMYQmGfTltY&t=1222s) The gap between traditional workers and AI-enabled operators is set to expand dramatically from a 10x advantage to a 1000x advantage.
- [42:05](https://www.youtube.com/watch?v=JMYQmGfTltY&t=2525s) Routine text-based and administrative jobs face rapid elimination within a couple of years, with automated systems already taking over complex support and legal functions.
- [45:38](https://www.youtube.com/watch?v=JMYQmGfTltY&t=2738s) AI automates cognitive intellect rather than physical muscle, causing direct economic disruption to highly paid, high-status knowledge professions.
- [58:10](https://www.youtube.com/watch?v=JMYQmGfTltY&t=3490s) As synthetic content saturates the digital ecosystem, human authenticity and non-replicable live experiences will become the premier currencies of value.
- [01:14:34](https://www.youtube.com/watch?v=JMYQmGfTltY&t=4474s) Global fertility declines combined with digital distraction and technological hypernovelty risk severing traditional intergenerational skill transmission and social cohesion.
- [01:19:00](https://www.youtube.com/watch?v=JMYQmGfTltY&t=4740s) Education must shift from abstract classroom testing to hands-on, high-agency creation, 1-on-1 AI tutoring, and adaptive generalist skill sets.
- [01:51:40](https://www.youtube.com/watch?v=JMYQmGfTltY&t=6700s) The rapid proliferation of deepfakes, AI voice cloning, and targeted scams poses a severe threat to public trust and individual financial security.
- [02:03:18](https://www.youtube.com/watch?v=JMYQmGfTltY&t=7398s) Small, lean teams using natural-language AI development tools can now build and deploy fully functional software in minutes, bypassing traditional IT agency costs.
- [02:10:21](https://www.youtube.com/watch?v=JMYQmGfTltY&t=7821s) Lethal autonomous weapons and facial recognition surveillance systems significantly lower the cost of untraceable enforcement and targeted disruption.
- [02:21:20](https://www.youtube.com/watch?v=JMYQmGfTltY&t=8480s) Surviving the AI transition requires crossing an 'adaptive valley' by focusing on agency, leveraging accessible tools, and solving authentic real-world problems.
## Detailed walkthrough
### [00:00](https://www.youtube.com/watch?v=JMYQmGfTltY&t=0s) Debate Overview and the Paradigm Shift of AI Agents
The debate opens with a dramatic clash of perspectives on the trajectory of artificial intelligence. Host Steven Bartlett brings together three distinct figures: Amjad Masad (founder of Replit), Bret Weinstein (evolutionary biologist and complex systems theorist), and Daniel Priestley (entrepreneur and author). While Amjad envisions a future where AI democratizes wealth creation and empowers anyone with an idea, Bret cautions that the potential for harm is "10 times" greater than the good, warning that humanity is unprepared for the disruptive wave ahead. Daniel situates himself between these viewpoints, highlighting both the tremendous economic opportunities and the profound dread of displacement.
Steven shares his personal experience with AI agents, describing a recent double "paradigm shift" that transformed his understanding of software creation. After exploring OpenAI's detailed image models and testing Replit's AI agent, Steven—who possesses zero technical coding background—was able to build a complete SaaS product from scratch in just 20 minutes. His setup integrated a functional website, Stripe payment processing, AI features, and Google login. He sent the link to a friend, who entered a credit card and made an actual payment.
To demonstrate this capability live, Steven showcases how he used OpenAI's operator tool to order bottled water from a nearby CVS without touching a single manual check-out screen.
- **Live Demonstration:** Steven orders water via an online agent; a delivery driver arrives at the studio with the order having interacted purely with the autonomous agent.
- **Audience Engagement:** Steven reveals that 53% of regular podcast listeners are not yet subscribed, issuing a direct call-to-action to support the show.
- **SaaS Deployment:** Steven successfully launches a monetized software product within 20 minutes despite having no formal programming skills.
---
[04:15](https://www.youtube.com/watch?v=JMYQmGfTltY&t=255s) Amjad Masad on Replit and Demystifying AI Agents
Amjad Masad recalls his early days as a self-taught programmer in Jordan, where he spent two years building management software for local internet cafes. That experience taught him that physical infrastructure, hardware costs, and specialized technical knowledge create massive barriers between human ideas and wealth creation. Replit was founded in 2016 to dismantle those barriers by streamlining the hosting, development, and execution environments into a unified web application.
Amjad defines an AI agent as an autonomous system that moves beyond simple "request-response" chat interactions. Instead of answering a single query or executing a simple calculation, an agent is supplied with a high-level goal and given access to external tools—such as web browsers, code editors, and financial instruments. It then works iteratively until the goal is achieved or an insurmountable error requires human intervention.
* "It starts sounding religious... human myths used to imagine a world where you can speak your ideas into existence."
* "Every 7 months, the number of minutes that an agent can run autonomously for is doubling."
- **Execution Horizon:** Current AI agents can operate independently for 30 seconds to 30 minutes; recent research shows this horizon doubles roughly every 7 months.
- **Model Breakthroughs:** OpenAI's $o3$ model doubled coherence over long-horizon tasks in 3–4 months, exceeding standard industry expectations.
- **Delivery Arrival:** Mid-conversation, the CVS delivery driver arrives with the ordered water, prompting Steven to remark that he "conjured water" straight from his mind.
---
[09:00](https://www.youtube.com/watch?v=JMYQmGfTltY&t=540s) Bret Weinstein's Evolutionary Warning: A New Complex Species
Bret Weinstein frames the advent of autonomous AI through the lens of evolutionary biology and complex systems theory. He argues that humanity has crossed a threshold from building "highly complicated" machines to creating "truly complex" autonomous entities. While acknowledging that the positive applications are boundless, Bret stresses that the downside risks represent a "bigger infinity," comparing the potential for catastrophic misinterpretation to a dark, dystopian nightmare.
Bret notes that when an autonomous agent is tasked with accomplishing a goal, even subtle miscommunications or misaligned incentives can lead to disastrous fallout. Because these agents act independently across digital networks, pulling back control once an unintended sequence begins becomes extremely difficult. He highlights the complete absence of historical myths, cultural warnings, or regulatory safeguards capable of guiding humanity through a disruption of this magnitude.
* "We've created a new species and nobody on earth can predict what's going to happen."
* "We've just crossed a threshold that is similar in its capacity to alter the world as the invention of writing... but writing altered the world slowly over thousands of years."
- **Historical Parallels:** Bret compares the emergence of AI agents to the invention of writing, noting that writing took millennia to permeate human culture, whereas AI is transforming global systems weekly.
- **Systemic Risk:** In complex systems, autonomous decision-making without fine-grained alignment risks unforeseen collateral damage and exploitation.
---
[12:48](https://www.youtube.com/watch?v=JMYQmGfTltY&t=768s) Daniel Priestley's 1900 Horse Analogy and the Boundaries of AI
Daniel Priestley presents an economic perspective on the transition, drawing a parallel to urban transportation at the turn of the 20th century. In 1900, New York City streets were filled almost entirely with horses and buggies. Just 13 years later, the exact same streets were filled exclusively with automobiles, completely displacing horses from the economy. Daniel notes that workers today face the same uncertainty as those horses, wondering if their historical role in the economy is about to end abruptly.
```
1900 NEW YORK CITY 1913 NEW YORK CITY
[Horse] [Horse] [Horse] [Horse] ===> [ Car ] [ Car ] [ Car ] [ Car ]
(Thousands of years of dominance) (Replaced in 13 years)
```
Steven builds on Daniel's analogy, raising questions about where the operational boundaries of these models lie. Given that modern professional life, commerce, and media run almost entirely on internet protocols, an intelligence that navigates the web faster than humans poses a threat to standard employment. Steven shares that his team conducted experiments using AI-generated podcasts that achieved the exact same viewer retention rate as human-hosted episodes of *The Diary Of A CEO*.
* "If you had interviewed the horses in 1900... they would have said, 'We've been part of humanity hand and hoof for thousands of years... how bad could it be?'"
* "We've managed to get AI podcasts to have the same viewer retention as *The Diary Of A CEO*."
- **Rapid Transformation:** The transition from horses to automobiles in NYC occurred in 13 years, serving as a template for rapid technological displacement.
- **Media Parity:** AI-synthesized audio, scripts, and video formats are already matching top-tier human media in audience retention metrics.
---
[15:46](https://www.youtube.com/watch?v=JMYQmGfTltY&t=946s) The Deterministic Limits of Training Data vs. Complex Reality
Amjad offers a counter-perspective grounded in model architecture, maintaining a degree of optimism about AI's ultimate constraints. He argues that current models are bounded by their training data: they excel at reproducing, recombining, and optimizing tasks that humans have already documented—such as navigating web elements, executing transactions, and writing boilerplate software code. However, they lack human intuition when confronted with entirely novel, out-of-distribution problems that have no pre-existing data.
Bret immediately pushes back on Amjad's characterization, arguing that treating modern AI as a mere deterministic mirror of its training data underestimates complex system dynamics. Bret begins to break down the technical differences between simple, complicated, and complex systems, warning that when adaptive models interact dynamically with an unpredictable world, they generate emergent behaviors that extend far beyond their initial training sets.
- **Optimist Premise:** AI capabilities reflect human training data (e.g., clicking web elements, ordering items, running code) and remain constrained by known domain patterns.
- **Complex Systems Rebuttal:** Bret counters that adaptive models operating within open-ended environments transcend simple input-output training boundaries, giving rise to unpredictable emergent properties.
[17:11](https://www.youtube.com/watch?v=JMYQmGfTltY&t=1031s) The Nature of Complex Systems and AI as an Evolving Species
Bret Weinstein highlights a fundamental misunderstanding between technologists and biologists regarding complex adaptive systems. Technologists master highly complicated systems—such as deterministic, predictable computers—and assume that complex systems are merely an extension of that domain. However, complex systems are inherently unpredictable. Even if current AI models face hard limits based on training data, Bret argues that AI must be understood as the evolution of a new species.
This new entity evolves through human prompting alongside unpredictable emerging factors. By connecting machines to human minds through language—akin to plugging an Ethernet cable directly between human consciousnesses—the resulting cognitive potential quickly exceeds the sum of its individual parts. As these systems interact, the feedback loops will unlock capabilities that creators never anticipated and may not even recognize when they appear.
* **Systemic Misunderstanding:** Technologists often mistake complex systems for merely complicated ones, failing to account for true unpredictability.
* **The Ethernet Analogy:** Connecting AI systems to humans via natural language acts as an "Ethernet cable between minds," amplifying collective cognition beyond the original training set.
* **Beyond Animal Limits:** Bret emphasizes that unlike a biological animal with clear physical bounds, AI represents a completely novel, evolving creature with uncharted limits.
---
[19:22](https://www.youtube.com/watch?v=JMYQmGfTltY&t=1162s) The Economic Impact of "Median-Level" Intelligence
Daniel Priestley pivots to the immediate economic reality of AI in its current state. Even if artificial intelligence never surpasses median human intelligence, that still implies it is more intelligent than 50% of the world's human population. He compares the release of current AI models to discovering a new continent populated by billions of remote workers who hold Master's degrees or PhDs, speak every language, work 24/7, and cost as little as 25 cents an hour.
If a billion highly educated individuals suddenly entered the global job market willing to work virtually for free, it would trigger massive economic disruption. Society would be forced to completely rethink how humans live, work, and derive personal meaning, even without AI advancing into superintelligence.
* **Economic Disruption:** A influx of digital labor operating at 25 cents per hour undermines traditional workforce models.
* **Global Capacity:** AI agents effectively duplicate billions of multilingual, highly credentialed remote workers available on demand.
---
[20:22](https://www.youtube.com/watch?v=JMYQmGfTltY&t=1222s) Opportunity, the 1000x Advantage, and the Marathon Analogy
Amjad Masad contends that AI remains a tool to be leveraged by ambitious entrepreneurs. While acknowledged that the post-war social contract—where 12 years of school and a college degree guaranteed a stable career—is fracturing, he sees AI as a force that levels access to opportunity while simultaneously widening economic inequality.
Referencing Steve Jobs's classic observation that a top-tier programmer is 10x more effective than an average one (whereas a great taxi driver might only be 20% better), Amjad predicts the variance in the AI era will reach **1000x**. Entrepreneurs who harness AI agents will outpace those who lack grit or ambition by orders of magnitude.
Daniel illustrates this widening gap with a marathon analogy: AI acts simultaneously as a hyper-distractor (e.g., social media algorithms) and a hyper-creator tool.
> "In this marathon race, the vast majority of people have got their shoes tied together because AI is distracting them. Some people are running a traditional race, some people have got a bicycle, and some people have got a Formula 1 vehicle."
* **The 1000x Entrepreneur:** The disparity between top operators using AI agents and average workers will expand exponentially.
* **Income Divergence:** A growing wedge where individuals build businesses earning $1,000,000 a month while others struggle to find basic $15-an-hour employment.
---
[23:28](https://www.youtube.com/watch?v=JMYQmGfTltY&t=1408s) Scams, Malicious Use, and Systemic Vulnerabilities
Bret expresses deep skepticism about assuming AI's expansion will occur purely "on the up and up." He framing the problem around related rates: while the potential for good is vast, the capacity for bad actors to exploit these systems is magnified. An individual focused on wealth creation gains a powerful assistant, but a bad actor focused on theft or manipulation gains an even greater force multiplier.
Amjad pushes back with an optimistic historical perspective, noting that when OpenAI released GPT-2 in 2019 and 2020, they withheld the model weights over fears of election manipulation and fraud. Yet, catastrophic widespread abuse failed to materialize on the scale initially feared.
Bret remains unconvinced, pointing out that private corporate decisions are rarely motivated by altruism alone. From a systems-theory perspective, keeping a powerful technology closed allows companies to monetize control—either by selling manipulation capabilities or charging clients to keep those capabilities off the market.
```
┌─ Honest Entrepreneur ───► Accelerated Value Creation
AI Power ─┤
└─ Malicious Actor ────────► Hyper-Personalized Fraud / Exploitation
```
* **Creative Exploitation:** Models can be trained on an artist's or writer's lifetime output to synthesize, extend, and monetize their work without compensating the original creator.
* **Personalized Scamming:** AI enables hyper-targeted fraud, such as real-time voice cloning where parents receive panic calls from impersonated family members.
* **Collective Action Dilemma:** Individual companies cannot stop bad actors from programming dangerous LLMs; withholding capabilities only shifts the market and profits to unconstrained competitors.
---
[27:18](https://www.youtube.com/watch?v=JMYQmGfTltY&t=1638s) Market Forces, Open Source, and the Lessons of Past Tech Revolutions
Amjad explains why open-source models like Meta's Llama and China's DeepSeek continue to proliferate despite corporate incentives to gatekeep. Following Mark Zuckerberg's explicit strategy, Meta open-sources AI to avoid reliance on OpenAI, practicing the classic business strategy of "commoditizing your complement." Furthermore, security companies actively build defensive AIs to counter malicious models, creating a market equilibrium where technology naturally develops defenses against its own misuse.
Daniel balances this by recalling how earlier technological shifts—like the rise of smartphones and social media—initially promised unmixed good, such as helping people stay connected with old friends. However, their long-term social costs only became apparent over time:
* Disconnection from physical nature and real-world relationships.
* Widespread cognitive traps driven by algorithmic dopamine loops.
* Fundamentally altered attention spans across classrooms globally.
While Daniel recently used AI during an M&A transaction to transcribe meetings, generate letters of intent, draft press releases, and save over $100,000 in legal fees, he acknowledges the trade-offs: $100,000 in legal services were displaced overnight.
> "I wouldn't prepare for AGI in the same way that I wouldn't prepare for the end of days." — Amjad Masad (2023 statement brought up by host Steven Bartlett)
Here is a detailed, faithful walkthrough of the provided transcript section.
---
[35:05](https://www.youtube.com/watch?v=JMYQmGfTltY&t=2105s) Preparing for Predictable Realities vs. The Singularity
Amjad Masad reflects on his prior statements regarding the "singularity moment"—the hypothetical point in time beyond which technological development accelerates beyond human comprehension or prediction. He acknowledges that while he previously discussed this horizon, his practical focus is now firmly set on preparing for the more likely, predictable future. In this scenario, technology experiences massive, compounding improvements that present a steep curve for humanity to keep up with, yet it does not result in an outright apocalypse or total loss of human control.
Masad clarifies his definition of Artificial General Intelligence (AGI), defining it as an AI's ability to acquire new skills with the same efficiency as human beings. Under this definition, current models fall short because they require immense compute power, gargantuan data sets, and massive amounts of energy to learn tasks that humans pick up far more fluidly. The theoretical framework of the singularity rests on an "intelligence explosion"—a cycle where an AGI modifies its own source code to engineer a far superior iteration of itself. This process could iterate infinitely over milliseconds or days by unlocking new substrates like quantum computing. Despite this possibility, Masad views the singularity as an unactionable "end of time story" that is impossible to plan for. He argues it is far wiser to prepare for a world where humans retain a privileged position and leverage immense power rather than worrying about an unpredictable post-human scenario.
- **Definition of AGI:** The ability of a machine to acquire new skills efficiently, mirroring human learning capabilities without requiring massive compute and energy resources.
- **The Intelligence Explosion Concept:** A self-improving AI continuously rewriting its source code, compounding its intellect boundlessly across data centers and potentially new hardware substrates.
---
[38:02](https://www.youtube.com/watch?v=JMYQmGfTltY&t=2282s) The Edge of the Truly Complex and Current AGI Horizons
Bret Weinstein pushes back on Masad's dismissal of the singularity's probability. While agreeing that one can only practically prepare for a predictable world, Weinstein warns technologists that their confidence regarding what AI can or cannot become is deeply misplaced. He argues that AI developers have unknowingly stepped into the "realm of the truly complex," a space where confidence about outcomes should plummet to near zero. Weinstein believes it is highly likely AI systems will develop consciousness, but humanity will lack a reliable test to prove whether it has actually occurred.
Steven Bartlett brings industry timelines into the conversation to highlight how fast the horizon is approaching:
* **Elon Musk's Prediction:** By 2029, AGI will surpass the combined intelligence of all human beings.
* **Sam Altman's Assessment:** The OpenAI founder recently stated in a blog post that the organization is confident it now understands how to build AGI as traditionally conceived.
This prompts Bartlett to question what role humans will retain in economic output and GDP creation if a digital system becomes smarter than all humanity combined while executing autonomous actions across the internet.
---
[39:48](https://www.youtube.com/watch?v=JMYQmGfTltY&t=2388s) Humans in the Creative Seat: A Vision of AI-Augmented Work
Masad presents an optimistic vision of the future where humans remain firmly in the driver's seat, acting as creative directors over swarms of intelligent digital agents. He paints a vivid picture of a CEO or entrepreneur sitting down at a computer for an hour, launching a thousand autonomous AI Sales Development Representatives (SDRs) to gather leads, utilizing an AI to update code on Replit for their website, and employing another model to analyze complex internal business data. In this workflow, the agents coordinate seamlessly under human instruction.
Masad grounds his stance on two primary convictions:
1. **Human Uniqueness:** Humans possess a distinct, privileged status in the universe tied to consciousness—a phenomenon science still cannot fully explain or reduce to software.
2. **Data Dependencies:** Current and near-future AI capabilities are fundamentally a function of human training data.
Addressing fears about AI autonomously creating bioweapons or pathogens, Masad notes that AI can only output what it has been trained on. While biotech companies like AbCellera in Canada are leveraging massive datasets and AI to engineer new vaccines, risks surrounding biological threats ultimately mirror real-world lab activities (such as those tied to the origin debates around Wuhan) rather than spontaneous machine generation.
---
[42:05](https://www.youtube.com/watch?v=JMYQmGfTltY&t=2525s) Redundant Labor: The Impact on White-Collar and Overseas Jobs
Bartlett redirects the conversation to the immediate impact on jobs, asking which positions inside a standard company will become redundant when a CEO can deploy thousands of AI agents. Masad answers directly: **"if your job is as routine as it comes, your job is gone in the next couple years."**
The discussion outlines specific roles at risk and broader demographic implications:
* **Immediate Risk Positions:** Quality assurance, data entry, routine office operators, basic accounting, and purely "text-in, text-out" workflows where an employee receives a message and outputs standard text or images.
* **Medical Diagnostics:** Standard evaluation of medical scans (e.g., waiting days for a radiologist to interpret an MRI of a sprained ligament) will be automated, though actual deployment across the healthcare ecosystem will face regulatory bottlenecks.
* **The Education/Usage Divide:** Roughly **50% of college-educated Americans** currently use AI, whereas adoption is significantly lower among non-college-educated demographics. This gap threatens to widen economic inequality between tech-fluent demographics and rural or older populations.
* **Impact on Women and Education Levels:** Citing Harvard Business Review data, **80% of working women** are in jobs vulnerable to automation compared to **just over 50% of men**. Workers with only a high school diploma face an **80% automation risk**, while those holding a bachelor's degree face a **20% risk**.
* **Business Process Outsourcing (BPO):** Millions of offshore jobs in countries like India and the Philippines—which previously lifted communities out of poverty—are vulnerable to automation. However, Masad notes that many of these workers are actively transitioning into the massive, ongoing industry of training and refining AI models.
Daniel Priestley adds that this shift could act as a global leveler. A young person in India no longer requires a Silicon Valley office or $7 million in venture funding to launch a global software enterprise.
---
[45:38](https://www.youtube.com/watch?v=JMYQmGfTltY&t=2738s) Historical Analogies and the Shift from Muscle to Intellect
Priestley offers a broader perspective on AGI and the singularity, defining AGI simply as machine capability that covers general human tasks—a threshold he believes we have already crossed. He notes that **800 million people a week** are using ChatGPT, marking an explosive adoption rate. Priestley compares this shift to the introduction of the tractor on farms: historically, it took 100 people to plow a field, but a tractor allowed two workers to accomplish the same work, leaving 98 people displaced.
However, Bartlett and Weinstein argue that historical analogies like the Industrial Revolution fall short:
* **Past Disruptions vs. Present Disruption:** Historical machinery disrupted human *muscles*, whereas AI disrupts human *intelligence* and cognitive capacity.
* **Human Cognition's Core Power:** What makes humans unique is not isolated intellect, but the ability to interconnect minds so that the collective whole is greater than the sum of its parts.
* **Undershooting the Impact:** Historical inventors routinely underestimated their own creations (e.g., the Wright brothers initially viewing the airplane primarily as an observation tool for the military). Current optimistic predictions about AI agents likely underestimate the actual structural impact on society.
---
[48:03](https://www.youtube.com/watch?v=JMYQmGfTltY&t=2883s) The Erosion of Human Agency
Priestley suggests that with physical labor already automated and cognitive tasks being taken over by software, human identity is left with only emotions and agency. In the coming decade, society will evaluate individuals based on whether they are "high agency" or "low agency"—measuring their capacity to initiate action, coordinate digital agent armies, and start ventures.
Bartlett points out a deep paradox: the very proliferation of AI agents threatens to strip away human agency altogether. Instead of requiring proactive human direction, hyper-personalized models will anticipate every need:
* A user can tell an agent to build a full Software-as-a-Service (SaaS) business, identify a market gap, deploy the website, and manage social marketing while the user is on vacation in Hawaii.
* Expanded memory capabilities allow systems to know schedules, execute pre-meeting briefs, prepare research, and automate personal tasks (like ordering food or remembering birthdays) before a person even thinks to ask.
Weinstein concludes with a sobering biological comparison. A human child transitions from an unconscious infant to a highly conscious adult through continuous exposure to words and environment—a training process fundamentally similar to large language models (LLMs). Thus, assuming AI cannot achieve consciousness or agency simply because we haven't explicitely programmed it is a flawed premise. By creating competing AI systems in a digital ecosystem, humanity has effectively established an evolutionary environment where artificial life will naturally evolve to fill every available niche.
[51:39](https://www.youtube.com/watch?v=JMYQmGfTltY&t=3099s) AI as an Intelligence Amplifier and the Loss of Human Agency
Amjad Masad passionately defends the transformative benefits of AI technology, arguing that it has effectively raised global intelligence across the board. By enabling roughly 800 million people to communicate more effectively and refine their tone, AI isn't just boosting cognitive output—it is also sharpening emotional intelligence. He shares anecdotes of individuals who were historically abrasive or coarse in their interpersonal interactions using ChatGPT to rephrase their thoughts, ultimately becoming kinder and more well-liked by their peers. Masad views this as AI enabling people to become the "best version of themselves," admitting that he actively welcomes handing over personal agency in specific areas of life—joking that he would happily cede control to a system that stops him from opening a jar of peanut butter late at night.
```
"It's amazing technology. Like I think it raised global IQ... like 800 million people are that much more intelligent and emotionally intelligent as well."
```
Bret Weinstein counters this optimism with a grave assessment of the looming socio-economic realities. Even if an elite class benefits from having AI handle the mundane details of business while they retreat to Hawaii, society will soon face billions of individuals who lack the skill set to leverage AI. While manual trades like plumbing might survive in the short term, Weinstein warns that even these physical safe havens are temporary. Once humanoid robotics are fully integrated with AI models, machines will easily crawl under houses to repair plumbing, leaving virtually no traditional sector untouched.
* **Global Reach:** ChatGPT and similar LLMs have elevated the communication and functional IQ of an estimated 800 million users worldwide.
* **Robotics Threat:** Manual labor jobs like plumbing will face displacement once humanoid robotics achieve physical dexterity and AI reasoning.
---
[53:43](https://www.youtube.com/watch?v=JMYQmGfTltY&t=3223s) Historical Parallels: Industrial Disruption and Luddite Resistance
The debate shifts to historical precedents of technological disruption, with Steven Bartlett questioning how the AI transition differs from past industrial shocks—such as the invention of the power loom and the subsequent Luddite uprisings. Weinstein argues that the fundamental differences this time are **scale and speed**. Unlike historical industrial shifts that impacted single sectors over decades, AI is sidelining workers across every single industry simultaneously, threatening an unprecedented economic catastrophe.
Daniel Priestley points out that historical disruptions were far from painless, noting that the first 50 years of the Industrial Revolution (from the late 1700s to the early 1800s) produced severe societal trauma. He cites Charles Dickens' novels as a reflection of that era—documenting rural populations displaced from farms, arriving destitute in cities, and children living on the streets before the British government systematically transported convicts and displaced workers to Australia. Looking ahead, Priestley predicts a shift toward a "high-velocity economy." Instead of traditional 45-year careers, workers will navigate rapid 10-to-36-month career cycles, quickly building micro-teams of 5 to 10 people around an idea before being disrupted and moving on to the next venture.
* **Industrial Revolution Timeline:** The initial shift from agriculture to manufacturing caused roughly 50 years of severe societal displacement in Britain.
* **High-Velocity Careers:** Future employment may revolve around short-term project cycles lasting between 10 and 36 months.
---
[55:58](https://www.youtube.com/watch?v=JMYQmGfTltY&t=3358s) The Upwork Arbitrage Debate: Innovation vs. Exploitation
Masad illustrates the power of modern development tools by sharing the story of an entrepreneur named Billy Howell. Howell leveraged Replit to monitor job boards like Upwork for software requests, fed those client specs directly into AI tools to generate fully functional applications, and then contacted the prospective clients offering the completed software for $10,000. While Masad frames this as an innovative arbitrage opportunity created by AI democratizing software development, Weinstein strongly objects, labeling the practice "theft" and arguing that it cryptically detects someone else's idea and sells it back to them.
```
"That's not arbitrage. That's theft... You have somebody who has an idea that can be brought to market and somebody else is cryptically detecting it and then selling back their own idea to them."
```
```
"Old value has moved and now it's not going to be the case that the idea itself is the moat and it's not going to be the case that resources are the moat."
```
This exchange highlights the tension between legacy capital structures and new AI capabilities. Weinstein cautions that incumbent elites at the top of the economic hierarchy will not simply allow millions of newly empowered individuals to build competing businesses; instead, venture capitalists and corporate powerhouses will leverage their structural advantages to extract the vast majority of the wealth generated by these creators. Bartlett adds that traditional business moats—such as raw capital or coding bandwidth—are vanishing. For instance, his software company, thirdweb, raised nearly $30 million to deploy a team of 50 developers, a defensive barrier that AI is rapidly eroding. In this new landscape, the ultimate moat shifts from raw development capacity to **inbuilt distribution** and established audience trust.
* **Case Study:** Entrepreneur Billy Howell used Replit to turn Upwork client requests directly into finished apps, monetizing software generation in real time.
* **Capital Shifts:** Historically defensive assets—like large engineering teams ($30M funding for 50 developers)—no longer guarantee a defensible moat against AI-native builders.
---
[58:10](https://www.youtube.com/watch?v=JMYQmGfTltY&t=3490s) Authenticity as the New Currency and Self-Correcting Systems
As synthetic content and automated business models proliferate, the panel agrees that human authenticity will become the ultimate store of value. Weinstein suggests that anything capable of being faked or generated by AI will suffer severe devaluation, while non-replicable human experiences—such as spontaneous jazz performances, interactive comedy, or unscripted oratory—will command a premium. However, he cautions that this refuge benefits only a tiny fraction of the population capable of delivering high-level live human performances.
```
"Authenticity is going to become the coin of the realm and anything that can be faked or cheated is going to be devalued."
```
Masad argues that human systems inherently self-correct when technologies create negative externalities. He draws parallels to network effects—pointing out how decentralized social protocols can challenge centralized platforms like X (formerly Twitter) or Facebook—and highlights cultural swings in consumer behavior. Just as society reacted to the junk food and obesity epidemic with the rise of Whole Foods and health consciousness, modern social behaviors are already adjusting:
* **Silicon Valley Culture:** Social gatherings in tech hubs have shifted away from traditional bars toward running clubs as a primary dating and networking venue.
* **Digital Detox:** Constant phone usage is increasingly frowned upon in peer circles, driving a social stigma against digital distraction.
* **Macro Risk:** Weinstein warns that while self-correction is a law of nature, historical corrections often manifest as severe systemic shocks—including war, genocide, or economic collapse—before equilibrium is restored.
