Fei-Fei Li Rejects 'Mother AI' Concept to Safety-Proof Superintelligence
Ai4
AI pioneer Fei-Fei Li argues that humanity must maintain strict human-centered governance over AI systems rather than surrendering dignity and agency to a benevolent, superintelligent safety architecture.
## Background
This content was published on YouTube by the channel Ai4 on August 24, 2025. According to the page description, it features a forward-looking fireside chat with Dr. Fei-Fei Li, Co-Founder & CEO of World Labs, exploring spatial reasoning, AI's understanding of the physical world, and human-centered design across technical, ethical, and societal dimensions.
## TL;DR
In this fireside chat, Dr. Fei-Fei Li reframes AI not as a competitive sports game, but as a foundational civilizational technology akin to electricity. She emphasizes the underappreciated role of AI in transforming education, advocating that tools should empower human agency through Socratic inquiry rather than being viewed solely through the lens of cheating. Li details her vision at World Labs regarding spatial intelligence and 3D world generation, which she views as a historic evolutionary leap into multi-universe interactions. Additionally, she rejects Geoffrey Hinton's 'mother AI' concept, asserting that humanity must retain its agency, while addressing key risks such as energy consumption, democratic integrity, and ensuring shared economic prosperity.
## Atmosphere & style
The discussion carries a warm, intellectually engaging, and deeply pragmatic atmosphere as CNN's Matt Egan guides Dr. Fei-Fei Li through high-level topics. Li brings a thoughtful, measured perspective—occupying what she calls the 'boring middle' between apocalyptic doomsday scenarios and delusional optimism. The dynamic between host and speaker is respectful yet candid, blending visionary scientific insights with grounded reflections on ethics, human dignity, and societal responsibility.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=4s) Artificial intelligence should be viewed as a foundational, long-term civilizational technology akin to electricity, rather than a short-term competition or sports game.
- [02:02](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=122s) The field of AI originated purely from human scientific curiosity, driven by pioneers like Alan Turing and John McCarthy asking if machines can think.
- [02:02](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=122s) Education is the most under-resourced and critically underappreciated domain in AI, encompassing K-12, higher education, and lifelong workforce re-skilling.
- [05:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=304s) Educational frameworks should reject the narrow lens of treating AI as a tool for cheating, reframing it as a mechanism for student empowerment.
- [05:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=304s) Prompt engineering mirrors the Socratic method of learning through structured questioning, where AI output serves as a baseline for students to build upon.
- [07:55](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=475s) There is no fundamental distinction between AI and AGI; AI was conceived from its beginning by its founders as general machine intelligence.
- [07:55](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=475s) Human-Centered AI centers on the core principle that technology must be developed, governed, and deployed to augment and elevate human condition and agency.
- [10:48](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=648s) Spatial intelligence—the ability of AI models to reason about, generate, and interact with 3D physical and digital worlds—forms the foundation for World Labs and advanced robotics.
- [15:43](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=943s) Advances in spatial AI enable humanity to transition from living strictly within a single physical universe to creating and interacting across multiple 3D digital worlds.
- [15:43](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=943s) AI's societal impact should be evaluated from a pragmatic 'boring middle' perspective, rejecting both apocalyptic doomsaying and delusional utopianism.
- [21:33](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1293s) Automation should target dangerous, tedious, or uncreative tasks to expand overall economic and creative potential without replacing core human capabilities.
- [22:06](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1326s) The proposal of a protective 'mother AI' is an incorrect paradigm; society must never surrender its dignity or agency to autonomous machines.
- [24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) Frontier AI model training demands nation-state scale energy consumption, reaching power levels comparable to the electricity grid of a medium-sized European country.
- [24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) Increases in overall AI productivity do not automatically yield shared economic prosperity without intentional societal alignment and planning.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=4s) AI as a Civilizational Technology
CNN Senior Reporter Matt Egan opens the session by thanking Dr. Fei-Fei Li for sharing her insights, calling her "literally one of the most qualified people on the planet" to discuss the ongoing AI revolution. Dr. Li warmly accepts the welcome, thanking Matt and event host Ai4, while jokingly noting that as a professor she naturally dreads early morning classes that start with an "A."
Matt kicks off the discussion by asking Dr. Li to contextualize where humanity currently stands in the AI revolution using a classic analogy: if this were a nine-inning baseball game, what inning would we be playing? Dr. Li politely rejects the premise of the comparison. She explains that AI is not a sports competition where there are winners and losers at the end; rather, humanity as a whole needs to win together. Instead of a game, she views AI as a shared journey or a civilizational revolution. Drawing a historical parallel to the rise of electricity 150 to 180 years ago, she asks what answer someone in 1870 could have possibly given if asked what "inning" electricity was in. Just as electricity continues to power human civilization over 150 years later, AI is at the very beginning of its own trajectory as a foundational civilizational technology.
<Image alt="Dr. Fei-Fei Li speaking at an AI conference" caption="Dr. Fei-Fei Li on stage discussing AI's civilizational impact" src="image_agent_tag_15834982630954969526"/>
---
[02:02](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=122s) The Scientific Origins and Underappreciated Role of AI in Education
When asked to identify the most underappreciated aspect of artificial intelligence, Dr. Li begins by reflecting on the field's roots. AI started purely as a scientific endeavor driven by human curiosity. Daring pioneers—from Alan Turing to AI founding fathers like her Stanford colleague John McCarthy—dared to ask the fundamental question, "Can machines think?" Over time, that curiosity expanded into massive scientific breakthroughs, including deep learning, data-driven machine learning, and generative AI, which are now actively altering the course of human history.
Dr. Li highlights that while AI dominates media coverage, congressional hearings in Washington D.C., and corporate boardrooms, the most under-resourced and critically underappreciated topic is **AI and education**. This spans K-12 schooling, personal education, and continuous lifelong re-skilling in the workforce. She emphasizes that AI should never replace human agency; rather, leveraging AI to transform education and empower teachers and students represents this decade's most important challenge and opportunity.
- **Foundational question of AI:** Dared by pioneers like Alan Turing and John McCarthy: *"Can machines think?"*
- **Scope of AI education:** Encompasses personal learning, K-12 systems, higher education, and corporate re-skilling.
- **Core goal:** Elevating human agency and capabilities rather than replacing human effort.
---
[05:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=304s) Reframing AI in the Classroom: From Cheating to Empowerment
Matt asks whether current educational systems are taking the right approach to AI and how prompt engineering fits into learning. Dr. Li responds candidly, sharing that just the day prior she met with senior congressional staffers from both parties at Stanford, and it became clear that society lacks a unified strategy for AI-powered learning. Because the technology is so new, the knee-jerk reaction around AI in education has been dominated by doomsday headlines and a narrow focus on one negative word: **cheating**.
Dr. Li argues that "cheating" is completely the wrong lens. The defining word should be **empowerment**. Addressing prompt engineering, she points out that humanity's very first "prompter" was Socrates over 2,000 years ago, who taught people that learning happens by asking the right questions. AI acts as a powerful tool that supercharges this Socratic process. Instead of warning students not to cheat, educators should frame AI output as a baseline—whether for a math problem, science query, or essay—and challenge students to use their unique human agency and intelligence to build beyond that baseline.
- **Dominant narrative error:** Treating AI in schools purely as a tool for cheating rather than learning enhancement.
- **Historical parallel:** Socrates as the original prompter who established learning through structured inquiry and questioning.
- **Pedagogical framework:** Treating AI outputs as the initial baseline and focusing student effort on higher-order critical thinking and agency.
---
[07:55](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=475s) Clarifying AGI and Defining Human-Centered AI
Addressing past media reports that she has pushed back on the concept of Artificial General Intelligence (AGI), Dr. Li clarifies her position. She notes that she has never claimed AGI is bad; rather, as a scientist, she sees no fundamental distinction between "AI" and "AGI." The founding fathers of AI—Turing, McCarthy, Marvin Minsky—never aimed to build narrow, non-general machines; they set out with the audacious vision of creating machines that think. Thus, AI and AGI are essentially synonymous.
Dr. Li goes on to explain why she founded Stanford's Human-Centered AI Institute (HAI) and authored a seminal 2018 *New York Times* op-ed advocating for this framework. Because AI is a civilizational technology impacting jobs, businesses, and child development, human benevolence must remain at the core. Technology is created by people, used by people, and must be governed by people. Human-Centered AI is guided by the principle of responsibly using technology to augment, enhance, and elevate the human condition rather than placing technology itself at the center.
