Robots, Machine Learning Threaten 45% of Modern Workforce

CGP Grey

Just as mechanical engines rendered horses economically obsolete after 1915, cognitive software and robotics will permanently depress demand for human labor across 45 percent of occupations without creating replacement jobs.

AI-assisted summary of What Happened to Horses Is Happening to Us — original source. Check the original for full context.

Source published: 2014-08-13

Key ideas

Unlike the historic Industrial Revolution that replaced physical muscle with machines, modern automation builds mechanical minds that directly diminish the economic demand for human cognitive labor. (0:03)
Autonomous transportation threatens to disrupt approximately 3 million jobs in the United States and at least 70 million worldwide, driven by the fact that driver wages represent roughly one-third of shipping costs. (4:57)
High-skilled professions like law, oncology, and algorithmic finance are just as vulnerable to automation as low-wage physical labor, with legal discovery and medical diagnostics already shifting to self-taught software systems. (9:12)
The occupational sectors most immediately threatened by current automation technologies comprise approximately 45 percent of the modern workforce—nearly double the 25 percent peak unemployment rate of the Great Depression. (12:43)

Detailed notes

## Background Published on YouTube by educational creator CGP Grey on August 13, 2014, the video addresses the economic disruption driven by advanced robotics and artificial intelligence. The accompanying description provides extensive documentation linking to real-world automation projects across manufacturing, transportation, healthcare, and machine-learning applications.

## TL;DR CGP Grey argues that modern advances in robotics and artificial intelligence represent an economic revolution distinct from previous industrial shifts. While historical machinery augmented or replaced physical human muscle—allowing workers to specialize in intellectual labor—emerging 'mechanical minds' are actively displacing human cognitive work across transportation, retail, law, medicine, and creative arts. Using the historical obsolescence of working horses after 1915 as a cautionary parallel, Grey demonstrates that market incentives systematically favor tirelessly efficient, self-learning algorithms over human labor. Because the vast majority of modern employment remains concentrated in occupations currently targeted by automation, society faces a structural displacement crisis that could surpass the unemployment levels of the Great Depression.

## Atmosphere & style The presentation carries a calm, methodical, and chillingly pragmatic tone, delivered through CGP Grey's characteristic deadpan, rapid-fire narration. Rather than resorting to sensationalist science-fiction tropes, the video maintains an analytical, matter-of-fact delivery supported by simple diagrams and real-world commercial footage. The emotional arc progresses from familiar historical economic reassurance into an unsettling, inescapable realization of vulnerability, leaving viewers with a haunting sense of systemic inevitability as traditional reassurances about human uniqueness are systematically dismantled.

## Bulletin - [00:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=3s) Unlike the historic Industrial Revolution that replaced physical muscle with machines, modern automation builds mechanical minds that directly diminish the economic demand for human cognitive labor. - [04:57](https://www.youtube.com/watch?v=7Pq-S557XQU&t=297s) Autonomous transportation threatens to disrupt approximately 3 million jobs in the United States and at least 70 million worldwide, driven by the fact that driver wages represent roughly one-third of shipping costs. - [09:12](https://www.youtube.com/watch?v=7Pq-S557XQU&t=552s) High-skilled professions like law, oncology, and algorithmic finance are just as vulnerable to automation as low-wage physical labor, with legal discovery and medical diagnostics already shifting to self-taught software systems. - [12:43](https://www.youtube.com/watch?v=7Pq-S557XQU&t=763s) The occupational sectors most immediately threatened by current automation technologies comprise approximately 45 percent of the modern workforce—nearly double the 25 percent peak unemployment rate of the Great Depression.

## Key takeaways - [00:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=3s) Historically, technological tools acted as mechanical muscles that relieved human physical toil and allowed the labor force to specialize into higher-value fields. - [00:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=3s) Modern computer science is creating mechanical minds, fundamentally disrupting the economic assumption that cognitive work will always remain uniquely human. - [01:05](https://www.youtube.com/watch?v=7Pq-S557XQU&t=65s) Modern general-purpose robots like Baxter are cheap, learn by demonstration, and can run for pennies of electricity per hour, making them economically superior even if they work slower than humans. - [03:31](https://www.youtube.com/watch?v=7Pq-S557XQU&t=211s) The parable of the horse demonstrates that working populations can reach an economic peak and face permanent obsolescence once mechanical alternatives surpass biological utility. - [04:57](https://www.youtube.com/watch?v=7Pq-S557XQU&t=297s) Autonomous transport does not need to achieve zero accidents; it only needs to outperform human drivers, who cause roughly 40,000 traffic fatalities annually in the United States alone. - [07:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=423s) Because white-collar office employees command higher compensation and software carries minimal deployment overhead, companies have massive financial incentives to automate mental labor. - [07:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=423s) Machine-learning systems have fundamentally changed programming: algorithms now teach themselves to trade stocks, generate corporate reports, and draft articles without manual human rule-coding. - [09:12](https://www.youtube.com/watch?v=7Pq-S557XQU&t=552s) Legal discovery has largely shifted away from human associate attorneys to specialized research software that evaluates millions of documents in hours with higher accuracy. - [09:51](https://www.youtube.com/watch?v=7Pq-S557XQU&t=591s) Networked diagnostic platforms like IBM's Watson ingest global clinical research and patient histories across millions of cases, threatening to depress baseline demand for human general practitioners. - [11:23](https://www.youtube.com/watch?v=7Pq-S557XQU&t=683s) The creative arts cannot absorb displaced cognitive workers because the arts constitute a popularity-driven sector that cannot support an entire economy, and algorithms like Emily Howell already compose human-indistinguishable art. - [12:43](https://www.youtube.com/watch?v=7Pq-S557XQU&t=763s) The assumption that new industries will rescue displaced labor fails because the vast majority of workers remain concentrated in legacy sectors, with newly invented job categories not appearing until rank 33 by workforce size. - [12:43](https://www.youtube.com/watch?v=7Pq-S557XQU&t=763s) Automation is an unavoidable economic engine that produces abundance, creating an urgent societal need to prepare for a world where large shares of the population are economically unemployable.

## Detailed walkthrough ### [00:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=3s) Mechanical Muscles to Mechanical Minds

Throughout early human history, every single individual had to hunt or gather merely to stay alive. Driven by a desire to make labor easier, humans systematically invented tools, moving from basic sticks to agricultural plows and eventually to modern motorized tractors. This progression transformed society from one where virtually everyone had to produce food to an age of modern agriculture where almost no one farms, yet food exists in massive abundance. For thousands of years, society has engineered tools to alleviate physical exertion. These early inventions served as mechanical muscles—machines that proved significantly stronger, vastly more reliable, and far more tireless than biological human bodies could ever hope to be.

Far from being catastrophic, replacing physical human labor with mechanical muscles historically benefited humanity. Displacing physical toil enabled the labor force to specialize into new fields, raising overall standards of living and growing economies so that even those remaining in physical roles were left better off. Today, however, that specialization has produced engineers and computer programmers whose purpose is not to augment physical capacity, but to construct mechanical minds. Just as physical machines reduced the demand for human muscle labor, mechanical minds are now systematically diminishing the demand for human cognitive labor. CGP Grey points out that we are entering an unprecedented disruption: “This is an economic revolution. You may think we've been here before, but we haven't. This time is different.”

```mermaid flowchart TD accTitle: Historical Progression of Labor Automation accDescr: Flowchart illustrating the transition from agricultural physical muscle to cognitive mechanical minds.

subgraph Muscle ["Phase 1: Physical Muscle"] A["Hunting & Gathering"] --> B["Sticks & Plows"] B --> C["Tractors & Modern Agriculture"] C --> D["Mechanical Muscles (Physical Labor Relieved)"] end

subgraph Mind ["Phase 2: Cognitive Labor"] D --> E["Specialization into Programming & Engineering"] E --> F["Mechanical Minds (Cognitive Labor Relieved)"] F --> G["Economic Revolution (Decline in Demand for Human Brain Labor)"] end ```

[01:05](https://www.youtube.com/watch?v=7Pq-S557XQU&t=65s) Physical Labor and General-Purpose Robots

Traditional industrial automation is typically visualized as massive, custom-built, exorbitant, and blind factory arms. While undeniably terrifying in scale and remarkably fast, these older robots never took over the entire global labor force because they were intensely rigid, required millions of dollars to implement, needed expert operators, and were only cost-effective in extremely narrow manufacturing contexts. The modern landscape, however, has transitioned to a completely different paradigm embodied by robots like Baxter. Unlike fixed machinery, Baxter possesses optical vision, costs less than the single annual salary of a standard human worker, and can be programmed simply by moving its arms so it can learn a task by direct demonstration.

Baxter represents the advent of general-purpose robotics. Just as early computing evolved from rare, custom-built mainframe machines into ubiquitous general-purpose microcomputers that transformed every industry through swappable software, general-purpose robotics can perform any physical task within physical reach. Even if a robot like Baxter operates at a fraction of human physical speed, it operates at pennies of electricity per hour against human workers earning minimum wage, meaning that being a tenth as fast is still wildly cost-effective when it costs one-hundredth of the price. Simpler automation has already slashed jobs: in supermarkets, single attendants oversee banks of thirty automated checkout registers where thirty workers once stood, and networked commercial barista systems are actively replacing human baristas worldwide by remembering customer preferences across thousands of sites. As Grey emphasizes: “We think of technological change as the fancy new expensive stuff, but the real change comes from last decade's stuff getting cheaper and faster.”

- **Baxter vs. Legacy Robotics**: Legacy units require millions of dollars, pre-programmed code, and dedicated technicians; Baxter operates with optical vision, costs less than an entry-level worker's annual pay, and learns via guided movement. - **Economic Math**: Even running at one-tenth the speed of human labor, a robot whose operating cost is mere pennies of electrical power per hour easily out-competes human minimum-wage staffing. - **Retail and Service Examples**: Modern supermarket configurations consolidate 30 human checkout cashiers down to a single human supervising 30 self-service terminals.

[03:31](https://www.youtube.com/watch?v=7Pq-S557XQU&t=211s) The Parable of the Luddite Horses

To understand why human labor is vulnerable, consider an imaginary conversation between two working horses in the early twentieth century. One horse nervously frets that the sudden arrival of mechanical engines and automobiles will soon render equine labor entirely obsolete. The second, more optimistic horse dismisses these concerns by pointing to history: past technological developments always improved life for horses. Farm work was backbreaking, galloping across continents delivering mail was exhausting, and charging into warfare was lethal. In the eyes of the optimistic horse, relocating to urban centers offered cushy work, and with human cities expanding, there would surely be more equine employment than ever before, alongside unimaginable new jobs created by the automobile.

```mermaid flowchart LR accTitle: The Equine Population Parable accDescr: The historical arc of equine labor peaking in 1915 followed by terminal decline.