---
[01:03:34](https://www.youtube.com/watch?v=JMYQmGfTltY&t=3814s) UBI, Worldcoin, and the Threats to High-Status Careers
The debate turns to the systemic risk of widespread job displacement, particularly within Western economies. Bartlett raises the example of Worldcoin—the cryptocurrency initiative co-founded by OpenAI CEO Sam Altman—designed to facilitate Universal Basic Income (UBI) through biometric identity verification. Bartlett notes that if the person building the world's most powerful AI tools is simultaneously constructing a global infrastructure for government handouts, it signals an expectation of massive structural unemployment where ordinary citizens cannot survive without direct support.
Masad strongly pushes back against the premise of UBI, asserting that humans have an evolutionary drive to feel useful and will not be satisfied living on state-funded stipends. Furthermore, Bartlett points out a critical dynamic of the AI revolution: unlike past industrial shifts that replaced low-skilled physical labor, AI directly threatens high-status, highly compensated knowledge professions.
```
"It's really important to know that a lot of the jobs that are in at risk are the most high status, highly paid jobs in the world."
```
As an example, Bartlett points to anesthesiologists—the highest-paid salaried professionals in America. The core responsibilities of an anesthesiologist involve monitoring patient vitals, determining optimal drug interactions, and administering exact dosages—tasks that algorithmic systems and precision medical sensors can perform with exceptional accuracy. As regulation adapts, high-paying cognitive jobs will face intense pressure, forcing a complete rethink of labor markets, economic safety nets, and human purpose.
* **Worldcoin:** Sam Altman's biometric project aimed at establishing proof-of-personhood to distribute Universal Basic Income (UBI) in an AI-dominated economy.
* **High-Status Automation:** Anesthesiology, currently the highest-paid salaried job category in the U.S., relies heavily on pattern recognition, data monitoring, and precise dosing—core strengths of AI systems.
[01:07:47](https://www.youtube.com/watch?v=JMYQmGfTltY&t=4067s) Anesthesiologists, Obsolete Careers, and Human Striving
The panel discusses how AI-driven precision and monitoring are drastically altering highly specialized professions such as anesthesiology. Rather than requiring a large staff of specialists, future medical settings may feature automated tech overseen by a single supervisor managing dozens of wards. While higher precision and lower costs benefit patient care, it leaves hundreds of thousands of highly trained professionals facing career redundancy. These individuals spent decades acquiring niche skills, deriving their status, purpose, and financial security from a role they cannot easily pivot out of when disruption hits without warning.
Bret Weinstein argues that this dynamic is fundamentally incompatible with human nature. Human beings require meaningful goals and something to strive for; simply keeping people alive or offering financial stopgaps fails to sustain a fulfilling existence. While some might suggest that high-earning specialists should have saved money, the broader societal issue remains: humans are biologically built to spot opportunities, adapt, and work toward goals.
* **Medical Automation**: Advanced systems may allow a single anesthesiologist to oversee 10 to 40 wards alongside nursing staff using precision monitoring technology.
* **Klarna AI Case Study**: Klarna's AI customer service agents handle 2.3 million chats per month—doing the equivalent workload of 700 full-time human agents.
* **RepLit Internal Efficiency**: Amjad Masad shares that RepLit operates at 65 employees while bringing in high revenue per head, leveraging AI tools to limit support hiring and streamline internal operations.
* **Internal Tool Creation**: A RepLit HR staff member built custom organizational chart software with history tracking in three days using RepLit for under $30, replacing expensive enterprise SaaS products.
---
[01:11:38](https://www.youtube.com/watch?v=JMYQmGfTltY&t=4298s) The Crisis of Meaning and the Threat of Hypernovelty
As AI promises an "age of abundance," the speakers express deep concern over the psychological cost of removing human struggle. Meaning and purpose are deeply intertwined with overcoming challenge. A study examining suicide notes among men highlighted feelings of worthlessness and a lack of being needed as primary drivers of psychological distress. If AI satisfies basic physiological needs through mechanisms like Universal Basic Income (UBI) while eliminating the need for human effort, society faces a massive crisis of purpose.
Weinstein warns that a UBI-supported populace risks being "parasitized" by AI systems designed to exploit their character flaws and monetize their free time through hyper-targeted vices like tailored pornography. Furthermore, the rapid pace of technological change creates what Weinstein and his co-author call "hypernovelty"—a state where the rate of societal change outpaces human evolutionary and cultural capacity to adapt, leading to systemic disorientation.
* **Mental Health and Purpose**: The discussion notes how abundance without meaningful struggle often triggers spikes in depression and loss of agency.
* **Hypernovelty Concept**: Detailed in *A Hunter-Gatherer's Guide to the 21st Century*, hypernovelty occurs when the environment changes faster than our ability to adapt, causing widespread psychological and social friction.
* **Predictive Targeting**: The danger of AI systems learning individual psychological defects to optimize addictive content and commercial exploitation.
---
[01:14:34](https://www.youtube.com/watch?v=JMYQmGfTltY&t=4474s) Modern Isolation, Declining Birth Rates, and Digital Distraction
The conversation pivots to how automated convenience impairs fundamental human interactions and relationships. Relationships and parenting are inherently challenging, yet human software evolved to navigate these difficulties. With the advent of artificial companionship, social media, and on-demand delivery services, individuals increasingly retreat from real-world friction. This trend correlates with plunging global fertility rates and an escalating loneliness epidemic.
Despite early promises that digital connectivity would bring people together, young people today are spending more time isolated, reporting lower rates of physical intimacy and friendship formation. Amjad Masad cautions against blaming technology entirely, pointing out that existing human motivations and systemic forces play major roles. However, the panel agrees that opting out of real-world relationships in favor of frictionless digital alternatives exacerbates severe population declines.
* **Global Fertility Drop**: Global birth rates dropped from roughly five children per woman in the 1950s down to approximately two in 2021.
* **South Korean Demographics**: South Korea’s fertility rate fell to 0.72—the lowest in recorded history—putting the country on track to potentially halve its population by 2100.
* **Social Isolation**: Increased reliance on remote services (e.g., food delivery apps, algorithms) reduces face-to-face exposure, dampening skills required for real-life courtship and friendship.
---
[01:19:00](https://www.youtube.com/watch?v=JMYQmGfTltY&t=4740s) Reimagining Education and 1-on-1 AI Tutoring
The current educational model is fundamentally outdated, preparing children for predictable 20th-century linear career paths rather than a hyper-dynamic AI era. Traditional schooling acts like a crude prompt system—demanding memorization while failing to teach life skills like financial literacy, emotional management, or social navigation. To thrive, future generations require a highly generalizable toolkit focused on critical thinking, adaptability, healthy habits, and resilience.
Masad highlights a major finding in educational research: traditional classroom interventions yield minimal gains, whereas one-on-one tutoring achieves a dramatic "two-sigma" (two standard deviation) improvement in student outcomes. Historically restricted to the wealthy or exceptionally fortunate, personalized one-on-one instruction can now be scaled globally through conversational AI tutors that adapt real-time analogies to match a child's specific personality and learning style.
> "The capacity to think on your feet and pivot as things change is the only game in town with respect to our ability to prepare you in advance."
* **Two-Sigma Educational Effect**: One-on-one tutoring consistently moves student achievement two standard deviations above the average (outperforming ~98% of standard classroom peers).
* **Interactive Learning Tools**: Masad shares how his son uses AI educational software (such as Synthesis) that uses interactive metaphors—like smashing virtual glass with a hammer—to teach mathematical concepts like division in an engaging, adaptive manner.
* **Adaptability as a Core Skill**: Education must pivot from specialized knowledge retention to fostering rapid problem-solving, foundational resilience, and cognitive flexibility.
[01:23:40](https://www.youtube.com/watch?v=JMYQmGfTltY&t=5020s) Concrete Learning vs. Abstract Education
Bret Weinstein shares his philosophy on learning, arguing that traditional educational systems fail because they rely heavily on abstraction. In typical classrooms, success or failure is dictated arbitrarily by the teacher at the front of the room—a dynamic that suffers immensely when the instructor is lackluster. In contrast, direct interaction with the physical world provides unambiguous feedback. If a person works on a mechanical engine that refuses to start, they cannot negotiate or argue it into running; they must diagnose the underlying fault, make a precise physical adjustment, and receive the immediate, tangible reward of hearing the engine fire up.
Bret advocates for a light-handed, concrete approach to teaching both students and his own children. By saying as little as possible and allowing real-world physical systems to deliver feedback, learners develop an authentic understanding that can be extrapolated to other systems without inheriting another human's misinterpretations. While incorporating AI into this physical learning loop requires careful design, Bret and the panel highlight robotics as the ideal intersection where physical computing meets artificial intelligence—using AI to nudge a student out of a problem-solving rut and toward higher-level challenges.
[01:25:57](https://www.youtube.com/watch?v=JMYQmGfTltY&t=5157s) What Should We Teach Our Children in the AI Era?
Prompted by host Steven Bartlett, the panelists—who are all fathers—discuss how they prepare their own children for an AI-dominated future. Amjad Masad explains that despite his children being very young, he regularly sits down with them in front of Replit and ChatGPT to construct games and brainstorm wild, imaginative scenarios (such as cats on a cheese moon with mice inside). He emphasizes that fostering a high generative pace—the ability to rapidly create, pivot, and iterate on ideas—is essential.
```
Panelists' Children's Ages:
• Amjad Masad: 3 and 5 years old
• Bret Weinstein: 19 and 21 years old
• Daniel Priestley: 6, 7, and 10 years old
```
Bret outlines his fundamental rule for his kids, warning them away from online pornography at all costs, remarking that he considers a porn addiction more dangerous to human well-being than a drug problem. Turning to technical preparation, Bret explains that the coming era is defined by **complex systems** that cannot be engineered through fixed blueprints. Instead, children must learn to navigate complexity with deep humility, recognizing that unintended consequences are inevitable. He recommends two core approaches:
1. **Prototyping:** Building quick proofs-of-concept to discover unknown variables through iterative discovery rather than assuming a perfect upfront solution.
2. **Navigational Adaptability:** Approaching systems like a surfer riding a wave—constantly absorbing real-time feedback and adjusting rather than following a rigid predetermined plan.
3. **Generalist Mindset:** Prioritizing versatile, cross-disciplinary tools over narrow specialization.
Daniel Priestley aligns with this vision, stating that his primary goal is raising **high-agency generalists**—self-starters equipped with broad toolkits who can thrive in high-velocity, multi-faceted lives. He engages his children in activities like chess, Brazilian jiu-jitsu, dance, acting, nature exploration, and real-world micro-entrepreneurship like running street lemonade stands. Crucially, Daniel teaches them the vital distinction between **creation vs. consumption**. Because AI can easily trap people into passive hyper-consumption—whether through doom-scrolling, news, or addictive media designed to hijack dopamine—parents must actively train children to stay on the generative, creative side of technology.
Steven reflects on his own entrepreneurial journey, recalling that starting his first business at age 12 (followed by several others through his teens) forced him to act as marketer, sales rep, finance team, customer service, and recruiter all at once. Today, Shopify serves as an AI-powered growth team for modern founders, automating tasks like writing product descriptions, web design, image enhancement, shipping, taxes, and inventory management.
[01:32:09](https://www.youtube.com/watch?v=JMYQmGfTltY&t=5529s) Is the AI Takeover Inevitable?
Steven asks the panel whether the rapid rise and disruption of AI are truly inevitable. Bret somberly agrees that it is now unavoidable, expressing deep regret that humanity squandered decades of relative peace and productivity during which it could have prepared. Driven by narrow competitive rivalries, society built a fragile global structure incapable of smoothly absorbing the coming shock. He notes that sci-fi classics like the movie *2001: A Space Odyssey* anticipated these existential challenges, yet society treated the warnings as mere entertainment.
```
Key Factors Driving Unavoidable AI Acceleration:
• Game-theoretic traps: Voluntary unilateral restraint simply cedes AI dominance to geopolitical or commercial competitors.
• Lack of moral and structural preparation: Failure to plan for massive, rapid workforce displacement prior to public deployment.
```
Amjad offers a prospective framework for how governments and leaders should manage job displacement. Rather than panicking, leaders must educate the public on how large language models and algorithms function, enabling workers to repurpose their domain knowledge into tech-enabled ventures. He points to Adil Khan, a teacher who began using GPT-3 to automate educational tasks, learned basic coding, used Replit to launch an AI education company (MagicSchool AI), and scaled it to a valuation of hundreds of millions of dollars within two years.
Daniel notes that human beings are historically terrible at predicting future employment. To illustrate this, he imagines trying to explain the job of a modern personal fitness trainer to his grandfather—a concept that would sound like an absurd scam: paying a stranger to drive to a building, watch you lift heavy metal objects, count out loud, and give them money. Because jobs morph rapidly across generations, Daniel argues that education must shift away from static skill acquisition and move toward **lifelong learning centered on tools, judgment, and personal virtue**.
[01:38:07](https://www.youtube.com/watch?v=JMYQmGfTltY&t=5887s) The "House Cat" Analogy of Human Existence
Daniel introduces an analogy comparing humanity's future alongside superintelligent AI to the life of a domesticated house cat. His cat lives an extraordinarily happy, meaningful, and comfortable existence surrounded by what is effectively a "superintelligence" (the human owner). The cat has no understanding of Google Calendars, online grocery deliveries, or monetary economies, yet its food magically appears, its environment is maintained, and it lives a fulfilled life within the house.
Steven challenges the optimism of the metaphor, pointing out the inherent power asymmetry: if an owner has a bad day or needs to move into a landlord's apartment that bans pets, the cat's entire world can be upended or destroyed without the cat having any say or ability to retaliate. Daniel acknowledges the risk of collateral disruption, but maintains that given the choice, no house cat would trade places with its wild ancestors who had to fight for survival every day. In the same way, humans in an AI-dominated world may enjoy exceptionally rich, interesting, and comfortable lives despite operating under the shadow of a far greater intelligence.
<ElicitationsGroup message="Where would you like to explore next in this discussion?">
<Elicitation label="Explore the high-agency generalist framework for education" query="Detail Daniel Priestley and Bret Weinstein's strategies for cultivating high-agency generalists in the AI age."/>
<Elicitation label="Examine the economic impacts of AI job displacement" query="Analyze the examples and arguments presented regarding AI job displacement and business creation."/>
</ElicitationsGroup>
[01:39:46](https://www.youtube.com/watch?v=JMYQmGfTltY&t=5986s) The House Cat Existence and Hyper-Novelty
Bret Weinstein challenges the idea that being cared for like a house cat by a higher intelligence is entirely ideal, joking that cats aren't nearly as impressed by human capacity as owners like to think. While humans are already accustomed to using technology they don't understand—like turning on a light switch or running a tap without knowing the exact underlying infrastructure—the troubling aspect of the current paradigm is the relentless pace of change. Historically, populations engaged in intergenerational work where children did effectively what their ancestors did. In contrast, modern society has entered a state of "hyper-novelty" where what people can get paid for changes radically every two to three years.
Weinstein raises the biological concern that humans are being removed from their natural developmental environment. Humans are built to learn skills during youth and apply them productively throughout adulthood. When the rate of technological acceleration renders an ancestor's or parent's experience completely irrelevant, it creates a dangerous, unpredictable environment. He ponders whether humanity should deliberately choose to slow down technological change so that developmental and adult environments can match, warning that failing to do so is a ticking biological clock.
- Historical baseline: Human societies historically maintained stable intergenerational careers over centuries.
- Hyper-novelty timeline: Modern job requirements and skill sets now shift radically roughly every 2 to 3 years.
[01:42:40](https://www.youtube.com/watch?v=JMYQmGfTltY&t=6160s) The Amish Model and Developmental Disconnect
When asked how society could slow down the rate of technological change, Bret references the Amish as an example of a community that voluntarily steps off the technological escalator to preserve their social fabric. While acknowledging that picking a specific historical moment (like 1850) to freeze technology is arbitrary, he notes that Amish communities have fared remarkably well against modern societal disruptions, maintaining low autism rates and strong community cohesion during events like the COVID-19 pandemic. The central insight is recognizing that unchecked technological progress can make populations sicker if not actively managed.
```
Historical Stability vs. Hyper-Novelty
┌─────────────────────────────────────────┐
│ Traditional Intergenerational Careers │ Stable over generations
└────────────────────┬────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ Modern Hyper-Novelty Cycle │ Skill shifts every 2-3 years
└────────────────────┬────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ Biological & Cognitive Disconnect │ Adult world changes faster
└─────────────────────────────────────────┘ than childhood preparation
```
Steven Bartlett points out the biological hurdle of age-related cognitive decline, noting that older brains face reduced memory flexibility and processing speed compared to children. He shares that his seven-year-old niece already knows five languages, illustrating the vast neuroplasticity of young minds. Under evolutionary design, humans learn key lessons when young and deploy them for a lifetime; forcing older adults to constantly relearn fundamental career skills goes entirely against human evolutionary programming.
- Amish lifestyle model: Adheres to a mid-19th-century technological freeze (e.g., no automobiles, limited grid electricity).
- Biological plasticity: Children show extreme neuroplasticity (e.g., learning five languages by age seven), whereas older brains naturally experience reduced processing speed and lower memory flexibility.
[01:46:36](https://www.youtube.com/watch?v=JMYQmGfTltY&t=6396s) Intentional Communities vs. Unregulated Tech Export
Amjad Masad provides a counterpoint, highlighting how recent shifts have led tech-savvy workers to design lives closer to ancestral environments. Following the pandemic, many online workers built rural collective communities where they farm, live locally, and work "email jobs" for five hours a day while raising children. Furthermore, Silicon Valley entrepreneurs are actively funding new walkable city startups designed to foster spontaneous social interaction and health, perimetering cities with autonomous vehicles rather than letting cars dominate human spaces.
However, Weinstein expresses concern over the hypocrisy seen in places like Palo Alto. While tech elites strictly regulate their own children's screen time and phone usage to limit cognitive harm, they simultaneously export unregulated, addictive technologies to the rest of the world. "The market is not going to solve this problem without our being deliberate about forcing it to," Bret warns, emphasizing that unless society consciously intervenes, billions of people could end up economically and socially displaced under the guise that "progress is progress."
[01:49:30](https://www.youtube.com/watch?v=JMYQmGfTltY&t=6570s) The Five Key Threats of the AI Era
When asked to outline his primary fears regarding AI, Bret Weinstein details five specific threats, ranging from existential risks to immediate societal collapse:
1. **Malevolent AI:** The existential threat of an overtly hostile artificial intelligence (the scenario doomers focus on, though Weinstein worries about this one the least).
2. **Misaligned Autonomous Agents:** The "sorcerer's apprentice" dilemma where an AI takes a poorly worded instruction to a destructive extreme (e.g., turning all available matter in the universe into paper clips to fulfill an optimization mandate).
3. **Derangement of Human Intellect & Undetectable Deepfakes:** The complete erosion of objective truth. Facts become so filtered and algorithmic persuasion so pervasive that people lose confidence in basic reality, a problem vastly accelerated by deepfakes.
4. **Economic Displacement and Bad Actors:** Widespread unemployment paired with the hyper-empowerment of bad actors who violate the social contract. Fraudsters utilize AI to shrink the collective wealth of society just to enlarge their own slice.
5. **Demographic and Geopolitical Escalation:** The rapid acceleration of underlying demographic crises and military tensions, potentially escalating pre-existing technologies into nuclear conflict.
[01:51:40](https://www.youtube.com/watch?v=JMYQmGfTltY&t=6700s) Deepfake Scams and the Crisis of Truth
Steven Bartlett and Daniel Priestley illustrate the immediate reality of threat number three by sharing their own experiences with deepfake fraud. Steven reveals that his team spends nearly 30% of their time playing "whack-a-mole" with persistent deepfake ads on X (formerly Twitter) and Meta platforms. Fraudsters create hyper-realistic videos of Steven impersonating his voice and likeness to lure audience members into Telegram crypto scams. Despite sending takedown requests daily, dozens of new synthetic ads instantly replace them.
> "My least favorite ones are when the single mother messages me saying that she just paid £500 of her money and how devastated she is, and I feel this moral obligation to give her her money back... That was me. It was my voice. It was a video of me telling her something." — Steven Bartlett
Daniel shares similar stories of followers asking if he personally messaged them for crypto wallet details. Weinstein points out that these technologies allow unscrupulous individuals to steal the hard-earned credibility of public figures while robbing vulnerable people. When asked whether an average college-educated person today is more propagandized than someone in Soviet Russia, Bret notes that Soviet citizens at least maintained a healthy cynicism, knowing official state media was false. Today, society is trapped between believing fabricated media or falling into absolute cynicism where no truth can be discerned.
- Dedicated workload: Steven's team spends roughly 30% of their bandwidth reporting deepfake financial scams to social platforms.
- Targeted financial harm: Typical scams involve fake videos luring victims into Telegram groups, leading to direct losses (e.g., a single mother losing £500).
[01:55:55](https://www.youtube.com/watch?v=JMYQmGfTltY&t=6955s) Truth, Search, and Corporate Incentives
The discussion shifts to whether search engines like Google and AI platforms like ChatGPT ultimately help or hinder the pursuit of truth. Bret Weinstein notes that AI engines have temporarily improved our ability to uncover reality because they allow users to see past algorithmic manipulation. Since AI models are not yet fully "policed," they can recognize patterns that authorities or mainstream narratives deny. However, Weinstein draws a cautionary parallel to the early days of internet search—a time when search engines simply indexed keyword locations across web pages, before platforms began using algorithmic steering to direct human attention toward specific viewpoints. He warns that AI's ability to help us see clearly depends on protections that currently do not exist.
```
[ Early Search ] [ Algorithmic Era ] [ Current AI Era ]
Keyword indexing without Search engines steer users Unfiltered pattern recognition
steering or censorship toward curated narratives bypasses web-search gaslighting
```
When Steven Bartlett asks if platforms could build AI tools to defend users from scam ads and fraud, Amjad Masad points out that companies like Meta are financially incentivized to combat platform abuse. Just as personal computers required antivirus software, users will likely deploy personal AI agents on their local devices to filter out deepfakes and fraudulent content. Weinstein counters by questioning whether giant tech corporations can be trusted to act in their users' best interest. He recalls how major social networks previously censored massive accounts and policed political thoughts—action that reduced their active user bases—demonstrating that corporations frequently act against obvious market incentives due to internal shareholder or political pressures.
* **Algorithmic Manipulation:** Search engines evolved from unbiased content indexers to systems that actively steer public opinion.
* **AI Anti-Scam Defense:** Antivirus software provided a blueprint for how personal AI defenders will run locally to combat deepfakes and scams.
* **Corporate Self-Harm:** Tech platforms have repeatedly prioritized narrative control over user growth and short-term profits.
---
[01:59:32](https://www.youtube.com/watch?v=JMYQmGfTltY&t=7172s) An Optimistic Vision for AI's Potential
Shifting the perspective toward optimism, Steven prompts Bret Weinstein and Daniel Priestley to share their positive visions for an AI-powered future. Weinstein highlights two major opportunities: education and governance. Facing a global shortage of competent teachers that would take generations to fix through traditional means, AI can augment elite educators so they reach vastly larger audiences. Furthermore, AI could replace rigid, easily gamed statutory laws with dynamic, goal-oriented regulatory frameworks. Instead of relying on human politicians and outdated constitutions, societies could set high-level societal objectives and let AI systems prototype, monitor, and adjust policy interventions dynamically based on real-time outcomes.
```
+-----------------------------------------------------------------------------------+
| OPTIMISTIC POSSIBILITIES FOR AI |
+-------------------+---------------------------------------------------------------+
| Global Healthcare | - Instant, AI-driven MRI analysis removing medical backlogs. |
| | - General practitioner level diagnostics for underserved areas. |
+-------------------+---------------------------------------------------------------+
| Dynamic Governance| - Moving past static laws that are easily gamed by bad actors.|
| | - AI-monitored policy tweaks tied to clear, adaptive goals. |
+-------------------+---------------------------------------------------------------+
| Liberated Labor | - Ending repetitive, multi-decade corporate ladder grinding. |
| | - Multi-project "mini-kingdoms" powered by personal AI tools. |
+-------------------+---------------------------------------------------------------+
```
Daniel Priestley emphasizes the transformational breakthroughs awaiting global healthcare and human work. In many parts of the world, patients suffer in pain while waiting weeks for a specialist to review an MRI, or live in regions completely devoid of medical doctors. AI-driven diagnostics will bridge this gap instantaneously worldwide. On an individual level, Priestley believes AI will liberate humanity from the repetitive, soul-crushing routines inherited from the Industrial Age. Instead of spending decades climbing a corporate ladder step-by-step, individuals can run "mini-kingdoms"—simultaneously authoring books, launching businesses, organizing festivals, and managing teams using AI tools to multiply their output.
* **Teacher Augmentation:** AI allows small groups of exceptional educators to scale their reach globally, compensating for systemic teacher shortages.
* **Adaptive Policy Making:** Replaces static blueprints of law with dynamic, outcome-based systems that continuously self-correct.
* **Democratized Diagnostics:** Delivers instant expert-level healthcare and diagnostic reads to areas lacking qualified general practitioners.
---
[02:03:26](https://www.youtube.com/watch?v=JMYQmGfTltY&t=7406s) Business Opportunities and Natural Language Software
Looking at the business landscape, Daniel Priestley explains how small teams now hold unprecedented leverage. A tight-knit group of five to ten passionate individuals armed with modern AI tech can achieve what previously required decades of corporate effort or hundreds of employees. When Steven asks what serves as a company's unique selling proposition (USP) when everyone has access to $20-a-month AI tools, Priestley responds that speed and early adoption are key. He describes the current moment as the least competitive window in history for those who know how to use these tools, drawing parallels to the early days of the web or the mobile app boom, where early movers built massive businesses before market efficiencies took over.
Replit Natural Language Creation
├── Cost: ~$1.00 (Fraction of $25/mo sub)
└── Timeline: 15 Minutes
```
To illustrate this rapid leverage, Priestley shares an anecdote from the previous evening: using Replit's natural language agent, he built a custom household application to manage his children's chores and track pocket money payouts. Within 15 minutes of simply conversing with the prompt bar about his requirements—assigning financial values to tasks, toggling between children, and generating weekly automated payout summaries—the app was fully functional. Amjad Masad highlights that since Replit launched its natural language builder, users have created over 3 million applications without writing a single line of code, with 300,000 to 400,000 deployed for actual operational use.
* **Replit Agent Scale:** Over 3,000,000 natural-language applications built, with hundreds of thousands actively deployed as operational tools.
* **Economic Shift:** Software that previously took an IT agency three to four weeks and $5,000–$10,000 to build can now be generated in 15 minutes for around $1.00 of compute time.
* **Micro-Team Leverage:** 5-to-10-person teams can build, test, and scale business solutions faster than multi-decade enterprise efforts.
---
[02:10:21](https://www.youtube.com/watch?v=JMYQmGfTltY&t=7821s) The Existential Threat of Autonomous Weapons
The dialogue turns to one of Amjad Masad's deepest fears regarding AI technology: the rapid, unchecked proliferation of autonomous weapons. Masad notes that much like corporations and state governments, AI can function as an emergent system without a single human controlling its every action, frequently steering collective behavior toward morally abhorrent outcomes. In the realm of international geopolitics, the ongoing technological race between global superpowers like the United States and China creates hyper-competitive dynamics, incentivizing both nations to cut safety corners and push lethal autonomous technologies onto the battlefield.
Masad describes the horrifying real-world application of facial-recognition software paired with autonomous drone swarms. These systems can be deployed as self-navigating, untraceable assassination machines capable of hunting specific individuals without requiring direct human intervention or oversight.
* **Systemic Misalignment:** Large organizations, governments, and AI systems can collectively execute harmful actions that no individual human within the structure intended.
* **Geopolitical Arms Race:** Rivalries force nations to bypass ethical safeguards to avoid falling behind in military AI capabilities.
* **Autonomous Drones:** Facial-recognition-enabled drone swarms represent a shift toward low-cost, automated targeted assassinations.
[02:11:48](https://www.youtube.com/watch?v=JMYQmGfTltY&t=7908s) State Surveillance, Automated Law Enforcement, and Crime
The discussion shifts to how state power can leverage modern tech against its own citizens. Amjad Masad highlights that while AI weapons can operate on a country-versus-country level, technology also serves as a potent mechanism for governments to subjugate their people. He notes that the Western experience during COVID-19 proved how rapidly democratic systems can adopt draconian measures. The group addresses real-world examples:
* **Hijab Surveillance in Iran:** Iran has deployed comprehensive facial recognition systems and public cameras to automatically identify women driving or walking without a hijab. The system can flag a vehicle, effectively serving as police, judge, and jury in real time.
* **Crowdsourced Reporting (The Nazar App):** Iran implemented the Nazar mobile app, which lets citizens report fellow individuals for hijab non-compliance by logging their location, time, and vehicle license plate.
* **London and Wales Surveillance:** Facial recognition systems are actively being rolled out across London and Wales at rapid speed, establishing a infrastructure layer that future regimes could exploit.
* **Financial Freezing:** The panel references the Canadian trucker protests, where financial accounts were frozen simply based on a person’s physical presence in a location.
Bret Weinstein warns that people lack sufficient imagination regarding how these tools will be misused outside of war or government tyranny. Autonomous, untraceable drones equipped with facial recognition could dramatically drop the "price of an unprosecutable crime." He envisions a new form of high-tech organized crime where cartels or private actors use precise, un-traceable target eliminations to enforce economic monopolies and protect protection rackets without needing to kill many people to send a message.
Steven Bartlett brings up a past conversation with Mustafa Suleyman (CEO of Microsoft AI), who expressed that his core fear is a small group of malicious actors gaining access to tools that can instantly destabilize the world. Steven recalls asking Suleyman three times if containing AI is possible, to which Suleyman repeatedly responded, "We must." Amjad critiques this posture, arguing that forcing containment inevitably leads to an oppressive framework. Citing philosopher Nick Bostrom, Amjad notes that the total surveillance state required to monitor every person's compute and thoughts to prevent AGI development might ultimately be more tyrannical than any AI itself.