---
[10:48](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=648s) World Labs, Spatial Intelligence, and the Multiverse
Matt pivots to World Labs—the startup co-founded by Dr. Li—and asks about her intriguing vision that AI will allow humans to generate infinite universes and live in a multiverse. Dr. Li grounds this concept through a concrete walkthrough of a future doctor's day, demonstrating how digital and physical worlds will seamlessly blend.
```
[ Physical World Reality ]
│
┌───────────────┼───────────────┐
▼ ▼ ▼
[ Precision ] [ Automated ] [ Immersive ]
[ AR Surgery] [ Autonomous] [ 3D Social ]
[ Co-Pilot ] [ Transport ] [ Commerce ]
│ │ │
└───────────────┼───────────────┘
│
[ Spatial Intelligence ]
[ (3D World Generation) ]
```
In this envisioned future, a surgeon wears AR devices or interacts with a digital co-pilot to perform life-saving surgery with microscopic precision beyond human vision. After work, they ride home in an autonomous car operating within a digital mapping universe. Back home, as a parent with no time to shop, they enter a 3D e-commerce world to try on outfits with their children, later transitioning to an interactive, immersive movie experience. To end the day, they join a transatlantic digital living room where European grandparents can join in real-time to sing happy birthday.
All these experiences rely on **spatial intelligence**—the capability of AI models to create, reason about, understand, and interact with generated 3D worlds. World Labs is building this fundamental generative technology from the ground up, which also serves as the foundational backbone for future robotics and autonomous physical agents.
<Image alt="World Labs Spatial Intelligence conceptual graphic" caption="Spatial intelligence powers 3D world generation and physical reasoning" src="image_agent_tag_15834982630954969939"/>
---
[15:43](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=943s) A Historical Evolutionary Leap and the Mid-Ground Perspective
Reflecting on the timeline for this vision, Dr. Li notes that technology is rapidly transitioning from academic research inside the "ivory tower" to commercial products. She highlights a profound evolutionary fact: for nearly one billion years, all living organisms on Earth existed within a single 3D physical universe. Through spatial AI and digital world generation, humanity is poised to become the very first species in history capable of creating and experiencing multiple 3D universes, unlocking unbounded creative and practical possibilities.
When asked about her outlook on AI's broader impact on medicine, science, and society, Dr. Li clarifies her position amid today's polarized rhetoric. She rejects both doomsaying and "delusional utopianism," positioning herself squarely in the pragmatic, "boring middle." She stresses that every human tool—from fire and the stone axe to modern AI—is an inherently double-edged sword that must be managed responsibly.
- **Evolutionary milestone:** Transitioning from 1 billion years of single-universe existence to multi-universe 3D interaction.
- **Positioning on AI risk:** Occupying the "boring middle" between apocalyptic doomsday rhetoric and delusional utopian optimism.
- **Exciting focus areas:** Overhauling standardized, memory-based education systems; supercharging human creativity across scientific discovery, art, and storytelling; and providing spatial foundations for advanced manufacturing and robotics.
[21:33](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1293s) Increasing the Global Pie and Redefining Human Work
Fei-Fei Li discusses the division between human labor and automated machine tasks, arguing that society shouldn't use technology to take over jobs that humans do best. Instead, machines ought to be deployed for roles that pose risks to human safety, lack human creativity, or are repetitive and tedious. By appropriately assigning these tasks to automation, society can unlock opportunities to expand overall economic and creative potential—effectively "increasing the global pie." She emphasizes that these possibilities are what truly excite her about the future of AI.
- Automation should focus on harmful, uncreative, or repetitive tasks rather than replacing core human capabilities.
- The overarching goal of adopting AI should be to enlarge the global economic and creative output.
[22:06](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1326s) Pushing Back on "Mother AI" and Demanding Human-Centered Agency
Matt Egan brings up a statement made on the same stage the previous day by Geoffrey Hinton—often hailed as the "Godfather of AI"—who suggested that building a protective "mother AI" might be humanity's single chance to prevent superintelligent systems from controlling or wiping out humankind. While Li expresses deep personal respect for Hinton as a friend and longtime mentor of over two to three decades, she respectfully pushes back against his premise. Framing the solution around a benevolent, superintelligent "mother AI" with maternal instincts is the wrong paradigm.
Instead, Li insists that the focus must remain squarely on human-centered AI. She argues that at no point should an individual, community, or society be expected or asked to relinquish their dignity or agency simply because a tool is powerful. Speaking from her perspective as an inventor, educator, and mother herself, she stresses that it is humanity's collective responsibility at every level to create, govern, and deploy technology in a manner that preserves human control and self-determination.
- Geoffrey Hinton previously proposed a "mother AI" concept to manage superintelligent systems and avoid human extinction.
- Li rejects the "mother AI" framing, asserting that humanity must never surrender its dignity or agency to a machine.
- Responsibility for governing AI ethically rests entirely on humans at every societal level.
[24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) Real-World AI Risks: Democracy, Labor, Energy, and Shared Prosperity
When asked about her biggest concerns regarding superintelligent AI systems, Li notes that while she often highlights the positive potentials of technology, humanity has a historical "dark underbelly" capable of using tools—from the primitive stone axe and fire to cutting-edge AI—to inflict harm. She acknowledges several real-world issues that keep her awake at night:
1. **Information and Democracy:** AI systems can be manipulated to distort public information and threaten the stability of democratic institutions.
2. **Labor Landscape Shifts:** Software engineering in Silicon Valley is already undergoing seismic changes. Rather than making engineers obsolete, AI is superpowering top talent; however, companies and workers face massive friction around reskilling and human capital investment.
3. **Nation-State Scale Energy Consumption:** Recounting a conversation with the CFO of OpenAI, Li notes that training the newest frontier models consumes roughly two gigawatts of power—a level of energy consumption comparable to the entire power grid of a medium-sized European country.
4. **Productivity vs. Prosperity:** Silicon Valley frequently celebrates gains in productivity, but Li cautions that "increasing productivity does not guarantee shared prosperity." Ensuring that AI-driven wealth and progress benefit all of society requires deliberate planning and economic alignment.
- **Energy Scale:** Training the latest AI models requires up to 2 GW of power, matching the total power consumption of some European nations.
- **Software Engineering Shift:** AI tools act as force multipliers for skilled engineers, but require major updates in corporate upskilling and training.
- **Democratic Integrity:** AI-driven information manipulation presents direct risks to democratic public life.
- **Economic Principles:** Li emphasizes that overall economic productivity gains do not automatically result in shared prosperity across society.
## Notable quotes
> [02:02](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=122s) Can machines think? — Fei-Fei Li
> [05:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=304s) cheating — Fei-Fei Li
> [05:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=304s) empowerment — Fei-Fei Li
> [15:43](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=943s) boring middle — Fei-Fei Li
> [15:43](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=943s) delusional utopianism — Fei-Fei Li
> [21:33](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1293s) increasing the global pie — Fei-Fei Li
> [22:06](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1326s) mother AI — Matt Egan
> [24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) dark underbelly — Fei-Fei Li
> [24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) increasing productivity does not guarantee shared prosperity — Fei-Fei Li
## Mentioned
Fei-Fei Li, Matt Egan, World Labs, Ai4, CNN, Stanford University, Stanford Human-Centered AI Institute (HAI), Alan Turing, John McCarthy, Socrates, Marvin Minsky, Geoffrey Hinton, OpenAI, New York Times
## Video description
Fei-Fei Li, Co-Founder & CEO, World Labs
Matt Egan, Senior Reporter, CNN
Abstracts
Join Fei-Fei Li for a forward-looking conversation exploring the intersection of AI’s understanding of the physical world, spatial reasoning, and the pursuit of human-centered design. This broad discussion will touch on technical, ethical, and societal dimensions shaping AI’s role in the world.
Join Fei-Fei Li for a forward-looking conversation exploring the intersection of AI’s understanding of the physical world, spatial reasoning, and the pursuit of human-centered design. This broad discussion will touch on technical, ethical, and societal dimensions shaping AI’s role in the world.