A["Historic Hardship (Plowing, War, Mail Delivery)"] --> B["Urban Migration (Cushy City Transport)"] B --> C["1915 Population Peak ('New Jobs' Fallacy)"] C --> D["Rapid Displacement by Mechanical Muscles"] D --> E["Permanent Economic Irrelevance"] ```

Modern viewers know that the optimistic horse was entirely mistaken. The equine population reached its absolute historical peak in 1915, entering an irreversible decline as the internal combustion engine took over. There was no magical mechanism built into the rules of economics ensuring that newer technology would provide better, higher-paying jobs for horses. Swapping the horse for a human worker reveals the exact same flawed optimism: “As mechanical muscles pushed horses out of the economy, mechanical minds will do the same to humans.” Technology accelerates, drops in price, and improves in capability at exponential rates that human biology cannot match, and societies remain entirely unprepared for the displacement to come.

- **Equine Population Peak**: Peaked in 1915 before mechanical transportation rendered equine economic utility virtually obsolete. - **The Economic Fallacy**: The assumption that technological advancements inherently create more jobs for the biological entities they displace is contradicted by history. - **Biological Limits**: Human cognitive and physical development is bounded by biology, whereas automation evolves along exponential technological cost-and-performance curves.

[04:57](https://www.youtube.com/watch?v=7Pq-S557XQU&t=297s) Automobiles and Transportation Automation

Autonomous vehicles are no longer a distant theoretical concept; they are already navigating real-world roads, having completed hundreds of thousands of accident-free test miles along the California coastline and across complex urban streets. Autonomous vehicles do not have to achieve absolute perfection to dominate the market—they simply have to perform better than biological humans. Human motorists currently cause approximately 40,000 traffic fatalities every single year in the United States alone. Unlike humans, automated systems never text while driving, do not drift off to sleep, do not blink, and do not make reckless decisions behind the wheel.

Calling these systems "self-driving cars" understates their scope in the same way calling early automobiles "horseless carriages" missed the transformation of transportation. A more accurate term is "Autos"—autonomous logistical solutions engineered to solve the fundamental problem of moving an object from Point A to Point B. These range from tiny units scurrying through logistics warehouses to colossal automated dump trucks traversing open-pit mines. In the United States, roughly 3 million people work directly in transportation; scaled internationally, this accounts for at least 70 million livelihoods. While some argue that organized labor unions will halt the transition, history demonstrates that labor movements cannot withstand insurmountable economic incentives. Wages represent roughly one-third of overall operating costs for transport firms, not including the losses incurred from mandatory human driver sleep schedules, careless driving, and vehicle collisions. Grey states the reality bluntly: “Economics always wins and there are huge incentives across wildly diverse industries to adopt autos.”

- **Human Driving Fatalities**: Human drivers cause roughly 40,000 vehicular deaths annually in the United States alone. - **Transportation Labor Scale**: Transportation employs roughly 3 million American workers, translating to at least 70 million transportation jobs globally facing direct disruption. - **Corporate Incentives**: Driver salaries constitute approximately 33% of a carrier's costs; switching to automated hauling eliminates mandatory rest breaks, driver error, and crash liabilities. - **Insurance Perspective**: Insurers favor autonomous fleets because the ideal policyholder consistently pays premiums without ever crashing.

[07:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=423s) The Shape of Things to Come: White-Collar Bots

A common defense against automation anxieties suggests that machines only eliminate low-skilled physical labor that workers shouldn't be doing anyway, freeing people to move into educated, white-collar professions. Yet even setting aside the logistical impossibility of retraining over a hundred million workers for advanced intellectual careers, white-collar office work offers zero refuge. For anyone whose daily responsibilities consist of sitting in front of a computer monitor typing and clicking, software bots represent an immediate economic competitor. Software bots carry neither physical fabrication costs nor hardware wear-and-tear, making them vastly faster and cheaper to deploy than industrial machines. Because white-collar personnel command higher corporate salaries, businesses face an even larger financial incentive to replace office workers than low-wage manual laborers.

Automation engineers specialize in writing code designed to eliminate these office positions. Crucially, contemporary software engineering has moved beyond manually hardcoding procedural rules; the technological frontier relies on building bots that learn on their own. By presenting algorithms with target goals and datasets of completed tasks, bots discover solutions that humans could never explicitly write instructions for. Financial markets serve as a prominent example: stock trading is overwhelmingly conducted by self-taught algorithms executing trades with other self-taught algorithms, turning the historic trading floor of the New York Stock Exchange into little more than a televised media backdrop. Furthermore, bots have already moved into journalism and report generation, producing sportsrecaps, administrative TPS reports, and financial summaries. Regarding the vulnerability of cognitive work, Grey warns: “Paper work, decision making, writing -- a lot of human work falls into that category and the demand for human mental labor in these areas is on the way down.”

- **Cost Incentives for White-Collar Automation**: Educated office employees cost companies significantly more in compensation and benefits than low-skill laborers, magnifying the cost savings realized through automation. - **Machine Learning Mechanics**: Automation programmers no longer manually write deterministic rules for every edge case; they construct machine-learning frameworks that train themselves on sample outcomes. - **Wall Street Transformation**: Equities trading has transitioned from human pit brokers to high-frequency self-taught trading algorithms, rendering physical trading floors ceremonial. - **Automated Text Generation**: Self-learning bots routinely author structured newspaper articles, corporate performance overviews, and recurring status reports without human journalists.

[09:12](https://www.youtube.com/watch?v=7Pq-S557XQU&t=552s) Professional Bots: Discovery and Legal Work

Elite professions that demand years of rigorous postgraduate education, such as the practice of law, are equally vulnerable to machine disruption. Popular perception associates legal work with dramatic courtroom arguments and jury trials, but in truth, the overwhelming bulk of a lawyer's time is dedicated to administrative groundwork: drafting standard contracts, forecasting potential litigation outcomes, and managing the labor-intensive process known as discovery. During legal discovery, vast archives of paper documents, corporate correspondence, and transaction records are examined to pinpoint hidden inconsistencies, suspect accounting entries, or pivotal evidentiary patterns.

This tedious pattern recognition is ideally suited for cognitive algorithms. In major contemporary law firms, discovery has already shifted away from human associate attorneys. The issue is not an absence of legal documentation—discovery workloads have swelled exponentially into millions of internal emails, corporate memos, and accounting spreadsheets. Specialized legal research bots systematically comb through this documentation in mere hours rather than the weeks or months required by human paralegals, outperforming human legal teams not only on cost and turnaround time, but also in structural thoroughness and error rates.

- **Core Legal Labor**: The vast majority of billable legal hours are spent parsing contracts, assessing liability risks, and conducting document discovery rather than litigating before a judge. - **Algorithmic Discovery Superiority**: Specialized software analyzes millions of legal exhibits, electronic communications, and corporate memos in hours instead of weeks, outperforming human legal associates.

[09:51](https://www.youtube.com/watch?v=7Pq-S557XQU&t=591s) Doctor Bots and High-Skilled White-Collar Automation

Beyond mundane document review and filtering emails without getting tired, artificial intelligence is pushing directly into elite professions that once seemed impervious to automation. IBM’s Watson famously demolished human champions on *Jeopardy!*, but winning game shows was merely a televised proof-of-concept. Watson’s actual day job is functioning as an elite diagnostic tool—trained to understand natural human language and produce precise, evidence-based medical diagnoses. At Memorial Sloan Kettering Cancer Center, Watson has already been put to work providing specialized guidance for lung cancer treatments. Automated diagnostic systems do not need to be flawless; they simply have to make fewer errors than their human counterparts.

```mermaid flowchart TD accTitle: Medical Diagnostic Automation Flow accDescr: Comparison of human doctor limitations versus networked doctor bot capabilities. subgraph HumanDoctors["Human Doctors"] HD1["Individual Experience Only"] HD2["Limited Drug Interaction Recall"] HD3["Fatigue & High Misdiagnosis Rates"] end subgraph DoctorBots["Networked Doctor Bots"] DB1["Collective Data from Every Bot"] DB2["Ingests Latest Global Research"] DB3["Correlates Complex Patient Histories"] end HumanDoctors -.->|"Outpaced by"| DoctorBots DoctorBots --> Outcome["Everyday Diagnostics Delivered via Smartphones"] ```

Human medicine is deeply flawed, plagued by alarming rates and severities of misdiagnosis. No single human physician can possibly track the exponential complexity of an individual’s complete medical history, let alone memorize every pharmaceutical compound and cross-interaction, especially when high-throughput research robots test thousands of novel drugs simultaneously. While human doctors only grow through their own individual case histories, software can network the collective clinical experience of every connected machine on Earth, digest up-to-the-minute global literature, and uncover statistical correlations across millions of patient records that human minds could never identify. As Grey argues: “Not all doctors will go away, but when the doctor bots are comparable to humans and they're only as far away as your phone -- the need for general doctors will be less.” Consequently, professionals, specialized knowledge workers, and low-skilled laborers all share the same mounting vulnerability.

- Watson’s clinical role at Memorial Sloan Kettering centers on evaluating patient records and recommending lung cancer therapies. - Automated laboratory research robots run parallelized testing on thousands of new chemical compounds at once, far exceeding the cognitive bandwidth of human pharmacologists. - The proliferation of software doctors directly accessible via smartphones threatens to sharply depress the baseline demand for general practitioners.

---

[11:23](https://www.youtube.com/watch?v=7Pq-S557XQU&t=683s) Creative Bots and the Limits of an Artistic Economy

For those who assume that intellectual and physical automation leaves room to retreat into human originality, Grey warns that being a “special creative snowflake” offers no safe harbor. Creativity is not divine inspiration or magic; the biological brain is an astonishingly complex physical machine, yet humanity is actively learning how to simulate its underlying mechanisms. Some assume that just as the industrial revolution relieved human muscle so labor could transition into cognitive tasks, cognitive automation will simply shift the workforce into imaginative, creative pursuits. Even granting the premise that human imagination possesses some irreducible spark, an economy cannot run on artistic expression alone. Full-time writers, poets, actors, directors, and visual artists make up an infinitesimal slice of the labor market, bound intrinsically to popularity contests where only a tiny fraction can ever earn a living wage.

As Grey flatly states: “There can't be such a thing as a poem and painting based economy.” Furthermore, algorithms are already invading the creative sphere. The musical accompaniment playing in the video itself was composed by an artificial intelligence program named Emily Howell, capable of generating an infinite stream of brand-new musical scores at zero marginal cost—scores that listeners routinely fail to distinguish from human compositions in blind trials. While philosophical debates about the nature of artificial creativity quickly turn abstract, historical precedent provides a straightforward trajectory: people once considered chess the absolute pinnacle of human intellectual creativity, right up until software decisively conquered world champions.

- Emily Howell is an algorithmic composition engine whose music proves indistinguishable from human composers under blind testing conditions. - The share of the population capable of sustaining a livelihood through creative arts is inherently restricted by the economics of audience attention and popularity. - The shift from mechanical muscle to cognitive work cannot simply repeat as a wholesale migration of the workforce into the arts.

---

[12:43](https://www.youtube.com/watch?v=7Pq-S557XQU&t=763s) The Horse Analogy and the Looming Economic Shock

Skepticism toward apocalyptic technological predictions is natural, but Grey stresses that this wave of automation is not speculative science fiction—it is already operational inside research facilities and commercial warehouses worldwide. Drawing the central historical parallel, Grey explains that horses did not vanish from the working economy because they grew lazy; they were displaced because combustion engines made them entirely unemployable. There was simply no commercial task a horse could perform that generated enough economic value to offset the price of its housing and hay. In the emerging economy, millions of capable, intelligent human beings will face the exact same predicament through no individual fault of their own.