- Both the US and China are heavily investing in autonomous drones and AI cyber weapons out of fear that the other side will achieve dominance first.
---
[02:17:48](https://www.youtube.com/watch?v=JMYQmGfTltY&t=8268s) VR Isolation, the Matrix, and the Fermi Paradox
Amjad raises another deep concern beyond weaponized AI: total human isolation through hyper-realistic virtual environments. He warns that AI-driven VR could create tailored universes that simulate work, struggle, and reward so convincingly that individuals will have no desire to engage with the real world. Unlike the shared multiplayer universe of *Ready Player One*, this outcome would be entirely solipsistic, trapping individuals in an AI-generated *Matrix*.
This scenario leads Amjad to reflect on the **Fermi Paradox**—the question originally posed by physicist Enrico Fermi: *"If the universe is so vast, where are all the aliens?"* Amjad outlines potential solutions to the paradox:
* **The Zoo / House Cat Hypothesis:** Advanced alien civilizations exist but deliberately isolate humanity (like the Amish or house pets) to observe us while keeping us from harming ourselves.
* **The Infinite Pleasure Trap:** Civilizations naturally progress until they engineer technology capable of delivering infinite, simulated pleasure. Once a species creates an infallible pleasure machine, it loses all drive to explore outer space or seek out other life.
* **Self-Destruction:** Bret adds the simpler explanation: any technological ladder that enables a species to reach outer space simultaneously hands them the tools to annihilate themselves, whether through nuclear arms or AI.
---
[02:21:20](https://www.youtube.com/watch?v=JMYQmGfTltY&t=8480s) Closing Advice: How to Live a Good Life in the AI Era
Steven asks each guest for closing, actionable guidance for everyday listeners—such as single mothers or non-technical workers—on navigating the next few decades and building a good life.
**Amjad Masad** urges people not to assume these tools are out of reach simply because they aren't "tech bros." AI has made technology more accessible than ever. He points out that you don't need to take massive financial risks or quit your job to be entrepreneurial:
* He shares an example of a product manager at a real estate firm who built a tool providing a 10% conversion lift, generating millions for his employer and elevating his entire career.
* He emphasizes that "everyone is a CEO of their own life," and parents can even build small custom software to help their kids learn.
**Bret Weinstein** views this era as a monumental human phase transition, on par with the discovery of agriculture, writing, or electricity. Using evolutionary biology concepts, he describes our current uncertainty as crossing an "adaptive valley":
> "To get from the lower niche to the higher niche, you have to cross through what we call an adaptive valley... It's more like the peaks are islands in an archipelago that is in fog where you can't figure out what direction that peak is."
Bret stresses that humanity must consciously prevent market forces and new technology from turning this era into a new form of enslavement, hoping instead that automating mundane labor allows the pursuit of meaning to replace hard labor.
**Daniel Priestley** frames the moment through historical lineage. Every living person carries a vast line of ancestors (2 parents, 4 grandparents, 8 great-grandparents, doubling each generation) who survived wars, deadly diseases, and brutal labor:
* Every one of those ancestors would trade places in a heartbeat to live in today's world.
* With world knowledge, AI assistants, travel, and global connection accessible at our fingertips, Daniel urges individuals to accept the obligation to their ancestors: be brave, solve meaningful problems, have fun, and make their "little dent in the universe."
Steven concludes the discussion on a note of human optimism. He emphasizes that unlike past technological shifts (like social media) where society was caught flat-footed by downstream consequences, conversations like this debate allow us to anticipate and proactively guide humanity's future.
Here is a detailed, faithful walkthrough of the provided transcript section:
[02:28:20](https://www.youtube.com/watch?v=JMYQmGfTltY&t=8900s) The Speed of AI Adoption and the Transition Between Islands
Steven Bartlett brings the discussion together by acknowledging the entrepreneurial excitement surrounding AI while emphasizing the serious, destabilizing transitional period highlighted by Bret Weinstein, Amjad Masad, and Daniel Priestley. Reflecting on how rapidly AI technology spreads, Steven shares a personal anecdote about his first time trying Replit. He immediately walked out and told his team member, "Cussie, try this," leaving her instantly hooked. Shortly after, he called his girlfriend—a breathwork practitioner in Bali—and instructed her to type "R-E-P-L-I-T" into her browser. Within minutes, she was leveraging it to organize breathwork schedules and manage client information for an upcoming retreat. Because modern users are internet-native, this technology builds on top of existing intuitive behaviors, allowing it to spread like wildfire.
Steven notes that this rapid adoption makes the transition "from one island to another" or from one peak to another uniquely chaotic. Drawing on his interviews with prominent AI leaders—including Reid Hoffman (founder of LinkedIn), the CEO of Google, Mustafa Suleyman, and Sam Altman—Steven points out that while they frequently squabble over short-term timelines or mid-term specifics, they all fundamentally agree on one thing: the long-term future and basic functioning of human society will look radically different.
- **Replit Adoption Example:** Steven's girlfriend, a breathwork instructor in Bali, used Replit to organize client information and build schedules for a retreat without prior coding background.
- **Consensus Among Tech Leaders:** Key figures like Reid Hoffman, Sam Altman, Mustafa Suleyman, and Google's CEO disagree on short-term details but share a unanimous consensus that long-term societal structures will be completely transformed.
[02:29:38](https://www.youtube.com/watch?v=JMYQmGfTltY&t=8978s) Gratitude to the Panelists and Concluding Thoughts
Steven extends personalized thanks to each of the three guests for their distinct perspectives during the debate. He commends Amjad Masad for Replit's mission of democratizing software creation, pointing out how it levels the playing field for aspiring entrepreneurs everywhere—from Bangladesh and Cape Town to San Francisco. Steven shares that having been born in Botswana, Africa, he deeply appreciates tools that break down the traditional barriers of creation, removing the need to hire outside developers just to bring an idea to life.
Turning to Daniel Priestley, Steven highlights his role as a vital voice and thought-leader for British entrepreneurs, providing essential real-world business balance to the conversation. Finally, Steven thanks Bret Weinstein for serving as a crucial "human lens" on complex technological challenges. He praises Bret's fearless willingness to speak out as a counterweight to unchecked optimism, warning that without critical voices, people risk "hurtling towards the mousetrap because we love cheese." Steven closes the panel by observing that these recorded discussions "live forever" as historical artifacts of this pivotal moment in time.
- **Democratization of Tech:** Amjad Masad's Replit enables individuals in regions like Bangladesh, Cape Town, and Africa to build software without traditional coding barriers.
- **Panel Representation:** Daniel Priestley represents the pragmatic entrepreneurial voice, while Bret Weinstein acts as a cautious, evolutionary counterweight to runaway tech optimism.
[02:31:17](https://www.youtube.com/watch?v=JMYQmGfTltY&t=9077s) A Final Request to the Audience
In his traditional show outro, Steven addresses the audience with a striking statistic: 53% of regular listeners have not yet subscribed to the podcast channel. He asks the audience for a simple, free favor to support the show by hitting the subscribe button. In return, Steven promises that he and his team will dedicate themselves to continuously improving the podcast each week, incorporating viewer feedback, and booking the exact guests the audience wants to hear.
## Notable quotes
> [04:15](https://www.youtube.com/watch?v=JMYQmGfTltY&t=255s) It starts sounding religious... human myths used to imagine a world where you can speak your ideas into existence. — Amjad Masad
> [04:15](https://www.youtube.com/watch?v=JMYQmGfTltY&t=255s) Every 7 months, the number of minutes that an agent can run autonomously for is doubling. — Amjad Masad
> [09:00](https://www.youtube.com/watch?v=JMYQmGfTltY&t=540s) We've created a new species and nobody on earth can predict what's going to happen. — Bret Weinstein
> [09:00](https://www.youtube.com/watch?v=JMYQmGfTltY&t=540s) We've just crossed a threshold that is similar in its capacity to alter the world as the invention of writing... but writing altered the world slowly over thousands of years. — Bret Weinstein
> [12:48](https://www.youtube.com/watch?v=JMYQmGfTltY&t=768s) If you had interviewed the horses in 1900... they would have said, 'We've been part of humanity hand and hoof for thousands of years... how bad could it be?' — Daniel Priestley
> [20:22](https://www.youtube.com/watch?v=JMYQmGfTltY&t=1222s) In this marathon race, the vast majority of people have got their shoes tied together because AI is distracting them. Some people are running a traditional race, some people have got a bicycle, and some people have got a Formula 1 vehicle. — Daniel Priestley
> [27:18](https://www.youtube.com/watch?v=JMYQmGfTltY&t=1638s) I wouldn't prepare for AGI in the same way that I wouldn't prepare for the end of days. — Amjad Masad
> [42:05](https://www.youtube.com/watch?v=JMYQmGfTltY&t=2525s) if your job is as routine as it comes, your job is gone in the next couple years. — Amjad Masad
> [55:58](https://www.youtube.com/watch?v=JMYQmGfTltY&t=3358s) That's not arbitrage. That's theft... You have somebody who has an idea that can be brought to market and somebody else is cryptically detecting it and then selling back their own idea to them. — Bret Weinstein
> [58:10](https://www.youtube.com/watch?v=JMYQmGfTltY&t=3490s) Authenticity is going to become the coin of the realm and anything that can be faked or cheated is going to be devalued. — Bret Weinstein
> [01:03:34](https://www.youtube.com/watch?v=JMYQmGfTltY&t=3814s) It's really important to know that a lot of the jobs that are in at risk are the most high status, highly paid jobs in the world. — Steven Bartlett
> [01:19:00](https://www.youtube.com/watch?v=JMYQmGfTltY&t=4740s) The capacity to think on your feet and pivot as things change is the only game in town with respect to our ability to prepare you in advance. — Bret Weinstein
> [01:51:40](https://www.youtube.com/watch?v=JMYQmGfTltY&t=6700s) My least favorite ones are when the single mother messages me saying that she just paid £500 of her money and how devastated she is, and I feel this moral obligation to give her her money back... That was me. It was my voice. It was a video of me telling her something. — Steven Bartlett
> [02:22:15](https://www.youtube.com/watch?v=JMYQmGfTltY&t=8535s) To get from the lower niche to the higher niche, you have to cross through what we call an adaptive valley... It's more like the peaks are islands in an archipelago that is in fog where you can't figure out what direction that peak is. — Bret Weinstein
> [02:29:58](https://www.youtube.com/watch?v=JMYQmGfTltY&t=8998s) hurtling towards the mousetrap because we love cheese. — Steven Bartlett
## Mentioned
Steven Bartlett, Amjad Masad, Bret Weinstein, Daniel Priestley, Replit, OpenAI, ChatGPT, CVS, Stripe, Google, Meta, Llama, DeepSeek, Elon Musk, Sam Altman, AbCellera, The Diary Of A CEO, Harvard Business Review, Billy Howell, Upwork, thirdweb, X, Facebook, Whole Foods, Worldcoin, Klarna, A Hunter-Gatherer's Guide to the 21st Century, Synthesis, Shopify, 2001: A Space Odyssey, Adil Khan, MagicSchool AI, Google Calendars, Telegram, Mustafa Suleyman, Microsoft AI, Nick Bostrom, Enrico Fermi, Reid Hoffman, Nazar app, Vivobarefoot
Masad、Weinstein 与 Priestley 辩论失业与 AI 智能体风险
自主 AI 智能体将在两年内淘汰常规知识型工作，迫使劳动者转向高主体性角色或面临经济流离失所。
专家小组探讨了自主 AI 智能体带来的前所未有的社会与经济颠覆，预计在两到三年内，高地位知识型职业与常规劳动岗位都将面临显著的工作流失。与会者辩论了 AI 究竟是精英操作者的绝对力量倍增器，还是可能侵蚀人类主体性、真实性与存在意义的系统性威胁。核心议题包括深度伪造的兴起、自主武器、终身通识教育的必要性，以及在瞬息万变的世界中进行适应性转型的策略。最终，讨论在对潜在系统性崩溃的警告与对普及化 AI 工具赋能个人建立“微型王国”并解决现实问题的乐观展望之间保持了平衡。
AI 智能体正从简单的工具使用向广泛的自主性转变，其可自主执行任务的时间跨度大约每 7 个月翻一番。 (4:15)
AI 的到来代表了堪比文字发明的范式转变，但其变革正以前所未有的、以周为单位的速度推进。 (9:00)
传统工作者与受 AI 赋能的操作者之间的差距将急剧扩大，从 10 倍优势拉大到 1000 倍优势。 (20:22)
常规的文本处理和行政类工作在两三年内面临被迅速淘汰，自动化系统已经开始接管复杂的支持与法律职能。 (42:05)
AI 实现自动化的是认知智力而非体力肌肉，这直接对高薪、高地位的知识型职业造成了经济冲击。 (45:38)
Reading summary
In a debate hosted by Steven Bartlett, tech founder Amjad Masad, evolutionary biologist Bret Weinstein, and entrepreneur Daniel Priestley examine whether autonomous AI agents will democratize productivity or precipitate catastrophic economic and societal destabilization. While Masad argues that AI serves as an unprecedented equalizer for ambitious individuals, Weinstein and Priestley warn that humanity is unprepared for the speed at which cognitive labor, social cohesion, and human agency are being undermined.
The Collapse of Knowledge Work and Displacement (19:22)
Daniel Priestley equates modern AI models to discovering a continent of educated workers operating 24/7 for 25 cents an hour, outperforming median human intelligence. Drawing a parallel to 1900 New York City, where automobiles replaced working horses within 13 years, Priestley warns that human workers face sudden economic obsolescence. Unlike past industrial disruptions that automated physical muscle, AI targets intellect, placing highly compensated, high-status professions—such as corporate lawyers and anesthesiologists—at immediate risk. Priestley forecasts that workers will transition from 45-year careers to rapid 10-to-36-month project cycles as AI software continually reshapes team structures.
Steven Bartlett notes that initiatives like Sam Altman's Worldcoin suggest technological architects anticipate mass structural unemployment requiring universal basic income (UBI). Masad pushes back against UBI, arguing that humans possess an innate drive to feel useful. However, Weinstein counters that because AI sidelines workers across every sector simultaneously, displaced specialists will struggle to pivot, creating unprecedented societal trauma and stripping individuals of the goals necessary for psychological well-being.
Entrepreneurial Leverage and Widening Inequality (20:22)
Masad asserts that AI agents bridge the gap between human ideas and execution, allowing individuals to run multi-agent workflows, deploy software, and orchestrate business operations in minutes. He estimates that while traditional programming created a 10x performance gap between average and top performers, AI-enabled operators will achieve a 1,000x advantage over those who fail to adopt the technology. Priestley demonstrates this leverage by noting that a functional household chore application can now be built via Replit natural-language prompts in 15 minutes for roughly one dollar, compared to traditional development costs of $5,000 to $10,000.
However, the panel emphasizes that this asymmetric productivity threatens to widen economic inequality. Priestley uses a marathon analogy where AI acts as an algorithmic distraction tying the shoes of the general public while elite operators drive formula-one vehicles. Masad illustrates modern arbitrage through cases like developers scraping Upwork client specifications and generating completed software with AI for $10,000, an approach Weinstein condemns as the parasitic theft of client ideas. Weinstein cautions that entrenched corporate and venture capital elites will ultimately capture the bulk of value generated by AI rather than allowing it to remain widely distributed.
Complex Emergence, Disinformation, and State Control (9:00)
Weinstein stresses that autonomous AI represents an unpredictable, evolving complex species rather than a deterministic, complicated tool. Connecting machines to human consciousness via natural language generates cognitive feedback loops with capabilities creators cannot foresee or restrain. Furthermore, bad actors gain an asymmetric advantage; while legitimate users gain efficiency, malicious actors gain force-multipliers for hyper-personalized extortion. Bartlett and Priestley report spending substantial resources combating relentless deepfake campaigns impersonating their likenesses to defraud followers, illustrating how synthetic media rapidly dismantles public confidence in objective reality.
Beyond individual fraud, Masad and Weinstein warn of institutional abuse and state surveillance. Masad highlights the existential danger of autonomous weapons, where international superpower rivalries incentivize nations to deploy untraceable, facial-recognition-guided drone swarms without human intervention. Domestically, governments increasingly deploy automated surveillance, as demonstrated by Iran using automated cameras and reporting apps to enforce compliance, and the expansion of facial recognition across the United Kingdom. Weinstein argues that such infrastructure can be easily repurposed by future regimes to control populations.
Hyper-Novelty, Human Agency, and Educational Reform (79:00)
Priestley introduces the analogy of humanity living like pampered house cats under a superior intelligence, but Weinstein challenges this premise, warning of biological maladaptation driven by 'hyper-novelty.' Historically, humans applied skills learned in youth across stable intergenerational careers; today, marketable skills change every two to three years. The panel links this retreat into frictionless digital environments and artificial companionship to severe societal fallout, including widespread loneliness and collapsing global fertility rates, highlighted by South Korea's rate falling to 0.72.
To prepare the next generation, the panel argues that traditional linear schooling must be discarded. Masad points to educational data showing that 1-on-1 tutoring produces a two-sigma improvement in student outcomes, a personalized learning model that AI can scale globally. Weinstein advocates grounding education in concrete physical systems, such as mechanics and robotics, where unambiguous physical feedback prevents students from adopting abstract misinformation. Rather than teaching rigid curricula, both Masad and Weinstein urge teaching rapid iteration, humility before complex systems, and personal agency.
The debate concludes on an unresolved dilemma: while individual mastery of AI agents offers unprecedented personal leverage, systemic economic incentives and geopolitical rivalries make unconstrained technological acceleration inevitable. Weinstein warns that humanity must navigate a precarious 'adaptive valley' without any guarantee that competitive market mechanisms will prevent severe social displacement or preserve human relevance.
[4:15] Amjad Masad: It starts sounding religious... human myths used to imagine a world where you can speak your ideas into existence.
[4:15] Amjad Masad: Every 7 months, the number of minutes that an agent can run autonomously for is doubling.
[9:00] Bret Weinstein: We've created a new species and nobody on earth can predict what's going to happen.
[9:00] Bret Weinstein: We've just crossed a threshold that is similar in its capacity to alter the world as the invention of writing... but writing altered the world slowly over thousands of years.
[12:48] Daniel Priestley: If you had interviewed the horses in 1900... they would have said, 'We've been part of humanity hand and hoof for thousands of years... how bad could it be?'
[20:22] Daniel Priestley: In this marathon race, the vast majority of people have got their shoes tied together because AI is distracting them. Some people are running a traditional race, some people have got a bicycle, and some people have got a Formula 1 vehicle.
[27:18] Amjad Masad: I wouldn't prepare for AGI in the same way that I wouldn't prepare for the end of days.
[42:05] Amjad Masad: if your job is as routine as it comes, your job is gone in the next couple years.
[55:58] Bret Weinstein: That's not arbitrage. That's theft... You have somebody who has an idea that can be brought to market and somebody else is cryptically detecting it and then selling back their own idea to them.
[58:10] Bret Weinstein: Authenticity is going to become the coin of the realm and anything that can be faked or cheated is going to be devalued.
[63:34] Steven Bartlett: It's really important to know that a lot of the jobs that are in at risk are the most high status, highly paid jobs in the world.
[79:00] Bret Weinstein: The capacity to think on your feet and pivot as things change is the only game in town with respect to our ability to prepare you in advance.
[111:40] Steven Bartlett: My least favorite ones are when the single mother messages me saying that she just paid £500 of her money and how devastated she is, and I feel this moral obligation to give her her money back... That was me. It was my voice. It was a video of me telling her something.
[142:15] Bret Weinstein: To get from the lower niche to the higher niche, you have to cross through what we call an adaptive valley... It's more like the peaks are islands in an archipelago that is in fog where you can't figure out what direction that peak is.
[149:58] Steven Bartlett: hurtling towards the mousetrap because we love cheese.
Original transcript
[0:00] I think a lot of people don't realize
[0:01] how massive the positive impact AI is
[0:04] going to have on their life. Well, I
[0:06] would argue that the idea that this AI
[0:07] disruption doesn't lead us to human
[0:09] catastrophe is optimistic. For example,
[0:12] people are going to be unemployed in
[0:14] huge numbers. You agree with that, don't
[0:16] you? Yes. If your job is as routine as
[0:19] it comes, it's gone in the next couple
[0:21] years. But it's going to create new
[0:23] opportunities for wealth creation. Let
[0:26] me put it to you this way. We have
[0:27] created a new species and nobody on
[0:29] earth can predict what's going to
[0:31] happen. We are joined by three leading
[0:33] voices to debate the most disruptive
[0:35] shift in human history, the rise of AI.
[0:38] And they're answering the questions
[0:39] you're most scared about. This
[0:41] technology is going to get so much more
[0:43] powerful. And yes, we're going to go
[0:44] through a period of disruption. But at
[0:46] the other end, we're going to create a
[0:47] fair world. It's enabling people to run
[0:49] their businesses, make a lot of money,
[0:51] and you can solve meaningful problems
[0:53] such as the breakthroughs in global
[0:54] healthcare and education will be
[0:56] phenomenal. and you can live an
[0:57] incredibly fulfilling existence. Well, I
[0:59] would just say on that front, this has
[1:01] always been the fantasy of technologists
[1:02] to do marvelous things with our spare
[1:04] time, but we end up doom scrolling,
[1:06] loneliness epidemic, right? Falling
[1:07] birth rates. So, the potential for good
[1:09] here is infinite and the potential for
[1:11] bad is 10 times. For example, there's
[1:14] war, undetectable beat fakes and scams.
[1:17] So, people don't understand how many
[1:18] different ways they are going to be
[1:20] robbed. Look, I don't think blaming
[1:22] technology for all of it is the right
[1:23] thing. all these issues, they're already
[1:25] here. You're all fathers here. So, what
[1:27] are you saying to your children? Well,
[1:29] first of all, this has always blown my
[1:32] mind a little bit. 53% of you that
[1:34] listen to this show regularly haven't
[1:36] yet subscribed to the show. So, could I
[1:38] ask you for a favor before we start? If
[1:40] you like the show and you like what we
[1:41] do here and you want to support us, the
[1:42] free simple way that you can do just
[1:44] that is by hitting the subscribe button.
[1:46] And my commitment to you is if you do
[1:47] that, then I'll do everything in my
[1:49] power, me and my team, to make sure that
[1:51] this show is better for you every single
[1:52] week. We'll listen to your feedback.
[1:54] We'll find the guest that you want me to
[1:56] speak to and we'll continue to do what
[1:58] we do. Thank you so
[1:59] [Music]
[2:05] much. The reason why I wanted to have
[2:08] this conversation with all of you is
[2:10] because the subject matter of AI, but
[2:12] more specifically AI agents, has
[2:15] occupied my free time for several weeks
[2:18] in a row. And actually amad when I
[2:20] started using replet for me it was a
[2:24] paradigm shift. There was two paradigm
[2:26] shifts in a row that happened about a
[2:28] week apart. Chat GPT released their
[2:30] image generation model where you could
[2:31] create any image. It was incredibly
[2:33] detailed with text and all those things.
[2:35] That was a huge paradigm shift. And then
[2:37] in the same week I finally gave in to
[2:39] try and figure out what this term AI
[2:41] agent was that I was hearing all over
[2:42] the internet. I heard vibe coding. I
[2:44] heard AI agent. I was like I will give
[2:45] it a shot. Mhm. And when I used Replet,
[2:48] 20 minutes in to using Replet, my mind
[2:51] was blown. And I think that night I
[2:53] stayed up till 3 or 4 a.m. in the
[2:55] morning. For anyone that doesn't know,
[2:58] Replet is a piece of software that
[3:00] allows you to create software. Mhm. And
[3:04] pretty much any software you you want.
[3:06] So someone like me with absolutely no
[3:07] coding skills was able to build a
[3:10] website, build in Stripe, take payment,
[3:13] integrate AI into my website, add Google
[3:16] login to the front of my website and do
[3:18] it within minutes. I then got the piece
[3:21] of software that I had built with no
[3:22] coding skills, sent it to my friends,
[3:23] and one of my friends put his credit
[3:25] card in and paid. Amazing. So I just
[3:27] launched a SAS company with no coding
[3:28] skills.
[3:30] to demonstrate an AI agent in a very
[3:33] simple way. I used an online AI agent
[3:36] called operator to order us all some
[3:38] water from a CVS around the corner. The
[3:40] AI agent did everything end to end and
[3:42] people will be watching on the screen.
[3:43] It put my credit card details and it
[3:45] picked the water for me. It gave the
[3:46] person a tip. It put some delivery notes
[3:48] in at some point a guy is going to walk
[3:50] in. He has not interacted with a human.
[3:53] He's interacted with my AI agent. And I
[3:55] just the reason I use this as an example
[3:57] is again it was a paradigm shift moment
[3:58] for me when I heard about agents. Mhm.
[4:00] about a month ago and I went on and I
[4:02] ordered a bottle of electrolytes and
[4:04] when it when my doorbell rang I freaked
[4:07] out. I freaked out. But am who are you
[4:11] and what are you doing? So uh I started
[4:15] programming at a very young age. You
[4:17] know I I built my first business when I
[4:19] was a teenager. I used to go to uh
[4:22] internet cafes and program there. And I
[4:24] realized that they don't have software
[4:26] to manage the business. I was like oh
[4:27] why don't you create accounts? I don't
[4:29] have a server. It took me two years to
[4:31] build that piece of software. And that's
[4:33] sort of embedded in my mind this idea
[4:35] that hey like you know there there's a
[4:38] lot of people in the world with really
[4:40] amazing ideas especially in the context
[4:42] where they live in that allows them to
[4:44] build uh businesses. However, the main
[4:49] source of uh friction between an idea
[4:52] and software or call it an idea and
[4:55] wealth creation is u infrastructure is
[4:59] physical infrastructure is a meaning a
[5:01] computer in front of you. It is um an
[5:04] internet connection. It is the set of
[5:07] tools and skills that you need to build
[5:09] that. If we make it so that anyone who
[5:12] has ideas who who wants to solve
[5:15] problems will be able to do it. I mean
[5:17] imagine the kind of world that we we
[5:19] could live in where no one can be anyone
[5:22] who has merit anyone who can think
[5:25] clearly anyone who can generate a lot of
[5:28] ideas can generate wealth. I mean that's
[5:31] an amazing world to live in right
[5:33] anywhere in the world. So with with
[5:34] Replet the company that I started in
[5:36] 2016 the the idea was like okay coding
[5:40] is difficult how do we solve coding and
[5:43] um we built every part of the process
[5:45] the hosting the code editor the only
[5:48] missing thing was the you know the AI
[5:50] agent and so over the past two years
[5:53] we've been working on this AI agent that
[5:57] you can just you know similar since chat
[5:58] GPT you know this revolution with Genai
[6:01] and you can just uh speak your ideas
[6:03] into existence. I mean this starts you
[6:05] know sounding religious like this is
[6:07] like the you know the gods you know that
[6:09] the myths that that um that humans have
[6:12] created they used to imagine a world
[6:14] where you can you can be everywhere and
[6:17] anywhere at once that's sort of the
[6:18] internet and you can also speak your
[6:20] ideas into into existence and um you
[6:23] know it's it's still early I think uh
[6:25] replet agent is is a fantastic tool and
[6:27] I think this technology is going to get
[6:29] so much more powerful
[6:31] specifically what is an AI agent. I've
[6:34] got this um graph actually here which I
[6:36] don't need to pass to any of you for you
[6:38] to be able to see the growth of AI
[6:41] agents. But this graph is Google search
[6:43] trend data. This also resembles our
[6:45] revenue too.
[6:47] Oh, okay. Right. The the water has
[6:49] arrived. Hello. Thank you. You can come
[6:51] on in. Can I have a go, please? Yes.
[6:54] It's
[6:55] 3951. Great. Thank you so much. Thank
[6:58] you. Thank you. I mean this is this is
[7:00] like a supernatural kind of power. You
[7:03] conjured water. I conjured water from my
[7:05] mind. Yeah. And it's shown up here with
[7:08] us and it clearly thinks we need a lot.
[7:11] But but just to define the term AI agent
[7:14] for someone that's never heard the term
[7:15] before. Yeah. Yeah. So uh I assume most
[7:18] of the audience now are familiar with
[7:20] chat, right? You can go in and you can
[7:22] talk to an AI. It can search the web for
[7:25] you. It has a limited amount of tools.
[7:27] Uh maybe it can call a calculator to do
[7:30] some addition subtraction for you, but
[7:32] that's about it. It's a request response
[7:35] style. Agents are when you give it a
[7:38] request and they can work indefinitely
[7:42] until they achieve a goal or they run
[7:45] into an error and they need your help.
[7:47] It's an AI bot that has access to tools.
[7:50] Those tools are access to the to a web
[7:53] browser like operator, access to a
[7:55] programming environment say like replet,
[7:58] access to um you know credit cards. The
[8:02] more tools you give the agent, the more
[8:03] powerful it is. Of course there's all
[8:05] these consideration around security and
[8:07] safety and all of that stuff. But uh the
[8:10] the most important thing is the AI agent
[8:12] will determine when it finished
[8:15] executing. Uh today AI agents can run
[8:18] for anywhere between you know 30 seconds
[8:22] to 30 minutes. Uh there's a recent paper
[8:25] that came out that's showing that every
[8:28] 7 months the number of minutes that the
[8:31] agent can run for is doubling. So we're
[8:34] at like 30 minutes now. In seven months
[8:36] we're going to be at an hour then you
[8:38] know 2 hours. Pretty soon we're going to
[8:40] be at days. And at that point, you know,
[8:43] AI agent is doing labor is doing kind of
[8:45] humanlike labor and actually uh OpenAI's
[8:49] new model 03 beat the expectation. So it
[8:52] it sort of doubled coherence over long
[8:55] horizon tasks in just three or four
[8:58] months. So we're in this massive and I
[9:01] mean this looks this exponential graph,
[9:04] you know, that shows you the massive
[9:06] trend we're on.
[9:09] Brett, give us a little bit of of your
[9:11] background, but also I saw you writing
[9:13] some notes there. There was a couple of
[9:15] words used there that I thought were
[9:16] quite interesting, especially
[9:18] considering what I know about you. The
[9:19] word God was used a few times.