李飞飞反对以“母亲AI”构想保障超级智能安全
AI先驱李飞飞认为，人类必须对AI系统保持严格的以人为本的治理，而非将尊严与自主权让渡给仁慈的超级智能安全架构。
在这场炉边谈话中，李飞飞博士将AI重新定义为类似于电力的基础性文明技术，而非一场竞技体育赛事。她强调了AI在变革教育领域方面被低估的作用，主张技术工具应通过苏格拉底式探究来赋予人类能动性，而非仅仅从作弊视角来看待。李飞飞详细阐述了她在World Labs关于空间智能与3D世界生成的愿景，并将其视为迈向多重宇宙交互的历史性演化飞跃。此外，她拒绝了Geoffrey Hinton提出的“母亲AI”概念，坚称人类必须保留自身的主体性，同时探讨了能源消耗、民主完整性及确保共享经济繁荣等关键风险。
李飞飞将 AI 视为如同 150 多年前电力兴起般的长期文明底层技术，而非分出胜负的体育比赛，强调 AI 的本质应是全人类共同受益的漫长历程。 (0:04)
李飞飞指出教育是当前 AI 最被低估且资源匮乏的领域，应摒弃将 AI 狭隘定义为“作弊工具”的负面视角，将其重塑为类似苏格拉底式提问、赋能学生思考能力的基准工具。 (5:04)
谈及初创公司 World Labs，李飞飞阐释了“空间智能”的核心理念，即让 AI 具备理解、推理并生成 3D 物理与数字世界的能力，为下一代机器人和多重宇宙体验奠定底层基石。 (10:48)
针对 Geoffrey Hinton 提出的由仁慈“母体 AI”保护人类的设想，李飞飞明确表示反对，强调人类绝不能向机器出让尊严与自主权，技术的研发与治理必须始终以人为中心。 (22:06)
谈及现实威胁，李飞飞警示前沿大模型训练能耗已高达 2 吉瓦（相当于中等欧洲国家的电网规模），并特别提醒生产力的提升绝不自动等同于社会成果的普惠共享。 (24:04)
Reading summary
Dr. Fei-Fei Li frames artificial intelligence not as a competitive race with clear winners and losers, but as an foundational civilizational technology comparable to the advent of electricity. She argues that society must anchor AI development in human dignity, agency, and empowerment rather than yielding control to speculative machine overseers or succumbing to polarized narratives.
Civilizational Impact and Human-Centered Governance (0:04)
Rejecting analogies that equate AI development to a nine-inning baseball game, Li compares the current era to the dawn of electricity 150 to 180 years ago, asserting that humanity is only at the beginning of a prolonged civilizational journey. From a scientific perspective, she views 'AI' and 'AGI' as essentially synonymous, noting that pioneering figures like Alan Turing and John McCarthy set out from the start to build machines capable of general thought rather than narrow tools.
Li emphasizes that technological advancement must remain strictly human-centered, a philosophy she codified through the Stanford Institute for Human-Centered AI (HAI). While acknowledging former mentor Geoffrey Hinton's suggestion that a protective 'mother AI' could shield humanity from superintelligence, Li explicitly rejects that premise. She insists that humanity must never surrender its dignity, self-determination, or governing responsibility to automated systems, asserting that technology must always be directed and held accountable by people.
Reframing Education from Cheating to Socratic Inquiry (5:04)
Li identifies education—spanning K-12 classrooms, higher education, and corporate workforce re-skilling—as the single most under-resourced and critically overlooked domain in AI policy. She contends that mainstream discourse has been distorted by sensationalism and an excessive institutional fixation on student cheating. Instead of viewing generative models as academic threats, educators should recognize their capacity for empowerment.
Drawing a historical parallel to Socrates establishing structured questioning over 2,000 years ago, Li describes prompt engineering as a modern continuation of the Socratic method. Under this pedagogical model, AI-generated outputs serve as an initial baseline for essays, math problems, or scientific inquiries. The educational objective then shifts toward training students to apply their critical thinking and human agency to analyze, interrogate, and build beyond that automated baseline.
Spatial Intelligence and Generative 3D Universes (10:48)
Detailing the technological focus of her startup, World Labs, Li outlines the emergence of spatial intelligence—the capacity for AI models to understand, generate, and interact within dynamic 3D environments. She observes that while biological life has been confined to a single physical 3D universe for roughly one billion years, generative spatial models will make humans the first species capable of creating and experiencing multiple virtual 3D universes.
Li grounds this concept in practical applications that blend physical and digital realities: surgeons utilizing high-precision augmented reality co-pilots during operations, autonomous transport systems navigating detailed spatial maps, immersive 3D e-commerce, and real-time interactive digital living rooms bridging families across continents. Beyond consumer and medical software, Li stresses that spatial reasoning provides the essential architectural foundation for future physical robotics and autonomous agents.
Energy Scale, Workforce Friction, and Systemic Vulnerabilities (24:04)
Positioning herself in the pragmatic 'boring middle' between apocalyptic doomsaying and utopian optimism, Li notes that all transformative tools carry inherent risks. She advocates automating hazardous, repetitive, and uncreative tasks to enlarge the overall global economic pie, while cautioning against using automation to displace work where human judgment and connection excel. In professional sectors like software engineering, AI is currently acting as a force multiplier for top talent, yet it introduces severe friction around corporate upskilling and workforce transitions.
Li highlights several urgent practical concerns, including the manipulation of information ecosystems that destabilize democratic institutions. She also cites a conversation with OpenAI's CFO indicating that training frontier models requires approximately two gigawatts of power—a level of consumption matching the national electrical grid of a medium-sized European country. Crucially, Li warns that enterprise productivity gains do not inherently lead to shared societal prosperity without intentional economic policies.