```mermaid flowchart LR accTitle: Employment Threat Profile accDescr: Historical Great Depression unemployment compared against the vulnerable modern workforce. subgraph GreatDepression["Historical Crisis"] GD["25% Peak Unemployment"] end subgraph ModernRisk["Current Automation Exposure"] MR["Top Existing Occupations: 45% of Total Workforce"] end ModernRisk -->|"Surpasses"| GreatDepression ```

Relying on novel industries to rescue displaced workers ignores the structural realities of the labor market. While the 1776 United States census tracked only a small handful of basic occupations, the hundreds of new job categories created since then employ only a trivial slice of modern workers. Ranking today's occupations by raw employee counts reveals a chilling pattern: the vast majority of people are still employed in traditional roles—topped by transportation, retail, and service positions—that existed in recognizable forms a century ago, almost all of which are primary automation targets. A newly invented job category does not even appear until number 33 on the list. Addressing the scale of the crisis, Grey warns: “The unemployment rate during the great depression was 25%. This list above is 45% of the workforce.”

Automation itself is neither inherently evil nor avoidable; it is simply an extraordinary engine designed to generate widespread abundance with minimal human exertion. The urgent challenge is societal: humanity must reckon immediately with a world where substantial shares of the population are economically redundant, preparing for a future where, for most occupations, humans need not apply.

- The peak unemployment rate during the Great Depression was 25 percent. - The top legacy occupational categories currently targeted by working automation account for roughly 45 percent of the modern workforce. - The single largest employment sector at the top of the job rankings is the transportation industry. - On the ranking of U.S. occupations by employment size, an entirely modern job type does not appear until rank 33.

## Notable quotes > [03:31](https://www.youtube.com/watch?v=7Pq-S557XQU&t=211s) As mechanical muscles pushed horses out of the economy, mechanical minds will do the same to humans. — CGP Grey

> [11:23](https://www.youtube.com/watch?v=7Pq-S557XQU&t=683s) There can't be such a thing as a poem and painting based economy. — CGP Grey

> [01:05](https://www.youtube.com/watch?v=7Pq-S557XQU&t=65s) We think of technological change as the fancy new expensive stuff, but the real change comes from last decade's stuff getting cheaper and faster. — CGP Grey

> [09:51](https://www.youtube.com/watch?v=7Pq-S557XQU&t=591s) Not all doctors will go away, but when the doctor bots are comparable to humans and they're only as far away as your phone -- the need for general doctors will be less. — CGP Grey

> [07:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=423s) Paper work, decision making, writing -- a lot of human work falls into that category and the demand for human mental labor in these areas is on the way down. — CGP Grey

## Mentioned Baxter, CGP Grey, Emily Howell, IBM, Jeopardy!, Memorial Sloan Kettering Cancer Center, New York Stock Exchange, Watson

## Video description Support Grey making videos: https://www.patreon.com/cgpgrey

Robots, Etc:

Terex Port automation: http://www.terex.com/port-solutions/en/products/new-equipment/automated-guided-vehicles/lift-agv/index.htm

Command | Cat MieStar System.: http://www.catminestarsystem.com/capability_sets/command

Bosch Automotive Technology: http://www.bosch-automotivetechnology.com/en/de/specials/specials_for_more_driving_safety/automated_driving/automated_driving.html

Atlas Update: https://www.youtube.com/watch?v=SD6Okylclb8&list=UU7vVhkEfw4nOGp8TyDk7RcQ

Kiva Systems: http://www.kivasystems.com

PhantomX running Phoenix code: https://www.youtube.com/watch?v=rAeQn5QnyXo

iRobot, Do You: https://www.youtube.com/watch?v=da-5Uw8GBks&list=UUB6E-44uKOyRW9hX378XEyg

New pharmacy robot at QEHB: https://www.youtube.com/watch?v=_Ql1ZHSkUPk

Briggo Coffee Experience: http://vimeo.com/77993254

John Deere Autosteer ITEC Pro 2010. In use while cultivating: https://www.youtube.com/watch?v=VAPfImWdkDw&t=19s

The Duel: Timo Boll vs. KUKA Robot: https://www.youtube.com/watch?v=tIIJME8-au8

Baxter with the Power of Intera 3: https://www.youtube.com/watch?v=DKR_pje7X2A&list=UUpSQ-euTEYaq5VtmEWukyiQ

Baxter Research Robot SDK 1.0: https://www.youtube.com/watch?v=wgQLzin4I9M&list=UUpSQ-euTEYaq5VtmEWukyiQ&index=11

Baxter the Bartender: https://www.youtube.com/watch?v=AeTs9tLsUmc&list=UUpSQ-euTEYaq5VtmEWukyiQ

Online Cash Registers Touch-Screen EPOS System Demonstration: https://www.youtube.com/watch?v=3yA22B0rC4o

Self-Service Check in: https://www.youtube.com/watch?v=OafuIBDzxxU

Robot to play Flappy Bird: https://www.youtube.com/watch?v=kHkMaWZFePI

e-david from University of Konstanz, Germany: https://vimeo.com/68859229

Sedasys: http://www.sedasys.com/

Empty Car Convoy: http://www.youtube.com/watch?v=EPTIXldrq3Q

Clever robots for crops: http://www.crops-robots.eu/index.php?option=com_content&view=article&id=62&Itemid=61

Autonomously folding a pile of 5 previously-unseen towels: https://www.youtube.com/watch?v=gy5g33S0Gzo#t=94

LS3 Follow Tight: https://www.youtube.com/watch?v=hNUeSUXOc-w

Robotic Handling material: https://www.youtube.com/watch?v=pT3XoqJ7lIY

Caterpillar automation project: http://www.catminestarsystem.com/articles/autonomous-haulage-improves-mine-site-safety

Universal Robots has reinvented industrial robotics: https://www.youtube.com/watch?v=UQj-1yZFEZI

Introducing WildCat: https://www.youtube.com/watch?v=wE3fmFTtP9g

The Human Brain Project - Video Overview: https://www.youtube.com/watch?v=JqMpGrM5ECo

This Robot Is Changing How We Cure Diseases: https://www.youtube.com/watch?v=ra0e97Wiqds

Jeopardy! - Watson Game 2: https://www.youtube.com/watch?v=kDA-7O1q4oo

What Will You Do With Watson?: https://www.youtube.com/watch?v=Y_cqBP08yuA

Other Credits

Mandelbrot set: https://www.youtube.com/watch?v=NGMRB4O922I&list=UUoxcjq-8xIDTYp3uz647V5A

Moore's law graph: http://en.wikipedia.org/wiki/File:PPTMooresLawai.jpg

Apple II 1977: https://www.youtube.com/watch?v=CxJwy8NsXFs

Beer Robot Fail m2803: https://www.youtube.com/watch?v=N4Lb_3_NMjE

All Wales Ambulance Promotional Video: https://www.youtube.com/watch?v=658aiRoVp6s

Clyde Robinson: https://www.flickr.com/photos/crobj/4312159033/in/photostream/

Time lapse Painting - Monster Spa: https://www.youtube.com/watch?v=ED14i8qLxr4

Music: Melodica Demo by Kevin McCloud, CC-BY

[Source](https://www.youtube.com/watch?v=7Pq-S557XQU)

中文简报

机器人与机器学习威胁现代45%就业岗位正如机械引擎在1915年后使马匹在经济上被淘汰一样，认知软件和机器人技术将在不创造替代工作的情况下，永久压低45%职业对人类劳动力的需求。 CGP Grey指出，机器人和人工智能领域的现代进展代表了一场不同于以往工业转型的经济革命。历史上的机器增强或取代的是人类的体力，使劳动者得以专门从事脑力劳动；而新兴的“机械大脑”正在交通运输、零售、法律、医疗和创意艺术等领域主动取代人类的认知劳动。Grey以1915年后役用马在历史上遭到淘汰作为警示对比，论证了市场激励机制系统性地偏向不知疲倦、高效且具有自主学习能力的算法，而非人类劳动力。由于绝大多数现代就业仍集中在目前正被自动化针对的职业中，社会正面临一场结构性失业危机，其严重程度可能超过大萧条时期的失业水平。与历史上用机器取代肌肉力量的工业革命不同，现代自动化正在打造“机械大脑”，直接削减了经济对人类认知劳动的需求。 (0:03) 受司机薪酬约占运输成本三分之一这一因素驱动，自动驾驶运输威胁将颠覆美国约300万个工作岗位以及全球至少7000万个岗位。 (4:57) 法律、肿瘤学和算法金融等高技能职业与低薪体力劳动一样易受自动化冲击，法律证据开示和医疗诊断目前已在向自学型软件系统转变。 (9:12) 目前直接受到自动化技术威胁的职业领域约占现代劳动力的45%——几乎是大萧条时期25%最高失业率的两倍。 (12:43)