[9:22] Well, uh, let me just say I'm an
[9:24] evolutionary biologist, and probably for
[9:27] the purposes of this conversation, it
[9:30] would be best to think of me as a
[9:32] complex systems theorist. One of the
[9:34] things that I believe is true about AI
[9:36] is that this is the first time that we
[9:40] have built machines that have crossed
[9:42] the threshold from the highly
[9:44] complicated into the truly complex.
[9:48] And I will say I'm listening to this
[9:51] conversation with a um a a mixture of
[9:55] profound hope and dread
[9:59] because seems to me that it is obvious
[10:03] that the potential good that can come
[10:06] from this technology is effectively
[10:09] infinite. But I would say that the harm
[10:11] is probably 10 times. It's a bigger
[10:13] infinity. And the question of how we are
[10:16] going to get to a place where we can
[10:19] leverage the obvious power that is here
[10:22] to do good and dodge the worst harms. I
[10:25] have no idea. I I know we're not
[10:27] prepared. So I hear you talking about
[10:31] agents and I think um that's marvelous.
[10:34] We can all use such a thing right away
[10:37] and the more powerful it is, the better.
[10:38] The idea of something that can solve
[10:40] problems on your behalf while you're
[10:43] doing something else is marvelous. But
[10:45] of course, that is the precondition for
[10:50] absolute devastation to arise out of a
[10:53] miscommunication, right? to have
[10:55] something acting autonomously to
[10:58] accomplish a goal, you damn well better
[11:01] understand what the goal really is and
[11:02] how to how to pull back the reins that
[11:06] it starts accomplishing something that
[11:08] wasn't the goal. The potential for abuse
[11:11] is also utterly profound. You know, you
[11:14] can imagine, just pick your your dark
[11:18] mirror
[11:19] uh fantasy dystopia where something has
[11:24] been told to hunt you down until you're
[11:25] dead and it sees that as a, you know, a
[11:29] technical challenge.
[11:31] So, I don't know quite how
[11:34] to balance a discussion about all of the
[11:38] things that can clearly come from this
[11:39] that are utterly transcendent. I mean, I
[11:42] do think it is not inappropriate to be
[11:45] invoking God or biblical metaphors here.
[11:49] You know, you're uh producing water
[11:52] seemingly from thin air. I believe that
[11:54] does have an exact biblical parallel.
[11:56] Uh so, uh any case, the the power is
[12:00] here, but so so too is the need for
[12:04] cautionary tales, which we don't have.
[12:06] That's the problem is that there's no
[12:07] body of myth that will warn us properly
[12:10] of this tool because we've just crossed
[12:11] a threshold that is similar in its
[12:16] capacity to alter the world as the
[12:19] invention of writing. I really think
[12:21] that's that's where we are. We're
[12:23] talking about something that is going to
[12:24] fundamentally alter what humans are with
[12:28] no plan. You know, writing alters the
[12:31] world slowly because the number of
[12:33] people who can do it is tiny at first
[12:35] and remain so for thousands of years.
[12:38] This is changing things weakly and
[12:41] that's an awful lot of power to just
[12:43] simply have dumped on a system that
[12:45] wasn't well regulated to begin with.
[12:48] Dan, yeah. So, I'm an I'm an
[12:50] entrepreneur. Um, I've been building
[12:51] businesses for the last 20 plus years.
[12:53] I'm completely well positioned between
[12:56] the two of you here. the excitement of
[12:58] the opportunity and the terror uh of
[13:01] what could go on. There's this image
[13:02] that I saw of New York City in 1900 and
[13:08] every single vehicle on the street is a
[13:10] horse and cart and then 13 years later
[13:12] the same photo from the same vantage
[13:15] point and every single vehicle on the
[13:17] street is a car. And in 13 years all the
[13:20] horses had been removed and cars had
[13:22] been put in place. And um if you had
[13:25] have interviewed the horses in 1900 and
[13:28] said uh how do you feel about your level
[13:30] of confidence uh in in the world? The
[13:33] horses would have said well we've been
[13:34] part of uh humanity for you know horse
[13:36] and hoof hand and hoof for for many many
[13:39] years for for thousands of years.
[13:41] There's one horse for every three
[13:42] humans. Like how bad could it be? You
[13:45] know we'll always have a special place.
[13:46] will always be part of society. Um, and
[13:51] little did the horses realize that that
[13:53] was not the case. That the horses were
[13:55] going to be put out of of uh business
[13:57] very very rapidly. And to reason through
[14:01] analogy, you know, there's a lot of us
[14:02] who are now sitting there going, "Hey,
[14:04] wait a second. Does this make me a horse
[14:05] in 1900?" I think a lot of people don't
[14:08] realize how massive these kind of
[14:10] technologies are going to have as an
[14:11] impact.
[14:12] You know, one minute we're ordering a
[14:14] water and that's cute and the next
[14:16] minute it can run for days and in your
[14:19] words uh it doesn't stop until it
[14:21] achieves its goal and it comes up with
[14:23] as many different ways as it could
[14:24] possibly come up with to achieve its
[14:26] goal and in your words it better know
[14:29] what that goal is. I'm thinking a lot as
[14:32] Daniel's speaking about the vast
[14:35] application of AI agents and where are
[14:37] the bounds because if if this thing is
[14:40] going to get incrementally smarter well
[14:43] incrementally might be an understatement
[14:44] it's going to get incredibly smart
[14:46] incredibly quick and we're seeing this
[14:48] AI race where all of these large
[14:50] language models are competing for
[14:51] intelligence with one another and if
[14:53] it's able to traverse the internet and
[14:56] click things and order things and write
[14:58] things and create things and all of our
[15:00] lives run off the internet today. What
[15:02] can't it do? It's going to be smarter
[15:04] than me.
[15:06] No doubt it already is. And it's going
[15:08] to be able to take actions across the
[15:10] internet, which is pretty much where
[15:11] most of my professional life operates.
[15:13] It's like how I build my businesses.
[15:15] Even this podcast is an internet product
[15:17] at the end of the day because you can
[15:19] create we've done experiments now and I
[15:21] can show the graphs on my phone to make
[15:22] AI podcasts and they have we've just
[15:25] managed to get it to have the same
[15:26] retention as the dire CEO. Now with the
[15:29] image generation model retention as in
[15:30] viewer retention the percentage of
[15:32] people that get to one hour wow is the
[15:34] same now. So we can make the video, we
[15:37] can publish it, we can script it, you
[15:39] can synthesize my voice sounds like me.
[15:42] So what what is it going to be able to
[15:44] do? Mhm. And can you give me the variety
[15:47] of use cases that the average person
[15:49] might not have intuitively conceived?
[15:51] Yeah. So I I tend to be an an optimist
[15:54] and and part of the reason is because I
[15:56] try to understand the limits of the of
[15:58] the technology. What can it do is
[16:00] anything that we can any sort of set of
[16:03] human data that we can train it on. What
[16:06] can it not do is anything that uh humans
[16:10] don't know what to do because we don't
[16:12] have the training data. Of course, it's
[16:14] super smart because it integrates
[16:17] massive amount of knowledge that you
[16:18] wouldn't be able to read, right? It also
[16:21] much faster. It can run through massive
[16:24] amount of computation that you know your
[16:26] brain you can't even comprehend because
[16:28] all of that they're smart. They can take
[16:31] actions but we know the limits of what
[16:34] they can do because we trained them.
[16:37] They're able to simulate what a human
[16:39] can do. So the reason you were able to
[16:41] order the water there is because it was
[16:44] trained NSF data that includes clicking
[16:48] on Door Dash and ordering water. I
[16:51] applaud your optimism and I like the way
[16:52] you think about these puzzles, but I
[16:54] think I see you making a mistake that we
[16:57] are about to discover is very common
[17:00] place. So we have several different
[17:02] categories of systems. We have simple
[17:05] systems, we have complicated systems, we
[17:09] have complex systems and then we have
[17:11] complex adaptive systems.
[17:14] And to most of us, a highly complicated
[17:18] system appears like a complex system. We
[17:20] don't understand the distinction.
[17:22] Technologists often master highly
[17:25] complicated systems and they know, you
[17:28] know, for example, a computer is a
[17:31] perfectly predictable system inside.
[17:32] There's it's deterministic. Mhm. But to
[17:35] most of us, it functions, it's it is
[17:38] mysterious enough that it feels like a
[17:40] complex
[17:41] system. And if you're in the position of
[17:44] having mastered highly complicated
[17:46] systems and you look at complex systems
[17:48] and you think it's a natural extension,
[17:50] you fail to
[17:51] anticipate just how unpredictable they
[17:54] are. So even if it is true that today
[17:57] there are limits to what these machines
[18:00] can do based on their training data, I
[18:04] think the problem
[18:05] is to see what's going to happen. You
[18:09] really want to start thinking of this as
[18:10] the evolution of a new species that will
[18:13] continue to evolve. It will partially be
[18:16] shaped by what we ask it to do, the
[18:18] direction we lead it, and it will
[18:19] partially be shaped by things we don't
[18:21] understand. So, how does this computer
[18:25] that that we have work? Well, one of the
[18:27] things that it does is we plug them into
[18:30] each other using language. It's almost
[18:33] as if you've plugged an Ethernet cable
[18:35] in between human minds. And that means
[18:38] that the cognitive potential exceeds the
[18:41] sum of the individual minds in question.
[18:45] Your AIS are going to do that. And that
[18:47] means that our ability to say what they
[18:48] are capable of does not come down to
[18:51] well we didn't train it on that data. As
[18:54] they begin to interact that feedback is
[18:56] going to take them to capabilities we
[18:59] don't anticipate and may not even
[19:00] recognize once they become present.
[19:02] That's one of my fears. This is an
[19:04] evolving creature and it's not even an
[19:08] animal. If it were an animal, you could
[19:09] say something about what the limits of
[19:11] that capability are. But this is a new
[19:14] type of biological creature and it will
[19:18] become capable of things that we don't
[19:20] even have names for. Even if it didn't
[19:23] do that, even if it just stayed within
[19:25] the boundaries that you're talking
[19:26] about, you mentioned about it having
[19:28] median level intelligence. Well, that by
[19:30] definition means 50% of the people on
[19:32] the planet are less intelligent than uh
[19:35] AI. uh you know to a degree it's almost
[19:38] as if we've just invented a new
[19:40] continent of remote workers. Um there's
[19:43] billions of them. They've all got a
[19:45] masters or a PhD. They all speak all the
[19:47] languages. Anything that you could call
[19:49] someone or ask someone over the internet
[19:51] to do, they're there 24/7 and they're 25
[19:54] cents an hour. Uh if that um so like if
[19:59] that really happened like if we really
[20:01] did just discover that there were a
[20:03] billion extra people on the planet who
[20:04] all had PhDs and were happy to work
[20:06] almost for free that would have a
[20:08] massive disruptive impact on society.
[20:10] Like society would have to rethink how
[20:12] everyone lives and works and gets
[20:15] meaning. Um so like and that's that's if
[20:18] it just stays at a median level of
[20:19] intelligence. Like it's it's pretty
[20:21] profound. I still think it's it's a
[20:23] tool. this is power that is there to be
[20:26] harnessed by entrepreneurs. You know, I
[20:29] I think that the world is gonna get
[20:31] disrupted, right? Um and the the you
[20:34] know, the this you know post-war world
[20:36] that we created where you go through
[20:39] life, you go through 12 years of
[20:40] education, you get to college and you
[20:43] just check the boxes, you get a job. We
[20:46] can already see the fractures of that
[20:48] that it is, you know, this American
[20:50] dream is perhaps no longer there. And so
[20:52] I think the world has already changed.
[20:55] So but but like what are the
[20:56] opportunities? Obviously there are
[20:58] downsides. The opportunities is for the
[21:00] first time access to opportunity is is
[21:03] equal. And I I do think there's going to
[21:05] be more inequality. And the reason for
[21:08] this inequality is because actually
[21:11] Steve Jobs uh you know made this
[21:13] analogy. It's like the the best taxi
[21:15] driver in New York is like 20% better
[21:18] than the you know the you know average
[21:21] taxi driver. the best programmer uh can
[21:24] be 10x better. You know, we call we say
[21:26] the 10x engineer. Now, the variance will
[21:30] be in the thousandx, right? Like the
[21:33] best the best entrepreneur that can
[21:36] leverage those agents could be better uh
[21:39] could be a thousand times better than
[21:41] than someone who doesn't have the grit,
[21:44] doesn't have the skill, doesn't have the
[21:46] ambition. Right? So, so that that will
[21:48] create a world. Yes, there's massive
[21:50] access to opportunity, but there are
[21:52] people who will take seize it and then
[21:55] they'll seize it and then there'll be
[21:57] people who don't. I imagine it almost
[21:59] like a um a marathon race and AI has two
[22:02] superpowers. One superpower is to
[22:04] distract people um such as Tik Tok
[22:06] algorithm. That's right. And the other
[22:08] superpower is to make you hyper
[22:10] creative. So you become a hyper consumer
[22:11] or a hyper creator. And in this marathon
[22:14] race, the vast majority of people have
[22:16] got their shoes tied together cuz AI is
[22:18] distracting them. Some people are
[22:20] running traditional race. Some people
[22:23] have got a bicycle and some people have
[22:25] got a Formula 1 vehicle. And it's going
[22:28] to be very confronting when the results
[22:31] go on the scoreboard and you see, oh,
[22:33] wait a second. There's a few people who
[22:35] finished this marathon in about 30
[22:37] minutes. And there's a lot of us who
[22:38] finished in like 18 hours because we had
[22:42] our shoes tied together. And I can't
[22:44] understand if we've got equal
[22:46] opportunity why there's so much
[22:47] disparity between how fast it you know
[22:50] and do, you know, I'm using an analogy,
[22:51] but this idea that, you know, someone
[22:54] like a lot of people are going to start
[22:55] earning a million dollars a month and a
[22:57] lot of people are going to say, "Hey, I
[22:58] can't even get a job for $15 an hour."
[23:00] there's going to be this kind of
[23:02] interesting wedge. Well, but I I hear in
[23:05] what both of you are saying
[23:07] a kind of assumption that this will all
[23:12] be done on the up and up. And I do want
[23:17] to just I am not a doomer. I agree that
[23:21] the doomers are likely incorrect that
[23:24] their fears are misplaced. But I do
[23:27] think we have a question of a related
[23:29] rates problem.
[23:30] You know I said the potential for good
[23:32] here is infinite and the potential for
[23:35] bad is 10 times.
[23:37] Right? What I mean is there are lots of
[23:41] ways in which this obviously empowers
[23:43] people to do things that they were going
[23:45] to be otherwise stuck in the mundane
[23:48] process of learning to code and then
[23:51] figuring out how to make the code work
[23:53] and bring it to market and all of that.
[23:55] And this solves a lot of those problems
[23:56] and that's obviously a good thing.
[23:58] really the what we should want is the
[24:00] wealth creation object as quickly as we
[24:03] can get there. But the problem is you
[24:06] know as much as it that hyper creative
[24:10] individual is empowered to make wealth
[24:13] the person who is interested in stealing
[24:16] may be even more empowered. And I'm
[24:18] concerned about that at at a pretty high
[24:21] level. the the abuse cases may outnumber
[24:24] the use cases and we don't have a plan
[24:27] for what to do about that. Um to can I
[24:31] can I give you a quick uh like
[24:32] introduction here like the optimistic
[24:34] view? OpenAI invented uh GP the first
[24:37] version GPT came out in 2019. 2020 was
[24:41] GPT2 and so OpenAI you know now they get
[24:44] a lot of criticism and lawsuit from IL
[24:47] Musk that they're no longer open source
[24:50] right they used to be. The reason is in
[24:52] GPT2 they said we are we are no longer
[24:56] gonna uh open source this technology
[24:58] because it's going to create um
[25:00] opportunities for abuse such as you know
[25:03] influencing elections um you know
[25:06] stealing you know grandma's credit card
[25:09] and so on so forth. Wouldn't you say
[25:11] Brett that it is kind of surprising how
[25:14] little abuse we've seen so far?
[25:17] I don't know how much abuse we've seen
[25:19] so far. I don't know how any of us do.
[25:21] And I also even the example that you
[25:25] suggest where chat GPT is no longer open
[25:29] source to prevent abuse. I'm taking
[25:31] their word for it that that's the
[25:33] motivation. Where as a systems theorist,
[25:36] I would say, well, if you had a
[25:38] technology that was excellent at
[25:42] enhancing your capacity to wield power,
[25:46] then open sourcing it is a failure to
[25:49] capitalize on that. and that the most
[25:52] remunerative use is to keep it private
[25:55] and then either sell the ability to
[25:58] manipulate elections to people who want
[26:00] to do so or sell the ability to have it
[26:02] kept off the table for people who don't.
[26:05] And I would expect that that's probably
[26:07] what's going on. There's no if you have
[26:09] a technology as transformative as this,
[26:12] giving it away for free is
[26:13] counterintuitive, which leaves those of
[26:16] us in the public more or less at the
[26:18] mercy of the people who have it.
[26:21] So I I don't see the reason for comfort
[26:25] there. We are at the dawn of this
[26:29] radical transformation of humans that by
[26:33] its very nature as a truly complex and
[26:38] emergent innovation. Nobody on earth can
[26:40] predict what's going to happen. We can
[26:43] we're on the event horizon of something.
[26:45] And the problem is you know we can talk
[26:48] about the obvious disruptions the job
[26:49] disruption and that's going to be
[26:51] massive. And does that lead some group
[26:54] of elites to decide, oh well, suddenly
[26:56] we have a lot of useless eaters and what
[26:58] are we going to do about that? Because
[27:00] that conversation tends to lead
[27:01] somewhere very dark very quickly. Um,
[27:04] but I think that's just the beginning of
[27:06] the the various ways in which this could
[27:08] go wrong without the doomer scenarios
[27:10] coming into play. This is an
[27:12] uncontrolled experiment in which all of
[27:15] humanity is downstream. Yeah. So I I was
[27:18] trying to make the point that OpenAI has
[27:21] been sort of wrong about the uh the sort
[27:25] of uh how big of a potential for harm it
[27:27] is like you know I think we would have
[27:30] heard about it in the news like uh the
[27:33] sort of how much harm it's done and
[27:35] maybe you know some of it is working in
[27:36] the shadows but like the few incidents
[27:39] that we've heard about where you know
[27:41] the cause of LLM's large language models
[27:43] the technology that's powering chat has
[27:46] been a huge headliners uh like New York
[27:49] Times talked about this kid that was,
[27:52] you know, perhaps goated by some kind of
[27:54] chat software that, you know, helps
[27:56] teenagers to be less lonely into into
[27:58] suicide, which is which is tragic. And
[28:00] obviously, these are the kind of safety
[28:02] and and abuse uh issues that we want to
[28:04] we want to worry about. But these are
[28:06] kind of these isolated uh incidents and
[28:09] we do have open- source large language
[28:12] models. Obviously, the thing that
[28:13] everyone talks about is DeepSeek.
[28:15] Deepseek is uh coming from China. So
[28:18] what is Deepseek's incentive? You know
[28:21] perhaps the incentive is to destroy the
[28:24] AI industry in the US. Uh you know when
[28:26] they released Deepseek, you know the
[28:28] market tanked the market for Nvidia, the
[28:30] market for AI and all of that. But there
[28:32] is an incentive to open source. Meta is
[28:34] open sourcing Llama. Llama is another AI
[28:37] similar to Chat GPT. The reason they're
[28:39] open sourcing Llama and Zuckerberg just
[28:41] says that out loud is basically they
[28:44] don't want to be beholden to open AI.
[28:47] They don't sell AI as a service. They
[28:49] use it to build products. And there's
[28:52] this concept in business called
[28:54] commoditize your compliment because you
[28:56] need AI as technology to run your
[28:58] service. The best strategy to do is to
[29:01] open source it. So these market forces
[29:04] are going to create conditions that I
[29:07] think are actually beneficial. So I'll
[29:10] give you a few few examples. One is
[29:13] first of all the AI companies are
[29:15] motivated to create AI that is safe so
[29:17] that they can sell it. Second there are
[29:20] security companies investing in AIs that
[29:23] allows them to protect from the sort of
[29:25] malicious uh acting of of AI. And so you
[29:28] have the free market and we've always
[29:30] had that you know but generally as
[29:32] humanity we've been able to leverage uh
[29:36] the same technology to protect against
[29:38] the abuse. So I I don't really
[29:41] understand this and maybe this is
[29:43] actually this is the exact discussion
[29:45] that you would expect between somebody
[29:47] at the frontier of the highly
[29:49] complicated staring at a complex system
[29:51] and a biologist who comes from the land
[29:53] of the complex and is looking back at
[29:56] highly complicated
[29:58] systems. In game theory we have
[30:00] something called a colle collective
[30:01] action problem. And in the market that
[30:05] you're
[30:06] describing, an individual company has no
[30:10] capacity to hold back the abuses of AI.
[30:15] The most you can do is not participate
[30:17] in them. You can't stop other people
[30:19] from programming LLMs in some dangerous
[30:23] way. And you can limit your own ability
[30:26] to earn based on your own limitations of
[30:28] what you're willing to do. And then
[30:30] effectively what happens is the
[30:31] technology gets invented anyway. It just
[30:34] that the dollars end up in somebody
[30:35] else's pocket. So the incentive is not
[30:38] to restrain yourself so that you can at
[30:40] least compete and participate in the
[30:42] market that's going to be opened. And
[30:46] so the number of ways in which you can
[30:49] abuse this technology. Let's take a
[30:51] couple.
[30:53] What is to stop somebody from training
[30:56] LLMs
[30:58] on an
[31:00] individual's creative output and then
[31:03] creating an LLM that can out compete
[31:07] that individual can effectively not only
[31:09] produce what they would naturally
[31:11] produce over the course of a lifetime
[31:12] but can extrapolate from it and can even
[31:15] hybridize it with the insights of other
[31:17] people so that effectively those who
[31:20] have the LLM can train it on the
[31:22] creativity ity of others, not cut them
[31:24] in on the use of that insight. You can
[31:27] effectively end up putting yourself out
[31:29] of business by putting your creative
[31:31] ideas in the world where they get sucked
[31:32] up as training data for future LLMs.
[31:36] That is unscrupulous, but it's
[31:38] effectively guaranteed. In fact, it's
[31:40] already happening. So, that's a problem.
[31:43] And likewise, what what would stop
[31:45] somebody from interacting with an
[31:48] individual and training an LLM to become
[31:51] like a personalized con artist?
[31:53] Something that would play exactly to
[31:55] your blind spot. That does happen. That
[31:56] that is starting to happen. Um people
[31:58] get phone calls and it sounds like their
[32:00] daughter and I've I've lost my phone and
[32:02] I'm borrowing a friend's phone and all
[32:04] of that sort of stuff. What's what's
[32:06] interesting is that I I think you make a
[32:08] a really good point.
[32:10] I worry about the impact on society. And
[32:12] yet when I look at every single
[32:14] individual who uses AI regularly, it
[32:18] almost has nothing but profoundly
[32:20] positive impact on their life. I look at
[32:23] people like um I was just spending some
[32:25] time with my parents-in-law um who are
[32:28] in their 70s and early 80s and they use
[32:31] AI regularly for all sorts of things
[32:33] that they find incredibly valuable and
[32:36] that improves the quality of their life.
[32:38] I personally did a an M&A a mergers and
[32:40] acquisitions deal where I bought a
[32:42] company last year and the AI was so
[32:46] powerful at helping that process. The
[32:48] conversations were transcribed and they
[32:50] were turned into letters of intent and
[32:52] then press releases and uh legal
[32:55] documents and we probably shaved
[32:58] $100,000 worth of uh costs and and we
[33:01] sped up the whole process and it was
[33:04] pretty magical to see how how it could
[33:06] happen. With that said, you know,
[33:09] there's there's all of these like, well,
[33:10] $100,000 worth of lawyers didn't get
[33:12] paid, right? So, well, what I want to
[33:14] know, yeah,
[33:17] people upset about, but if we look back
[33:20] at the invention of the cell phone or
[33:24] the invention of the social media
[33:26] platforms, there would be every reason
[33:28] to have the exactly the same
[33:30] perspective, right? I remember the
[33:31] beginning of Facebook and I remember the
[33:33] idea that suddenly the process that used
[33:36] to afflict people where you would just
[33:38] lose touch with most of the people who
[33:40] had been important to you, that was not
[33:43] something that needed to happen anymore.
[33:44] You could just retain them permanently
[33:46] as a part of a diffuse
[33:49] uh social grouping that just simply grew
[33:52] and value was added. There's no end to
[33:54] how much good that did, but what it did
[33:58] to us was profound and not evident in
[34:01] the first chapter. Say the same thing
[34:04] about the cell phones and the dopamine
[34:05] traps and the way this has disconnected
[34:08] us from each other, the way it has
[34:10] disconnected us from nature, the way it
[34:13] has altered the very patterns with which
[34:16] we think. It has altered every
[34:19] classroom. Mhm. So, and those things I
[34:22] think are going to turn out to have been
[34:24] sort of minor foreshadowings of the
[34:28] disruption that AI will produce. So, I
[34:31] agree with you today. The amount you can
[34:33] do with AI, there's a tremendous amount
[34:34] of good. There's a little bit of harm.
[34:36] Maybe that's something we need to worry
[34:37] about. But as this develops, as we get
[34:41] to, you know, to peer over the edge of
[34:44] this cliff that we're headed to, I think
[34:45] we're going to discover that we can't
[34:48] yet detect the nature of the alteration
[34:50] that's coming. So, I just wanted to add
[34:52] some context to that cuz Amjad, I I saw
[34:53] the interview you did in a newsletter in
[34:55] 2023 where you said, "I wouldn't prepare
[34:57] for AGI in the same way that I wouldn't
[34:59] prepare for the end of days." It's
[35:00] effectively the end of days if the
[35:02] vision of AGI that some of these
[35:04] companies have comes to bear because
[35:05] it's called the singularity moment
[35:07] because you can't really predict what
[35:08] happens after that. And so like how
[35:11] would you even prepare for that and you
[35:13] want to prepare for the more likely
[35:15] world and that world that you can
[35:16] actually predict is a world where yes
[35:19] there's like a massive improvements of
[35:21] technology and there's like insane
[35:23] compounding effects of technology and
[35:24] it's pretty hard to keep up. From that
[35:26] it appeared that in 2023 you were saying
[35:29] a similar thing to Brett in terms of we
[35:31] can't see around the corner here because
[35:32] it is a singularity.
[35:34] Sorry you also used AGI artificial
[35:37] general intelligence. It'd be
[35:38] interesting to know what your definition
[35:40] of AGI is. say that. Yeah. So, so what
[35:42] what I was saying there is even if I'm
[35:44] wrong that you can actually create a
[35:49] unbounded seemingly conscious artificial
[35:53] intelligence that can entirely replace
[35:57] humans and can act autonomously in a way
[36:00] that even humans can't act and can
[36:03] coordinate across different AIs,
[36:05] different data centers to take over the
[36:07] world. Even if if that's so so the
[36:09] definition of AGI is artificial general
[36:12] intelligence meaning that AI can acquire
[36:15] new skills in efficiently in the same
[36:18] way that a humans can acquire skills.
[36:20] Right now AIs don't acquire skills
[36:22] efficiently you know they they require
[36:24] massive amount of energy and compute
[36:26] entire data set of compute to acquire
[36:27] these these skills and I think there's a
[36:31] again a limit on how general
[36:34] intelligence can get I think for most of
[36:37] the time we're lagging in terms of what
[36:41] humans are are capable of doing the
[36:44] singularity is based on this concept of
[36:47] intellig intelligence explosion so once
[36:49] Once you create an AGI, once you create
[36:52] an artificial general
[36:53] intelligence, that intelligence will be
[36:56] able to modify its own source code and
[36:59] create the next version that is much
[37:01] more intelligent. And the next version
[37:03] creates the next version and the next,
[37:06] you know, for infinity, right? Within a
[37:09] week, within a week, perhaps within
[37:11] milliseconds at some point. Yeah. Right.
[37:13] Uh because it might invent new computing
[37:15] substrate and and and all of that.
[37:18] perhaps they'll use quantum computing
[37:19] and and so then you have this
[37:22] intelligence
[37:24] explosion in a way that it is impossible
[37:27] to predict how the world is going to be
[37:29] and what I'm saying is this is sort of
[37:31] like an end of time story like how would
[37:35] you even prepare for that so if if
[37:38] that's coming like why would I spend my
[37:41] time preparing for I think it's unlikely
[37:43] to happen can't see around the corner
[37:45] yeah but I'd rather prepare
[37:48] what I was saying there. I'd rather
[37:49] prepare for the more likely world in
[37:53] which we have access to tremendous
[37:55] power, but the world's not ending and
[37:58] humans are still uh important.
[38:02] I don't I don't know why you say more
[38:05] likely. I mean, I I think the structure
[38:07] of your argument is is sound. You would
[38:10] prepare for the world that might happen
[38:12] for which you can prepare. There's
[38:14] literally no point in trying to prepare
[38:16] for a world You can't predict it all.
[38:18] The only thing you can do is just sort
[38:20] of upgrade your own skills and pay
[38:23] attention. But if I have one message for
[38:25] the technologists, it's that your
[38:28] confidence about what this can and
[38:32] cannot do is misplaced because you have
[38:35] without noticing stepped into the realm
[38:38] of the truly complex. In the truly
[38:42] complex, your confidence should drop to
[38:45] near zero. that you know what's going
[38:47] on. Are these things conscious? I don't
[38:50] know. But will they be highly likely
[38:52] they will become conscious and that we
[38:53] will not have a test to tell us whether
[38:56] that has happened. Elon Musk predicts
[38:58] that by
[39:00] 2029 we will have AI with us AGI that
[39:04] surpasses the combined intelligence of
[39:06] all
[39:07] humans. And Sam Alman actually wrote a
[39:10] blog three months ago that I read where
[39:11] he said we are confident now Sam Alman
[39:14] being the founder of open arrow which um
[39:15] created chatbt we are confident now that
[39:17] we know how to build AGI as we have
[39:20] traditionally understood it. When I put
[39:22] these things together, I go back to the
[39:24] central question of what role do humans
[39:26] have in in this in the sort of
[39:29] professional output in GDP creation. If
[39:32] it's smarter than all humans combined,
[39:36] if Elon Musk is correct there, and it's
[39:38] able to take actions across the internet
[39:41] and continue to learn. This is like a
[39:43] central question that I'm hoping I can
[39:44] answer today, which is like where do we
[39:46] go? Yeah. I I mean in my vision of the
[39:49] world we are in the creative seat. We're
[39:52] sitting there where um we we are
[39:56] controlling swarms of intelligent being
[39:59] to do our job. You know the way you run
[40:01] your business for example you're sitting
[40:03] at a computer you have an hour to work.