Li concludes that realizing the promise of artificial intelligence depends on avoiding extreme rhetorical poles and directly addressing material bottlenecks in energy, democratic integrity, and human capital. The central unresolved challenge remains establishing deliberate governance mechanisms that translate raw productivity gains into broad, shared prosperity without compromising human agency.
## Background
该视频由频道 Ai4 于 2025 年 8 月 24 日发布在 YouTube 平台上。视频描述指出，本次对谈由 World Labs 联合创始人兼 CEO 李飞飞与 CNN 资深记者麦特·埃根（Matt Egan）展开，旨在展望 AI 对物理世界的理解、空间推理能力以及以人为本的设计理念，并深入探讨技术、伦理与社会层面的多维影响。
## TL;DR
李飞飞博士在对谈中指出 AI 是一项类似于电力的“文明级技术”，正处于极其早期阶段，并呼吁重塑 AI 在教育领域的作用。她阐述了 World Labs 在“空间智能”与三维数字生成领域的探索，描绘了人类在多重宇宙与物理世界融合交互的未来。针对“母体 AI”及“末日论”等观点，她明确予以反驳，坚持“以人为本”的理念，强调技术应放大人性能动性、维护人类尊严，同时警惕能源消耗、劳动力剧变及共同繁荣等现实挑战。
## Atmosphere & style
对话氛围理性而深邃，既有学术大家对技术本质的透彻剖析，又不乏接地气的幽默与温情。李飞飞以务实且富有远见的“中道”视角，在回应“末日论”与“狂热乌托邦”时展现出从容与坚定，与主持人的机敏提问形成良好互动。整场对谈充满了对人类命运的人文关怀，让观众在感受前沿技术震撼的同时，也能深刻体会到技术伦理与人类尊严的沉甸甸分量。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=4s) 李飞飞认为 AI 并非有输赢的棒球比赛，而是一项类似于电力的“文明级技术”，正处于长期的极早期阶段。
- [00:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=4s) 人类纯粹的科学好奇心与求知欲，是推动 AI 从早期探索走向深度学习与生成式 AI 突破的根本原动力。
- [04:10](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=250s) AI 对教育的重塑严重缺乏资源和重视；社会应摆脱将其等同于“作弊”的偏见，转而将其视为赋能工具。
- [04:10](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=250s) 提倡“苏格拉底式提示”，利用 AI 给出答案基线，引导学生发挥人类独特的能动性、创造力与深度思考。
- [07:55](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=475s) AI 与 AGI 在科学语境下本就是同义词；应摆脱对 AGI 概念的狂热追逐，回归“以人为本的 AI”核心原则。
- [07:55](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=475s) 以人为本的 AI 包含三大支柱：由人创造、由人使用、由人治理，旨在增强和提升人类而非取代人类。
- [10:48](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=648s) 空间智能（Spatial Intelligence）是理解与生成三维世界的底层技术，将使人类能在物理与数字融合的“多重宇宙”中交互。
- [10:48](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=648s) World Labs 旨在从底层科学入手开发生成式 3D 空间技术，为未来创作、机器仿真及多元交互提供基础设施。
- [15:40](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=940s) 反对 AI 末日论与盲目乐观的乌托邦主义，主张立足务实中道，将 AI 作为拓展教育、创造力和制造业的增强工具。
- [15:40](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=940s) 建议改变教育评估体系，摆脱死记硬背与标准化考试，重点培养孩子独有的想象力和创造力。
- [21:33](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1293s) 发展 AI 的目标应是承担危险、重复的工作并扩大全球经济与生产力的“蛋糕”，而非取代人类最擅长的工作。
- [22:06](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1326s) 明确反对杰弗里·辛顿关于建立“母体 AI”的设想，强调人类在任何时候都不应让渡自身的尊严与自主权。
- [24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) AI 工具会映射出人性的复杂与混乱，需警惕其对民主与信息安全可能带来的滥用和操纵风险。
- [24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) 劳动力市场正在经历剧烈转变，企业与社会必须重新思考技能提升（upskill）以及对人力资本的投资。
- [24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) 顶尖 AI 模型训练耗电量已高达 2 吉瓦（gigawatts），规模堪比欧洲中等国家的用电量，提出严峻能源挑战。
- [24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) 必须明确区分“生产力增长”与“共同繁荣”，生产力的提升并不自动保证共同繁荣的实现。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=4s) AI不是有输赢的棒球比赛，而是人类文明的“电能”革命
在 AI4 大会的对话开场中，主持人 CNN 资深记者麦特·埃根（Matt Egan）打趣道，台下的观众显然不是为了看他，而是为了听“人工智能界最具备发言资质的大咖之一”李飞飞博士（Dr. Fei-Fei Li）分享前沿洞察。李飞飞首先向主办方和观众致谢，并开玩笑说作为一名大学教授，她最害怕教名称以字母“A”开头的早课，因此对大清早到场的观众表示由衷敬佩。
当被问及如果将当前的 AI 革命比作一场 9 局制的棒球比赛，我们目前处于第几局时，李飞飞明确否定了这种比喻。她强调 AI 绝非一场存在输家与赢家的体育竞技，因为人类必须在这场技术演进中取得共同胜利。在她看来，AI 是一项“文明级技术”（civilizational technology），其性质更类似于 150 至 180 年前电力的崛起——从基础科学走向技术应用，再到最终催生出各种产品。
> "如果回到 1870 年去问当时的人：‘电力的崛起处于棒球比赛的第几局？’ 你能给出什么答案？直到今天，电力依然在为我们的文明提供动能。我认为 AI 也是如此，我们正处于这项将为人类未来赋能的文明级技术的极早期阶段。"
```
[科学发现/基础探索] ──► [技术突破/应用开发] ──► [产品化/文明级赋能]
(阿兰·图灵 / 麦卡锡) (深度学习 / 生成式 AI) (长期全行业赋能)
```
李飞飞重申，AI 的起点是人类纯粹的求知欲。从阿兰·图灵（Alan Turing）到约翰·麦卡锡（John McCarthy，李飞飞在斯坦福的昔日同事），这些先驱提出了“机器能否思考”这一大胆设想。作为见证并推动深度学习、数据驱动机器学习以及生成式 AI（Gen AI）演进的一代学者，她感到无比幸运，因为这项技术正以前所未有的方式重塑历史，全方位增强人类的工作与生活。
- **文明级技术定位**：AI 并非短期赛道竞争，而是像电力一样长期支撑文明运转的基础设施。
- **驱动力来源**：从早期探索到如今的突破，人类对科学的“好奇心”始终是 AI 发展的根本原动力。
### [04:10](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=250s) 被严重低估的课题：AI 对教育的重塑与“苏格拉底式提示”
李飞飞指出，虽然当下的 AI 讨论铺天盖地，从华盛顿特区的政策制定者到各家公司的董事会，再到各类播客和社交媒体，无一不在探讨 AI，但有一个极其关键的领域却处于“资源严重不足、被深度低估”的状态，那就是 **AI 与教育**。这种忽视涵盖了 K-12 基础教育、高等教育，以及企业的持续终身学习与技能培训。
李飞飞分享了她前一天在斯坦福大学的经历：当时她与来自美国国会两院及两党的高级幕僚围坐交流，发现无论是在国家政策层面、教育系统内部，还是在家长和教师个人层面，大家普遍缺乏一个清晰的“AI 赋能学习”框架。由于缺乏认知，加上外界过度宣传“AI 末日论”，导致人们对教育中的 AI 产生了膝跳反射式的负面排斥。
> "当人们想到 AI 与教育时，脑海中浮现的第一词汇往往是‘作弊’（cheating）。这完全是个错误的词！我们最应该想到的词是‘赋能’（empowering）。"
在谈到如何通过 Prompt（提示词）进行学习时，李飞飞追溯了人类历史上的“提示词第一人”——苏格拉底（Socrates）。苏格拉底在两千多年前就通过不断的追问与提问来引导学习，这正是提示词的核心精髓：**提出正确的问题**。AI 应该为学生的学习提供基线（baseline），而人类独特的能动性与智慧则可以在此基线之上飞跃。
```
[传统的膝跳反射式认知] ─────────► 聚焦于“作弊”与防范限制
[苏格拉底式提问/提示哲学] ─────────► 借助 AI 提供基线 ──► 发挥人类能动性与创造力
```
- **教育重塑的紧迫性**：利用 AI 转型教育、赋予师生更多能力，是本十年最具挑战性也最具价值的机遇。
- **转变教育观念**：学校和家长不应仅告诫学生“不要用 AI 作弊”，而应引导他们利用 AI 给出的基线答案，去探索更具深度的思考。
### [07:55](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=475s) 从“AGI 竞赛”转向“以人为本的 AI”
针对硅谷、华尔街以及政策圈对“通用人工智能”（AGI）的狂热追逐，主持人提及李飞飞过去曾对 AGI 概念持保留态度。李飞飞现场澄清，她从未认为 AGI 是坏事，只是在她看来，**AI 与 AGI 在科学语境下本就是同义词**。当阿兰·图灵、约翰·麦卡锡和马文·明斯基（Marvin Minsky）等 AI 奠基人最初设想“能够思考和行动的机器”时，他们胸怀的就是最大格局的想象力，并未将其局限在狭隘的、非通用的领域。
基于这一视角，李飞飞解释了自己当年为何创办斯坦福大学“以人为本人工智能研究院”（HAI），并在 2018 年发表《纽约时报》专栏文章倡导“以人为本的 AI”（Human-Centered AI）：
> "AI 是由人创造的，是由人使用的，也应当由人来治理。我们必须把人类置于中心，而不是将技术置于中心。‘以人为本’的核心原则，就是负责任地使用 AI 来增强（augment）和提升人类，改善人类的生存状况。"
李飞飞强调，无论是引导企业决策、学术研究还是公共政策制定，“以人为本”都应当成为核心指导思想。AI 的价值在于放大人自身的 agency（能动性），帮助人们提高工作效率、降低成本、改善子女教育，而不是用技术去取代人。