Reading summary

CGP Grey argues that just as mechanical engines permanently displaced working horses from the economy after 1915, the rapid development of mechanical minds and general-purpose robotics threatens to cause unprecedented, structural unemployment for biological human labor. The Equine Fallacy and Cognitive Automation (0:03) Historically, technological advancement substituted mechanical muscles for human physical effort, which allowed displaced workers to specialize into cognitive fields and elevated overall living standards. Grey contends that modern computer science has fundamentally inverted this trajectory: engineers are now constructing mechanical minds designed to systematically diminish the demand for human brain labor. Believing that technological progress will always generate better replacement jobs reflects the flawed optimism of working horses in the early twentieth century. Equine populations reached their peak in 1915 before the internal combustion engine rendered their labor commercially obsolete; horses were not displaced because they became lazy, but because they could no longer produce enough economic value to offset the cost of their food and housing. Grey stresses that human biology cannot compete with the exponential cost reductions and performance gains of automated systems. As mechanical minds evolve along technological curves that human biology cannot match, society faces an economic revolution that past historical precedents cannot adequately soothe. General-Purpose Robotics and Logistics Disruption (1:05) Unlike traditional industrial automation, which required millions of dollars for rigid, blind machinery, modern robotics embodies adaptable, low-cost general-purpose designs like Baxter. Baxter utilizes optical vision, costs less than an entry-level worker's annual salary, and learns tasks through physical demonstration. Even when operating at a tenth of human speed, machines running on mere pennies of electricity per hour easily out-compete minimum-wage labor. Simpler automation has already slashed jobs in commercial environments, such as supermarket banks where a single human supervisor oversees thirty automated checkout registers, or networked barista machines that automate drink preparation across thousands of sites. This economic pressure is poised to transform transportation through autonomous vehicles or "Autos." Autonomous systems do not need to be flawless to displace human workers; they simply need to outperform biological drivers, who cause approximately 40,000 traffic fatalities annually in the United States alone. With driver wages accounting for roughly one-third of shipping operating costs—excluding the financial toll of mandatory sleep schedules and vehicle collisions—employers and insurers have overwhelming economic incentives to adopt autonomous fleets, threatening roughly 3 million transportation livelihoods in the U.S. and at least 70 million globally. Machine Learning in White-Collar and Elite Professions (7:03) White-collar office work offers no permanent haven from disruption. Software bots involve neither physical fabrication costs nor hardware wear-and-tear, making them cheaper and faster to deploy than factory robots. Because educated office staff earn higher corporate salaries than manual laborers, firms face an even stronger financial incentive to automate desk jobs. Rather than writing rigid procedural rules, engineers increasingly build machine-learning models that learn independently from target data. This transition is evident in financial markets, where high-frequency self-taught algorithms execute trades and relegate physical trading floors to media backdrops, as well as in automated corporate report writing and sports journalism. Specialized professions demanding graduate training face identical vulnerabilities. In the legal sector, billable work is heavily concentrated in contract review and legal discovery. Specialized research software combs through millions of corporate emails and spreadsheets in hours instead of the weeks needed by human paralegals, outpacing human teams in both cost and error reduction. Similarly, in healthcare, diagnostic platforms like IBM's Watson have evaluated records and advised on lung cancer therapies at Memorial Sloan Kettering Cancer Center. Unlike individual human physicians, who are constrained by fatigue and limited recall of pharmaceutical interactions, networked software pools the clinical insights of every connected system, pointing toward a future where smartphone-accessible diagnostics sharply reduce the baseline demand for general practitioners. Creative Limits and Structural Economic Shocks (11:23) Grey dismisses the assumption that workers can broadly transition into the arts or creative careers once cognitive labor is automated. True full-time artists, writers, and performers represent an infinitesimal fraction of the economy governed by popularity dynamics where only a few earn a living wage; a national economy cannot survive on poetry and painting alone. Furthermore, artificial intelligence is actively making inroads into creative fields. Algorithms like Emily Howell generate original musical compositions at zero marginal cost that blind listeners routinely cannot distinguish from human work. The scale of this labor displacement threatens severe economic instability. While hundreds of specialized job titles have been created since the 1776 U.S. census, a modern job title does not appear until rank 33 on the list of largest contemporary occupations. The top legacy occupations—principally transportation, retail, and service roles—account for roughly 45 percent of the modern workforce and represent the primary targets of operational automation. By comparison, peak unemployment during the Great Depression reached 25 percent. Grey concludes that while automation is inherently designed to create abundance with minimal human exertion, society is wholly unprepared for the resulting economic shock, which will leave large portions of the population economically redundant as machine capabilities outstrip human labor.

精炼阅读摘要

CGP Grey 提出，当前正在发生的是一场前所未有的经济革命：正如过去的“机械肌肉”将马匹彻底排除出劳动力市场一样，如今能够自我学习的“机械心智”正系统性地替代人类的脑力与体力劳动，使大批劳动力面临失去经济实用价值的风险。从机械肌肉到通用机器人的成本倒逼 (0:03) 人类历史上曾通过发明工具演变为机械肌肉，将劳动力从繁重的农业体力中解放出来，促使人们转向专业化分工，提升了整体生活水平。然而，CGP Grey 指出，当今的工程师和程序员所制造的不再是强化肉体的机器，而是能够替代人类认知劳动的机械心智。过去的大型工业自动化设备造价高达数百万美元、程序固定且需要专职技术人员操作，因此应用场景非常受限；但以 Baxter 为代表的现代通用机器人打破了这一局限。Baxter 具备光学视觉，单机成本低于普通工人的年薪，且通过人工引导其手臂即可完成动作学习。通用机器人正在复刻早期计算机从专用大型机向通用微机演进的轨迹，理论上能执行物理触及范围内的任何常规任务。CGP Grey 强调，真正的技术变革来自于十年前的技术变得更便宜、更迅速。即使通用机器人的动作速度只有人类工人的十分之一，但其运行所消耗的电费每小时仅需几美分，与领取最低工资的人类员工相比具有压倒性的成本效益。这种替代已在日常服务业发生，例如部分现代超市中一名监督员即可管理多达三十台自助结账机，联网商业咖啡系统也在凭借跨店记忆偏好的能力逐步取代人工咖啡师。卢德派马匹的寓言与技术乐观主义谬误 (3:31) CGP Grey 引用了一个二十世纪初两匹工作马的虚构寓言来反驳技术乐观主义。其中一匹马担忧内燃机汽车将终结马匹劳动，另一匹马则依据历史经验反驳，坚信技术进步总会改善马的生活——过去农田耕作、跨洲送信与参与战争都极其艰苦，而迁入城市拉车更为舒适，城市扩张理应带来更多马匹工作岗位。然而事实表明，乐观的马彻底判断失误：马匹数量在 1915 年达到历史顶峰后便断崖式下跌，内燃机的普及使马匹劳动力几乎彻底丧失了经济用途。 CGP Grey 警告，现代人类工人正在重蹈这种乐观主义的逻辑谬误。经济学规律中并不存在“新技术必然会为被淘汰的生物体创造更好、更高报酬岗位”的保证。人类在认知和生理层面的进化受制于生物学极限，而自动化技术的成本与性能曲线呈指数级发展。当马匹被机械肌肉赶出经济体系，机械心智也可能以同样的方式对待人类，马匹退役并非因为懒惰，而是因为它们创造的商业价值已无法覆盖维持其生存的草料成本。交通运输与高成本白领岗位的自动化替代 (4:57) 自动驾驶系统无需做到绝对完美，只需比人类表现更稳定即可占据优势。在全美每年约四万起人为交通事故致死的背景下，自动驾驶软件不存在注意力分散、打瞌睡或鲁莽操作的问题。CGP Grey 认为将其称为“自动驾驶汽车”低估了其影响，更准确的定义是解决点对点搬运问题的“Autos”，形态涵盖物流仓储小车到露天矿场重型自卸卡车。全美约有三百万名运输从业人员，全球对应约七千万个生计。由于司机薪资占运输企业运营成本的三分之一左右，再加上强制休息、疲劳驾驶与事故损失，企业和保险机构拥有极强的动力采用自动化车队，工会抗争难以阻挡这一趋势。与此同时，高学历白领并非安全的避难所。从事键盘与屏幕交互的白领员工薪酬与福利显著高于底层体力工人，且软件机器人没有硬件磨损或物理制造成本，使企业获得更庞大的自动化替代收益。更关键的是，现代软件工程已转向让机器人自主学习的模式，只需提供目标和样本数据，算法就能自行发现人类无法穷举编码的解决方案。在华尔街，自学算法之间的自主高频交易已让实体纽交所交易大厅沦为媒体取景背景；在文字工作领域，软件也已能自动撰写体育战报、绩效报告和财务摘要，使人类文案与决策工作的需求进入下行区间。法律、医疗与创意领域的算法渗透 (9:12) 需要长年高等教育的专业领域同样受到冲击。在法律行业中，出庭辩论仅占少数，大量可计费工时被用于合同起草与证据开示阶段。随着现代商业纠纷产生数以百万计的电子邮件与账表，人工初级律师与律师助理的翻阅效率已无法应对，而专业法律研究机器人不仅能在数小时内处理海量数据，且在错误率和完整性上均优于人工团队。医疗领域中，IBM Watson 曾赢得智力竞赛，其实际任务是作为精英诊断工具应用在纪念斯隆-凯特琳癌症中心协助肺癌治疗建议。联网的医疗程序能同步全球临床经验、文献和高通量药物筛选机器人的数据，克服人类医生个人经验狭窄、遗忘复杂药物交叉作用及误诊率偏高的问题，未来通过智能手机提供日常诊断将大幅削减普通全科医生的需求。退守至人类独创性或艺术领域的设想同样受到质疑。CGP Grey 指出，大脑虽然高度复杂，但本质依然是物理机器，其工作机制正被逐步模拟。该视频的配乐便由人工智能 Emily Howell 创作，在双盲测试中听众通常无法将其与人类作品区分开来。更现实的阻碍在于宏观经济结构，全职作家、画家与音乐家依靠受众注意力竞争生存，仅占劳动力市场的极小比例。CGP Grey 直言不可能建立一个完全以诗歌和绘画为基础的经济体，将人类心智全面迁往创意产业无法化解就业危机。 CGP Grey 强调自动化技术不是科幻假设，其基础设施已经在现实机构与仓储中运转。大萧条时期的失业率峰值为 25%，而当前自动化技术所直接瞄准的传统骨干职业（以运输、零售和服务业为主）占全美劳动力的 45% 左右，完全新型的工种在按就业人数排序的榜单中直到第 33 位才出现。自动化本身旨在不依赖人类繁重劳力即可创造普遍丰裕，但当大量人口面临失去经济实用价值的现实时，社会必须直面并尽快化解这种制度性冲击。