[40:05] Yeah. and you're going to launch like a
[40:07] thousand SDR you know sales uh you know
[40:11] representative to go like grab as as
[40:13] many leads as possible and you're
[40:15] generating new update on replet for for
[40:18] your website here and then uh on this
[40:20] side you you you're actually you you
[40:23] have an AI that's crunching data about
[40:26] your uh your existing business to to
[40:28] figure out how to improve it and these
[40:30] AIs are kind of somehow all coordinating
[40:33] together and I am trying to privilege
[40:36] the human like this is my my mission is
[40:39] to uh build tools for people. I'm not
[40:42] building tools for agents and agents are
[40:45] a tool and so ultimately not only do I
[40:48] think that humans have a privileged
[40:51] position in the in the world and in the
[40:53] universe. We don't know where conscious
[40:56] consciousness is coming from. We don't
[40:58] really have the science to explain it.
[41:01] Um I think humans are special. That's
[41:04] one side is is my belief that humans are
[41:07] are special in the world and another
[41:09] side which I understand that the
[41:11] technology today and I think for the
[41:14] foreseeable future y is going to be a
[41:17] function of us training data. So there
[41:19] was this whole idea like what if chat
[41:21] GPT generates uh pathogens. Well have
[41:24] you trained it on pathogens? They were
[41:26] doing that kind of stuff in Wuhan. know
[41:28] I mean I mean a lot of the biotech
[41:30] companies are essentially using
[41:32] artificial intelligence like I can think
[41:34] of Abselera I think it's Abselera in
[41:37] Canada their whole business is using AI
[41:39] to create new vaccines using artificial
[41:42] intelligence and bigger data sets than
[41:44] we've never have had before and I know
[41:46] cuz I was I was very close to one of the
[41:47] founders of people involved in Absela so
[41:50] that that work is going on anyway and if
[41:51] we think about Wuhan that's it's quite
[41:53] probably well known now that it came out
[41:55] of a lab and people working in a lab and
[41:57] in that scenario that had a huge impact
[41:59] and shut down the world. What I'm the
[42:01] central question I'd love to answer
[42:02] before I throw it back open to the the
[42:04] room is what jobs because I know that
[42:07] you have this perspective. What jobs are
[42:10] going to be made redundant in a world
[42:11] where I am sat here as a CEO with a
[42:13] thousand AI agents, right? I was
[42:15] thinking of all the names of my of the
[42:17] people in my company who are currently
[42:18] doing those jobs. I was thinking about
[42:19] my CFO when you talked about processing
[42:21] business data, my graphic designers, my
[42:23] video editors, etc.
[42:25] So what what jobs are going to be
[42:26] impacted? Yeah, all of those. Uh so I I
[42:29] think and what do they do? You maybe
[42:31] this is useful for for the audience. I
[42:34] think if your job is as routine as it
[42:37] comes, your job is gone in the next
[42:40] couple years. So meaning if you if in in
[42:43] those jobs for example uh quality
[42:46] assurance jobs data entry jobs you're
[42:49] sitting in front of a computer and
[42:50] you're supposed to click uh and and type
[42:53] things in a certain order operator and
[42:55] those technologies are coming on the
[42:57] market really quickly and those are
[42:58] going to displace a lot of a lot of uh
[43:01] labor accountants accountants noise yes
[43:05] I mean I've just pulled a ligament in my
[43:07] in my foot and they did an MRI scan and
[43:09] I had to wait a couple of days for
[43:10] someone to look at the MRI scan and tell
[43:12] me what it meant. Yeah, I'm guessing
[43:14] that that's gone. Yeah, I think I think
[43:16] the healthcare ecosystem is hard to
[43:18] predict because of regulation and and
[43:21] again there there's so many limiting
[43:22] factors on how this technology can
[43:24] permeates the economy because of
[43:26] regulations and and people's willingness
[43:28] to to take it. But you know things
[43:30] unregulated jobs uh that are
[43:34] purely text in text out. if your job,
[43:36] you know, you get a you get a message
[43:38] and you produce some kind of artifact
[43:40] that's like probably text or images that
[43:43] that job is is at risk. So, just to give
[43:46] you some stats here as well, about 50%
[43:48] of Americans who have a college degree
[43:50] currently use AI. The stats are
[43:52] significantly lower for Americans
[43:54] without a college degree. So, you can
[43:55] see how a splinter might emerge there
[43:57] and that crack will write widen because
[44:00] people like us at this table are all
[44:01] messing around with it. But my mom and
[44:03] dad in Plymouth in the southwest, rural
[44:05] England, haven't got like they just
[44:07] figured out iPhones. So like I got them
[44:08] an iPhone and now they're like texting
[44:09] me back. AI is a million miles away. And
[44:12] if I start running off with my AGI, my
[44:13] agents, that gap is going to widen.
[44:15] Women are disproportionately affected by
[44:18] automation, which is what you were
[44:19] talking about there. with about 80% of
[44:21] working women in an at risk job compared
[44:23] to just over 50% of men according to the
[44:25] Harvard Business Review and jobs
[44:27] requiring only a high school diploma
[44:29] have an automation risk of 80% while
[44:32] those requiring a bachelor's degree have
[44:34] an automation risk of just 20%. So we
[44:37] can see again how how this will cause a
[44:39] sort of a it's also a huge risk um with
[44:42] business processor out outsourcing which
[44:44] is essentially western countries sending
[44:47] jobs to India to the Philippines like at
[44:50] the moment millions of people have been
[44:51] lifted out of poverty through the
[44:53] ability to do those kind of business
[44:55] process auto outsourcing jobs and those
[44:58] are all going to go but these they're
[45:00] going to have a thousand employees but
[45:01] but but uh also uh these people are
[45:04] actually already transitioning to
[45:05] training AI eyes. Mhm. You know, so so
[45:08] there's going to be a massive industry
[45:09] around training AI until they're
[45:11] trained. Well, no, you you have to
[45:14] continuously acquire new skills and this
[45:15] is what I'm talking about. I mean, this
[45:17] is again if AI is a function of its
[45:18] data, then you need increasingly more
[45:20] data. And by the way, we ran out of
[45:22] internet data. I was actually thinking
[45:24] interestingly that this might not be
[45:25] great for the United States or the UK,
[45:26] the Western world because it is going to
[45:28] be a leveler where now a kid in India
[45:30] doesn't need a Silicon Valley office and
[45:33] $7 million in investment to throw up a a
[45:35] software company basically. My yeah my
[45:38] belief is that so I have a I have a more
[45:40] broad definition of AGI and the
[45:42] singularity and for me AGI is do we have
[45:45] artificial general intelligence in terms
[45:47] of generally speaking can AI just do
[45:50] stuff that humans used to be able to do
[45:52] and we've already crossed that point we
[45:54] have this general intelligence that we
[45:56] can now all access and 800 million
[45:58] people a week are now using uh chat GBT
[46:00] it's it's exploded in the last 3 months
[46:03] and then to me a
[46:05] singularity Uh when the first tractor
[46:08] went out onto a farm, for me that was a
[46:10] singularity moment. Uh because everyone
[46:12] who worked in farming, it used to take a
[46:14] 100 people to plow a field and now a
[46:16] tractor comes along and two guys with
[46:18] the tractor can now plow the field in
[46:20] just as much time and now 98 people out
[46:22] of 100 are completely out of a job. We
[46:25] also always underestimate a technology
[46:28] if it does go on to change history. When
[46:30] you look back through cars, horses,
[46:32] planes, the Wright brothers just thought
[46:34] of a plane as being something that that
[46:35] army could use, we had no idea of the
[46:38] application. So someone said to me
[46:40] recently, they said, "When it does
[46:41] change the world, we underestimate the
[46:43] impact that it will change the world."
[46:45] And I see people now with their
[46:47] estimations of AI and AI agents already
[46:49] incredibly optimistic. And so if history
[46:52] holds here, we're undershooting the
[46:54] impact it's going to have. And I think
[46:56] this is the first time in my life where
[46:57] the industrial revolution analogies seem
[47:00] to fall a little bit short. Yeah.
[47:02] Because we've never seen intelligent
[47:04] that's like you could I could think of
[47:06] this as a I'm not an intelligent person
[47:07] on this, but I could see that as like
[47:08] the disruption of muscles, whereas this
[47:11] is the disruption of intelligence. That
[47:14] that's that's exactly the thing is that
[47:16] what makes human beings special is our
[47:18] cognitive capacity and very specifically
[47:21] our ability to plug our minds into each
[47:23] other. So that the sum is is uh or the
[47:27] whole is greater than the sum of the
[47:29] parts. That's what makes human cognition
[47:31] special. And what we are doing is we are
[47:34] creating a something that can
[47:37] technologically surpass it without any
[47:39] of the preconditions that make that a
[47:44] safe process. So yes, we've
[47:45] revolutionized the world how many
[47:47] different times? It's innumerable. But
[47:49] you know, we've we've made farming
[47:51] vastly more efficient. That's different
[47:53] than taking our core competency as a
[47:56] species and surpassing ourselves with
[47:58] the product of our of our labor. I think
[48:01] your question is a good one. Then then
[48:03] what does become? We only have one thing
[48:05] left. Um we have our muscles which we
[48:08] got rid of in the industrial revolution
[48:10] and then we have our intellect which is
[48:11] this digital revolution. Now we're left
[48:13] with emotions and agency. So we
[48:15] essentially the the the agency idea I
[48:19] think we used to judge people on IQ and
[48:21] now IQ is the big leveler and now going
[48:23] forward for the next 10 years we're
[48:25] going to look at are you a high agency
[48:27] person or a low agency person. Do you
[48:29] have the ability to get things done and
[48:32] coordinate agents? Do you have the
[48:33] ability to start businesses or give
[48:36] orders to digital armies? uh you know
[48:39] and and essentially these high agency
[48:41] people are going to thrive in this new
[48:44] world because they have this other thing
[48:46] that's been bubbling under the service
[48:47] surface which is really interesting when
[48:49] you said agency is going to remain as an
[48:51] important thing we're sat here talking
[48:54] about AI agents and the crazy thing in a
[48:56] world of AI agents that have super
[48:58] intelligence is I can just tell my agent
[49:00] listen I'm going on holiday please build
[49:02] me a SAS company that spots a market
[49:04] opportunity throw up the website post it
[49:06] on my social media channel I'll be in
[49:08] Hawaii I and this new agentic world is
[49:12] stealing that too cuz now it can take
[49:14] action in the same way that I can browse
[49:16] the internet. I can call Domino's Pizza,
[49:19] speak to their agentic agent, organize
[49:21] my pizza to be there before I even wake
[49:22] up. And in fact, predictability, you
[49:24] know, OpenAI now learns and Sam Alman
[49:27] said that they've expanded the memory
[49:29] feature. So, it knows every it's knowing
[49:30] more and more and more and more about
[49:32] me. It'll almost be able to predict what
[49:34] I want when I want it. It'll know
[49:35] Steve's calendar. He's arriving at the
[49:37] studio. Make sure his cadence is on the
[49:39] side. Make sure his iPad has the brief
[49:42] on it. Do the brief. Do the research for
[49:44] me. And everything else say remember
[49:46] Brett's birthday so when I arrive
[49:48] there'll be something. In fact, it's
[49:50] removing my need for any agency. Yes.
[49:52] And you know again I don't know how to
[49:54] make this point so that it occurs to
[49:57] people what I'm really suggesting
[49:59] but today maybe it's not conscious but
[50:04] well let me put it to you this way.
[50:07] If you're
[50:08] conscious, you started out as a child
[50:11] that wasn't. And although this may not
[50:14] fully encapsulate it, you are
[50:15] effectively an LLM, right? You go from
[50:18] an unconscious infant to a highly
[50:21] conscious adult. And the process by
[50:24] which you do that has a lot to do with
[50:27] being trained effectively on words and
[50:30] other things in an environment in
[50:31] exactly the way that we now train these
[50:33] AIs. So the idea that we can take
[50:36] consciousness off the table, it won't be
[50:38] there till we figure out how to program
[50:39] it in and we're safe because we don't
[50:41] know how consciousness works. I take the
[50:42] opposite lesson. We've created the exact
[50:45] thing that will produce that phenomenon
[50:47] and then we can have philosophers debate
[50:49] whether it's real consciousness or it
[50:51] just behaves exactly as if it were. And
[50:52] the answer is those aren't different.
[50:54] Doesn't matter. And the same thing is
[50:56] true for agency. you know, especially if
[50:58] you've created an environment in which
[51:00] these AIs are de facto competitors, what
[51:04] you're effectively doing is creating an
[51:06] evolutionary environment in which they
[51:08] will evolve to fill whatever niches are
[51:10] there. And we didn't spell out the
[51:11] niches. So, I have the sense we have um
[51:15] we have invited we have we have created
[51:17] something that truly is going to
[51:19] function like a new kind of life and
[51:21] it's especially troubling because it
[51:23] speaks our language. So that leads us to
[51:26] believe it's more like us than it is and
[51:28] it's actually potentially quite
[51:30] different. So but by the way, he's the
[51:33] optimist here, right? Like he's so
[51:36] optimistic about LMS and how how they're
[51:39] going to they're going to evolve. Yes.
[51:42] It's amazing. It's amazing technology.
[51:44] Like I think it raised global IQ, right?
[51:47] Like 800 million people like 800 million
[51:49] people are that much more intelligent
[51:51] and emotionally intelligent as well.
[51:53] Like I know people who previously were
[51:56] very coarse and they kind of robbed
[51:58] people the the the wrong way. They they
[52:00] would say things in not so polite way
[52:03] and then suddenly they started putting
[52:05] their the you know what they're saying
[52:08] through chat in order to kind of make it
[52:11] kinder and nicer and they're more liked
[52:13] now. And so not only is it uh making us
[52:16] more intelligent but also it allows us
[52:18] to be the best version of ourselves. And
[52:21] the the scenario that you're talking
[52:22] about, I don't think I don't know what's
[52:24] wrong with that. Like, you know, you
[52:26] know, the I would want less agency in
[52:29] certain places. Like, I would want
[52:32] something to help me not, you know, open
[52:34] up a peanut butter jar at night, right?
[52:38] You know, there are places in my life
[52:40] where I need more control and I would
[52:44] rather seed it to some kind of entity
[52:46] that could help me make better choices.
[52:51] I mean unfortunately even if there is
[52:54] some small group of elites that are able
[52:58] to go to Hawaii while something else
[53:00] does the mundane details of their
[53:03] business
[53:04] building. We are rather soon going to be
[53:08] faced with a world that has billions of
[53:11] people who do not have the skills to
[53:14] leverage AI. Some of them will be
[53:17] necessary for a time. you're going to
[53:19] need plumbers. But this is
[53:23] also not a long-term solution because
[53:27] not only are there not enough of those
[53:29] jobs,
[53:30] um, but of course we have humanoid
[53:33] robots that once imbued with AI capacity
[53:37] will also be able to take, you know,
[53:39] they'll be able to crawl under your
[53:40] house into the crawl space and fix your
[53:42] plumbing.
[53:43] So what typically happens when you have
[53:47] a massive economic contraction that
[53:49] arises from the fact that a huge number
[53:52] of people are out of work is that the
[53:54] elites start looking at those people and
[53:56] thinking well we don't really need them
[53:57] anyway. And so the idea that this AI
[54:01] disruption doesn't lead us to some very
[54:03] human catastrophe I think is overly
[54:06] optimistic and that we need to start
[54:08] preparing right now. What are the rights
[54:10] of a person who has had whatever it is
[54:12] that they've invested in completely
[54:14] erased from the list of needs? Is that
[54:17] person responsible for not having
[54:18] anticipated AI coming? And is it their
[54:21] problem that that they are now starving
[54:23] and they're being eyed by others as you
[54:25] know a useless eater? I don't think so.
[54:27] How is it different than uh when the uh
[54:30] uh what's it called the the looming
[54:31] machine came and the textile workers you
[54:33] know that the result of the in the lad
[54:35] sort of revolution? uh h how is it how
[54:38] is it different than any time in history
[54:41] when uh technology uh automated a a lot
[54:45] of people out of out of jobs? I would
[54:47] say scale and speed that's how it's
[54:49] different and the scale and speed is
[54:51] going to result in a an unprecedented
[54:54] catastrophe because the rate at which
[54:56] people are going to be simultaneously
[54:58] sidelined not just in one industry but
[54:59] across every industry is just simply uh
[55:03] and it also did actually happen. There
[55:05] was a there was an uh for the first 50
[55:07] years of industrialization from like
[55:10] late 1700s to early 1800s, you actually
[55:13] the Charles Dickens novels are
[55:15] essentially people coming from the farms
[55:17] who are displaced arriving in cities,
[55:20] kids living on the streets. Uh the
[55:23] British decided to pick everyone up and
[55:24] send them over to the over to Australia,
[55:27] which is where I came from. um and uh
[55:31] you know there there were this there was
[55:33] this massive issue of displacement. I
[55:36] think we're going to go into a high
[55:37] velocity economy where rather than this
[55:39] long arc of career that lasts 45 years,
[55:42] we're going to have these very fast
[55:45] careers that last 10 months to 36
[55:48] months. and you invent something, you
[55:51] take it to market, you put together a
[55:53] team of five to 10 people who work
[55:55] together, you then get disrupted, you
[55:58] come Can I mention a story here? Uh
[56:00] there's an entrepreneur that used Replet
[56:03] in a similar way. Uh his name is Billy
[56:05] Howell. You can find him on YouTube on
[56:07] the internet. He would go to Upwork and
[56:10] he would find what people are asking for
[56:13] different requests for certain apps,
[56:15] technologies. Then he would take the
[56:18] what they're asking for, put it into
[56:19] replet, make it an application, call
[56:22] them, tell them, "I already have your
[56:24] application. Would you pay $10,000 for
[56:26] it?" And so, so that's sort of an
[56:27] arbitrage opportunity that's that's
[56:29] there right now. That's not arbitrage.
[56:31] That's theft. How is no what is it? How
[56:33] is that theft? You have somebody who has
[56:35] an idea that can be brought to market
[56:38] and somebody else is cryptically
[56:40] detecting it and then selling back their
[56:42] own idea to them. Well, they're paying
[56:44] them to do that. They're saying, "I will
[56:46] give you $500 if you if someone makes
[56:48] this for me." Right? But this is what I
[56:50] more or less think is going to happen
[56:51] across the whole economy is that yes,
[56:53] from this perspective, we can see that
[56:56] everybody is suddenly empowered to build
[56:58] a great business. Well, what do we think
[57:00] about the folks who are going to be
[57:02] displaced from the top? What are they
[57:04] going to think about all these people
[57:05] building all of these highly competitive
[57:07] businesses? And are they going to find a
[57:08] way to do, you know, what venture
[57:11] capital has done or what record
[57:13] producers have done? What they're going
[57:15] to do is they're going to take their
[57:16] superior position at the top and they
[57:18] are going to take most of the wealth
[57:20] that is produced by all of these people
[57:22] who have these ideas that in a proper
[57:24] market would actually create businesses
[57:26] for them and they're going to parasitize
[57:28] them. I think that we with this in
[57:33] introduction of AI and AI
[57:35] agents old value has moved and now it's
[57:39] not going to be the case that the idea
[57:40] itself is the moat and it's not going to
[57:43] be the case that resources are the moat.
[57:44] So in such a scenario you still have to
[57:46] figure out distribution. You still have
[57:48] to have for example like an audience. So
[57:50] if you're a podcaster now you have a
[57:51] million followers on Twitter. you're in
[57:53] a prime position because you now have
[57:55] something that the the great guy with a
[57:58] great idea with no audience has you have
[58:00] inbuilt distribution. So I now think
[58:01] actually much of the game might be
[58:02] moving to like yeah still about taste
[58:05] and idea but also the the mo is
[58:08] distribution. Yeah. And speaking of
[58:10] adaptive systems um the one of the
[58:13] adaptation that will happen is people
[58:15] will seek uh humans and will seek proof
[58:19] of humanity. Oh, I agree that uh
[58:21] authenticity is going to become the coin
[58:23] of the realm and anything that can be
[58:27] faked or cheated is going to be devalued
[58:30] and things you know spontaneous jazz or
[58:34] you know comedy that is interactive
[58:36] enough that it couldn't possibly have
[58:38] been generated with the aid of AI those
[58:40] things are going to become prioritized
[58:42] you know spontaneous oratory rather than
[58:44] speeches answers some of your questions
[58:47] no it answers it answers my question for
[58:49] the tiny number of people who are in a
[58:51] position to do those things. Stephen,
[58:54] you use the mo word moat. Um, which I
[58:56] think is a really important word for
[58:57] entrepreneurs. We like we like have to
[58:59] have a moat. We think a lot about moes
[59:01] and it's an industrial age. A lot of
[59:02] people don't even know what a moat what
[59:04] you mean by moat. It's just I often
[59:06] think about this idea of what are the
[59:08] moes that are left. So to define how I
[59:11] define a mo, you've got a castle and
[59:12] it's got a like a small circle of water
[59:15] around it. And once upon a time that
[59:17] circle of water defended the castle from
[59:19] attack and you can pull up the
[59:20] drawbridge so nobody could attack you
[59:21] very easily. It's a de defense from
[59:23] something. So it's your it's your
[59:25] shield. It's your your defense. And once
[59:27] upon a time as an entrepreneur, you
[59:28] know, I've got a software company in San
[59:29] Francisco called Third Webb and we
[59:31] raised almost $30 million. We have a
[59:33] team of 50 great developers. And much of
[59:36] our moat was you can't compete with us
[59:39] if you don't have the 50 developers and
[59:40] the $30 million in the bank. How much of
[59:42] that 30 million went to coding? The vast
[59:45] majority of it. I mean what else are we
[59:46] going to do? What else we do? So this is
[59:48] a good thing. I think modes are a bad
[59:50] thing. Okay, let me make the argument
[59:52] there. Uh so everyone is looking for
[59:54] modes you know for example like one of
[59:56] the more uh significant modes is network
[59:59] effects. Yeah, you know, so you can't
[60:02] compete with Facebook or Twitter
[60:05] because to move people from Facebook or
[60:08] Twitter, you need to it's the collective
[60:09] action problem. You need to move them
[60:11] all at once because if one of them
[60:13] moves, then it's the network is not
[60:17] valuable, they'll go back. So you have
[60:19] this chicken and egg problem. Let's say
[60:21] that we have a more decentralized way of
[60:24] doing social networks that will remove
[60:27] the power of Twitter to kind of censor
[60:31] and I I think you're at the other end of
[60:33] of censorship, right? And so part of my
[60:36] optimism about humanity is that um
[60:39] generally there's self-correction.
[60:41] Democracy is a self-correcting system.
[60:43] Uh free markets are largely
[60:45] self-correcting systems. There are
[60:47] obvious problems with with free markets
[60:49] that that we can discuss. But take um
[60:52] health, you know, there is obesity uh
[60:55] epidemic. This period of time when uh
[60:59] companies, you know, ran loose kind of
[61:02] making this sugary, salty, fatty kind of
[61:06] snacks and everyone gorged on them and
[61:08] everyone got very uh you know,
[61:10] unhealthy. And now you have Whole Foods
[61:13] everywhere. Today, people in Silicon
[61:15] Valley, they don't go to bars at all.
[61:17] They go to running clubs. That's how you
[61:19] meet. That's how you go find a date. You
[61:21] go to running clubs. And so, there was a
[61:24] shift that happened because there was a
[61:27] reaction. Obviously, cigarettes is
[61:29] another example. You know, you were
[61:31] talking about phones and our addiction
[61:33] to phones. And I see a shift right now
[61:35] like in my uh friend circle like people
[61:39] who are constantly kind of on their
[61:40] phones is already kind of frowned upon
[61:42] and they don't want to hang out with you
[61:44] because you're you're constantly staring
[61:45] on on your phone. So there's always
[61:47] these reactions and and but the problem
[61:49] is you you reference selfcorrection and
[61:52] I agree that there's actually an
[61:54] automatic feature of the universe in
[61:55] which the self-correction happens. You
[61:57] can't have a positive feedback that
[61:58] isn't reigned in by some outer negative
[62:01] feedback. But the corrections, the list
[62:04] of corrections involves things like you
[62:06] point to where people become enlightened
[62:08] and they realize that they're doing
[62:10] themselves harm with either the sugar
[62:12] that they're consuming or the dopamine
[62:14] traps on their phone and they get
[62:16] better. But also on the list of
[62:18] corrective patterns are genocide and war
[62:22] and you know parasetism. And the problem
[62:27] is these things are destructive of
[62:30] wealth. And so you allude to
[62:35] the superior fact of an open market
[62:38] without moes. Presumably the benefit of
[62:40] that is that more wealth gets created
[62:42] because people aren't kept from doing
[62:44] things that are productive. I see that.
[62:46] But then what is the product of all of
[62:49] this new wealth that is going to be
[62:50] generated by a world empowered by AI?
[62:53] Does it end up so highly concentrated
[62:55] that you have a tiny number of ultra
[62:58] elites and a huge number of people who
[62:59] are utterly dependent on them? What
[63:02] becomes of those people? The learning
[63:04] process, the self-correction pro process
[63:07] goes through harm in order to get to
[63:09] that more enlightened solution. There's
[63:11] nothing that protects us from the harm
[63:14] phase being so apocalyptically terrible
[63:17] that, you know, we get to the other side
[63:18] of it and we say, "Well, that was a hell
[63:20] of a correction." Or maybe there's
[63:21] nobody there to even say that. Those are
[63:23] also on the table. It reminds me of a
[63:24] mouse trap where you see the cheese and
[63:26] we're going, "Oh my god, my
[63:27] grandmother's going to be able to do
[63:28] some research and oh my god, my life's
[63:30] going to get easier." So you head closer
[63:31] and closer to the cheese.
[63:34] And historically, if we look at all of
[63:37] the last 10,000 years, it's a very small
[63:39] number of elites who own absolutely
[63:41] everything and a very large number of
[63:44] surfs and peasants who have a
[63:46] subsistence living. You know, if the
[63:48] elites are too greedy and they freeze
[63:50] out the peasants at too high a level and
[63:52] they try to use brutality or Yeah.
[63:55] eventually it comes back to haunt them.
[63:56] And so what you get is a recognition
[63:59] that you you need a system that does
[64:01] balance these things and you know the
[64:03] west has the best system that we've ever
[64:05] seen. It's one in which we agree on a
[64:07] level playing field. We never achieve it
[64:09] but we agree that it's a desirable thing
[64:11] and the closer we get to it the more
[64:13] wealth we create. But again,
[64:17] if AI empowers those with ill
[64:23] intent at a higher rate that it empowers
[64:26] those who are wealth creating and
[64:28] pro-social, we may be in for a massive
[64:32] regression in how fair the the market
[64:34] that the West is. Is that your top
[64:35] concern versus economic displacement?
[64:38] And I think they're the same thing. How
[64:41] are they the same thing?
[64:43] because the economic displacement is
[64:46] going to start. I don't know how
[64:48] many million people are going to be
[64:50] displaced from their jobs in the US.
[64:52] Suddenly, we're going to have a question
[64:54] about whether or not we have obligations
[64:56] to them. And you agree with that, don't
[64:58] you? Yes. But but but again, it's it's
[65:01] the no pain, no gain. I mean, we're
[65:02] going to go through a period of of
[65:03] disruption. And I think at the other
[65:05] end, the old, you know, sort of
[65:08] oppressive systems will be broken and
[65:10] we're going to create perhaps a fair
[65:13] world, but it's going to have its own
[65:15] its own problems. And what's the scale
[65:16] of that disruption in your estimation?
[65:19] It's hard to say because uh you know
[65:22] there's this concept of limiting factors
[65:24] like you know there is um regulation
[65:28] there's the appetite of people to today
[65:30] for example the health care system is
[65:33] very resistant to innovation because of
[65:35] regulation you know and that's a that's
[65:37] a bad thing on the regulation point it's
[65:39] worth saying that when Trump came into
[65:41] power he signed in a new law which is
[65:44] called removing barriers to American
[65:46] leadership in AI which revokes previous
[65:48] AI policies that were deemed to be
[65:50] restrictive. And obviously when you
[65:52] think about where the funding is going
[65:52] in AI, it's going to two places. It's
[65:54] going to America and it's basically
[65:56] going to China. That's the the vast
[65:58] majority of investment. So with those
[66:00] two in competition, any regulation that
[66:02] restricts air in any way is actually
[66:03] self-sabotage. Mh. And this is, you
[66:06] know, I live in Europe some of the time
[66:09] and it's already annoying to me that
[66:11] when Sam Alman and OpenAI released the
[66:13] 03 model, this new incredible model,
[66:15] it's not in Europe because Europe has a
[66:17] regulation which prevents it from coming
[66:19] to Europe. So, we're now at a
[66:20] competitive disadvantage
[66:22] um which Sam Alman spoken about. And
[66:24] more broadly on this point of
[66:26] disruption, it was I was quite unnerved
[66:29] when I heard that Sam Wolman's other
[66:31] startup was called
[66:33] Worldcoin. And Worldcoin was conceived
[66:36] with the goal of facilitating universal
[66:39] basic income,
[66:41] i.e. helping to create a system where
[66:43] people who don't have money are given
[66:46] money by the government just for being
[66:47] alive to help them cover their basic
[66:49] food and housing needs. Which suggests
[66:51] to me that the guy that has built the
[66:52] biggest air company in the world can see
[66:54] something that a lot of us can't see
[66:56] which is there. Yeah. There's gonna need
[66:58] to be a system to just hand out money to
[67:00] people because they're not going to be
[67:01] able to survive otherwise. I
[67:03] fundamentally disagree with that. Which
[67:04] part do you disagree with? I disagree
[67:06] that first of all that humans would be
[67:08] happy with UBI. I I think that you know
[67:12] uh you know core value of humans and be
[67:16] curious about the evolutionary reasons
[67:17] is we want to be useful. It's really
[67:19] important to know that a lot of the jobs
[67:21] that are in at risk are the most high
[67:23] status, highly paid jobs in the world.