- **概念厘清**：AI 与 AGI 在早期学术定义上并无本质割裂，过于执念于 AGI 词汇容易偏离技术服务于人的本质。
- **以人为本的三大支柱**：由人创造（Created by people）、由人使用（Used by people）、由人治理（Governed by people）。
### [10:48](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=648s) 空间智能与 World Labs：在“多重宇宙”中交互的未来
谈及她共同创办的新兴初创公司 **World Labs**，李飞飞详细阐述了**空间智能（Spatial Intelligence）**概念，并展示了一个令人惊叹的未来图景：AI 将有能力生成三维数字世界，让人类生活在物理与数字融合的“多重宇宙”（Multiverse）之中。
为了让这一概念更具象，李飞飞构建了一位未来女外科医生的日常场景：
```
[外科手术场景] ──► 借助 AR/数字助手进入三维高精微观世界，精准处理血管
│
[日常通勤场景] ──► 乘坐智能驾驶汽车，车辆在道路 traffic 的数字世界中平稳行驶
│
[商业购物场景] ──► 接入电商三维空间，与孩子远程虚拟试穿各种服装
│
[跨国社交场景] ──► 将远在欧洲的祖父母“数字传送”至客厅，跨大西洋共度生日
```
李飞飞指出，上述场景中的每一个数字世界都需要理解、推理和生成三维空间的底层技术，这就是 **空间智能**。World Labs 正是从基础科学做起，构建能够生成三维世界的生成式能力。此外，机器人也是这一生态的关键一环。机器人并不一定非要像科幻电影中那样庞大或呈人形，汽车、辅助设备或任何形态的机械，本质上都是帮助人类更好生活的机器人。
- **三维空间的独特性**：近十亿年来，地球上的生命仅存在于一个可触摸的物理三维世界中；而凭借数字技术与空间智能，人类将成为**首个能够同时在多个 3D 虚拟宇宙中体验与交互的物种**。
- **World Labs 的定位**：从底层科学入手，开发生成式 3D 空间技术，为未来的创作、机器人仿真及多元交互提供基础设施。
### [15:40](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=940s) 告别“末日论”与“狂热乌托邦”：做理性务实的“中间派”
对于何时能实现上述“多重宇宙”的愿景，李飞飞打趣主持人依然偏爱棒球比赛式的进度预测。她透露，自己之所以选择走出学术界的“象牙塔”，与昔日的学生和同事共同创业，正是因为这项技术已经到了迈向现实与产品落地（如赋能创作者、机器人仿真等）的临界点。
在讨论 AI 对社会的正面影响时，李飞飞特别说明自己既不是“AI 末日论者”（Doomsayer），也不是“盲目乐观的乌托邦主义者”（Delusional utopian），而是坚守在**务实的中道（Boring middle）**：
> "人类非常擅长制造双刃剑。从火、石斧一直到 AI，我们制造的每一种工具都是双刃剑。技术本身并不完美，但它是一项文明级技术，能够为无数行业注入强劲动力。"
李飞飞再次表达了她对 AI 赋能人类四大核心领域的无限期待：
1. **教育系统重塑**：李飞飞希望未来几百年后的历史学家记录 21 世纪时，能将“人类因 AI 重塑了教育体系”写进历史。人类不应再将儿童宝贵的早期青春消耗在记忆性考试和标准化测试上——因为这些测试当下的 AI 都能轻松通过。
2. **人类创造力爆棚**：人类在生理结构上不占优势（跑不快、飞不起来、游泳也不出色），但人类的核心特质是**具备创造力**。AI 能作为超级助手，全面赋能艺术家、科学家、音乐家、医生及创业者。
3. **制造与机器人产业**：空间智能作为机器人的物理与认知基础，将推动新一代制造与服务机器人的诞生，帮助人类更好地工作与生活。
- **工具的双刃剑属性**：承认 AI 的潜在风险，但重点应放在如何将其作为增强人类能力的工具。
- **教育改革呼吁**：摆脱背诵与应试教育，转向培养人类独有的想象力、好奇心与创造力。
<ElicitationsGroup message="想要进一步探索李飞飞博士的 AI 理念与空间智能技术？你可以：">
<Elicitation label="深入了解 World Labs 的空间智能与 3D 生成技术原理" query="请详细介绍李飞飞创办的 World Labs 及其核心技术“空间智能”（Spatial Intelligence）的工作原理与应用场景。"/>
<Elicitation label="探究 Stanford HAI“以人为本 AI”的具体框架与政策建议" query="斯坦福大学以人为本人工智能研究院（HAI）提出的“以人为本 AI”（Human-Centered AI）具体包含哪些核心原则和政策建议？"/>
</ElicitationsGroup>
### [021:33] 扩大全球蛋糕与机器的分工
李飞飞（Fei-Fei Li）强调，我们不应该让机器去取代人类最擅长的工作，而是应该利用机器去承担那些对人类有伤害、无法展现人类创造力、或是高度重复性的工作。在她看来，发展 AI 的核心目标应当是借此机会扩大全球经济与生产力的“蛋糕”（increase the global pie）。她指出这些前景令人无比兴奋，并表示关于这一点自己可以说上三天三夜。
### [022:06] 回应杰弗里·辛顿：反驳“母体 AI”设想，坚持以人为本
主持人麦特·埃根（Matt Egan）提及，前一天被尊称为“AI教父”的杰弗里·辛顿（Geoffrey Hinton）也在同一舞台上分享了他的担忧。辛顿提出，为了防止未来比人类更聪明且更强大的 AI 控制或消灭人类，唯一的办法是赋予系统一种类似于“母性本能”（maternal instinct）的“母体 AI”（mother AI）。
李飞飞对此回应道，辛顿是她认识并敬重了二三十年的好友与导师，能与他并称为“AI教父/教母”感到非常荣幸。然而，她明确表示不同意辛顿关于建立“母体 AI”的框架。她认为将 AI 塑造成一个比人类聪明且具备某种本能的统治者或保护者，本身就是一种错误的思考视角。
李飞飞提出尊严与主权不可让渡的观点：
* **以人为本（Human-Centered AI）：** 必须确保 AI 在个体、社区和社会的各个层面都尊重并维护人类的尊严（human dignity）与自主权（human agency）。
* **人类的责任：** 无论技术多么强大，人类都不应该在任何时刻主动放弃或被要求让渡自身的尊严与主权。作为母亲、教育家和发明家，她坚信这是 AI 发展的核心基石。
### [024:04] 超级智能与实际风险：深夜焦虑与社会挑战
针对主持人询问关于超智能 AI 系统带来的失控风险以及究竟是什么让她“夜不能寐”时，李飞飞表示，虽然她常谈及技术积极的一面，但人类历史的阴暗面同样不容忽视。从火、石斧到现代最顶尖的技术，人类一直有能力使用工具伤害彼此与社会。AI 作为一种强大工具，必然会映射出人性中复杂和混乱的一面。
具体而言，让李飞飞感到担忧和警惕的现实风险包括：
* **民主与信息安全：** AI 可能在信息传播及民主社会的公共生活中被滥用或操纵，而维护强健的民主制度对国家至关重要。
* **劳动力市场的剧烈阵痛：** 技术的转型必然带来就业与劳动力结构的巨大变革。以硅谷为例，软件工程正经历地震般的剧变（seismic change）。但这并不意味着人们不应成为软件工程师——在她的公司 World Labs，最顶尖的软件工程师正借助 AI 获得巨大的能力提升（superpowered by AI）。重点在于企业和社会如何思考技能提升（upskill）以及对人力资本的投资。
* **国家级能源消耗：** AI 模型的训练对能源提出了前所未有的要求。李飞飞提到前一天与 OpenAI 的首席财务官（CFO，即 Sarah Friar）交流，得知最新模型的训练消耗了高达 **2 吉瓦（gigawatts）** 的能源，这相当于欧洲一个中等规模国家单数字级别的总用电量。
* **生产力增长与共同繁荣的背离：** 科技界（尤其是硅谷）非常热衷于讨论全球生产力（productivity）的提升，但**“生产力的提高并不保证共同繁荣（shared prosperity）”**。这是两个截然不同的概念，社会在 AI 时代必须深思并落实如何同时实现这两者。
### [027:36] 结语
对话在主持人麦特·埃根对李飞飞博士的精彩分享与深刻洞察的感谢声中圆满结束。
---
#### 重点事实与观点速览
- **人物互动：** 杰弗里·辛顿（Geoffrey Hinton）建议通过培养具有“母性本能”的“母体 AI”来应对比人类更聪明的超级 AI；李飞飞明确对此表示反对。
- **核心理念：** 强调“以人为本的 AI”（Human-Centered AI），认为人类在任何情况下都不应让渡自身尊严（human dignity）与自主权（human agency）。
- **训练能耗数据：** 最新的顶尖 AI 模型训练耗电量已达 **2 Gigawatts**，规模堪比中等欧洲国家的运行能耗。
- **经济与社会警示：** 明确区分了“生产力增长”（productivity increase）与“共同繁荣”（shared prosperity），提醒社会不能将二者混为一谈。
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=4s) 如果回到 1870 年去问当时的人：‘电力的崛起处于棒球比赛的第几局？’ 你能给出什么答案？直到今天，电力依然在为我们的文明提供动能。我认为 AI 也是如此，我们正处于这项将为人类未来赋能的文明级技术的极早期阶段。 — 李飞飞
> [04:10](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=250s) 当人们想到 AI 与教育时，脑海中浮现的第一词汇往往是‘作弊’（cheating）。这完全是个错误的词！我们最应该想到的词是‘赋能’（empowering）。 — 李飞飞
> [07:55](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=475s) AI 是由人创造的，是由人使用的，也应当由人来治理。我们必须把人类置于中心，而不是将技术置于中心。‘以人为本’的核心原则，就是负责任地使用 AI 来增强（augment）和提升人类，改善人类的生存状况。 — 李飞飞
> [15:40](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=940s) 人类非常擅长制造双刃剑。从火、石斧一直到 AI，我们制造的每一种工具都是双刃剑。技术本身并不完美，但它是一项文明级技术，能够为无数行业注入强劲动力。 — 李飞飞
> [24:04](https://www.youtube.com/watch?v=WpfGwb7_xBM&t=1444s) 生产力的提高并不保证共同繁荣（shared prosperity）。 — 李飞飞
## Mentioned
李飞飞 (Dr. Fei-Fei Li), 麦特·埃根 (Matt Egan), 阿兰·图灵 (Alan Turing), 约翰·麦卡锡 (John McCarthy), 马文·明斯基 (Marvin Minsky), 杰弗里·辛顿 (Geoffrey Hinton), Sarah Friar, Ai4, CNN, 斯坦福大学 (Stanford University), 斯坦福大学以人为本人工智能研究院 (Stanford HAI), World Labs, OpenAI
## Video description
Fei-Fei Li, Co-Founder & CEO, World Labs
Matt Egan, Senior Reporter, CNN
Abstracts
Join Fei-Fei Li for a forward-looking conversation exploring the intersection of AI’s understanding of the physical world, spatial reasoning, and the pursuit of human-centered design. This broad discussion will touch on technical, ethical, and societal dimensions shaping AI’s role in the world.