Detailed notes — zh

## Background 该视频由知名科普频道 CGP Grey 于 2014 年 8 月 13 日发布在 YouTube 平台上。视频简介提供了大量关于各类自动化机器人、无人驾驶与人工智能系统（如 Baxter、Watson、Kiva 等）的外部参考链接，旨在佐证当前自动化浪潮对人类劳动力市场的全面冲击。 ## TL;DR 视频通过回顾人类制造“机械肌肉”替代纯体力的历史，指出如今“机械大脑”的诞生正带来前所未有的经济革命。如同 20 世纪初内燃机将马匹彻底挤出经济体系一样，新一代廉价通用机器人、自动驾驶车辆（Autos）和具备自我学习能力的软件智能体，正迅速蚕食从蓝领运输、零售到白领文书、法律、医疗乃至艺术创意的全方位岗位。作者强调这绝非虚构的科幻，面临自动化直接威胁的传统岗位占劳动力总量的 45%，社会亟需正视人类可能因非自身过错而整体丧失被雇佣价值的严峻未来。 ## Atmosphere & style 全片充斥着一种冷静克制却令人不寒而栗的清醒感。CGP Grey 采用逻辑严密、层层递进的叙事风格，以锐利的隐喻和无可辩驳的数据步步击碎观众对“人类不可替代”的盲目乐观。快节奏的剪辑与算法谱写的背景音乐相互交织，给观众带来一种目睹历史车轮不可逆转碾压而来的紧迫感与沉重宿命感。 ## Bulletin - [00:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=3s) 机械肌肉淘汰了纯体力劳动并催生了专业分工，而如今程序员正在制造“机械大脑”，让脑力劳动面临同样的替代危机。 - [01:05](https://www.youtube.com/watch?v=7Pq-S557XQU&t=65s) 通用机器人如 Baxter 的造价比普通工人的年薪更低，运营成本仅需电费，正在复制早期个人电脑席卷各行各业的普及路径。 - [04:57](https://www.youtube.com/watch?v=7Pq-S557XQU&t=297s) 全球约有 7000 万个驾驶与运输岗位正面临自动化运输工具（Autos）的直接替代，自动化无需做到完美，只要错误率低于人类即可全面普及。 - [12:47](https://www.youtube.com/watch?v=7Pq-S557XQU&t=767s) 大萧条时期的失业率峰值为 25%，而当前面临现有自动化技术直接替代威胁的传统工种已占总劳动力的 45%。 ## Key takeaways - [00:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=3s) 从农耕工具到拖拉机的演进解放了纯体力劳动，而现代软件工程师制造的“机械大脑”正使人类的智力劳动变得不再紧俏。 - [01:05](https://www.youtube.com/watch?v=7Pq-S557XQU&t=65s) Baxter 等通用机器人具备视觉学习能力且成本极低，如同 20 世纪 80 年代的计算机，拉开了体力劳动低成本替代的序幕。 - [03:31](https://www.youtube.com/watch?v=7Pq-S557XQU&t=211s) 马匹数量在 1915 年达到顶峰后断崖式下跌，经济规律从未保证新技术一定会为旧有劳动者创造出更多、更好的岗位。 - [04:57](https://www.youtube.com/watch?v=7Pq-S557XQU&t=297s) 自动化运输工具不仅限于自动驾驶乘用车，还涵盖矿区重载与仓储物流，将彻底颠覆全球数千万驾驶与物流岗位。 - [07:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=423s) 白领办公室工作并非安全避风港，由于白领员工薪酬更高，企业有更强烈的经济动机部署自我学习的软件智能体进行替代。 - [07:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=423s) 现代前沿编程已转变为让程序自主摸索掌握技能，金融交易与新闻撰写等领域已高度由算法程序主导。 - [09:12](https://www.youtube.com/watch?v=7Pq-S557XQU&t=552s) 法律行业的“证据开示”等繁琐信息检索工作已被分析智能体取代，其在处理海量文档时的准确率与效率远超人类律师团队。 - [09:51](https://www.youtube.com/watch?v=7Pq-S557XQU&t=591s) 医疗机器人如 IBM Watson 能实时吸取全球海量医学文献与跨机器病例数据，常规全科医生的市场需求将面临大幅萎缩。 - [11:23](https://www.youtube.com/watch?v=7Pq-S557XQU&t=683s) 创造力并非无法解析的魔法，算法作曲家 Emily Howell 的作品已在双盲测试中让听众无法分辨，人类独占的智力高地正被逐步攻破。 - [11:23](https://www.youtube.com/watch?v=7Pq-S557XQU&t=683s) 艺术创意产业属于无法容纳海量人口的“注意力经济”，人类社会不可能仅依靠写诗和画画构建就业基本盘。 - [12:47](https://www.youtube.com/watch?v=7Pq-S557XQU&t=767s) 主流职业榜单排名前列的工种大部分形成于百年前，指望占比极低的新兴科技工种吸纳庞大的失业人群并不切实际。 - [12:47](https://www.youtube.com/watch?v=7Pq-S557XQU&t=767s) 危机爆发并不需要等到全员失业，只要现存 45% 的易替代岗位受到冲击，社会失业率就可能超越大萧条时期的 25% 水平。 ## Detailed walkthrough ### [00:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=3s) 机械肌肉与机械大脑的交接从人类起源开始，每个人都必须狩猎或采集才能勉强生存。但人类天生“聪明地偷懒”，因此不断制造工具来减轻劳作负担。从最初的木棍、犁具一路发展到现代拖拉机，人类社会完成了一次不可思议的转变：原本全员皆农的社会形态，变成了如今极少数人从事农业即可养活全部人口，并创造出惊人的食物富足。这种转变绝不仅限于农业。过去数千年来，人类一直致力于制造各种工具以减少各类体力劳动，创造出了所谓的“机械肌肉”。机械肌肉比肉体凡胎的人类肌肉更强壮、更可靠、不知疲倦。这种淘汰人类纯体力劳动的过程是好事，因为它将人力解放出来去进行专业化分工，提升了所有人的生活水平，推动了经济增长。然而，随着分工进一步深化，一部分专业化人才变成了程序员与工程师，他们的任务是制造“机械大脑”。正如机械肌肉曾经让单纯的人类体力需求大幅下降一样，机械大脑也正在让原本专属于人类的大脑劳动变得不再紧俏。CGP Grey 强调这是一场不可逆转的经济革命，并直言： > “This is an economic revolution. You may think we've been here before, but we haven't. This time is different.” ```mermaid flowchart TD accTitle: 工具演进与自动化历史 accDescr: 展示从原始体力劳动到机械肌肉、再到分工产生程序员并制造出机械大脑的演变过程。 subgraph Past ["体力劳动革命（机械肌肉）"] A["木棍与原始耕作"] --> B["犁具与农业分工"] B --> C["拖拉机与现代农业"] C --> D["机械肌肉解放人类劳动力"] end subgraph Present ["智力劳动革命（机械大脑）"] D --> E["人力专业化分工\n（软件工程师 / 程序员）"] E --> F["构建机械大脑\n（降低对人类大脑劳动的需求）"] end ``` --- ### [01:05](https://www.youtube.com/watch?v=7Pq-S557XQU&t=65s) 通用机器人 Baxter 与体力劳动的新纪元过去人们谈及自动化时，脑海中浮现的往往是工厂里那些巨大、昂贵、高度定制化、极为高效却“极其愚笨”的机械臂。它们对周围环境和自身工作内容一无所知，虽然看起来令人生畏，但由于只能在极为狭窄的特定环境下实现成本效益，因此并未真正席卷整个世界。那是旧时代的自动化。新一代自动化的代表是 Baxter。不同于那些动辄数百万美元、需要高水平操作员和专业技术团队调校的传统机械臂，Baxter 自带视觉系统，只需要看着人类示范动作，就能学会该如何操作。更关键的是，Baxter 的造价比一名人类普通工人的年薪还要低。它不是为了某项单一任务预先编程的专用机，只要是在双臂触及范围内的活计，它几乎都能胜任——它是一台“通用机器人（general purpose robot）”。这种通用属性具有颠覆性，正如早期的计算机一样：最初的计算机昂贵且高度定制化，但当相对廉价的通用计算机登场后，它们凭借更换软件就能处理从计算找零、分配飞机座位到运行游戏等任何任务的能力，迅速渗透到了社会的方方面面，而巨大的市场需求反过来推动计算机以极快速度变得更强大、更便宜。当下的 Baxter 就相当于 20 世纪 80 年代的计算机，它绝非技术的顶点，仅仅是一个起点。即使 Baxter 动作笨拙缓慢，但它的运行成本每小时仅需几美分的电费，而肉体凡胎的工人却必须支付最低工资；当成本只有百分之一时，即使工作速度只有人类的十分之一，在商业上也极具成本效益。事实上，比 Baxter 笨拙得多的设备早就在取代人类岗位： - 在现代化超市中，过去需要 30 名收银员的场景，如今只需 1 名员工看管 30 台自助收银机。 - 全球数十万咖啡师正面临专用咖啡机网络的冲击；虽然部分顾客迷恋某位咖啡师调制的特定配方，但数以百万计的普通大众只想要一杯合格的咖啡，而联网咖啡机器人不仅能记住你的身份，还能在任何地方端出完全符合你口味的饮品。 - 真正的技术变革不在于昂贵的新奇玩意，而在于上一个十年的技术正在以惊人的速度变得更便宜、更快速。 --- ### [03:31](https://www.youtube.com/watch?v=7Pq-S557XQU&t=211s) 卢德派马匹的寓言 CGP Grey 虚构了一场发生在 20 世纪初两匹马之间的对话。其中一匹马满怀担忧，害怕新出现的各种机械肌肉（内燃机、拖拉机、汽车）会让马匹彻底失去存在的必要。而另一匹乐观的马则反驳道，历史证明所有的技术进步都让马的生活变得更轻松：想想过去那些繁重的农田拉犁、横穿大陆递送邮件的奔波、或是冲入战场出生入死，那才叫悲惨；如今城市里的马车拉车工作惬意得多，而且城市里涌入了这么多人类，对马匹的需求必然空前高涨；即便那个叫“汽车”的古怪玩意儿普及开来，也一定会为马匹创造出人类和马现在都想象不出的全新岗位。身处现代的人类当然知晓故事的结局：现代虽然仍有极少数工作马，但辉煌早已不再。马匹的数量在 1915 年达到顶峰，此后便是一路断崖式下跌。CGP Grey 借此指出了许多人在思考就业时常犯的一个逻辑盲区： > “There isn’t a rule of economics that says better technology makes more better jobs for horses.” 如果把这句关于马的荒谬论调套用到人类身上，即断言“更先进的技术一定会为人类创造出更多、更好的岗位”，人们却往往觉得理所当然。机械肌肉曾将马匹彻底挤出经济体系，而机械大脑也将以同样的方式把大量人类挤出劳动市场。这种替代不会一蹴而就，但其规模与速度将超出人类的适应能力。