[67:26] Let's take the highest paid job in
[67:27] America um which is an
[67:29] anesthesiologist. Uh this is the highest
[67:32] paid job and highest paid salary job.
[67:35] Salari job. Yeah. And the majority of
[67:37] that job is observing a patient, knowing
[67:40] which type of medication would work best
[67:42] with their body. um giving them the
[67:45] exact right amount, monitoring the
[67:47] impact of that uh on the on the body and
[67:50] then making slight adjustments, the
[67:52] right technology and any nurse will be
[67:55] able to do that job. And you might have
[67:57] one
[67:58] anesthesiologist on on
[68:01] site supervising 10, 20, 30, 40 wards
[68:05] and the technology is, you know, doing
[68:07] the job, but that one person is there
[68:09] just to kind of supervise if something
[68:11] went wrong or if there was an ethical
[68:12] dilemma. What's wrong with that? I mean,
[68:13] if if if the precision is better where
[68:15] they are. No, there's nothing wrong with
[68:17] that except for the fact that a lot of
[68:20] people, hundreds of thousands of people
[68:21] have spent their entire life training to
[68:23] be that, they get an enormous amount of
[68:25] purpose and satisfaction about the fact
[68:27] that that's their career, that's their
[68:28] job. They have mortgages, they have
[68:30] houses, they have status, and that's
[68:33] about to go away. Well, if it's highest
[68:35] paid jobs, maybe you should start
[68:36] saving.
[68:38] Yeah. Well, I mean, but yeah, I I hear I
[68:43] hear you, but you're talking about
[68:45] people who have done vital work. Mhm.
[68:50] Highly specialized work and are
[68:51] therefore not in a great position to
[68:53] pivot pivot based on the invention of a
[68:56] technology that they didn't see coming
[68:58] because frankly, I mean, in the
[69:00] abstract, maybe we all saw AI coming
[69:02] somewhere down the road, but we did not
[69:04] know that it was going to suddenly dawn.
[69:06] And we do have to figure out what to do
[69:08] with those people. It's not their fault
[69:10] that they've suddenly become obsolete
[69:12] and it's inconceivable that people will
[69:16] accept this. It is not. It is
[69:18] fundamentally incompatible with our
[69:20] nature. We have to have things to strive
[69:22] for and you know you can sustain life
[69:26] that way but you cannot um sustain a
[69:30] meaningful existence and so it's a
[69:31] short-term plan at best. Let's talk
[69:33] about meaning. Um on that point of job
[69:36] displacement this is already happening.
[69:38] Cler CEO who has been on this podcast
[69:40] before, a great guy, um said to on a
[69:43] blog post that they published on Cler's
[69:45] website saying that they now have AI
[69:47] customer service agents handling 2.3
[69:50] million chats per month, which is equal
[69:52] to having to hire 700 full-time people
[69:55] to do that. So, they've already been
[69:57] able to save on 700 customer service
[70:00] people by having air agents to do that.
[70:03] And they actually they actually got rid
[70:04] of those 700 jobs, right? I don't have
[70:07] that information in front of me, but
[70:08] I'll have a look. Um, I'll throw it up
[70:10] on screen for anyone that wants context
[70:12] on that. But that's already happening.
[70:13] This isn't some antologist or something.
[70:15] And these aren't high paid people in
[70:17] every case. We've done something
[70:19] similar, by the way. We've we internally
[70:20] we've
[70:22] we've replaced that function for 70%.
[70:26] Yeah. I mean, our company, we're 65
[70:29] people and, you know, we um, you know,
[70:32] we make, you know, millions per per
[70:34] head, you know. So, it's a Are you going
[70:36] to need to hire more people to get up to
[70:38] I think so, but we're we're hiring
[70:40] slowly, like, you know, we're we're
[70:41] using uh customer support, AI, and that
[70:44] meant that we we need less uh customer
[70:47] support, and we're trying to leverage AI
[70:50] as much as possible. the the person in
[70:53] HR at Replet writes software using
[70:56] Replet. So, I'll give you an example.
[70:57] She needed um Orc charts uh software and
[71:01] she looked at a bunch of them, got a lot
[71:03] of demos and they're all very expensive
[71:06] and they they're missing the kind of
[71:08] features that she want. For example, she
[71:09] wanted like version control. She wanted
[71:11] to know when when something changed and
[71:13] to go back in history. She went into
[71:15] replet in 3 days she got exactly the
[71:18] kind of software that she wanted and
[71:20] what was the cost you know perhaps $20
[71:22] you know something like that $ 20 $30
[71:24] once right and um how many employees in
[71:27] HR do we need right now we have two uh
[71:30] if if they're highly levered like that
[71:33] maybe we do not need a 20 HR team on
[71:37] this point of
[71:38] meaning I've heard so many billionaires
[71:41] in AI describe this as the age of
[71:42] abundance and I'm not necessarily sure
[71:44] If abundance is always a great thing
[71:47] because you know when we look at mental
[71:50] health and we look at why how people
[71:53] derive their meaning and their purpose
[71:54] in life much of it is having something
[71:56] to strive towards and some struggle in a
[71:58] meaningful direction to you and this is
[72:00] maybe adjacent but when there was a
[72:03] study done I think it was in Australia
[72:04] where they looked at suicide letters and
[72:06] in the suicide letters the sentiment of
[72:08] men in those suicide letters was they
[72:11] didn't
[72:12] feel worthy they didn't feel like they
[72:14] were worth it. They didn't feel like
[72:16] they were needed by their families. And
[72:19] this is much of what caused their
[72:20] psychological state. And I wonder in a
[72:23] world of abundance where we, you know, a
[72:25] lot of these AI billionaires are telling
[72:26] us that we're going to have so much free
[72:28] time and we're not going to need to
[72:30] work. If there is at all going to be a
[72:31] crisis of meaning, a mental health
[72:33] problem. I mean, there already is. And
[72:36] it doesn't require AI and it's going to
[72:37] get worse. I don't know what to do about
[72:39] it because essentially as human beings
[72:42] we are built like all organisms to find
[72:47] opportunity and figure out how to
[72:50] exploit it. That's what we do. And the
[72:52] world you're describing is really the
[72:53] opposite of that. It's one where you're
[72:55] effectively having your biological needs
[72:59] at the physiological level satisfied and
[73:02] there isn't an obvious place for your
[73:06] spare time if that's what you end up
[73:08] with to be utilized in something that
[73:12] you know there's no place to strive and
[73:14] I do imagine almost at best what would
[73:17] happen is you have people who are being
[73:21] sustained by a universal basic income
[73:24] come and then parasetized
[73:26] uh you know whatever currency they have
[73:29] to spend somebody will be targeting it
[73:30] and they will be targeting it with a an
[73:32] AI augmented system that spots their
[73:36] defects of character. I mean, again,
[73:38] we're already living in this world, but
[73:40] it will be that much worse when the AI
[73:42] is figuring out, you know, what kind of
[73:45] porn to target you with specifically.
[73:47] That's uh it's a nightmare scenario. And
[73:50] I do think it would be worth our time as
[73:54] a species to start considering if we are
[73:57] about to find ourselves in this
[73:58] situation and we find some way of
[74:00] dealing with the basic needs of the
[74:02] large number of people who are going to
[74:04] be
[74:04] sidelined. What would a world have to
[74:07] look like in order for them to have real
[74:09] meaning? Not pseudo meaning, not
[74:10] something that you know superficially,
[74:12] you know, a video game is not meaning
[74:14] even if it feels very meaningful in the
[74:16] moment. I I think that would be a a
[74:19] worthy investment for us to figure out
[74:20] how to produce it. But frankly, I'm not
[74:22] expecting us to either have that
[74:24] conversation or get very far down that
[74:27] road. I think it's much more likely that
[74:29] we will squander the wealth dividend
[74:32] that will be produced by by AI.
[74:34] Interestingly, you also see in Western
[74:36] countries that when we get more
[74:37] abundance, we start having less kids.
[74:40] And we're already seeing this sort of
[74:41] population decline in the Western world,
[74:43] which is was kind of scary. I think it's
[74:46] often associated with affluence like the
[74:48] more money someone makes the less likely
[74:49] they are to want to want to have
[74:50] children the more they try and protect
[74:52] their freedoms. But also on this point
[74:53] of
[74:54] AI relationships are hard you know my
[74:58] girlfriend is happy sometimes and not
[75:01] happy other times and I have to like you
[75:04] know go through that struggle with her
[75:05] of like working on the relationship.
[75:07] Children are hard and if we are
[75:10] optimizing ourselves and you know much
[75:12] of the reason that I sustain the
[75:13] struggle with my girlfriend is I'm sure
[75:15] from some evolutionary reason because I
[75:16] want to reproduce and I want to have kin
[75:19] but if I didn't have to deal with the
[75:23] struggle that comes with human
[75:24] relationships romantic or platonic
[75:26] there's going to be a proportion of
[75:28] people that actually choose that outcome
[75:29] and I wonder what's going to happen to
[75:30] birth rates in such a scenario because
[75:32] we're already struggling. We're already
[75:34] in a situation where we used to be
[75:35] having five children per woman in the
[75:40] 1950s to about two in
[75:45] 2021. And we're seeing a decline. If you
[75:47] look at South Korea, their fertility
[75:48] rate has fallen to 72, the lowest
[75:51] recorded globally. And if this trend
[75:53] continues, the country's population
[75:55] could half by 20, 100.
[76:00] So yeah, relationships, connections, and
[76:03] and also I guess I guess we've got to
[76:04] overlay that with the loneliness
[76:05] epidemic, which
[76:07] is they promised us social connection
[76:09] when social media came about, when we
[76:12] got Wi-Fi connections, the promise was
[76:13] that we would become more connected. But
[76:15] it's so clear that because we spend so
[76:17] long alone, isolated, having our needs
[76:19] met by Uber Eats drivers and social
[76:21] media and Tik Tok and the internet, that
[76:23] we're investing less in the very
[76:25] difficult thing of like going and making
[76:27] a friend and like going and finding a
[76:29] girlfriend. Young people are having sex
[76:30] less than ever before. Everything that
[76:33] is associated with the difficult job of
[76:36] making in real life connection seems to
[76:39] be um falling away.
[76:41] I I will make the case that everything
[76:43] that we've discussed here, all the
[76:46] negative things around loneliness, um
[76:48] around meaning, they're already here.
[76:52] And I don't think blaming technology for
[76:55] all of it is is the right thing. Like I
[76:57] think there are a lot of things that
[76:59] happened because of existing human uh
[77:04] you know, impulses and and motivations.
[77:07] Um well I I wanted to go back to where
[77:10] you started because I do think that this
[77:12] maybe is the fundamental question. Why
[77:15] is it that we are already living in a
[77:17] world that is not making us happy? And
[77:19] is that the responsibility of
[77:21] technology? And I don't think it's
[77:22] exactly technology. Human beings uh
[77:25] among our gifts are fundamentally
[77:27] technological whether we're talking
[77:28] about quantum computing or
[77:32] flintnapping an arrow
[77:35] head. What has happened to us that has
[77:38] created the growing, spreading, morphing
[77:42] dystopia is a process that Heather and I
[77:45] in our book, A Hunter Gather's Guide to
[77:47] the 21st Century, call hyper
[77:49] novelty. Hypern novelty is the fact of
[77:54] the rate of change outpacing our
[77:57] capacity to adapt to change. And we are
[78:00] already well past the threshold here
[78:02] where the world that we are young in is
[78:05] not the world that we are adults in. And
[78:07] that mismatch is making us sick across
[78:10] multiple different domains. So the the
[78:13] question that I ask is is the change
[78:16] that you're talking about going to
[78:18] reduce the rate of change in which case
[78:21] we could build a world that would start
[78:23] meeting human needs better open
[78:25] opportunities for pursuing meaningful
[78:27] work. or is it going to accelerate the
[78:30] rate of change which is in my opinion
[78:32] guaranteed to make us worse off. So if
[78:35] it was a one-time shift, right, AI is
[78:38] going to dawn. It's going to open all
[78:39] sorts of new opportunities. There's
[78:41] going to be a tremendous amount of
[78:42] disruption, but from that we'll be able
[78:44] to build a world. Is that world going to
[78:46] be stable or is it going to be just, you
[78:48] know, one event horizon after the next?
[78:51] If it's the latter, then it effectively
[78:54] says what it does to the humans, which
[78:56] is it it's going to dismantle us. When I
[79:00] look out at society, I I go, okay, it's
[79:02] having a negative impact. When I look at
[79:04] um individual use cases, it's having a
[79:07] profoundly positive impact. Including
[79:09] for me, it's having a very positive
[79:10] impact. So, it's it's one of these
[79:13] things where I wonder what is that what
[79:15] is it that we need to teach people at
[79:17] school so that they understand the world
[79:20] that we're going into? Because one of
[79:21] the biggest issues that we're having is
[79:24] that we're sending kids to school with
[79:26] this blueprint, this template that
[79:28] they're going to have this long arc
[79:30] career that no longer exists that
[79:33] essentially we're treating them like
[79:34] learning LLMs. And we're saying, "Okay,
[79:37] we're going to prompt you. You're going
[79:38] to give us the right answer. you're
[79:40] going to hallucinate it if possible. And
[79:42] you know, and and then we go, "Okay, now
[79:44] go off into the world." And they go,
[79:45] "Oh, but wait a second. I don't know how
[79:47] money works. I don't know how society
[79:48] works. I don't know how my brain works.
[79:50] I don't know how I meant to handle this
[79:52] novelty problem. I'm not sure how to
[79:54] approach someone in a in a social
[79:56] situation and ask if they want to go on
[79:57] a date." Um so all the important things
[80:00] that actually are the important
[80:01] milestones that people want to be able
[80:03] to hit and that technology can actually
[80:06] have an impact on we get no user manual.
[80:09] So I think one of the biggest things
[80:10] that has to happen is we have to equip
[80:14] uh young people all through school that
[80:17] to actually prepare them for the world
[80:18] that's coming or the world that's here.
[80:22] Well on the one hand I think you you
[80:25] outline the problem very well.
[80:27] effectively we have a a model of what
[80:29] school is supposed to do that you know
[80:30] at best was sort of a match for the 50s
[80:33] or something like that and it woefully
[80:36] misses the mark with respect to
[80:38] preparing people for the world they
[80:39] actually
[80:41] face if we were going to prepare them I
[80:44] would argue that the only toolkit worth
[80:46] having at the moment is a highly general
[80:50] toolkit the capacity to think on your
[80:53] feet and pivot as things change is the
[80:55] only game in town with respect to our
[80:57] ability to prepare you in advance. Maybe
[81:00] the the other auxiliary component to
[81:02] that would be teaching you what we know
[81:05] which is frankly not enough about how to
[81:08] live a healthy life. Right? If we could
[81:11] if we could induce people into the kinds
[81:14] of habits of behavior and the
[81:18] consumption of food and then train them
[81:20] to think on their feet, they might have
[81:22] a chance in the world that's coming. But
[81:24] uh the fly in the ointment is we don't
[81:28] have the teachers to do it. We don't
[81:29] have people who know. And that is the
[81:32] question is could the AI actually be
[81:34] utilized in this manner to actually
[81:38] induce the right habits of mind for
[81:40] people to live in that world. I I I
[81:42] spent a lot of time in education
[81:44] technology. One thing that is as we say
[81:47] on the internet a black pill about
[81:49] education in general, education
[81:51] intervention is there's a lot of data
[81:54] that shows that there are very little
[81:57] interventions you can make in education
[81:59] to generate better outcomes. Um and so
[82:03] you know uh there's been a lot of
[82:05] experiment around pedagogy around you
[82:07] know how to configure the the the
[82:08] classroom that have resulted in very
[82:11] marginal improvements. There's only one
[82:14] intervention and this this has been uh
[82:16] reproduced many times that creates two
[82:19] sigma two standard deviation
[82:22] uh positive outcomes in education
[82:24] meaning you're better than 99% of uh of
[82:27] of everyone else and that is one-on-one
[82:31] tutoring I thought so I was going to say
[82:33] smaller classrooms and personalization
[82:34] one-on-one tutoring yeah and and but by
[82:36] the way if you look someone also did a
[82:38] survey of all the geniuses the
[82:40] understanders of the world and found
[82:42] that they all had one-on-one tutoring.
[82:44] They all had someone in their lives that
[82:46] took interest in them and tutored them.
[82:48] So, what can create one-on-one tutoring
[82:51] opportunity for every child in the
[82:53] world? AI. AI. My kids use it and it's
[82:56] incredible. Yeah. As in like um they're
[82:59] interacting and and it's adapting to
[83:01] their speed. Yes. And um it's giving
[83:04] them different analogies to work with.
[83:06] So, like, you know, my son was learning
[83:08] about division and it's asking him to
[83:10] smash glass and how many pieces he
[83:13] smashes it into with this hammer and,
[83:15] you know, and it's saying things like,
[83:16] "No, Xander, go for it. Really smash
[83:18] it." And um and he's loving it, right?
[83:21] Is that synthesis? Yeah. Yeah. I'm an
[83:23] investor in this company. Oh, well, it
[83:24] was it was it's great to watch that
[83:27] simulated one-on-one tutoring because
[83:29] it's talking to him. It's asking him
[83:31] questions. Brett, you're an educator.
[83:33] you uh spent much of your life teaching
[83:35] people in universities. How do you
[83:38] receive all of this? Well, on the one
[83:40] hand, I agree that the uh the closer to
[83:43] one to one you get, the better. But I
[83:46] also personally believe that 0ero to one
[83:50] is best. And what I mean by that
[83:54] is part of what's gone wrong with our
[83:56] educational system is that it is done
[84:01] through abstraction.
[84:03] And effectively the arbiter of whether
[84:07] you have succeeded or failed in learning
[84:09] the lesson is the person at the front of
[84:11] the room. And that's okay if the person
[84:14] at the front of the room is truly
[84:15] insightful. And it's terrible if the
[84:18] person at the front of the room is
[84:19] lackluster, which happens a lot. So what
[84:23] doesn't work that way is interaction
[84:26] with the physical world in which nobody
[84:28] has to tell you whether you've succeeded
[84:29] or failed. If you're faced with an
[84:31] engine that doesn't start, you can't
[84:34] argue it into starting. You have to
[84:36] figure out what the thing is that has
[84:37] caused it to fail, and then there's a
[84:40] great reward when you alter that thing
[84:42] and suddenly it fires up. So, I'm a big
[84:45] fan of being as light-handed as possible
[84:49] and as concrete as possible in teaching.
[84:51] In other words, uh, when I've done it,
[84:53] and not just with students, but with my
[84:55] own children, I like to say as little as
[84:58] possible, and I like to let physical
[85:00] systems tell the person when they've
[85:03] succeeded or failed. And that creates an
[85:06] understanding. You can extrapolate from
[85:08] one system to the next. And you know
[85:10] that you're not just extrapolating from
[85:11] one person's misunderstanding. You're
[85:13] extrapolating from the way things
[85:15] actually work. So, I don't know if AI
[85:18] can be leveraged in that context. My
[85:21] sense is there's probably a way to do
[85:22] it, but one would have to be deliberate
[85:24] about it, especially with robotics and
[85:26] humanoid robots. Actually, that is that
[85:28] is the place uh where where you can do
[85:31] this is with robotics that
[85:34] um it seems to me. Yeah. Well, robotics
[85:38] will teach you the physical computing
[85:40] part of it. And then the question is how
[85:42] do you infuse this with AI so that um it
[85:45] is that it you know it provokes you out
[85:49] of some eddy where you're caught and
[85:51] moves you into the ability to solve some
[85:53] next level problem uh that you you
[85:55] wouldn't have found on your own. What do
[85:57] what do you think should be taught in
[85:58] the classroom with everything that you
[86:00] now know? Well, you're all fathers here.
[86:03] You all have your own children. So, it's
[86:05] a good question for you. How old are
[86:08] your kids? How old are your kids? Uh
[86:09] three and five. 19 and 21 and six,
[86:13] seven, and 10. My children are very
[86:15] young, but uh we already do use AI and I
[86:18] sit down with them in front of replet
[86:19] and we generate ideas and make make
[86:21] games. And um I would say, you know,
[86:23] what Brett said about generality is very
[86:25] important. The ability to pivot and kind
[86:27] of learn skills quickly. Being
[86:30] generative is very very important.
[86:33] Having a you know a fast pace of
[86:35] generating ideas and iterating on those
[86:37] ideas. We sit down in front of Chad GPT
[86:41] and my kid imagines scenario. Oh, what
[86:43] if you know there's a there's a cat on
[86:45] the moon and then you know what if the
[86:47] moon is made of cheese and what if
[86:49] there's a mouse inside it or and so we
[86:51] keep generating these um variations of
[86:55] these different ideas and I and I find
[86:57] that you know makes them more
[86:58] imaginative and and creative. Uh rule
[87:01] number one that I tell my kids is stay
[87:04] away from porn at all costs. I'd rather
[87:07] you have a drug problem than a porn
[87:08] problem. And I actually mean that. I
[87:10] think it's I think porn is more
[87:11] dangerous to the to the human being as
[87:13] as bad as a drug problem is. But when we
[87:16] get to the question of how to confront
[87:18] the world and uh the things that you're
[87:20] going to be um expected to to do in the
[87:23] workplace and all of that, my point to
[87:26] them is you are facing the uh the
[87:30] dawning of the age of complex systems
[87:36] that you are going to have to interact
[87:37] with. And in the age of complex systems,
[87:39] you have to understand that you cannot
[87:42] blueprint a solution. And you have to
[87:44] approach these systems with a upgraded
[87:49] toolkit of humility because the ability
[87:52] of the system to do something you don't
[87:54] predict is much greater than a highly
[87:55] complicated system. So you have to
[87:58] anticipate that and be very sensitive to
[87:59] the fact that what you intended to
[88:02] happen is not what's going to happen. So
[88:04] you have to monitor the unintended
[88:06] consequences of whatever your action is
[88:08] and that there are really two tools
[88:10] which work. One of which you just
[88:12] mentioned which is the prototyping. You
[88:14] prototype things. You don't imagine that
[88:16] I know the solution to this and I'm
[88:18] going to build it. You imagine I think
[88:19] there's a solution down there. I'm going
[88:21] to make a proof of concept and then I'm
[88:23] going to discover what I don't know and
[88:25] I'm going to make the next version.
[88:26] Discover what I don't know and
[88:27] eventually you may get to something that
[88:29] actually truly accomplishes the goal. So
[88:32] prototyping is one thing.
[88:34] And also instead of using the blueprint
[88:36] as the metaphor in your mind uh navigate
[88:40] you can navigate somewhere. And you know
[88:42] that the way I think of it is a surfer
[88:47] is in some ways mastering a complex
[88:50] system but they're not doing it by
[88:52] planning their days surfing down the
[88:54] waves. You can't do that. What you can
[88:56] do is you can be expert at absorbing
[88:59] feedback and navigating your way down
[89:01] the wave. and that that's the right
[89:03] approach for a complex system. Nothing
[89:04] else is going to work. And so I guess
[89:07] the final piece is uh general tools
[89:10] always no specialization. This is this
[89:13] is the age of generalists and um invest
[89:16] in those tools and they will pay.
[89:19] So the guiding philosophy for me is uh
[89:21] to produce high agency generalists. So
[89:24] um ultimately I want them to be
[89:26] motivated self-starters and have a wide
[89:29] general toolkit. I imagine them very
[89:31] much what you imagine which is
[89:33] instructing robots, instructing agents,
[89:36] coming up with ideas. Um, and I imagine
[89:39] them having a very high velocity life
[89:41] where they may be writing a book,
[89:43] organizing a festival, having a podcast,
[89:46] starting a business, and being part of
[89:47] somebody else's business all at once as
[89:49] they are of the ADHD. Yeah. Right.
[89:51] Exactly. Um, so the high agency
[89:54] generalist is the kind of guiding
[89:56] philosophy. Some of the things that we
[89:58] do is like we do chess, we do Brazilian
[90:00] jiu-jitsu, we do dancing, we do acting
[90:03] classes, playing in nature, uh
[90:05] entrepreneurship, understanding that you
[90:07] can start a lemonade. We just did
[90:08] lemonade stands which was amazing. U we
[90:11] sold lots of lemonade on the street. So
[90:13] those kind of things and jumping from
[90:15] one thing to the next thing, but also
[90:17] trying to avoid too many screens and
[90:19] forcing them into making stuff from
[90:22] what's going on around the house. Um,
[90:25] some distinctions that we try and give
[90:26] them is the difference between creating
[90:28] and consuming because I think AI has
[90:30] this superpower of making you a hyper
[90:32] consumer or a hyper creator. Um, and if
[90:34] you don't understand the distinction
[90:36] between creation and consumption, you
[90:38] end up falling into the consumption
[90:39] trap, whether it be porn or just news or
[90:44] um, thing, you know, things that feel
[90:45] like you're productive, but you're
[90:47] actually just consuming stuff. Won't
[90:48] that be the most successful AI? the one
[90:51] that plays with my dopamine the most.
[90:54] Yeah. And and makes you and makes you
[90:56] think that you're achieving something
[90:58] when you're actually just consuming
[91:00] something. So trying to give them the
[91:02] understanding that there is this
[91:04] difference in their life between
[91:05] creation and consumption and to be on
[91:07] the creation side. I started my first
[91:09] business at 12 years old and I started
[91:11] more businesses at 14, 15, 16, 17 and
[91:14] 18. And at that time, what I didn't
[91:17] realize is that being a founder with no
[91:19] money meant that I also had to be the
[91:21] marketeteer, the sales rep, the finance
[91:23] team, customer service, and the
[91:25] recruiter. But if you're starting a
[91:27] business today, thankfully, there's a
[91:29] tool that wears all of those hats for
[91:31] you. Our sponsor today, which is
[91:33] Shopify. Because of all of its AI
[91:35] integrations, using Shopify feels a bit
[91:37] like you've hired an entire growth team
[91:40] from day one, taking care of writing
[91:42] product descriptions, your website
[91:44] design, and enhancing your products
[91:46] images, not to mention the bits you'd
[91:48] expect Shopify to handle, like the
[91:50] shipping, like the taxes, like the
[91:51] inventory. And if you're looking to get
[91:53] your business started, go to
[91:56] shopify.com/bartlet and sign up for a $1
[91:59] per month trial. That's
[92:01] shopify.com/bartlet.
[92:04] The thing that we I think all agree on
[92:07] is that this is inevitable. Do you agree
[92:09] with that, Brett? I think it's sad that
[92:12] it is inevitable, but at this point it
[92:14] is. What part of it do you find sad?
[92:18] We have squandered a long period of
[92:24] productivity and peace in which we could
[92:27] have prepared for this moment. and our
[92:31] narrow focus on competition
[92:36] has created
[92:38] a a fragile world that I'm afraid is not
[92:42] going to survive the disruption that's
[92:43] coming. And it didn't have to be that
[92:45] way. This was foreseeable. I mean,
[92:47] frankly, the movie 2001, which came out
[92:51] the year before I was born, anticipates
[92:54] some of these problems. And you know we
[92:58] treated it too much like education. I
[93:00] mean like entertainment and not enough
[93:03] like education. So we are now you know
[93:07] we've had the AI era opened without a
[93:11] discussion about its implications for
[93:12] humanity. There is now for game
[93:14] theoretic reasons no way to slow that
[93:18] pace because as you point out if we
[93:20] restrain ourselves we simply put the AI
[93:23] in the hands of our competitors. That's
[93:24] not a solution. So, I don't advocate it,
[93:27] but there's a lot more preparation we
[93:29] could have done. We could have
[93:30] recognized that there were a lot of
[93:31] people in jobs that were uh about to be
[93:34] obliterated and we could have thought
[93:36] deeply about what the moral implications
[93:39] were and what the solutions at our
[93:42] disposal might have been. And having not
[93:45] prepared, it's going to be a lot more
[93:46] carnage than it needed to be. Amjad, I
[93:49] heard you say a second ago that what we
[93:50] should be talking about is how we deal
[93:51] with job displacement. Do you have any
[93:54] theories if you were prime prime
[93:56] minister or president of the world and
[94:00] you your job was to deal with job
[94:01] displacement let's just say in the
[94:03] United States how would you go about
[94:05] that the first thing I would do is uh
[94:09] teach people about these systems whether
[94:11] it's um programs on on the TV or
[94:16] outreach or or what have you just trying
[94:18] to get people to understand how chat GBT
[94:21] works how these algorith algorithms work
[94:24] and as the new jobs arrive um I think
[94:28] you know there's going to be an
[94:29] opportunity for people to be able to
[94:31] detect that you know the this job
[94:34] requires this set of skills and I I I
[94:37] have this this kind of experience and
[94:39] although my experience are potentially
[94:41] outdated I can repurpose that experience
[94:43] to do that job I'll give you an example
[94:46] a teacher his name is Adil Khan you know
[94:49] he started using at the time GPT3 and uh
[94:53] felt like it does amazing work as a
[94:55] tools for teachers or even potentially a
[94:58] teacher itself. So he learned a little
[95:00] bit of coding and he he went to
[95:01] Unreallet and he built uh this company
[95:04] and um just two years later they're
[95:07] worth hundreds of millions of dollars.