Join Fei-Fei Li for a forward-looking conversation exploring the intersection of AI’s understanding of the physical world, spatial reasoning, and the pursuit of human-centered design. This broad discussion will touch on technical, ethical, and societal dimensions shaping AI’s role in the world.
[Source](https://www.youtube.com/watch?v=WpfGwb7_xBM)
Source quotes
[0:04] 李飞飞: 如果回到 1870 年去问当时的人：‘电力的崛起处于棒球比赛的第几局？’ 你能给出什么答案？直到今天，电力依然在为我们的文明提供动能。我认为 AI 也是如此，我们正处于这项将为人类未来赋能的文明级技术的极早期阶段。
[4:10] 李飞飞: 当人们想到 AI 与教育时，脑海中浮现的第一词汇往往是‘作弊’（cheating）。这完全是个错误的词！我们最应该想到的词是‘赋能’（empowering）。
[7:55] 李飞飞: AI 是由人创造的，是由人使用的，也应当由人来治理。我们必须把人类置于中心，而不是将技术置于中心。‘以人为本’的核心原则，就是负责任地使用 AI 来增强（augment）和提升人类，改善人类的生存状况。
[15:40] 李飞飞: 人类非常擅长制造双刃剑。从火、石斧一直到 AI，我们制造的每一种工具都是双刃剑。技术本身并不完美，但它是一项文明级技术，能够为无数行业注入强劲动力。
[24:04] 李飞飞: 生产力的提高并不保证共同繁荣（shared prosperity）。
Source quotes
[2:02] Fei-Fei Li: Can machines think?
[5:04] Fei-Fei Li: cheating
[5:04] Fei-Fei Li: empowerment
[15:43] Fei-Fei Li: boring middle
[15:43] Fei-Fei Li: delusional utopianism
[21:33] Fei-Fei Li: increasing the global pie
[22:06] Matt Egan: mother AI
[24:04] Fei-Fei Li: dark underbelly
[24:04] Fei-Fei Li: increasing productivity does not guarantee shared prosperity
Original transcript
[0:04] Uh I know everyone's not here to see me
[0:05] but to see Dr. Lee. So thank you so much
[0:08] Dr. Lee for doing this and sharing your
[0:11] insights right now. I mean you are
[0:12] literally one of the most qualified
[0:15] people on the planet to discuss the AI
[0:18] revolution. So thank you.
[0:19] >> Thank you Matt. Thank you AI4. Uh as a
[0:22] professor I dread teaching classes
[0:24] starting at A. So kudos to all of you
[0:26] being here.
[0:29] Well, first I I want to ask you to talk
[0:31] a little bit about where we are right
[0:34] now in the AI revolution. If this were a
[0:38] nine inning baseball game, what inning
[0:41] would we be in?
[0:43] >> Interesting you asked the question that
[0:45] way, Matt. Um, I actually do not think
[0:48] AI is a baseball game because
[0:53] nothing against baseball. Um,
[0:56] AI is not a sports competition because I
[0:59] would not like to see losers at the end
[1:02] of this. Humanity needs to win and
[1:08] and it's much more like a journey we're
[1:11] taking on together. If you're a little
[1:13] more for excitement, I call this a
[1:15] revolution. It's a civilizational
[1:17] technology. And the an the analogy I
[1:21] draw is think 150 180 years ago
[1:25] electricity was on the rise from a
[1:27] science to a technology to a product.
[1:30] And if you ask people at that time 1870
[1:35] which inning are you in as a baseball
[1:38] game for electricity? What answer would
[1:40] you have? One nine seven. To this day,
[1:45] electricity
[1:47] continues to power our civilization. And
[1:50] I think that's what AI is. We're at the
[1:52] very beginning of the civilizational
[1:55] technology that would power the the the
[1:57] the future of humanity.
[2:00] >> Okay. So, the very beginning. Okay.
[2:02] Well, that's great. Now can you when you
[2:05] when you take a step back and you really
[2:06] think about the evolution of AI, what is
[2:10] the most important but most
[2:13] underappreciated aspect of artificial
[2:16] intelligence?
[2:19] >> Well, there are many right from the
[2:21] evolution of AI. I think it's very
[2:24] important to um to remember that this
[2:28] began as a science a group of daring
[2:32] individuals from Alan Turing to the
[2:34] founding fathers of uh uh artificial
[2:37] intelligence including my own colleague
[2:39] like John McCarthy just dared humanity
[2:43] to ask the question can machines think
[2:45] and why do we ask that question because
[2:48] we were curious and that's what humans
[2:50] are humanity has always chased
[2:53] innovation and discovery out of
[2:55] curiosity.
[2:57] But along the way, and I belong to that
[2:59] generation, I feel very lucky that this
[3:02] curiosity grew bigger. We not only
[3:05] discovered some of the most important
[3:07] scientific breakthroughs and innovations
[3:10] along the way, including deep learning,
[3:13] datadriven, um, machine learning and and
[3:17] now Gen AI. We also discovered that we
[3:20] are now changing the course of history
[3:23] because uh it's so powerful that as a
[3:27] technology it can literally superpower
[3:30] every aspect of human work, human life.
[3:33] And then looking to the future, we're
[3:36] not lacking AI discussion today. today's
[3:40] uh society, the airwaves and the
[3:42] podcasts, the the the the the tweets and
[3:46] the uh uh you know the videos and all
[3:49] that is filled with AI discussion from
[3:53] literally Washington DC to boardrooms.
[3:56] But there is one topic of AI in today's
[4:00] AI revolution that I think it's deeply
[4:04] underdised under resources resourced and
[4:07] underappreciated and that's AI and
[4:10] education
[4:11] >> and uh this goes from personal education
[4:14] to K12 education to uh uh continuous
[4:18] lifelong learning and skilling from both
[4:21] company point of view as well as just uh
[4:24] formal education point of you because
[4:26] this is a technology that should not and
[4:30] will not if we have the will replace
[4:33] humanity or uh human agency but this is
[4:36] a technology that empowers human agency
[4:38] and using AI to transform education to
[4:43] empower learning for students and
[4:45] teachers is to me this decade's most
[4:50] important um challenge and opportunity.
[4:54] Wow. And are we
[4:59] when it when it comes to AI and
[5:02] education,
[5:04] what should be is the is the current
[5:06] approach the right one? I mean, is there
[5:08] anything that you would say that we
[5:09] should be doing differently? And how
[5:11] does that relate to the importance of
[5:13] prompting and knowing how to prompt?
[5:16] >> Is there an approach right now? Yeah, I
[5:19] >> I'm I'm asking you because literally
[5:21] yesterday I was in a room with uh uh
[5:24] senior staffers from both houses as well
[5:27] as both parties at Stanford and I don't
[5:30] think as a country I don't think as an
[5:33] education system and I don't think even
[5:35] as individuals whether you're parents or
[5:38] teachers or many students have a
[5:41] actually a approach of AI powered
[5:44] learning because it's so early. I'm not
[5:47] blaming anybody. I just think it's so
[5:49] early and and because of all the
[5:53] sometimes too much noise that has this
[5:56] AI doomsday, we almost created a
[6:01] instinctive negative um reaction,
[6:04] knee-jerk reaction when it comes to AID
[6:07] education. The number one word when
[6:10] people think about AI and education is
[6:12] the word cheating. And that's just
[6:15] absolutely the wrong word. The number
[6:17] one word we should uh think is
[6:20] empowering is you you you talk about
[6:22] prompting.
[6:24] How to prompt to learn is literally
[6:29] you know who is the first humanity's
[6:31] first prompter?
[6:34] It's Socrates.
[6:37] He's the person more than 2,000 years
[6:40] ago taught us how to prompt. Of course,
[6:43] he didn't use that word prompt. I don't
[6:45] even know what's a Greek word for
[6:47] prompt, but
[6:49] but it's asking the right question. And
[6:51] that's the heart of human learning. And
[6:54] we now have created a powerful tool that
[6:58] can supercharge that process of
[7:02] learning. And yet we tell our students,
[7:05] don't cheat with AI. What we should tell
[7:08] our students is this is the baseline AI
[7:11] will give you. Whether it's a science
[7:13] question, a math question or English
[7:15] essay. Now you are a human being with
[7:19] agency, an intelligence that no AI, no
[7:22] supercomputer can ever achieve. Given
[7:26] this baseline, given this tool, where
[7:29] can you go?