生物学的进化速率根本无法与指数级提升、价格越来越便宜的机器相提并论。 --- ### [04:57](https://www.youtube.com/watch?v=7Pq-S557XQU&t=297s) 自主行驶工具（Autos）消灭运输就业自动驾驶汽车绝非遥不可及的未来，它们已经切实行驶在现实世界中，在加利福尼亚海岸线和繁华城市街道上无人干预地穿梭了数十万英里。自动驾驶不需要做到绝对完美，只要比人类驾驶员更出色即可——光是在美国，人类司机每年就会引发车祸导致 40,000 人丧生。机器从不眨眼，开车时不发短信，不会犯困，也不会做出愚蠢冲动的决策，在安全性上击败人类轻而易举。继续把自动驾驶车辆称作“汽车”，就像当年把汽车叫做“机械马”一样狭隘，会严重限制人们对其应用维度的想象。Grey 建议将其统称为“Autos（自动化运输工具）”——它们本质上是“将物体从 A 点移动到 B 点”的终极解决方案： - 人类体型的 Autos 用于运送乘客； - 微型 Autos 在仓储物流系统中穿梭运转； - 巨型 Autos 则可以直接部署在露天矿场中执行高危重载作业。整个交通运输行业正面临灭顶之灾。在美国，仅运输业就直接雇佣了大约 300 万人；若将这一数字推算至全球，至少意味着 7000 万个驾驶与运输岗位即将灰飞烟灭。面对这种剧变，CGP Grey 直截了当地宣称： > “These jobs are over.” 有人寄希望于工会抗争来阻挡这一进程，但回溯历史，工人反抗替代技术的斗争从未取得过最终胜利，经济规律永远是赢家。企业有着巨大的财务动力去拥抱 Autos：在许多运输公司中，人类劳动力直接占到了运营总成本的三分之一左右；这还只是账面工资，长途卡车司机在途中的睡觉休息时间在消耗成本，疏忽大意引发的事故更在耗费巨资。就连保险公司也会极力推动自动驾驶，因为它们最梦寐以求的客户就是那些按时缴纳保费、却永远不会出车祸的机器驾驶员。 --- ### [07:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=423s) 白领阶层与能自我学习的软件智能体许多人盲目乐观地认为，汽车和 Baxter 不过是在淘汰本就不该由人去干的低技能脏活累活，被替换掉的工人只需接受再教育、从事更高智力的白领工作即可，就像历史上一贯演变的那样。但暂且不论如何将上亿劳动力全部送入高等教育这一难题，白领办公室工作本身就不是避风港。如果你的日常工作只是整天坐在屏幕前打字、点击鼠标，那么软件智能体（Software Bots）正为你而来。从企业成本核算的角度看，白领员工往往比蓝领工人更昂贵、数量更庞大，因此企业对白领工作进行自动化的经济动机反而远高于低技能劳动力。大批“自动化工程师”的核心职责，就是编写软件脚本来取缔普通员工的日常职能。更具颠覆性的是，前沿技术并非依靠顶尖工程师去硬编码替代规则，而是让程序员构建具有自我学习能力的智能体，让它们学会连程序员自己都不知道该如何实现的操作。CGP Grey 总结道： > “The cutting edge of programming isn't super-smart programmers writing bots, it's super-smart programmers writing bots that teach themselves how to do things the programmer could never teach them to do.” 哪怕仅仅设定一个目标而完全不提供实现步骤，智能体也能自我摸索并掌握技能。 - 股票金融市场在很大程度上已经脱离了人类掌控：整个市场基本上是由自我学习的交易程序在与其他自我学习的交易程序互相对抗与博弈，纽约证券交易所如今的交易大厅与其说是工作场所，不如说更像一个供电视转播录制的摄影棚背景板。 - 智能体已学会了文字创作：媒体报刊上的新闻报道很多已出自算法之手，市场上存在专门开发自动写作智能体的公司，无论是体育快讯、日常业务报告还是企业季度财务总结，算法均能胜任。文书处理、日常决策和文案撰写等大量传统脑力劳动的市场需求正在剧烈下滑。 --- ### [09:12](https://www.youtube.com/watch?v=7Pq-S557XQU&t=552s) 专业技术岗位的防线失守：法律行业的取证变革即便连看似高不可攀的传统精英专业岗位（如律师），也难以阻挡软件智能体的蚕食。人们提起律师往往会联想到法庭上唇枪舌剑的辩护，但法律行业绝大部分的实际工作量，主要集中在起草法律文书、预测诉讼走向及影响，以及一项被称为“证据开示（Discovery）”的环节。在证据开示阶段，海量的文档、合同和票据会被倾倒给律师团队，律师们必须在成箱的材料中大海捞针，寻找蛛丝马迹、特定交易模式或违规操作。这项任务如今在许多律所已经不再属于人类工作： - 纸质与电子材料的数量比以往任何时候都要庞大； - 聪明的调研分析智能体能够在数小时内通读、过滤数百万封电子邮件、备忘录与财务账单，而传统的人类律师团队往往需要耗费数周时间； - 智能体不仅在耗时和金钱成本上对人类调查员形成了碾压之势，而且在检索准确性与系统性上也远超容易疲劳的肉体大脑。 ### [09:51](https://www.youtube.com/watch?v=7Pq-S557XQU&t=591s) 专业领域的智力替代：医疗机器人与 Watson 机器人在专业信息处理上拥有人类不可比拟的优势，那就是专注度与准确性——它们在翻阅上百万封电子邮件时绝不会打瞌睡。但这仅仅是极其基础的应用。IBM 开发的机器人 Watson 曾因在智力竞赛节目《危险边缘》（Jeopardy!）中击溃人类选手而声名大噪，但这不过是它的一项业余消遣。Watson 真正的“本职工作”是成为世界上最顶尖的医生：它能够听懂人类以日常语言陈述的病情，并给出精准的诊断结果。事实上，Watson 已经在纪念斯隆-凯特琳癌症中心（Memorial Sloan Kettering Cancer Center）实际投入使用，为肺癌治疗方案提供专业指导。面对医疗自动化，人们往往苛求其毫无瑕疵，然而事实并非如此。CGP Grey 明确指出：“Just as Auto don’t need to be perfect -- they just need to make fewer mistakes than humans -- the same goes for doctor bots.”（正如自动驾驶汽车不需要完美无瑕，它们只需要比人类犯更少的错误即可——医生机器人也是同一个道理。）人类医生绝非无懈可击，其误诊的频率与严重程度令人心惊胆战；此外，人类医生的认知带宽在面对复杂病史时显得极度受限。现代医学中，要理解每一种药物以及药物之间错综复杂的相互作用，早已超出了人类大脑的认知范畴，尤其是当自动化研究机器人已经能够同时测试成千上万种全新药物时。人类医生只能通过自身的有限经验缓慢精进，而医疗机器人却能瞬间吸取全球每一台医疗机器人的集体经验。它们能时刻跟进全球最新的医学科研进展，追踪分布在世界各地的所有患者情况，并挖掘出人类根本不可能发现的潜在关联。尽管这并不意味着所有人类医生都会彻底消失，但当医疗机器人的水平足以比肩人类，且只要掏出手机就能触手可及时，社会对全科普通医生的需求势必会大幅萎缩。从低技能工人、白领职员到专业技术人士，所有人都面临着自动化的威胁。 ```mermaid flowchart TD accTitle: 医疗机器人与人类医生能力对比 accDescr: 对比人类医生受限的单体学习模式与医疗机器人连接全球数据、跨机器经验共享的智能优势 subgraph Human["人类医生局限"] H1["单体经验积累缓慢"] H2["药物相互作用认知受限"] H3["误诊率与严重程度高"] end subgraph Bot["医疗机器人优势"] B1["跨机器全球实时经验共享"] B2["海量研究文献与药物实验即时消化"] B3["算法识别全球患者隐性关联"] end Human -.->|"被逐步替代"| Bot ``` - **司法与事务处理**：机器人在审核数以百万计的电子邮件等繁重工作中具备绝对准确性，永不疲劳。 - **临床实践**：Watson 在纪念斯隆-凯特琳癌症中心为肺癌治疗提供决策支持，展示了处理非结构化自然语言与临床指导的能力。 - **药物研发**：自动化实验室机器人已能同时测试数千种药物，信息产出速度彻底碾压个体医生的阅读上限。 --- ### [11:23](https://www.youtube.com/watch?v=7Pq-S557XQU&t=683s) 破除“特殊雪花”神话：创意领域的自动化部分专业人士或许自恃拥有独特的创造力，自封为“独特的创意雪花”而对这波浪潮不以为意。CGP Grey 直言不讳地打破了这种幻想：“Creativity may feel like magic, but it isn't.”（创造力感觉上像是魔法，但它并不是。）人类大脑确实是一台极其复杂的机器——或许是全宇宙最精密的机器——但这并未阻挡人类尝试去模拟它的步伐。过去有一种观点认为，正如“机械肌肉”解放了人类体力并推动人类转向脑力思考岗位一样，“机械大脑”也将促使人类转向纯粹的创意类工作。然而，即便退一步假定人类心智真的具有某种不可思议的创造力，支撑整个经济社会的也绝不是艺术创意。在现实劳动力市场中，能够单凭创作维生的作家、诗人、导演、演员和艺术家所占的比例微乎其微。鉴于这类职业高度依赖关注度与人气，它们在总人口中的占比注定极为有限——世界上根本不可能存在以“写诗和画画”为核心支柱的经济体。更具讽刺意味的是，自动化已经踏入了艺术殿堂。视频中所播放的背景音乐，正是由一个名为 Emily Howell 的算法机器人谱写的。她不仅能够全天候、免费创作海量全新音乐，而且在双盲测试中，听众根本无法分辨她的作品与人类作曲家之间的区别。CGP Grey 总结道：“People used to think that playing chess was a uniquely creative human skill that machines could never do right up until they beat the best of us. And so it will go for all human talents.”（人们过去曾认为下国际象棋是人类独有的创造性技能、机器永远无法胜任，直到机器把我们最优秀的棋手彻底击溃；人类的所有其他才能，未来也都将步此后尘。） - **经济结构现实**：艺术创作属于“注意力经济”，无法像制造业或服务业那样吸纳数以亿计的普通劳动力。 - **算法作曲实例**：Emily Howell 证明了机器人可以以零边际成本无限生成高质量艺术作品，并通过盲测验证。 - **认知边界退守**：从国际象棋被攻克开始，被人类视作“纯人类独占”的智力与创意高地正被算法逐步瓦解。 --- ### [12:47](https://www.youtube.com/watch?v=7Pq-S557XQU&t=767s) 历史的宿命：当人类成为无法被雇佣的“马” 面对这一严峻现实，人们很容易因为过往无数荒谬的末日预言而陷入冷嘲热讽与习惯性否认。但必须清醒认识到，这绝不是科幻小说：自动化设备目前已经真实运转在世界各地的实验室与仓储物流中心。人类历史上固然经历过多轮经济变革，但当前的机器人革命在本质上截然不同。马匹之所以在现代社会失去工作，并不是因为它们作为一个物种变得懒惰了，而是因为它们变得“彻底不具备雇佣价值”（unemployable）——马匹能做的事情太少，其产出早已无法覆盖饲料与马厩的成本。成千上万聪明能干的优秀人类，很快就会发现自己沦为了新时代的“马匹”，因完全非自身过错的原因而变得毫无雇佣价值。指望“新兴职业”拯救就业同样缺乏现实支撑。