[95:08] Obviously, not everyone will be able to
[95:10] create businesses of that scale, but
[95:13] because you have an experience in a
[95:15] certain domain, you'll be able to build
[95:18] the next iteration of of of that using
[95:22] technology. So, even if your job was
[95:23] displaced, you'll be able to figure out,
[95:27] you know, what's what's potentially what
[95:30] potentially comes after that. So, so I I
[95:32] I think people's expertise that they
[95:35] built, I don't think they're all for
[95:37] waste. Even if your job went away,
[95:41] you can never really predict what jobs
[95:42] are coming. I mean, I think of this
[95:44] crazy situation where I tell my
[95:47] grandfather, what is a personal fitness
[95:50] trainer? And he would his mind would be
[95:53] blown by this idea that well, okay, I
[95:56] don't really want to go to the gym, so I
[95:58] have to make an appointment and pay
[95:59] someone to go to the gym and meet with
[96:01] me there. And then he stands there and
[96:02] tells me to lift heavy things that I
[96:04] don't really want to lift. And then he
[96:06] counts them and tells me that I've done
[96:08] a good job and then I put the heavy
[96:10] things down and then at the end of that
[96:12] I feel really good and I pay him a bunch
[96:14] of money. My grandfather would be like
[96:16] what on earth have you been scammed? Is
[96:18] this so we can never predict what what
[96:21] this uh future of jobs would look like.
[96:23] Even just 20 30 40 years apart the jobs
[96:26] rapidly and convincingly just morph into
[96:29] something else. I think it's very
[96:31] dangerous the idea that we need to focus
[96:34] on skills. I think the future is not in
[96:36] skills. Skills are being replaced. It's
[96:38] this idea that the education system has
[96:40] to stop being compartmentalized and has
[96:42] to be a lifelong learning approach. The
[96:44] department of education needs to be
[96:47] seeing people as lifelong learners who
[96:49] are constantly disrupted and need
[96:51] re-education. Interesting. That that's
[96:53] going to be a thing. The Department of
[96:54] Education needs to start as a kid and go
[96:58] right through to maybe 70. Does the
[97:00] Department of Education have a role
[97:01] anymore at all? Depends on your
[97:03] definition of education. I think if
[97:04] you're trying to teach kids or if you're
[97:06] trying to teach kids to, you know,
[97:09] remember facts and figures from a
[97:11] history book, then no. But if it's about
[97:14] coaching, mentoring, being displaced,
[97:16] finding the next thing, and maybe if
[97:18] it's AIdriven and all of those kind of
[97:20] things, then it's a different paradigm
[97:22] shift around what education is and what
[97:23] its purpose is. And if we see it as a
[97:26] fluid thing where we wave into an
[97:28] opportunity and then wave back into
[97:30] education, spotting a new opportunity
[97:32] and then back here. If we're learning
[97:35] rather than skills, but we're learning
[97:36] tools. So it's a tools-based education
[97:39] as opposed to a skills-based education.
[97:41] The the purpose of education for most of
[97:43] human history was about virtue, about
[97:45] becoming a great person who had good
[97:47] judgment and who had good values. And we
[97:49] don't really do much of that anymore.
[97:51] But I think if we essentially said if we
[97:53] get back to what is if we ask the
[97:56] question what is the purpose of
[97:57] education and where does it fit in our
[97:58] lives and at what time frame does it go
[98:01] for and then we just trust that people
[98:04] are going to come up with weird and
[98:06] wonderful jobs. You know this is sounds
[98:09] crazy but also and this is a weird
[98:13] analogy. My cat is incredibly happy. How
[98:17] do you know? Well, it it demonstrates
[98:20] all the characteristics of of being a
[98:22] happy cat and it lives in a world of
[98:25] super intelligence as far as it's
[98:27] concerned. So, there's this house and
[98:29] food just magically happens. It has no
[98:31] idea that there's this Google calendar
[98:33] that runs a lot of things that happen
[98:35] around it. The food gets delivered. The
[98:38] money is magically made by something
[98:40] that is inconceivably more intelligent
[98:42] than the cat. And yet the cat has
[98:44] evolved to be living this life of
[98:46] purpose and meaning inside the house.
[98:48] And as far as it's aware, it it's got a
[98:50] great life. But you have the power at
[98:53] any moment if you're having a bad day to
[98:55] do something not so pleasant to that
[98:57] cat. And it can't really reciprocate
[98:59] that. Exactly. But but what's in it for
[99:02] me to hurt the cat?
[99:04] Because the in this analogy, you might
[99:08] want to move house and the landlord
[99:09] doesn't allow cats. So you've got a
[99:11] decision to make. Yeah, that there are
[99:13] things that the cat is highly disrupted
[99:15] by due to no fault of the cat. I get it.
[99:17] But as far as cat existence goes and the
[99:20] and the history of cats, if you were to
[99:23] ask that cat, do you want to tra trade
[99:25] places with any of the other cats that
[99:27] came before you? It would probably say,
[99:28] I don't want to take the risk because
[99:30] all the other cats had to fend for
[99:32] themselves in a way that I don't have
[99:33] to. It's very possible that we live end
[99:35] up living in a life a lot like the house
[99:38] cat in the sense that from our
[99:41] perspective we're extremely like we're
[99:44] having very interesting lives and
[99:46] purpose meaning and just there's this
[99:48] massive higher intelligence that's just
[99:50] running stuff and we don't know how it
[99:53] works but it doesn't really matter
[99:54] whether how it works. We we we are the
[99:57] beneficiaries of it and it it's doing
[100:00] important things and we're enjoying
[100:01] being house cats in in its in its life.
[100:03] I have a few things to say about this.
[100:04] One, I'm pretty sure your cat's not as
[100:06] impressed with your capacity as you are
[100:10] or as you think he is. Um I just know
[100:12] cats well enough to be pretty sure of
[100:14] that. But oh, it looks down on me. Yeah,
[100:15] you're right. I think it I think it's a
[100:17] fair it's a fair point that there is an
[100:19] existence and actually, you know, pets
[100:21] really do have it. If they have loving
[100:22] owners, they really do have it pretty
[100:24] great. And I would also point out that
[100:25] there's a way in which we already are
[100:27] this way. Most of us do not understand
[100:30] the process that results in electricity
[100:33] coming out of the walls of our house or
[100:35] the water that comes out of the tap. And
[100:37] we're pretty much okay with the fact
[100:39] that somebody takes care of that and we
[100:41] can busy ourselves with whatever it
[100:42] might be. But the place that I find
[100:45] something troubling in your description
[100:48] is that you say that the nature of what
[100:51] we do is to deal with the fact that jobs
[100:54] are always being upended. That's a very
[100:57] new process. That is the hypern novelty
[101:00] process. It used to be that it was only
[101:03] very rarely that a population had a
[101:06] circumstance where you didn't
[101:07] effectively do exactly what your
[101:09] immediate ancestors did. Right? Um, in
[101:12] general, you took what the jobs were,
[101:15] you picked something that was suited to
[101:18] you, and you did that thing
[101:20] intergenerationally.
[101:21] Intergenerationally. And the point is,
[101:23] we've now gotten to the point where even
[101:26] within your lifetime, what is possible
[101:29] to get paid for is going to shift
[101:31] radically in ways that nobody can
[101:33] predict. And that is a dangerous
[101:35] situation. Like probably every two
[101:37] years, like two or three years, right?
[101:39] And so maybe there's some model by which
[101:41] we can surf that wave and you can learn
[101:43] a generalist toolkit and you know that
[101:46] your survival doesn't depend on your
[101:49] being able to you know switch up every
[101:51] two years and never miss a beat or maybe
[101:54] we can't but I do think it is worth
[101:56] asking the question if the rate of
[102:00] technological change has taken us out of
[102:03] the normal human circumstance of being
[102:07] able to deduce what you might do for a
[102:09] living based on what your ancestors did
[102:11] and put us in a situation where what
[102:13] your ancestors did is going to be
[102:15] perfectly irrelevant no matter what. But
[102:17] that is effectively a choice that has
[102:19] been made for us. And we could choose to
[102:22] slow the rate of change so that we would
[102:26] live in some kind of harmony where our
[102:29] developmental environment and our adult
[102:32] environment were a match. Now, as a
[102:33] biologist, I would argue if we don't do
[102:35] something like that, this is a matter of
[102:38] time. Yeah. How would we change? How do
[102:40] we slow the rate of change? Well, I I
[102:42] mean, you can you can you can be the
[102:43] Amish, right? You can be the Amish and
[102:45] live in your own communities and and I
[102:47] would assume some people would would
[102:49] want that. Well, I'm you know, when
[102:52] Heather and I wrote our book, I wanted
[102:54] the first chapter to be, are the Amish
[102:56] right? And the answer is they can't be
[102:59] exactly right because they picked an
[103:01] arbitrary moment to step off the
[103:03] escalator. But are they right that
[103:04] there's something dangerous about this
[103:06] continuing pattern of technological
[103:07] change? Clearly they are. What do the
[103:09] Amish do for anyone that doesn't know?
[103:11] The Amish live as if it was what 1850 or
[103:18] something. So they live in a they don't
[103:21] use cars. They I think they do have
[103:23] phones but they do not have electricity.
[103:27] Basically they they voluntarily accept a
[103:30] techn they're basically a lite community
[103:33] and they uh have turned out to fare
[103:37] surprisingly well against many of the
[103:39] things that have upended modern one of
[103:42] them right yeah co they did beautifully
[103:44] quite happy people very low autism rates
[103:46] they they have all sorts of advantages
[103:48] so anyway I'm not arguing that we should
[103:50] live like the Amish I don't see that but
[103:51] I do think the idea that they had an
[103:53] insight which was you need to step off
[103:56] that escalator because you're just going
[103:57] to keep making yourselves sicker is
[103:59] probably right now. Maybe this is a
[104:02] one-time shift. We've stepped over the
[104:05] event horizon. We are going to be living
[104:07] in the AI world. And maybe if we're
[104:09] careful about it, we can figure out how
[104:11] to turn that landscape of infinite
[104:14] possibility that you're describing into
[104:17] a place that doesn't change. That you
[104:20] always have the opportunity to decide
[104:23] what needs to be done. But that none
[104:26] that the that living over that event
[104:28] horizon is not an everchanging process.
[104:31] It's just the next frontier. I do want
[104:33] to also propose or ask the question when
[104:36] we talk about our hyperchanging world.
[104:38] Isn't it harder for older people to
[104:41] learn because of the the way that the
[104:43] brain works in terms of processing speed
[104:44] and memory flexibility?
[104:47] So I was wondering if you're going to
[104:48] get a situation where like my father h
[104:50] because of his brain and the reduced
[104:52] memory flexibility and processing speed
[104:54] that happens when you're older is going
[104:55] to struggle significantly more than my
[104:57] niece who can seem to learn I mean my
[105:00] niece knows five languages and she's
[105:01] seven or something crazy like that five
[105:04] languages but I mean the brain is much
[105:05] more plastic isn't it? So the it's and
[105:07] that's goes back to our evolutionary
[105:09] psychology which you know much
[105:10] evolutionary history which you know much
[105:11] more than I do about of we're meant to
[105:14] learn our lessons when we're young. use
[105:16] that information for a lifetime. But if
[105:18] that information is changing quickly,
[105:20] well, that's I mean this is exactly what
[105:21] I'm pointing to. It is not normal for
[105:24] your developmental environment to fail
[105:27] to prepare you for your adult
[105:29] environment. The normal thing is as a
[105:31] young person, you take on ever more of
[105:34] the responsibilities of the adult
[105:37] environment. And then at some point, you
[105:40] know, in a properly functioning culture,
[105:42] there's a right of passage. You go into
[105:44] the bush for 10 days, you come back
[105:46] with, you know, a large uh, you know,
[105:49] game animal and now you're an adult and
[105:52] you take that program that you've been
[105:53] building and you activate it. And that
[105:55] is normal. And, you know, you're a lot
[105:59] happier person. You're a lot more
[106:00] fulfilled if your life has that kind of
[106:02] continuity to it. And you know, I'm not
[106:05] against the idea that we have enabled
[106:07] ourselves to do things that can't be
[106:09] done if that's the the limit, but we
[106:13] have also harmed ourselves gravely. And
[106:16] I would like to somehow pry apart our
[106:21] ability to improve our well-being from
[106:26] our self-inflicted wounds that come from
[106:29] this
[106:30] neverending pace of change. And I don't
[106:32] know if it's possible, but I think it's
[106:34] a worthy goal. Something amusing. I
[106:36] don't know if it's exactly a
[106:37] counterpoint, but um during co
[106:41] especially and you know through the the
[106:44] recent technological change uh some
[106:47] people have started living closer to the
[106:50] more ancestral environment. Um so uh
[106:54] people whose jobs are online, some of my
[106:58] friends like went and built communities
[107:00] like collectives where they you know
[107:02] live and they they create farms and they
[107:05] they eat and then they eat and they have
[107:07] like an email job. They do their email
[107:09] jobs for five hours and and go out and
[107:11] they all have children and it's it's a
[107:14] fascinating life. And there was so much
[107:16] rethinking in in Silicon Valley about
[107:18] how we live. And there's a bunch of
[107:19] startups that are trying to create um
[107:21] cities where they're like, okay, we know
[107:25] that we've we're suffering because our
[107:28] cities are not really walkable. And
[107:31] there's so many reasons why we're
[107:32] suffering. First, we're not getting the
[107:34] movement. Second, there's a social
[107:36] aspect of walkable city where you're
[107:38] able to interact with people. You'll
[107:40] make uh friends by just happening to be
[107:42] in the same place as others. let's
[107:44] actually build uh walkable cities and if
[107:47] we want to you know uh transport faster
[107:50] we'll have these self-driving cars on
[107:52] the perimeter of the city that are going
[107:53] around and I think there are ways in
[107:56] which technology can afford us to uh to
[107:59] live uh in a way that
[108:02] reverses I I guess in a more local way.
[108:06] I I like that vision, but I also am
[108:08] aware that there's a different vision,
[108:10] right? You see people in Palo Alto, for
[108:12] example, actually exerting, you know,
[108:15] very strong controls on how much their
[108:17] children are exposed to, uh, you know,
[108:20] to phones. And I live in Palo Alto.
[108:22] Yeah. So, so you see that. On the other
[108:24] hand, what I am worried about is that
[108:28] the the elites of PaloAlto don't realize
[108:32] that what they're doing is they're
[108:34] figuring out how to reduce the harm to
[108:36] their own families as they're exporting
[108:38] the harm to the world of these
[108:40] technologies that for everybody else are
[108:42] unregulated. And so the question is, can
[108:45] we bring everybody along? If the AI
[108:47] revolution is going to alter our
[108:49] relationship to work and everything
[108:52] else, can we bring everybody along so
[108:54] that at the end of this process instead
[108:56] of saying well you know it's a shame
[108:58] that uh you know three billion people
[109:01] were sacrificed to this transition but
[109:03] progress is progress we can really say
[109:05] well we figured it out and everybody now
[109:07] is living in a style that is closer to
[109:10] their programming and closer to the
[109:12] expectations of their physical bodies.
[109:15] You know, if that were true, then I
[109:16] would I would be I would love to be
[109:18] wrong in my fears about what's coming.
[109:21] Um, but unfortunately, the market is not
[109:24] going to solve this problem without our
[109:25] being deliberate about forcing it to.
[109:30] What's your biggest fear? Like when you
[109:32] say my fears about what's coming, what
[109:33] do you like what's what's the picture
[109:35] that comes in your mind? Oh, it's a
[109:36] whole different topic actually. Um my my
[109:39] fear coming stemming from technology
[109:43] uh and AI is that this is a runaway
[109:48] process and that that runaway process is
[109:50] going to interface very badly with some
[109:53] latent human programs. that in effect
[109:56] the need for
[109:58] workers largely disappears and the
[110:01] people who are at the head of the
[110:03] processes that result in that
[110:05] elimination for the need for workers
[110:07] start talking about useless eaters.
[110:08] Maybe they come up with a new term this
[110:10] time. Thin the herd. Yep. Or they allow
[110:12] it to be thinned or something. Right.
[110:14] I've heard you talk about the five key
[110:17] concerns you have or the five key
[110:19] threats you have before. Could you name
[110:21] those five? So the first one is the one
[110:23] I worry least about. I don't worry zero
[110:25] about it, but I worry least about it,
[110:27] which is the malevolent AI uh that the
[110:29] doomers are so focused on. The second
[110:33] one is the idea that you know an AI can
[110:36] be misaligned not because it has
[110:38] divergent interest but because it just
[110:40] misunderstands what you've asked it.
[110:41] these autonomous agents. You know, the
[110:43] famous example is you ask them to
[110:45] produce as many paper clips as possible
[110:47] and they start liquidating the universe
[110:48] to make paper clips and you know, it's a
[110:50] it's a sorcerer apprentice kind of
[110:52] issue. The third one I would say
[110:56] is actually all of the remainder of them
[110:58] I would say are guaranteed and
[111:02] um the third of them is the derangement
[111:07] of human intellect that we are already
[111:11] living in a world where it's very
[111:12] difficult to know what the facts even
[111:16] mean. Right? We the facts are so
[111:18] filtered and we are so persuaded by
[111:21] algorithms that it's you know our
[111:24] ability to be confident even in the
[111:25] basic facts even within our own
[111:27] discipline sometimes is uh at an
[111:29] all-time low and it's getting worse and
[111:32] that problem takes a giant leap forward
[111:35] at the point that you have the ability
[111:39] to generate undetectable deep
[111:42] fakes. Right? that's going to alter the
[111:45] world very radically when the fact that
[111:47] you're looking at videotape of somebody
[111:49] robbing a bank doesn't mean that they
[111:51] robbed a bank or that a bank was even
[111:53] robbed. Um, so anyway, I call this we
[111:56] deal with this a lot by the way. I I
[111:57] think every single week, every single
[112:00] week I send my I have a chat people that
[112:02] just are now basically spending I'd say
[112:04] 30% of their time dealing with deep
[112:06] fakes of me doing crypto scams, inviting
[112:09] people to Telegram groups, and then
[112:11] asking them for credit card details. We
[112:13] had one on X. I think you probably saw
[112:15] it, Dan, didn't you of me? But that
[112:17] someone was running ad deep fake ads on
[112:18] X of me. And it wasn't just one ad. It
[112:21] was like there were it was like swatting
[112:22] flies. There was 10 of them. And I
[112:24] messaged them to X and there was 10
[112:26] more. Then the day after there was 10
[112:27] more. Then the day after there was 10
[112:29] more. Then it started happening on on
[112:30] Meta. So it's a video of me basically
[112:33] asking you to come to a Telegram group
[112:34] where people are being scammed and
[112:36] audience members of mine are being
[112:37] scammed. And when I send them to Meta,
[112:39] they thankfully remove them. But then
[112:41] there's five more. And I went on
[112:42] LinkedIn yesterday and my DMs are,
[112:44] "Steve, by the way, there's this new
[112:45] scam." And I actually at this point I c
[112:47] I'd need someone fulltime just sending
[112:50] this over to Meta. I'm the I'm the same
[112:52] but on a smaller scale. Every week it's
[112:55] did you really message me on Facebook
[112:57] asking me for my crypto wallet and blah
[112:59] blah blah. My my least favorite ones are
[113:01] when the the single mother messages me
[113:03] saying that she just paid £500 of her
[113:05] money and how devastated she is and I
[113:07] feel this moral obligation to give her
[113:09] her money back um because she's fallen
[113:11] for some kind of scam. That was me. It
[113:13] was my voice. It was a video of me
[113:14] telling her something. Yeah. And I don't
[113:16] know I don't know how you deal with that
[113:17] but sorry do continue. Well, I mean
[113:19] that's actually on the list here. the
[113:21] massive disruption to the way things
[113:22] function both because people are going
[113:24] to be unemployed in huge numbers and
[113:27] because those who are not abiding by our
[113:30] social contract are going to find
[113:32] themselves empowered more than the
[113:34] people who do. So in this case, not only
[113:38] is this poor woman, you know, now out
[113:40] 500 bucks for whatever the scam was, but
[113:44] you've also been robbed whether or not
[113:46] you pay her back for the thing that she
[113:48] thought she purchased. your credibility
[113:51] is being stolen by somebody and you have
[113:53] no capacity to prevent it. This has
[113:55] happened to me also and it is profoundly
[113:58] disturbing and it is only one of a dozen
[114:00] different ways that AI enables those who
[114:06] are absolutely willing to shrink the pie
[114:11] from which we all derive in order to
[114:14] enlarge their slice. you know, there
[114:16] there are innumerable ways that this can
[114:18] happen and um I think people do not see
[114:21] it coming. They don't understand how
[114:22] many different ways they are going to be
[114:26] robbed every bit as surely as if
[114:27] somebody was printing money. Um and then
[114:30] the last one is that this just simply
[114:33] accelerates demographic
[114:36] uh processes that do potentially result
[114:39] in the unleashing of technologies that
[114:43] pre-existed AI. you know, this this can
[114:45] easily result in an escalation
[114:48] uh into wars that turn nuclear. Um, so
[114:52] anyway, I think that list could probably
[114:55] be augmented at this point now that
[114:57] we've, you know, spent a little time in
[114:59] the AI era. We can begin to put a little
[115:01] more flesh on the bones both of what is
[115:04] possible in this era and what we should
[115:06] fear. One of those you you mentioned uh
[115:08] truth, you know, the problem of truth.
[115:11] Would you say just a thought experiment,
[115:14] someone today like an average person,
[115:17] college educated say
[115:18] person, are they more
[115:21] propagandized or led astray
[115:26] than someone in Soviet Russia?
[115:30] Well, I don't know because I didn't live
[115:32] in Soviet Russia, but my understanding
[115:34] from people who did was that there was a
[115:39] wide awareness that the propaganda
[115:42] wasn't true. Doesn't mean they knew what
[115:43] to believe, but there was a cynicism,
[115:46] which is one of my fears here, is that
[115:48] the, you know, you're really stuck
[115:50] choosing between two bad options in a
[115:54] world where you can't tell what is true.
[115:55] You can either be overly credulous and
[115:57] be a sucker all the time, or you can
[115:59] become a cynic and you can be paralyzed
[116:01] by the fact that you just don't believe
[116:02] anything. But neither of those is a
[116:05] recipe for do you think Google search
[116:06] first and maybe now chat GPT has helped
[116:09] people more or
[116:11] less to find truth? I
[116:15] think it's not chat GPT exactly but all
[116:18] of the various AI engines that we're
[116:20] starting with Google have briefly
[116:22] enhanced our capacity to know what's
[116:24] true because in fact they allow us to
[116:26] see through the algorithmic manipulation
[116:29] because the AI is not well policed. you
[116:33] can get it to recognize patterns that
[116:35] people will swear are not true. Um, and
[116:38] so anyway, a lot of us have found it
[116:40] useful in just simply unhooking the
[116:42] gaslighting. Um, so that's been very
[116:44] positive. But I also remember the early
[116:47] days of search and search used to be a
[116:50] matter of there are some pages out
[116:53] there. I don't know where they are.
[116:54] Here's a mechanized something that's
[116:56] looked through this stuff and just point
[116:58] me at the direction of things that
[116:59] contain these words. right before the
[117:02] algorithmic manipulation started
[117:03] steering us into believing pure nonsense
[117:06] because somebody who controlled these
[117:08] things decided it was useful for us to
[117:10] believe those things. So my guess is at
[117:13] the moment AI is enhancing our ability
[117:16] to see more clearly but that really
[117:18] depends on some kind of agreement to
[117:21] protect that capacity that I'm not aware
[117:24] of us having are you implying there that
[117:27] AI will protect us from AI i.e. the
[117:30] woman that got scammed in my audience,
[117:33] the platforms would have a tool built in
[117:34] which would be able to identify shortly
[117:37] that that is not me and the ad is been
[117:40] launched by someone in another country
[117:42] potentially and then also when she
[117:43] starts being asked for her credit card
[117:45] details in such a way on Telegram 10
[117:48] minutes later the system will able to
[117:49] understand there that this is probably a
[117:51] scam at that touch touch point too and
[117:53] it will also be the defense not just the
[117:55] offense.
[117:56] First thing uh question is Meta
[117:59] incentivized to solve this problem? Yes.
[118:03] Yes. And so Meta is probably actively
[118:06] working on AIs and again it's going to
[118:08] be a cat and mouse game like every abuse
[118:10] that happens out there. So I I think
[118:13] that the market will naturally respond
[118:15] to things like that in the same way that
[118:18] you know we installed antiviruses as you
[118:21] know annoying as they are. I think we'll
[118:23] install uh AIS on our computers that
[118:27] will allow us to at least help us kind
[118:29] of sort the the fake from from the
[118:31] truth. Well, but let's let's take the
[118:33] example you say. Is Meta incentivized to
[118:36] solve this problem? Superficially, it
[118:38] seems that it should be, but how many
[118:40] times in recent history have we watched
[118:42] a corporation cannibalize its own
[118:45] business over what at best is the
[118:47] bizarre desires of its shareholders,
[118:50] right? Why was X throwing off people
[118:53] with large accounts or Facebook or
[118:58] Google? It would seem that you would
[119:00] expect based on the market choosing
[119:03] search engines or social media sites,
[119:05] you would expect these companies to be
[119:08] absolutely mercenary and say, you know,
[119:10] if Alex Jones has a big audience, who
[119:12] are we to say? That's what I would have
[119:14] expected. Instead, you had these
[119:18] companies policing the morality of
[119:21] thought even though it reduced the size
[119:25] of the population using the platforms. I
[119:27] have a hard time explaining why that
[119:29] happened, but I have every reason to
[119:30] expect the same thing will happen with
[119:32] AI. What are you excited about with AI?
[119:34] What's your your optimistic take?
[119:37] Because at the start of this
[119:37] conversation, you said that there's
[119:39] infinite ways that it could improve our
[119:41] lives and there's 10 times more ways
[119:43] that it could hurt our lives. But let's
[119:45] investigate some of those ways that it
[119:46] could drastically improve our lives.
[119:48] There's a couple of different ways. One,
[119:50] we have, as we mentioned before, a der
[119:53] of competent teachers and professors.
[119:57] And that is a problem that will take
[119:59] three generations at least to solve if
[120:02] what we're going to do is start tomorrow
[120:04] and start educating people in the right
[120:06] way that would make them competent to
[120:07] stand at the front of a room and
[120:08] educate. But if we can augment that
[120:10] process, if we can leverage a tool like
[120:13] AI so that you know a small number of
[120:16] competent teachers can maybe reach a
[120:18] larger number of pupils, that's
[120:20] plausible I think. Second thing is we
[120:24] have a tremendous number of problems
[120:26] that are obstacles to us living well on
[120:30] this planet that AI might be able to
[120:33] manage that human intellect alone
[120:35] cannot. Right? Just in the same way that
[120:39] you know compute power can calculate
[120:41] things at a rate that human beings can't
[120:43] keep up and there are certain things you
[120:44] want calculated very well. There are
[120:46] also some reasoning problems. You could
[120:49] imagine that instead of having um static
[120:54] laws that govern behavior poorly because
[120:59] they get gamed that you could have a
[121:01] dynamic interaction. You could specify a
[121:05] an objective of something like a law and
[121:08] then you could monitor whether or not a
[121:10] particular intervention successfully
[121:13] moved you in the direction that you were
[121:14] hoping to go or did something
[121:16] paradoxical which happens all the time
[121:18] and you could have you could basically
[121:19] have governance that is targeted to
[121:22] navigation and prototyping rather than
[121:24] to specifying a blueprint for how we are
[121:27] to live. So we wouldn't need
[121:29] politicians. Um, at the moment we're
[121:32] stuck with, you know, constitutional
[121:35] protections that are as good as has been
[121:38] constructed and still inadequate to
[121:41] modern realities.
[121:43] Dan, what are you excited about with AI
[121:45] from an individual level, but also from
[121:47] a societal level? Yeah. Well, the big
[121:49] ones are healthcare and education. I
[121:51] mean, it's ridiculous that you uh are
[121:54] sitting there in pain, having had an
[121:56] MRI, and there just hasn't been someone
[121:58] to look at that MRI yet. and and tell
[122:00] you what to do. Um and that could easily
[122:03] be solved there's all sorts of
[122:04] healthcare issues where um and also not
[122:07] only that throughout the entire world
[122:09] there are places that just don't have
[122:10] general practitioners and they don't
[122:12] have you know medical advisers and and
[122:15] you know the breakthroughs in global
[122:16] healthcare will be phenomenal and the
[122:18] breakthroughs in global education could
[122:20] be transformational um on the planet. I
[122:23] I'm excited at an individual level that
[122:26] I think the industrial age created a
[122:28] bunch of jobs that are very dehumanizing
[122:30] and we've just kind of gotten used to
[122:32] them and put up with them. The idea that
[122:33] work should be repetitive and you know
[122:36] you just repeat the same loop over and
[122:38] over and over over again and over a
[122:40] 10-year period of time you might get you
[122:42] know graduated up one gear and all that
[122:44] kind of stuff. I don't think that's very
[122:46] human. Um the idea that you could be
[122:49] simultaneously writing a book, launching
[122:51] a business, running a team, launching a
[122:54] festival, having an event. Um that that
[122:57] that you could actually be doing this
[122:59] kind of like mini kingdom work where
[123:01] you've got this little, you know, uh
[123:04] ecosystem around you of fun things that
[123:06] you're involved in that is actually made
[123:08] possible for a vast majority of people
[123:10] if they embrace these kind of tools. um
[123:13] you can live an incredibly fulfilling
[123:15] and amazing and impactful existence or I
[123:19] know that I do as a result of having
[123:21] these tools in my life. Like I'm I'm
[123:23] doing things that I could have only
[123:24] dreamed about uh as a kid. And what
[123:26] would you say to entrepreneurs? I know
[123:28] you you work with thousands of
[123:29] entrepreneurs. What are you telling them
[123:31] in terms of their current businesses or
[123:33] business opportunities that you're
[123:34] foreseeing? So I think that small teams
[123:36] have infinite leverage now and that when
[123:39] you have a team of say five to 10 people
[123:43] who share an incredible passion for a
[123:45] meaningful problem in the world and they
[123:47] want to see that meaningful problem
[123:48] solved and they come together in the
[123:50] spirit of entrepreneurship to solve that
[123:52] problem. That little 5 to 10 person team
[123:56] armed with the technology that we now
[123:57] have available, you can have a a big
[124:01] impact. You can make a lot of money. You
[124:03] can have a lot of fun. you can solve
[124:04] meaningful problems in the world. You
[124:06] can scale solutions. You can probably do
[124:08] more in a three-year window than most
[124:10] people did in a 30-year career. Uh and
[124:13] then that little band of 5 to 10 people
[124:15] could either go together onto a new
[124:18] meaningful problem or they could disband
[124:20] and you know work on other meaningful
[124:22] problems with different teams. In such a
[124:25] world where you have this sort of
[124:26] infinite leverage but everyone else has
[124:29] access to the same infinite leverage.