[7:31] That is the the question of uh prompting
[7:33] or education or learning. And I would
[7:36] encourage every parent, every teacher,
[7:39] every student to think in that way and
[7:42] think how ha how did I learn better or
[7:45] what have I learned better using AI as a
[7:48] tool.
[7:49] >> It's fascinating.
[7:55] Now I know in the past um you've pushed
[7:57] back on the term AGI. Can you explain
[8:01] why instead of a race to AGI, which is
[8:05] the focus right now in Silicon Valley
[8:07] and Wall Street and elsewhere in
[8:09] Washington, why do you think we should
[8:11] be focused instead on human centered AI?
[8:14] And what what is that? Can you explain
[8:16] to the audience what you mean by human-
[8:18] centered AI?
[8:19] >> Yeah, first of all, I Okay, to correct
[8:22] record, I never said AGI is bad. I just
[8:25] said I don't know the difference between
[8:28] the word AGI and AI. And kudos to Dan
[8:31] and friends. This is AI4, not AGI4.
[8:35] Because when Alan Turing and John
[8:37] McCarthy and Marvin Minsky, the founding
[8:40] fathers and and thinkers of AI, when
[8:42] they dared humanity with this
[8:44] possibilities of machines that think and
[8:47] do, in my opinion, they didn't say that
[8:50] machines that think narrowly and
[8:53] non-generally.
[8:55] They literally just had the biggest
[8:57] imagination. So for me as a scientist,
[9:00] AI was born with this audacious goal of
[9:06] understanding and creating machines that
[9:08] think. So AI and AGI are synonyms to me.
[9:11] So this is why I I tend to answer as a
[9:14] scientist and a scholar. I just don't
[9:16] know if the these are two differentiated
[9:18] terms. But beside that point, why did I
[9:22] found um the world's and Stanford's
[9:25] first human- center AI institute? Why
[9:28] did I write a New York Times oped back
[9:31] in 2018 actually calling this framework
[9:35] of human center uh AI uh concept is
[9:40] because this is a civilizational
[9:42] technology. Uh many of you sitting here
[9:44] here might be thinking about the future
[9:46] of jobs. Many of you sitting here might
[9:49] be thinking about how I can use AI to
[9:51] supercharge my business to make my
[9:54] product better to cut cost to to make
[9:58] work more efficient. Your parents,
[10:00] you're thinking about your kids
[10:01] learning. We just talked about this. At
[10:04] the end of the day, if we believe in the
[10:06] benevolence of human society and where
[10:08] we want to push for the future, we need
[10:12] to put humans in the center. We cannot
[10:14] put a technology in the center. the the
[10:17] the the
[10:18] AI is created by people. It is used by
[10:22] people and it should be governed by
[10:24] people. And that's why we should put
[10:26] humans in the center of this. And the
[10:28] principle of human- centered AI for me
[10:31] very much so is responsibly using AI to
[10:35] augment and enhance humanity to better
[10:38] human condition. And I think that's um
[10:41] that can guide business that can guide
[10:43] research that can guide uh policy
[10:46] making.
[10:48] Um, let's talk a little bit about World
[10:50] Labs, the startup that you co-founded.
[10:53] Um, and also about spatial intelligence.
[10:56] Um, I've heard you say that in the
[10:59] future
[11:00] AI can create infinite universes and
[11:04] that we could live in the multiverse.
[11:07] Uh, when I heard that, it sort of blew
[11:08] my mind. Can you explain what you mean
[11:11] here about this idea of infinite
[11:13] universes and how does that relate to
[11:15] the work you're doing right now?
[11:17] Yeah, this is a m thanks for asking this
[11:19] question. Um, this is a future I can
[11:22] imagine and um let's just say you're a
[11:28] doctor of the future. Okay, you're a
[11:31] surgeon. I can imagine your life at work
[11:34] in the following way. If you have to do
[11:36] surgery for a patient and to save their
[11:39] lives, you through a machine or wear a
[11:43] goggle, you transport yourself at your
[11:46] incredible skill set into a digital
[11:49] world that connects. I'm not going to
[11:52] get gory, but if you have to do surgery,
[11:54] you have to do certain things with your
[11:56] patient's body. But using a digital uh
[12:01] co-pilot or help or or or AR devices,
[12:05] you can get into that world in much more
[12:08] precision and accuracy than you can ever
[12:12] do with your own eyes, right? Whether
[12:14] you're looking for tiniest blood vessels
[12:16] or looking for things to to fix, that's
[12:21] one world. You finish the surgery
[12:24] successfully. You you you you drive
[12:27] home. We're not surprised. You might be
[12:30] too tired to drive yourself. Then you
[12:32] stay in the car that you know is works
[12:38] operates in the digital world of roads
[12:40] and traffics. And maybe you're
[12:43] co-driving or maybe you're letting it to
[12:45] drive. That's a different universe.
[12:48] You're a mom doctor, so you might even
[12:50] need to shop for your kids next prom
[12:53] dress. You go into a world that um that
[12:58] um is e-commerce world. It like as a
[13:02] mom, I really love this. I have no time
[13:04] for shopping. So I would love to go into
[13:06] a world that uh I can try on different
[13:09] clothes. Your kids can dry on different
[13:11] clothes, shop for whatever you want and
[13:13] that's a commercial world. Now you want
[13:16] to have a weekend with your kids and
[13:18] then you go to a movie experience. The
[13:20] future of movie is not just passive
[13:22] watching. You can weather through
[13:24] goggles or screens or whatever uh
[13:27] interactive media and you can actually
[13:30] experience a a you know a totally
[13:33] different outer space world based on
[13:35] whatever that story you're watching. uh
[13:39] or you have a loved one living in Europe
[13:41] and uh you guys want to share a birthday
[13:44] party in you know transatlantic
[13:47] and then you have a platform where you
[13:50] you you transport the grandparents from
[13:54] from Europe digitally to your living
[13:57] room sing the birthday songs together.
[13:59] Why am I experiencing all this? Every
[14:02] world I just experienced is a blend of
[14:05] physical and digital world. it it has
[14:08] its own universe. You can interact and
[14:11] and work and live in it in ways that can
[14:14] benefit your life. You can bring
[14:16] different people in. It can be social.
[14:18] It can be solo. And if this sounds like
[14:22] sci-fi, the question is what technology
[14:26] empowers that? And that technology to me
[14:29] is the digital creation of 3D worlds and
[14:34] uh as well as so many things that comes
[14:37] with it. The ability to understand to
[14:39] reason to interact and that's what I
[14:42] call spatial intelligence and um that's
[14:45] what world lab is doing. World Lab is
[14:48] creating that generative capability,
[14:51] you know, from a basic science uh ground
[14:54] up of technology that can empower and
[14:57] envision you usages and products like
[15:01] this. And also just to add to all what I
[15:05] described that robots will be part of
[15:07] this. Robots is not necessarily what you
[15:10] see in these um movies. They have to
[15:13] come in humanoid forms and they have to
[15:16] look really big and scary. You robots
[15:18] just is another form of machines that
[15:21] can help people whether it's driving in
[15:24] the shape of a car or if it's helping
[15:27] people in the shape of humanoid or if
[15:29] it's anything else. So the fundamental
[15:32] technology I envision here is spatial
[15:35] intelligence and that's why I see there
[15:37] are many universes we can create and
[15:40] live and work in.
[15:41] >> Wow. Well, that it does sound like
[15:43] science fiction. How far away are we
[15:46] from this vision that you have of of
[15:49] multiple universes?
[15:50] >> You just love your baseball game
[15:52] analogy.
[15:55] Um,
[15:57] we're we're beginning to see that. Um
[16:00] I'm excited to um take a step outside of
[16:03] the ivory tower from being a professor
[16:06] to uh be becoming an entrepreneur with
[16:09] my former students and colleagues
[16:12] because I think this technology is at
[16:14] the cusp of becoming reality and
[16:17] becoming uh uh products. uh whether
[16:20] we're talking about just superpowering
[16:23] creators uh robotic simulations all this
[16:27] is uh we're seeing that acceleration and
[16:31] everything I do I firmly believe um is
[16:35] for the purpose of augmenting humans and
[16:38] I think this technology is um is really
[16:41] part of that and this is one more thing
[16:43] that's really cool I want to share with
[16:45] you because most people probably don't
[16:47] don't appreciate that is that for
[16:52] as long as animals or living things on
[16:55] earth existed which is close to a
[16:57] billion years we existed in one universe
[17:01] it's the 3D universe the 3D world we can
[17:04] tangibly touch with digital revolution
[17:08] hum humanity will be the first species
[17:11] that can experience multiple universe in
[17:15] 3D and that's why it makes us so special
[17:18] it makes the future So incredible to
[17:20] imagine because the the the possibility
[17:23] is literally unbounded because of this.
