What this covers
David Autor, an economist at MIT, discusses how artificial intelligence poses a different and in some ways more diffuse economic threat than previous shocks, and argues for deliberately shaping AI as a tool that amplifies human expertise rather than replacing it. The conversation—recorded with Reid Hoffman—traces the contours of this risk by drawing parallels to the China trade shock of the early 2000s, examining how labor markets adjust to major disruptions, and exploring what education and policy might look like if the goal is to extend valuable decision-making capability to people without elite credentials.
The episode covers substantial ground across economics, work, and education. Early sections establish the China shock as a useful but imperfect comparison: sudden, geographically concentrated, it devastated specialized manufacturing towns even though a million displaced workers would barely register spread across a 150-million-worker labor force. The AI shock, Autor argues, will be different—distributed across industries and occupations, often felt by firms as productivity gain rather than pure loss. The heart of the argument centers on a simple economic principle: wages track the scarcity of valuable expertise, not the social importance of work. A crossing guard prevents collisions with identical stakes as an air traffic controller, yet earns a quarter the pay because certification is scarce for one and common for the other. Automation doesn't eliminate jobs so much as make valuable skills abundant. The conversation then turns to how AI might be designed differently: as a collaboration tool that lets non-experts filter and apply AI output, versus automation that deskills workers by removing the expert portions of their tasks. The final sections address education, inequality, demographic challenges, and why labor markets—despite their flaws—remain more intrinsically distributional than systems based on capital or machines.
Autor argues that the central risk of AI is not job loss but the devaluation of scarce expertise, and that the better path is to design AI as a collaboration tool that lets more people—especially those without elite credentials—acquire and apply valuable decision-making expertise.
- Wages depend on expertise being both valuable and scarce; automation threatens pay by making valuable skills abundant, not by eliminating work
- The AI shock differs from the China trade shock in that it lacks regional concentration, won't wipe out whole industries, and is often experienced by firms as productivity-enhancing
- AI used as a collaboration tool can extend expertise to non-elites and improve the middle class, whereas automation of expert tasks deskills workers
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
A crossing guard and an air traffic controller do fundamentally the same job—preventing collisions—yet crossing guards earn less than a quarter of what air traffic controllers do, because almost no training or certification is required to be a crossing guard; this illustrates that pay tracks the scarcity of required expertise, not social value.
“think of it like a crossing guard versus an air traffic controller... those are the same job... And yet, crossing guards make less than a quarter of what air traffic controllers do... because there's almost no training or certification required to become a crossing guard.”
Labor markets have a natural adjustment rate of roughly 3% per year (a ~30-year career), so a shock spread over 25 years is manageable through retirement and non-entry, but the same displacement occurring suddenly—e.g., autonomous vehicles replacing 2-3 million drivers on a single Labor Day—would be catastrophic; the slow capital replacement cycle naturally buffers such transitions.
“labor markets have a natural rate of adjustment... 3% of people will retire out of anything... If autonomous vehicles come Labor Day this year replaced all long-distance drivers... That would be a very serious problem cuz there are more than... 3 million... people who just do their living in driving vehicles... If it happened over 25 years... that's kind of a manageable problem... it takes decades to replace all that capital”
If half the people doing leisure, hospitality, and janitorial work were extracted from it, the remaining half would get a big pay increase because firms would have to compete for them—low wages in these jobs stem from oversupply, just as barbers' wages rise over time not because they cut hair faster but because they must be compensated to remain barbers rather than do something else.
“if we just extracted half of the people who were doing... leisure and hospitality and janitorial services from that work the... half who remained would get a big pay increase... why do the wages of barbers rise over time? They're not getting any faster cutting anyone's hair... they have to be compensated to be barbers as opposed to being something else”
The labor market has two virtues a capital-based distribution lacks: it is intrinsically more equal because everyone owns exactly one worker (themselves) in a non-slavery society, with about 60% of US income being labor income that goes first to workers; and in a democracy, if most people work it is easy to argue they are contributors who deserve a vote and a share, whereas if wealth pours from machines it becomes harder to justify everyone's share.