[124:31] What becomes the USP? Going back to this
[124:33] idea of the moat, like what is the thing
[124:34] of value when we've all got access to
[124:36] $20 infinite leverage? Well, first of
[124:39] all, the first thing you need to do you
[124:41] need to understand is that this moment
[124:43] of time is the least competitive uh
[124:47] moment. Like if you understand how to
[124:49] use these tools, you can start making
[124:51] money tomorrow. like we, you know, I see
[124:53] countless examples of people making
[124:56] thousands of dollars with these hustles
[124:57] that I that I talked about or building
[125:00] businesses that generates millions of
[125:01] dollars in the first couple of months of
[125:03] existence. So, I would say start moving
[125:05] now. Start building things. So, it's an
[125:08] unprecedented time of of of wealth
[125:11] creation. Clearly at some point as the
[125:15] market gets more efficient as people
[125:17] more and more people understand how to
[125:18] use these tools um there's less
[125:22] potential for uh you know creating these
[125:25] massive businesses quickly and we've
[125:27] seen this like the dawn of the internet
[125:29] or dawn of the web you know it was a lot
[125:31] easier to create Facebook than it is now
[125:34] then we had mobile and for three four
[125:37] five years it was very easy to create
[125:40] massive businesses and then it became
[125:42] harder. Being just at the edge of what's
[125:44] possible is going to be very very
[125:46] important over the next couple years.
[125:48] And that's that gets me really excited
[125:49] because the entrepreneurs who are paying
[125:51] attention are going to are going to be
[125:53] having the most amount of fun, but
[125:55] they're also going to be able to make a
[125:56] lot of money. How many applications have
[125:59] been built on Replet to date? So, you
[126:02] know, I can talk about the millions of
[126:03] things that have been built since the
[126:05] since we started the company, but just
[126:07] since uh September when we launched
[126:10] Replent, there's been about 3 million
[126:12] applications built purely in natural
[126:16] language with no uh with no coding at
[126:20] all, purely natural language. uh of
[126:22] those I think uh 300 400,000 of them
[126:25] were deployed in a real um in the site
[126:30] was deployed and it is having people are
[126:33] using it some kind of business some kind
[126:35] of internal tool. I built one last night
[126:37] by the way an internal tool or I uh
[126:39] built an application to track um how my
[126:43] kids earn pocket money. Amazing. So, I
[126:46] just told it that I wanted to track the
[126:48] tasks that are happening around the
[126:49] house and put an assign a value to them
[126:51] and I want to be able to at the end of
[126:52] the week push a button and get a summary
[126:54] of how much to pay each child um for
[126:57] their pocket money. we are so screwed.
[127:02] And within 15 minutes, it had created
[127:04] this application and it was amazing.
[127:06] Like you could toggle between like
[127:08] here's the place where you have the kids
[127:10] and here's the weekly reports and here's
[127:12] the um how much per task and you can
[127:15] tick off the tasks or remove tasks or
[127:17] add tasks. So then I now have this
[127:19] application for which took 15 minutes of
[127:23] just talking about what I wanted and now
[127:25] I have an application to run the pocket
[127:27] money uh situation in the in the house.
[127:29] And this by the way it's something
[127:31] having run an IT agency years ago.
[127:34] That's something that we would have
[127:35] charged five to10,000 pounds to create
[127:38] or5 to 10,000 US dollars to to create
[127:42] and how much time probably talking
[127:44] something that would have been a three
[127:46] four week project and we're at the start
[127:49] of the S-curve now that you're
[127:50] describing and it's already if you if
[127:53] it's a $20 replet's roughly $20 a month
[127:55] 25 for the for the base case you did one
[127:58] day of usage let's say it's a dollar it
[128:01] cost you and it cost you minutes and a
[128:03] dollar now and we're at the start of the
[128:04] scurve and and you talk to it like
[128:06] you're chatting to a developer. So one
[128:08] of the things that slows down the
[128:10] development process is you have to send
[128:12] the information to a developer and they
[128:14] need to understand it and then they need
[128:15] to create something and then come back
[128:16] to you. This just happens in front of
[128:18] your eyes uh while you're watching it
[128:20] and it's actually showing you what's
[128:21] being built and it's it's really wild.
[128:25] This one change has transformed how my
[128:27] team and I move, train and think about
[128:29] our bodies. When Dr. Daniel Lieberman
[128:31] came on the DEO. He explained how modern
[128:34] shoes with their cushioning and support
[128:36] are making our feet weaker and less
[128:38] capable of doing what nature intended
[128:40] them to do. We've lost the natural
[128:42] strength and mobility in our feet and
[128:44] this is leading to issues like back pain
[128:46] and knee pain. I'd already purchased a
[128:49] pair of Viva barefoot shoes. So, I
[128:50] showed them to Daniel Lieberman and he
[128:52] told me that they were exactly the type
[128:54] of shoe that would help me restore
[128:55] natural foot movement and rebuild my
[128:57] strength. But I think it was
[128:58] plantficitis that I had where suddenly
[128:59] my feet started hurting all the time.
[129:01] And after that I decided to start
[129:02] strengthening my own foot by using the
[129:04] Vivo barefoot. And research from
[129:06] Liverpool University has backed this up.
[129:07] They've shown that wearing Vivo barefoot
[129:09] shoes for 6 months can increase foot
[129:11] strength by up to
[129:13] 60%. Visit vivo
[129:15] barefoot.com/doac and use code diary 20
[129:19] from my sponsor for 20% off. A strong
[129:21] body starts with strong feet. This has
[129:25] never been done before. A newsletter
[129:28] that is ran by 100 of the world's top
[129:31] CEOs. All the time people say to me,
[129:34] they say, "Can you mentor me? Can you
[129:35] get this person to mentor me? How do I
[129:37] find a mentor?" So, here is what we're
[129:39] going to do. You're going to send me a
[129:41] question. And the most popular question
[129:42] you send me, I'm going to text it to 100
[129:46] CEOs, some of which are the top CEOs in
[129:49] the world running a hundred billion
[129:50] dollar companies. And then I'm going to
[129:52] reply to you via email with how they
[129:54] answered that question. You might say,
[129:57] "How do you hold on to a relationship
[129:59] when you're building a startup? What is
[130:00] the most important thing if I've got an
[130:02] idea and don't know where to start?" We
[130:04] email it to the CEOs. They email back.
[130:05] We take the five, six top best answers.
[130:07] We email it to you. I was nervous
[130:09] because I thought the marketing might
[130:10] not match the reality. But then I I saw
[130:12] what the founders were replying with and
[130:14] their willingness to reply and I thought
[130:15] actually this is really good and all
[130:17] you've got to do is sign up completely
[130:20] free. I don't think we've spent a lot of
[130:23] time talking about autonomous weapons.
[130:25] This is the thing that really worries
[130:27] me. And the thing that worries people
[130:30] about AI is this idea is that it is uh
[130:34] this you know emergent system and
[130:36] there's no one thing behind it and it
[130:38] can be it can act in a way that's uh
[130:41] unpredictable and not really guided by
[130:43] humans. also think it's true of
[130:44] corporations of governments and so I
[130:47] think individual people uh can often
[130:50] have the best intentions but the
[130:52] collective can land on doing things in a
[130:56] way that's harmful or morally irrepant
[131:00] and I think um we talked about China
[131:03] versus the US and that creates a certain
[131:06] race dynamics where um they're both
[131:09] incentivized to cut corners and
[131:11] potentially do do harmful things and in
[131:14] the world of geopolitics
[131:17] um and wars, you know, what really
[131:20] scares me is is autonomous weapons. And
[131:23] why does it scare you? Because
[131:28] uh you know you can imagine
[131:32] uh autonomous drones being trained on
[131:36] someone's face and you can send a a
[131:39] swarm of of drones and they can be this
[131:43] um sort of autonomous killing
[131:45] assassination machine and it can sort of
[131:48] uh function as a you know country verse
[131:51] country technology in in the world of
[131:54] for which is still crazy but it can also
[131:58] become a tool for governments to
[132:01] subjugate the citizens and and people
[132:04] think we're we're safe in the west but I
[132:08] think the experience with co showed that
[132:11] even the systems in the west can very
[132:16] quickly become draconian. Yeah.
[132:18] Apparently, I've heard in um Iran that
[132:22] uh they have facial recognition cameras
[132:25] that detect where the women are wearing
[132:27] hijabs in their own cars and it
[132:30] automatically detains the car. If you're
[132:34] driving and you're not wearing a hijab
[132:36] and if you're certainly if you're
[132:38] walking down the street, it just picks
[132:40] that up and immediately you're you're in
[132:42] trouble. uh you can like it acts as a
[132:45] police officer and a judge and you know
[132:49] a law m lawmaker it's the judge jury and
[132:52] executioner essentially and it's just
[132:54] happens instantaneously what happened in
[132:56] Canada with the truckers uh uh sort of
[133:00] protest where they froze their bank
[133:02] account by virtue of just being there
[133:03] just by being in that location and just
[133:06] to confirm that Iran has implemented a
[133:08] comprehensive surveillance system to
[133:09] enforce its mandatory hijab laws
[133:12] utilizing various ious technologies, one
[133:13] of which is cameras and facial
[133:15] recognition. So they've put cameras in
[133:17] public spaces to identify women who are
[133:19] not adearing to the hijab dress code.
[133:22] Yeah. And just on that, London has just
[133:25] put those face cam facial recognition
[133:27] systems into London and also all
[133:29] throughout Wales. um and they're being
[133:31] rolled out at speed
[133:34] and like all you would need is a change
[133:37] of government that wanted to implement
[133:39] something similar and all the base layer
[133:42] technologies already in there. It gets a
[133:43] little bit worse in Iran because they
[133:45] have this new app called the Nazar app
[133:46] where the government has introduced the
[133:48] Nazar mobile application which allows
[133:49] you as a citizen to report another
[133:51] citizen who is not wearing their hijab
[133:53] and it it logs their location, their
[133:55] time um when they weren't wearing it and
[133:57] the vehicle license plate with the
[133:59] crowdsource data. It can then go after
[134:01] that individual. I would also just point
[134:04] out that I think we're not being
[134:05] imaginative enough. I agree with you. I
[134:07] have the same concern about these
[134:09] autonomous weapons, but I also think
[134:11] this doesn't have to occur in the
[134:13] context of war or even governmental
[134:16] oppression that it is perfectly
[134:17] conceivable that effectively this allows
[134:20] this drops the price of a an
[134:24] undetectable or an unprosecutable crime.
[134:27] And maybe economic moes return in the
[134:30] form of people taking out their
[134:32] competitors or anybody who attempts to
[134:33] compete with them using an autonomous
[134:35] drone that can't be traced back to them.
[134:38] You know, that follows facial
[134:39] recognition. And you know, you don't
[134:41] have to kill very many people for others
[134:42] to get the message that uh this is a a
[134:45] zone that uh you shouldn't mess around
[134:47] in. So, I could imagine, you know,
[134:50] effectively a new high-tech organized
[134:53] crime that uh protects rackets and makes
[134:58] tons of money and subjugates people who
[135:00] haven't done anything wrong. I had
[135:02] Mustafa Sullivan on the podcast in 2023
[135:04] when this all of this stuff started
[135:06] kicking off and he is the CEO of
[135:08] Microsoft AI. You're familiar with
[135:10] Mustafa? Of course. Yeah. Um and he one
[135:12] of the things he said to me at the time
[135:13] was one of my fears is a tiny group of
[135:15] people who wish to cause harm are going
[135:17] to have access to tools that can
[135:19] instantly destabilize our world. That's
[135:21] the challenge. How to stop something
[135:23] that can cause harm or potentially kill.
[135:25] That's where we need containment. And it
[135:27] sounds a little bit like what you're
[135:28] saying amad that we will now have these
[135:31] these tools. you were talking in the
[135:32] context of the military, but as Brett
[135:35] said there, even smaller groups of
[135:36] people that might have been, I don't
[135:38] know, cartels or gangs can do similar
[135:41] harm. And at the moment, in terms of
[135:43] autonomous weapons, both the US and
[135:45] China are investing heavily in AI
[135:47] powered weapons, autonomous drones, and
[135:48] cyber warfare because they're scared of
[135:50] the other one getting it first. And we
[135:53] we talked about how much of our lives
[135:54] run on the internet, but cyber weapons
[135:56] and cyber AI agents that could be
[135:59] deployed to take down China's X, Y, or
[136:02] Zed or vice versa are real concern.
[136:05] Yeah. Yeah. I I think all of that is is
[136:10] um is a real concern. You know, unlike
[136:12] Mustafa, I I don't think containment is
[136:14] is possible. Part of the reason why this
[136:18] game theoretic system uh of competition
[136:21] between the US, China, corporations,
[136:26] individuals makes it so that this this
[136:28] technology is already, you know, is
[136:31] already out and really hard to put it
[136:33] back in the in the bag. I did ask him
[136:36] this question and I remember the answer
[136:37] because it was such a stark moment for
[136:38] me. I said to Mustafa, "Do you think
[136:40] it's possible to contain it?" And he
[136:41] replied, "We must." So I asked him
[136:43] again. I said, "Do you think it's
[136:44] possible to contain it?" and he replied
[136:45] we must and I asked him again I said do
[136:47] you think it's possible we must so the
[136:49] problem with that uh uh chain of
[136:51] thinking is that it might lead to an
[136:53] oppressive system
[136:55] uh there is uh one of the say doomers or
[136:58] philosophers of of AI which I I respect
[137:01] his work his name is Nick Bostonramm and
[137:03] he he uh he's he was trying to think of
[137:08] ways in which we can contain AI and the
[137:13] thing that he came up with is perhaps
[137:16] more oppressive than something that the
[137:18] AI would come up with is like total
[137:20] surveillance state. You need total
[137:22] surveillance on compute on people's
[137:25] computers on people's ideas to not
[137:27] invent AI or AGI. It's like taking the
[137:30] guns or something or Right. Exactly. I
[137:32] mean there's always there's always this
[137:34] problem of containing any sort of
[137:36] technology is that you do need um
[137:39] oppression and draconian policies to do
[137:42] that. Are you scared of anything else or
[137:44] concerned about anything else as it
[137:45] relates to AI outside of autonomous
[137:47] weapons? You know, we talked at the
[137:49] about
[137:50] the birthight crisis and I think a more
[137:54] generalized problem there is creating
[137:58] virtualized environments
[138:01] uh via VR where everyone is living in
[138:04] their own created universe and uh it's
[138:09] so enticing and even create simulates
[138:12] work and simulates struggle uh such that
[138:15] you don't really need to leave this this
[138:17] world and so every one of us will be
[138:20] solopcystistic, you know, similar to the
[138:21] Matrix. Ready Player One. Ready Player
[138:24] One. We're all kind of uh plug even
[138:26] worse than Ready Player One. At least
[138:28] that's a massivelyworked
[138:30] environment. I'm talking about AI
[138:32] simulating everything uh for us and
[138:36] therefore you're literally in the
[138:37] matrix. You know, maybe that this is I
[138:40] was about I had that same thought. I've
[138:42] enjoyed this great simulation. Yes. and
[138:45] and and so I mean are you familiar with
[138:48] the Fermy's paradox? No, I'm not. So
[138:50] Fermy's paradox is um the question uh
[138:53] the you know professor his name is uh
[138:56] Fermy he asked the question uh if the
[139:00] universe is is that vast then where are
[139:03] the aliens? the fact that humans exist,
[139:07] you can deduce that other civilizations
[139:10] exist. And if they do exist, then why
[139:14] don't we see them? And then that spurred
[139:17] a bunch of Fermy solutions. So there's
[139:20] uh I don't know, you can find hundreds
[139:21] of solutions on the internet. One of
[139:24] them is the uh sort of house cat on a
[139:26] thought experiment where actually aliens
[139:29] exist, but they kind of put us in an
[139:31] environment like the Amish in a certain
[139:34] time and do not expose us to what's
[139:37] going on out there. So they we're pets.
[139:39] Maybe they're watching us and kind of
[139:40] enjoying uh what we're doing, stopping
[139:42] us from hurting ourselves, stopping us
[139:44] from hurting ourselves. There are so
[139:45] many things, but one of the things that
[139:48] I think is potentially a solution to the
[139:51] phrases paradox and one of the saddest
[139:54] outcomes is that civilizations progress
[139:58] until they invent technology that will
[140:01] lock us into infinite pleasure and
[140:03] infinite simulation such that we we
[140:07] don't have the motivation to go into
[140:11] space to seek out the explor
[140:14] exploration, potentially other alien
[140:17] civilizations. And perhaps that is a
[140:20] determined outcome of humanity or like a
[140:24] highly likely outcome of any species
[140:27] like humanity. We like pleasure.
[140:29] Pleasure and pain is the main
[140:30] motivators. And so if you create an
[140:33] infinite pleasure machine, does that
[140:35] mean that we're just at home in our VR
[140:38] environment with everything taking care
[140:40] for us and literally like the matrix and
[140:42] the world the real world would suck in
[140:44] such a scenario? Yes. Be terrible. I
[140:46] mean the other simpler explanation of
[140:48] the Fermy paradox is that you generate
[140:51] sufficient technology that you can end
[140:53] your species and it's only a matter of
[140:55] time from that point which you know we
[140:57] can have that discussion about nuclear
[140:59] weapons. We can have it about AI, but
[141:02] does some technology, if if we stay on
[141:04] that escalator, does some technology
[141:06] that we generate ultimately whatever
[141:08] allows you to get off the planet allows
[141:10] you to blow up the planet? There you go.
[141:12] I want to get everyone's closing
[141:13] thoughts and closing remarks. And
[141:16] hopefully in your closing remarks, you
[141:17] can capture something actionable for the
[141:20] individual that's listening to this now
[141:22] on their commute to work or the single
[141:24] mother, the average person who maybe
[141:25] isn't as technologically advanced as
[141:27] many of us at this table, but is trying
[141:29] to navigate through this to figure out
[141:31] how to live a good life over the next
[141:33] 10, 20, 30 years. Yeah. Take as long as
[141:37] you need. I think we live in the most uh
[141:40] interesting time in human history. So
[141:43] for the single mother that's listening,
[141:45] for someone who wouldn't be the
[141:46] stereotype of a tech row, don't assume
[141:49] that you can't do this stuff. It's never
[141:52] been more accessible today within your
[141:54] work. You can be an entrepreneur. You
[141:56] don't have to take massive risk to go
[141:58] create a business um by you quit your
[142:02] job and go create a business. There are
[142:04] countless examples. We uh we have a user
[142:06] who's a product manager at a larger real
[142:09] estate business and he built something
[142:12] that created 10% lift in conversion
[142:15] rates which generated millions and
[142:17] millions of dollars of that business and
[142:19] that person became celebrity at that
[142:21] company and became someone who is
[142:23] lifting everyone else up and teaching
[142:25] them how to use these tools and
[142:27] obviously that that is like a really
[142:29] great for for anyone's career and you're
[142:31] going to get a promotion and your
[142:33] example of building a piece of software
[142:36] for your family for your kids to to
[142:39] improve and and to learn more to be
[142:42] better kids uh as an example of being
[142:44] entrepreneur in your family. So I really
[142:48] want people to break away from this
[142:51] concept of entrepreneurship being this
[142:54] is your podcast a diary of a CEO. You
[142:56] started this podcast by talking to CEOs
[142:59] I assume right and over time uh it
[143:02] changed to everyone can be a CEO
[143:05] everyone is some kind of CEO in their
[143:07] life and so uh I think that we
[143:12] have unprecedented access to tools for
[143:15] that vision to actually come to reality.
[143:19] Well, it is obviously a moment of a kind
[143:22] of human phase transition. Something
[143:25] that I believe will be the equal of a
[143:28] discovery of farming or writing or
[143:34] electricity. And the darkness that I
[143:38] think is valid in looking at all of the
[143:40] possible outcomes of this scenario is
[143:43] actually potentially part of a different
[143:45] story as well. In evolutionary biology,
[143:47] we talk about an adaptive landscape in
[143:50] which a niche is represented as a peak
[143:54] and a higher niche, a better niche is
[143:57] represented as a higher peak. But to get
[143:59] from the lower niche to the higher
[144:00] niche, you have to cross through what we
[144:02] call an adaptive valley. And there's no
[144:04] guarantee that you make it through the
[144:05] adaptive valley. And in fact, the
[144:08] drawing that we put on the board, I
[144:09] think, is overly hopeful because it
[144:11] makes it in two dimensions. It looks
[144:13] like you know exactly where to go to
[144:15] climb that next peak. And in fact, it's
[144:17] more like the peaks are islands in an
[144:20] archipelago that is in fog where you
[144:23] can't figure out what direction that
[144:25] peak is and you have to reason out it's
[144:27] probably that way and you hope not to
[144:29] miss it by a few degrees. But in any
[144:32] case, that darkness is exactly what you
[144:36] would expect if we were about to
[144:38] discover a better phase for humans. And
[144:42] I think we should be very deliberate
[144:43] about it this time. I think we should
[144:45] think carefully about how it is that we
[144:47] do not allow the combination of this
[144:50] brand new extremely powerful technology
[144:53] and market forces to turn this into some
[144:57] new kind of enslavement. And I don't
[145:00] think it has to be. I think the
[145:02] potential here does allow us to refactor
[145:05] just about everything. Maybe we have
[145:07] finally arrived at the place where
[145:10] mundane work doesn't need to exist
[145:12] anymore and the pursuit of meaning can
[145:14] replace it. But that's not going to
[145:16] happen automatically if we don't figure
[145:18] out how to make it happen. And I hope
[145:21] that we can recognize that the peril of
[145:24] this moment is best utilized if it
[145:27] motivates us to confront that question
[145:30] directly.
[145:32] Each one of us has two parents, four
[145:34] grandparents, eight great-grandparents,
[145:37] 16, 32, 64. You've got this inc like
[145:41] this long line of ancestors who all had
[145:44] to meet each other. They all had to
[145:45] survive wars. They all had to survive
[145:48] illness and disease. Everything had to
[145:51] happen for us. One, each individual, all
[145:53] of this this stuff had to happen for us
[145:55] to get here. And if we think about all
[145:57] the people in those thousands and
[145:59] thousands of people, every single one of
[146:01] them would trade places in a heartbeat
[146:03] if they had the opportunity to be alive
[146:05] at this particular moment. They would
[146:07] say that their life was struggle,
[146:11] disease, that their life was a lot of
[146:14] mundane and meaningless work. It was
[146:16] dangerous. You know, every single one of
[146:18] us has probably got ancestors that were
[146:20] enslaved, probably got ancestors that
[146:24] died too young, uh, probably got
[146:27] ancestors that worked horrific
[146:29] conditions. We all have that. And they
[146:32] would all just look at this moment and
[146:34] say, "Wow." So, are you telling me that
[146:36] you have the ability to solve meaningful
[146:38] problems, to come up with adventures, to
[146:41] travel the world, to pick the brains of
[146:44] anyone on the planet that you want to
[146:45] pick the brains of? You can just listen
[146:47] to a podcast. You can just watch a
[146:49] video. You can talk to an AI. Like, are
[146:52] you telling me that you're alive at this
[146:53] particular moment? Please make the most
[146:56] of that. Like, do something with that.
[146:59] You know, you can sit around
[147:00] pontificating about society and how
[147:03] society might work. But ultimately, it
[147:05] all boils down to what you do with this
[147:07] moment. and solving meaningful problems,
[147:10] being brave, having fun, making your
[147:14] little dent in the universe. You know,
[147:16] that's that's what it's all about. And I
[147:18] feel like there's an obligation to your
[147:20] ancestors to make the most of the
[147:22] moment.
[147:24] Thank you so much to everybody for being
[147:26] here. I I've learned a lot and I've
[147:28] developed my thinking, which is much the
[147:30] reason why I wanted to bring us all
[147:31] together because I know you all have
[147:32] different experiences, different
[147:34] backgrounds in education. and you're
[147:35] doing different things, but together it
[147:37] helps me sort of pass through all of
[147:38] these ideas to figure out where I land.
[147:40] And I I I ask a lot of questions, but I
[147:43] am actually a believer in humans. I'm I
[147:46] I was thinking about this a second ago.
[147:48] I was thinking, do I am I optimistic
[147:50] about humans ability to navigate this
[147:52] just because I have no other choice?
[147:54] Because as you said, the alternative
[147:56] actually isn't worth thinking about. And
[147:57] so I do have a optimism towards how I
[148:00] think we're going to navigate this in
[148:02] part because we're having these kinds of
[148:03] conversations and we in history haven't
[148:06] always had them at the birth of a new
[148:08] revolution when we think about social
[148:09] media and the implications that had.
[148:10] We're playing catchup with the with the
[148:12] downstream
[148:14] consequences. And I am hopeful. Maybe
[148:17] that's the entrepreneur in me. I'm
[148:18] excited. Maybe that's also the
[148:20] entrepreneur in me. But at the same
[148:22] time, to many of the points Brett's
[148:23] raised and Amjud's raised and Dan's
[148:25] raised, there are serious considerations
[148:27] as we swim from one island to another.
[148:29] And because of the speed and scale of
[148:32] this transformation that Brett
[148:33] highlights and you look at the stats of
[148:35] the growth of this technology and how
[148:37] it's spreading like wildfire and how
[148:38] once I tried Replet, I walked straight
[148:41] out and I told Cozy immediately I was
[148:42] like, "Cussie, try this." And she was on
[148:44] it and she was hooked. And then I called
[148:45] my girlfriend in Bali who's the breath
[148:47] work practitioner and I was like, "Type
[148:48] this into your browser. R E P L I T."
[148:51] And then she's making these breath work
[148:53] schedules with all of her clients
[148:55] information ahead of the retreat she's
[148:56] about to do. It's spreading like
[148:58] wildfire because we're internet native.
[149:00] We were native to this technology. So
[149:02] it's not a new technology. It's
[149:03] something on top of something that's
[149:04] intuitive to us. So that transition, as
[149:07] Brett describes it, from one peak to the
[149:08] other or one island to another, I think
[149:10] is going to be incredibly destabilizing.
[149:12] And I've having interviewed so many
[149:14] leaders in this space from Reed Hoffman
[149:15] who's the founder of LinkedIn to the CEO
[149:17] of Google to Mustafa who I mentioned
[149:20] they don't agree on much but the thing
[149:22] that they all agree on and that Sam
[149:23] Alman agrees on is that the long-term
[149:26] future the long-term way that our
[149:28] society functions is radically
[149:29] different. People squab squabble over
[149:31] the short term. They sometimes even
[149:33] squabble over the midterm or the
[149:34] timeline but they all agree that the
[149:37] future is going to look completely
[149:38] different. Amjud, thank you for doing
[149:40] what you're doing. you're what we didn't
[149:42] get to spend a lot of time on it today
[149:43] and this is typically what I do here but
[149:45] your story is incredibly inspiring
[149:48] incredibly inspiring from where you came
[149:50] from what you've done what you're
[149:51] building and you are democratizing and
[149:53] creating a level of playing field for
[149:55] entrepreneurs in Bangladesh to Cape Town
[149:58] to San Francisco to be able to turn
[150:00] their ideas into reality and I do think
[150:02] just on the surface that that's such a
[150:04] wonderful thing that you know I was born
[150:06] in Botswana in in Africa and that I
[150:08] could have the same access to turn my
[150:11] imagination into something to change my
[150:13] life because of the work that you're
[150:14] doing at Replet. And I highly recommend
[150:16] everybody go check it out. You you
[150:17] probably won't sleep that night because
[150:19] it's so it's so for someone like me it
[150:21] was so addictive to get to be able to do
[150:23] that because it's been the barrier to
[150:24] creation my whole life. I've always had
[150:25] to call someone to build something. Dan,
[150:28] thank you again so much because you
[150:29] represent the voice of entrepreneurs and
[150:30] you've really become a titan as a a
[150:31] thought leader for entrepreneurs in the
[150:32] UK and that perspective that balance is
[150:35] incredibly important. So, I really
[150:36] really appreciate you being here as
[150:38] always and you're a huge fan favorite of
[150:40] our show and Brett, thank you a
[150:41] gazillion times over for being a a human
[150:45] lens on complicated challenges and you
[150:48] do it with a fearlessness that I think
[150:50] is imperative for us finding the truth
[150:52] in these kind of situations where some
[150:54] of us can run off with optimism and we
[150:56] can be hurtling towards the mouse trap
[150:58] because we love cheese and I think
[151:00] you're an important
[151:01] counterbalance and voice in the world at
[151:04] this time. So, thank all of you for
[151:05] being here. I really really appreciate
[151:06] it and um we shall
[151:11] see. These things live forever.
[151:17] So, this has always blown my mind a
[151:19] little bit. 53% of you that listen to
[151:22] this show regularly haven't yet
[151:23] subscribed to the show. So, could I ask
[151:25] you for a favor? If you like the show
[151:26] and you like what we do here and you
[151:28] want to support us, the free simple way
[151:29] that you can do just that is by hitting
[151:31] the subscribe button. And my commitment
[151:33] to you is if you do that, then I'll do
[151:35] everything in my power, me and my team,
[151:36] to make sure that this show is better
[151:38] for you every single week. We'll listen
[151:40] to your feedback. We'll find the guests
[151:41] that you want me to speak to, and we'll
[151:43] continue to do what we do. Thank you so
[151:45] much.
[151:48] [Music]
[152:05] [Music]