1776 年美国人口普查时仅统计了极少数几种职业；如今虽然存在数百类职业，但新兴职业在整个劳动力大军中所占的比例微乎其微。纵观按从业人数排序的职业榜单，排在榜首的是体量庞大的运输业，顺着榜单往下看，绝大多数行业在一百年前就已经存在，且几乎全都在现代自动化的精准打击范围之内。整个榜单直到第 33 位，才终于出现了一种真正意义上的“现代新兴职业”。危机爆发并不需要等到每一位咖啡师或白领彻底失业。CGP Grey 警告道：“The unemployment rate during the great depression was 25%.”（大萧条时期的失业率是 25%。）而如今处于自动化直接威胁下的传统行业从业者，占到了总劳动力的 45% 之多。仅仅是依靠今天已经成熟运行的技术，就足以在短时间内把失业率推向超越大萧条的水平。自动化本身并不是为了作恶，它是以极低代价创造极高生产力的强大工具；真正的危机在于，整个社会必须从现在开始严肃思考：当庞大的人群非因自身过错而彻底丧失被雇佣价值时，面对一个绝大多数工作都写着“人类无需应聘”的未来，人类社会究竟该何去何从。 - **失业率参照**：1930年代美国大萧条峰值失业率为 25%；而当前面临自动化直接替代威胁的现存传统职业从业者占劳动力总量的 45%。 - **职业榜单结构**：排名前列的绝大多数工种（如运输、零售、后勤等）均形成于百年前，真正的新型技术岗位直到榜单第 33 位才初次现身。 - **核心本质**：自动化本身不可阻挡，它旨在以极低劳动成本创造丰饶物质，但社会分配模式与就业范式亟需彻底重构。 ## Notable quotes > [07:03](https://www.youtube.com/watch?v=7Pq-S557XQU&t=423s) The cutting edge of programming isn't super-smart programmers writing bots, it's super-smart programmers writing bots that teach themselves how to do things the programmer could never teach them to do. — CGP Grey > [11:23](https://www.youtube.com/watch?v=7Pq-S557XQU&t=683s) People used to think that playing chess was a uniquely creative human skill that machines could never do right up until they beat the best of us. And so it will go for all human talents. — CGP Grey > [11:23](https://www.youtube.com/watch?v=7Pq-S557XQU&t=683s) Creativity may feel like magic, but it isn't. — CGP Grey > [03:31](https://www.youtube.com/watch?v=7Pq-S557XQU&t=211s) There isn’t a rule of economics that says better technology makes more better jobs for horses. — CGP Grey ## Mentioned Baxter, Emily Howell, IBM, Watson, Memorial Sloan Kettering Cancer Center, 纽约证券交易所, 《危险边缘》（Jeopardy!）, CGP Grey ## Video description Support Grey making videos: https://www.patreon.com/cgpgrey ## Robots, Etc: Terex Port automation: http://www.terex.com/port-solutions/en/products/new-equipment/automated-guided-vehicles/lift-agv/index.htm Command | Cat MieStar System.: http://www.catminestarsystem.com/capability_sets/command Bosch Automotive Technology: http://www.bosch-automotivetechnology.com/en/de/specials/specials_for_more_driving_safety/automated_driving/automated_driving.html Atlas Update: https://www.youtube.com/watch?v=SD6Okylclb8&list=UU7vVhkEfw4nOGp8TyDk7RcQ Kiva Systems: http://www.kivasystems.com PhantomX running Phoenix code: https://www.youtube.com/watch?v=rAeQn5QnyXo iRobot, Do You: https://www.youtube.com/watch?v=da-5Uw8GBks&list=UUB6E-44uKOyRW9hX378XEyg New pharmacy robot at QEHB: https://www.youtube.com/watch?v=_Ql1ZHSkUPk Briggo Coffee Experience: http://vimeo.com/77993254 John Deere Autosteer ITEC Pro 2010. In use while cultivating: https://www.youtube.com/watch?v=VAPfImWdkDw&t=19s The Duel: Timo Boll vs. KUKA Robot: https://www.youtube.com/watch?v=tIIJME8-au8 Baxter with the Power of Intera 3: https://www.youtube.com/watch?v=DKR_pje7X2A&list=UUpSQ-euTEYaq5VtmEWukyiQ Baxter Research Robot SDK 1.0: https://www.youtube.com/watch?v=wgQLzin4I9M&list=UUpSQ-euTEYaq5VtmEWukyiQ&index=11 Baxter the Bartender: https://www.youtube.com/watch?v=AeTs9tLsUmc&list=UUpSQ-euTEYaq5VtmEWukyiQ Online Cash Registers Touch-Screen EPOS System Demonstration: https://www.youtube.com/watch?v=3yA22B0rC4o Self-Service Check in: https://www.youtube.com/watch?v=OafuIBDzxxU Robot to play Flappy Bird: https://www.youtube.com/watch?v=kHkMaWZFePI e-david from University of Konstanz, Germany: https://vimeo.com/68859229 Sedasys: http://www.sedasys.com/ Empty Car Convoy: http://www.youtube.com/watch?v=EPTIXldrq3Q Clever robots for crops: http://www.crops-robots.eu/index.php?option=com_content&view=article&id=62&Itemid=61 Autonomously folding a pile of 5 previously-unseen towels: https://www.youtube.com/watch?v=gy5g33S0Gzo#t=94 LS3 Follow Tight: https://www.youtube.com/watch?v=hNUeSUXOc-w Robotic Handling material: https://www.youtube.com/watch?v=pT3XoqJ7lIY Caterpillar automation project: http://www.catminestarsystem.com/articles/autonomous-haulage-improves-mine-site-safety Universal Robots has reinvented industrial robotics: https://www.youtube.com/watch?v=UQj-1yZFEZI Introducing WildCat: https://www.youtube.com/watch?v=wE3fmFTtP9g The Human Brain Project - Video Overview: https://www.youtube.com/watch?v=JqMpGrM5ECo This Robot Is Changing How We Cure Diseases: https://www.youtube.com/watch?v=ra0e97Wiqds Jeopardy! - Watson Game 2: https://www.youtube.com/watch?v=kDA-7O1q4oo What Will You Do With Watson?: https://www.youtube.com/watch?v=Y_cqBP08yuA ## Other Credits Mandelbrot set: https://www.youtube.com/watch?v=NGMRB4O922I&list=UUoxcjq-8xIDTYp3uz647V5A Moore's law graph: http://en.wikipedia.org/wiki/File:PPTMooresLawai.jpg Apple II 1977: https://www.youtube.com/watch?v=CxJwy8NsXFs Beer Robot Fail m2803: https://www.youtube.com/watch?v=N4Lb_3_NMjE All Wales Ambulance Promotional Video: https://www.youtube.com/watch?v=658aiRoVp6s Clyde Robinson: https://www.flickr.com/photos/crobj/4312159033/in/photostream/ Time lapse Painting - Monster Spa: https://www.youtube.com/watch?v=ED14i8qLxr4 Music: Melodica Demo by Kevin McCloud, CC-BY [Source](https://www.youtube.com/watch?v=7Pq-S557XQU)