[17:27] >> Wow, that's fascinating. Now, as you
[17:29] look into your crystal ball and you
[17:31] think about the future, what are you
[17:32] most excited about when it comes to the
[17:35] positive impacts that AI can have on
[17:39] humanity and society, whether it's
[17:42] medicine, science, or or elsewhere?
[17:46] Yeah, I I am excited by many things, but
[17:49] that I also want to make sure that um
[17:53] that I'm not delivering a wrong message.
[17:56] I'm not a delusional utopian of AI.
[18:00] There's a lot of extreme rhetoric of AI
[18:03] from doomsaying to delusional utopian. I
[18:07] am actually the boring middle in the
[18:10] sense that
[18:12] we are going to talk about the positive
[18:13] of this technology but I think no
[18:16] technology is own is perfect. Humanity
[18:19] is very good at creating double-edged
[18:21] sword. Every single piece of tool from
[18:23] fire to stone axe all the way to AI we
[18:27] have made our tool double-edged sword.
[18:30] So um we we we should definitely make
[18:33] sure that uh we acknowledge that but it
[18:37] is a tool. It is a civilizational
[18:40] technology that can superpower so many
[18:42] tech uh sectors. We already talk about
[18:44] education. I I know I'm a little
[18:47] obsessed with this topic partially
[18:48] because I think the public should be way
[18:50] more obsessed about AI and education
[18:53] than what it is now. is that 50 years
[18:57] from now, 100 years from now when
[18:59] historians writes the history of 21st
[19:02] century, I really hope a salient part of
[19:06] this history is that 21st century
[19:08] humanity reshaped its education system
[19:12] because of AI. We should not be using
[19:15] the most precious human capital which
[19:18] are our children for so many years of
[19:21] their early lives to just pass memory
[19:25] based tests, standardized tests that AI
[19:28] can pass today
[19:35] and we should really forge a completely
[19:38] different path for our future
[19:40] generations. So that is one sector I'm
[19:42] extremely excited about. I'm absolutely
[19:45] extremely excited about creativity. And
[19:48] when I say creativity, it's not just
[19:50] artists and musicians and writers.
[19:53] Humans are fundamentally creative
[19:55] animals. We actually kind of suck as
[19:58] animals. We don't run very fast. We
[20:00] don't see very far. We don't swim very
[20:03] very much. We don't fly. We we just
[20:05] don't do much physically compared to
[20:07] many other animals. Like I was in Africa
[20:10] for vacation. Oh my god, the animal
[20:12] world is just magical and we're kind of
[20:15] useless. But
[20:18] sorry, please don't write Fay says
[20:20] humans are useless. That's not what I
[20:22] mean. I I'm just saying as an animal.
[20:25] But we're creative beings and our power
[20:29] is trying to create for good. create for
[20:33] out of imagination, curiosity and uh AI
[20:37] can superpower artists, AI can
[20:39] superpower scientists, AI can superpower
[20:43] um you know um musicians uh doctors. So,
[20:47] alien creative business from scientific
[20:50] discovery to artistic expression to
[20:52] storytelling to business creation AI
[20:56] should be right there and I'm excited
[20:58] about I'll just give one last example. I
[21:01] think um AI can I mean there's so many
[21:04] possibilities but AI also can become a
[21:08] catalyst for um manufacturing
[21:13] um and uh robotics you know the spatial
[21:16] intelligence that I'm creating to me is
[21:19] the foundation of robotics and uh it's
[21:24] time that we we create machines to help
[21:27] us live better and uh work better. They
[21:30] are they we should not let machines take
[21:33] over jobs that humans do best but we
[21:37] should use machines to do jobs that harm
[21:40] human or does not um index of human
[21:44] creativity or is repetitive or or many
[21:49] other forms. So I think we should allow
[21:51] oursel that opportunity to increase the
[21:57] global pie. So these are the things that
[21:59] truly uh excite us. I I could go on and
[22:03] on, but I I I'll stop here.
[22:06] >> Well, so you're you're um you're often
[22:08] referred to as the godmother of AI, and
[22:11] yesterday the godfather of AI, Jeffrey
[22:14] Hidden, was on this stage talking about
[22:18] some of his concerns. Um he said that
[22:21] our one chance to ensure that AI doesn't
[22:24] gain control of humans is to have a
[22:26] mother AI so that when it's smarter than
[22:29] us, it won't wipe us out or try to
[22:32] control us. What do you make of this
[22:33] idea that we need to instill AI systems
[22:36] with a maternal instinct?
[22:40] >> I love Jeff. He's he's been my friend
[22:42] and mentor um personally for in this
[22:45] field for um for more than two three
[22:48] decades. It's honor to be even called
[22:51] godmother of AI when he's the godfather
[22:53] of AI. But u I disagree with Jeff that
[22:57] we should have a a a mother AI that um
[23:02] that's smarter than us and that has a
[23:05] whatever instinct because I think that's
[23:08] the wrong way to frame it. I would
[23:10] respective fully push back and say we
[23:13] need human center AI that gives gives
[23:17] back or preserves human dignity and
[23:20] human agency at every individual,
[23:23] community and societal level. It is our
[23:26] responsibility at every single level to
[23:30] create and and use this and govern this
[23:33] technology in the most responsible way.
[23:36] And at no moment, not a single human
[23:40] should be asked or should choose to let
[23:43] go our dignity and agency just because a
[23:46] tool is powerful. And as a mother, as a
[23:49] educator, and as a inventor, I really
[23:53] believe this is the core of what AI is
[23:56] is it should be human- centered.
[24:04] Now, I I know in the past you've said
[24:05] that we need to focus on science, not
[24:07] science fiction, but there are concerns
[24:11] about losing control of AI. What keeps
[24:14] you awake at night when it comes to
[24:17] super intelligent AI systems and the
[24:20] risks posed to humans?
[24:23] >> I mean, truthfully, a lot of things.
[24:25] This is where I I become the the middle,
[24:29] right? I do talk about the positive
[24:31] things but I do think that humanity also
[24:34] has a dark underbelly. We have in
[24:37] throughout the course of history have
[24:38] proven that we're very capable of
[24:41] hurting each other, hurting our society,
[24:44] hurting our hum uh community in using
[24:47] whichever tool again from fire to stone
[24:50] axe all the way to the most powerful
[24:52] technology. And I do foresee that that
[24:56] part of our DNA, part of the messiness
[24:58] of being human will actually transpire
[25:02] using AI technology. Um so things that
[25:06] um makes me lose sleep. Um I do think
[25:09] that AI can be a powerful tool that can
[25:13] be manipulated whether it's on the
[25:16] information side as well as in this um
[25:21] in the public life of a democratic
[25:24] society. I think AI can be uh used in
[25:28] the wrong way. um preserving democracy
[25:32] and a strong democracy for our country
[25:34] is very important. I do think that there
[25:37] is going to be incredible growing pain
[25:41] in job and labor landscape shift shifts
[25:46] and this has happened in every time
[25:48] technology transforms uh products and
[25:51] and services and we're going to see that
[25:54] software engine I come from Silicon
[25:56] Valley software engineering is going
[25:58] through a seismic change. It's not as
[26:01] simple as don't be a software engineer.
[26:04] That's not true. In my company, the best
[26:07] software engineers are superpowered by
[26:09] AI. But how do you learn? How do you
[26:11] upskill? How does companies think about
[26:14] human capital investment is a real it's
[26:18] a real thing. of course global energy
[26:21] you know yesterday I was uh talking with
[26:24] the CFO of open AI and and she was
[26:27] sharing that uh the latest models are
[26:30] trained on two gigawatts of uh energy
[26:33] and a country a middlesized country in
[26:36] Europe runs on singledigit gigawatts so
[26:39] this is nation state level energy usage
[26:43] and last but not the least I want to say
[26:46] say is especially we have seen this in
[26:48] the last 30 years is that um we've seen
[26:52] global productivity increase because of
[26:56] technology and we'll continue to see
[26:59] global productivity increase and I know
[27:03] my friends and colleagues in Silicon
[27:05] Valley love to talk about this
[27:08] increasing productivity. I think that
[27:11] what we need to keep in mind is
[27:14] increasing productivity does not
[27:17] guarantee shared prosperity.
[27:20] These are two different concepts and as
[27:23] a society we need both and this is
[27:26] something that we should really
[27:28] thoughtfully understand and implement in
[27:32] the age of AI.
[27:36] >> Well, thank you so much Dr. for uh
[27:38] talking to us and sharing all of your
[27:40] insights in this incredibly important
[27:42] moment. Thank you.
[27:43] >> Thank you, Matt.
[27:43] >> Thank you so much.
[27:45] >> Thank you.