Source quotes

[7:03] CGP Grey: The cutting edge of programming isn't super-smart programmers writing bots, it's super-smart programmers writing bots that teach themselves how to do things the programmer could never teach them to do. [11:23] CGP Grey: People used to think that playing chess was a uniquely creative human skill that machines could never do right up until they beat the best of us. And so it will go for all human talents. [11:23] CGP Grey: Creativity may feel like magic, but it isn't. [3:31] CGP Grey: There isn’t a rule of economics that says better technology makes more better jobs for horses.

Source quotes

[3:31] CGP Grey: As mechanical muscles pushed horses out of the economy, mechanical minds will do the same to humans. [11:23] CGP Grey: There can't be such a thing as a poem and painting based economy. [1:05] CGP Grey: We think of technological change as the fancy new expensive stuff, but the real change comes from last decade's stuff getting cheaper and faster. [9:51] CGP Grey: Not all doctors will go away, but when the doctor bots are comparable to humans and they're only as far away as your phone -- the need for general doctors will be less. [7:03] CGP Grey: Paper work, decision making, writing -- a lot of human work falls into that category and the demand for human mental labor in these areas is on the way down.

Original transcript

[0:03] Every human used to have to hunt or gather to survive. But humans are smart...ly lazy so [0:08] we made tools to make our work easier. From sticks, to plows, to tractors we’ve gone [0:13] from everyone needing to make food to, modern agriculture with almost no one needing to [0:18] make food — and yet, we still have abundance. [0:20] Of course, it’s not just farming, it’s everything. We’ve spent the last several [0:24] thousand years building tools to reduce physical labor of all kinds. These are mechanical muscles. [0:29] Stronger, more reliable, and more tireless than human muscles ever could be. [0:34] And that's a good thing. Replacing human labor with mechanical muscles frees people to specialize [0:38] and that leaves everyone better off - even those still doing physical labor. This is how economies [0:44] grow and standards of living rise. [0:46] Some people have specialized to be programmers and engineers whose job is to build mechanical [0:51] minds. Just as mechanical muscles made human labor less in demand so are mechanical minds [0:56] making human brain labor less in demand. [0:59] This is an economic revolution. You may think we've been here before, but we haven't. [1:03] This time is different. [1:05] ## Physical Labor [1:07] When you think of automation, you probably think of this: giant, custom-built, expensive, [1:11] efficient, but really dumb robots blind to the world and their own work. They were a [1:16] scary kind of automation but they haven't taken over the world because they're only [1:20] cost effective in narrow situations. [1:23] But they're the old kind of automation, this is the new kind. [1:27] Meet Baxter. [1:28] Unlike these things which require skilled operators and technicians and millions of dollars, [1:32] Baxter has vision and can learn what you want him to do by watching you do it. [1:37] And he costs less than the average annual salary of a human worker. Unlike his older [1:41] brothers he isn't pre-programmed for one specific job, he can do whatever work is within the [1:46] reach of his arms. Baxter is what might be thought of as a general purpose robot and [1:51] general purpose is a big deal. [1:53] Think computers, they too started out as highly custom and highly expensive, but when cheap-ish [1:58] general-purpose computers appeared they quickly became vital to everything. [2:02] A general-purpose computer can just as easily calculate change or assign seats on an airplane [2:07] or play a game or do anything just by swapping its software. And this huge demand for computers [2:13] of all kinds is what makes them both more powerful and cheaper every year. [2:18] Baxter today is the computer of the 1980s. He’s not the apex but the beginning. Even [2:23] if Baxter is slow his hourly cost is pennies worth of electricity while his meat-based [2:27] competition costs minimum wage. A tenth the speed is still cost effective when it's a [2:32] hundredth the price. And while Baxter isn't as smart as some of the other things we will [2:36] talk about, he's smart enough to take over many low-skill jobs. [2:40] And we've already seen how dumber robots than Baxter can replace jobs. In new supermarkets [2:45] what used to be 30 humans is now one human overseeing 30 cashier robots. [2:50] Or take the hundreds of thousand baristas employed world-wide? There’s a barista robot coming [2:54] for them. Sure maybe your guy makes the double-mocha-whatever just perfect and you’d never trust anyone [2:59] else -- but millions of people don’t care and just want a decent cup of coffee. Oh, and [3:05] by the way this robot is actually a giant network of robots that remembers who you are [3:09] and how you like your coffee no matter where you are. Pretty convenient. [3:13] We think of technological change as the fancy new expensive stuff, but the real change comes [3:17] from last decade's stuff getting cheaper and faster. That's what's happening to robots [3:22] now. And because their mechanical minds are capable of decision making they are out-competing [3:27] humans for jobs in a way no pure mechanical muscle ever could. [3:31] ## Luddite Horses [3:33] Imagine a pair of horses in the early 1900s talking about technology. One worries all [3:38] these new mechanical muscles will make horses unnecessary. [3:41] The other reminds him that everything so far has made their lives easier -- remember all [3:45] that farm work? Remember running from coast-to-coast delivering mail? Remember riding into battle? [3:50] All terrible. These city jobs are pretty cushy, and with so many humans in the cities there [3:54] will be more jobs for horses than ever. [3:57] Even if this car thingy takes off - he might say - there will be [4:00] new jobs for horses we can't imagine. [4:02] But you, dear viewer, from beyond 2000 know what happened -- there are still working horses, [4:08] but nothing like before. The horse population peaked in 1915 -- from that point on it was [4:13] nothing but down. [4:14] There isn’t a rule of economics that says better technology makes more better jobs [4:18] for horses. It sounds shockingly dumb to even say that out loud, but swap horses for humans [4:24] and suddenly people think it sounds about right. [4:27] As mechanical muscles pushed horses out of the economy, mechanical minds will do the [4:31] same to humans. Not immediately, not everywhere, but in large enough numbers and soon enough [4:37] that it's going to be a huge problem if we are not prepared. And we are not prepared. [4:42] You, like the second horse, may look at the state of technology now and think it can’t [4:46] possibly replace your job. But technology gets better, cheaper, and faster at a rate [4:50] biology can’t match. [4:52] Just as the car was the beginning of the end for the horse so now does the car show us [4:56] the shape of things to come. [4:57] ## Automobiles [5:02] Self-driving cars aren't the future: they're here and they work. Self-driving cars have [5:06] travelled hundreds of thousands of miles up and down the California coast and through [5:09] cities -- all without human intervention. [5:12] The question is not if they'll replaces cars, but how quickly. They don’t need to be perfect, [5:16] they just need to be better than us. Humans drivers, by the way, kill 40,000 people a [5:22] year with cars just in the United States. Given that self-driving cars don’t blink, [5:26] don’t text while driving, don’t get sleepy or stupid, it's easy to see them being better [5:30] than humans because they already are. [5:33] Now to describe self-driving cars as cars at all is like calling the first cars mechanical [5:39] horses. Cars in all their forms are so much more than horses that using the name limits [5:44] your thinking about what they can even do. Lets call self-driving cars what they really [5:48] are: [5:49] Autos: the solution to the transport-objects-from-point-A-to-point-B problem. Traditional cars happen to be human [5:55] sized to transport humans but tiny autos can work in warehouses and gigantic autos can [5:59] work in pit mines. Moving stuff around is who knows how many jobs but the transportation [6:04] industry in the United States employs about three million people. Extrapolating world-wide [6:09] that’s something like 70 million jobs at a minimum. [6:13] These jobs are over. [6:15] The usual argument is that unions will prevent it. But history is filled with workers who [6:19] fought technology that would replace them and the workers always lose. Economics always [6:24] wins and there are huge incentives across wildly diverse industries to adopt autos. [6:30] For many transportation companies, humans are about a third their total costs. That's [6:34] just the straight salary costs. Humans sleeping in their long haul trucks costs time and money. [6:39] Accidents cost money. Carelessness costs money. If you think insurance companies will be against [6:44] it, guess what? Their perfect driver is one who pays their small premiums and never gets [6:48] into an accident. [6:50] The autos are coming and they're the first place where most people will really see the [6:54] robots changing society. But there are many other places in the economy where the same [6:58] thing is happening, just less visibly. [7:00] So it goes with autos, so it goes for everything. [7:03] ## The Shape of Things to Come [7:06] It's easy to look at Autos and Baxters and think: technology has always gotten rid of [7:10] low-skill jobs we don't want people doing anyway. They'll get more skilled and do better [7:15] educated jobs -- like they've always done. [7:17] Even ignoring the problem of pushing a hundred-million additional people through higher education, [7:22] white-collar work is no safe haven either. If your job is sitting in front of a screen [7:27] and typing and clicking -- like maybe you're supposed to be doing right now -- the bots [7:31] are coming for you too, buddy. [7:32] Software bots are both intangible and way faster and cheaper than physical robots. Given [7:37] that white collar workers are, from a company's perspective, both more expensive and more [7:41] numerous -- the incentive to automate their work is greater than low skilled work. [7:46] And that's just what automation engineers are for. These are skilled programmers whose [7:51] entire job is to replace your job with a software bot. [7:54] You may think even the world's smartest automation engineer could never make a bot to do your [7:58] job -- and you may be right -- but the cutting edge of programming isn't super-smart programmers [8:03] writing bots, it's super-smart programmers writing bots that teach themselves how to [8:08] do things the programmer could never teach them to do. [8:11] How that works is well beyond the scope of this video, but the bottom line is there are [8:15] limited ways to show a bot a bunch of stuff to do, show the bot a bunch of correctly done [8:20] stuff, and it can figure out how to do the job to be done. [8:23] Even with just a goal and no knowledge of how to do it the bots can still learn. Take the [8:28] stock market which, in many ways, is no longer a human endeavor. It's mostly bots that taught [8:33] themselves to trade stocks, trading stocks with other bots that taught themselves. [8:38] As a result, the floor of the New York Stock exchange isn't filled with traders doing their [8:42] day jobs anymore, it's largely a TV set. [8:44] So bots have learned the market and bots have learned to write. If you've picked up a newspaper [8:48] lately you've probably already read a story written by a bot. There are companies that [8:53] teach bots to write anything: sports stories, TPS reports, even say, those quarterly [8:57] reports that you write at work. [8:59] Paper work, decision making, writing -- a lot of human work falls into that category [9:03] and the demand for human metal labor is these areas is on the way down. But surely the professions [9:09] are safe from bots? Yes? [9:12] ## Professional Bots [9:15] When you think 'lawyer' it's easy to think of trials. But the bulk of lawyering is actually [9:19] drafting legal documents, predicting the likely outcome and impact of lawsuits, and something [9:24] called 'discovery' which is where boxes of paperwork gets dumped on the lawyers and they [9:28] need to find the pattern or the one out-of-place transaction among it all. [9:32] This can be bot work. Discovery, in particular, is already not a human job in many law firms. [9:38] Not because there isn't paperwork to go through, there's more of it than ever, but because [9:42] clever research bots shift through millions of emails and memos and accounts in hours [9:46] not weeks -- crushing human researchers in terms of not just cost and time but, most [9:51] importantly, accuracy. Bots don't get sleepy reading through a million emails. [9:56] But that's the simple stuff: IBM has a bot named Watson: you may have seen him on TV [10:01] destroy humans at Jeopardy — but that was just a fun side project for him. [10:05] Watson's day-job is to be the best doctor in the world: to understand what people say [10:09] in their own words and give back accurate diagnoses. And he's already doing that at [10:14] Slone-Kettering, giving guidance on lung cancer treatments. [10:17] Just as Auto don’t need to be perfect -- they just need to make fewer mistakes than humans -- [10:21] the same goes for doctor bots. [10:23] Human doctors are by no means perfect -- the frequency and severity of misdiagnoses are [10:28] terrifying -- and human doctors are severely limited in dealing with a human's complicated [10:33] medical history. Understanding every drug and every drug's interaction with every other [10:37] drug is beyond the scope of human knowability. [10:40] Especially when there are research robots whose whole job it is to test thousands of new [10:45] drugs at a time. [10:47] And human doctors can only improve through their own experiences. Doctor bots can learn from [10:51] the experiences of every doctor bot. Can read the latest in medical research and keep track [10:54] of everything that happens to all their patients world-wide and make correlations that would [10:59] be impossible to find otherwise. [11:01] Not all doctors will go away, but when the doctor bots are comparable to humans and they're [11:06] only as far away as your phone -- the need for general doctors will be less. [11:10] So professionals, white-collar workers and low-skill workers all have things to worry about [11:15] from automation. But perhaps you are unfazed because you're a special creative snowflake. Well [11:21] guess what? You're not that special. [11:23] ## Creative Bots [11:27] Creativity may feel like magic, but it isn't. The brain is a complicated machine -- perhaps [11:32] the most complicated machine in the whole universe -- but that hasn't stopped us from [11:36] trying to simulate it. [11:38] There is this notion that just as mechanical muscles allowed us to move into thinking jobs [11:42] that mechanical minds will allow us to move into creative work. [11:45] But even if we assume the human mind is magically creative -- it's not, but just for the sake of argument -- [11:50] artistic creativity isn't what the majority of jobs depend on. The number of writers and poets [11:55] and directors and actors and artists who actually make a living doing their work is a tiny, [12:00] tiny portion of the labor force. And given that these are professions dependent [12:04] on popularity they'll always be a very small portion of the population. [12:08] There can't be such a thing as a poem and painting based economy. [12:12] Oh, by the way, this music in the background that you're listening to? It was written by a bot. [12:17] Her name is Emily Howell and she can write an infinite amount of new music all day for free. [12:21] And people can't tell the difference between her and human composers [12:25] when put to a blind test. [12:27] Talking about artificial creativity gets weird fast -- what does that even mean? [12:31] But it's nonetheless a developing field. [12:33] People used to think that playing chess was a uniquely creative human skill that machines [12:37] could never do right up until they beat the best of us. And so it will go for all human talents. [12:43] ## Conclusion [12:47] Right: this may have been a lot to take in, and you might want to reject it -- it's [12:51] easy to be cynical of the endless and idiotic predictions of futures that never are. So [12:55] that's why it's important to emphasize again that this stuff isn't science fiction. The robots [13:00] are here right now. There is a terrifying amount of working automation in labs and warehouses [13:05] around the world. [13:07] We have been through economic revolutions before, but the robot revolution is different. [13:12] Horses aren't unemployed now because they got lazy as a species, they’re unemployable. [13:17] There's little work a horse can do that do to pay for its housing and hay. [13:21] And many bright, perfectly capable humans will find themselves the new horse: unemployable [13:26] through no fault of their own. [13:28] But if you still think new jobs will save us: here is one final point to consider. The [13:33] US census in 1776 tracked only a few kinds of jobs. Now there are hundreds of kinds of [13:38] jobs, but the new ones are not a significant part of the labor force. [13:42] Here's the list of jobs ranked by the number of people who perform them - it's a sobering [13:46] list with the transportation industry at the top. Continuing downward, all of this work existed [13:52] in some form a hundred years ago and almost all of them are targets for automation. Only [13:58] when we get to number 33 on the list is there finally something new. [14:02] Don't that every barista or white collar worker need lose their job before things are a problem. [14:07] The unemployment rate during the great depression was 25%. [14:10] This list above is 45% of the workforce. Just what we've talked about today, the stuff that [14:17] already works, can push us over that number pretty soon. And given that even in our modern [14:22] technological wonderland new kinds of work aren't a significant portion of the economy, [14:28] this is a big problem. [14:29] This video isn't about how automation is bad -- rather that automation is inevitable. It's [14:34] a tool to produce abundance for little effort. We need to start thinking now about what to [14:39] do when large sections of the population are unemployable -- through no fault of their own. [14:44] What to do in a future where, for most jobs, humans need not apply.