“it's intrinsically a lot more equitable... than the capital market because everybody... owns no more than one worker... They just own themselves... 60% of the income in the United States is labor income that goes first to workers... in a democratic society... if most people are working then it's easy for people to say... they have a vote... Whereas if... all the money... comes out of a fountain... then it's much harder to say that everybody deserves their share”
Workers initially displaced by the China shock largely did not move up or out; 20 years later, many remain in relatively low-paid manufacturing or other low-paid work, while it is their children who never entered manufacturing—showing labor markets adjust across cohorts rather than mid-career.
“the workers who were initially in those locations actually have not moved on. They've not kind of moved up or out uh and many of them are still in relatively low-paid manufacturing or other low-paid work.”
We have lost more clerical jobs over the last 30 years than manufacturing jobs, yet no one talks about a 'clerical shock'—because there was never a 'clerical capital' of the US, since clerical workers existed in every industry rather than concentrated in specialized towns.
“we've lost more clerical jobs uh over the last 30 years probably than we have lost manufacturing jobs. But, no one talks about the clerical shock. Uh why not? Well, one reason is there was never a clerical capital of the United States.”
Demand effects can offset automation: getting cheaper and faster at colonoscopies won't make people line up for more, but cheaper/faster software coding meets vast latent demand (every appliance now has embedded software), so coding work may grow even as the skill set shifts—just as 1990s website-building moved from technical HTML skill to design skill.
“if we... got like really really good and cheap and fast at colonoscopies, people still wouldn't be lining up at their proctologist's office... if we get better, cheaper, faster at coding... There's a lot of demand for software... when people started developing websites... that was all about skill in HTML... Now... it's not really a technical skill, it's design”
Americans work far less than they used to—average annual hours fell from about 3,000 a year 120 years ago to about 1,900 today, with the weekend, vacations, later workforce entry, and earlier retirement meaning we work a much smaller fraction of our healthy lives—so the 'overworked society' narrative is a myth and we have handled increased leisure well.
“at the beginning of the 20th century... people worked on average in the United States about 3,000 hours a year. Uh now we work on average about 1,900 hours of the year... we've invented the weekend... people enter the labor force... 16 18 20 25... and then they retire when they have... 20 years of health remaining... That's kind of a myth.”
The last 40-50 years have been the best in human history by key welfare measures—an unprecedented global middle class, China reducing poverty from about 70% to a couple of percent (over a billion people lifted out), and strong growth and improving health in Central/South America and sub-Saharan Africa—progress that people in the West tend to overlook.
“this has been the best 40 or 50 years that humanity has ever experienced... We've never had a... global middle class until now... China has... reduced its poverty level from 70% to... a couple percentage points... more than a billion people... growth in sub-Saharan Africa has been really strong”
For expertise to have market value it must satisfy two conditions: it must produce a service people value (data science, not card tricks) and it must be scarce, because if everyone is expert no one is expert and the skill won't pay; automation threatens wages not by eliminating the need for a skill but by making the machine able to do it cheaper and faster, devaluing the human skill.
“for expertise to have market value, uh, it needs to have two things... it needs to produce a service that people value... The second thing is it needs to be scarce. Uh, because if everyone is expert, no one is expert... it can devalue a skill set very quickly, not because no one needs the skills anymore, but because the machine can do it... cheaper, faster”
Over the last 40 years computerization hollowed out skilled production and office work that followed codified rules, pushing workers who would have done that work into low-paid services (food service, cleaning, security, home health aides) that are socially valuable but require little training and therefore won't pay well—the threat of automation is not running out of work but making valuable skills abundant.
“computerization has led to a lot of hollowing out of both production work and office work... it was feasible to turn them into computer code... a lot of the people who would have been doing office work and manufacturing work 40 years ago now find themselves doing services... it's not expert work... that means it won't pay well.”
Technologies are often complementary to expertise, acting as force multipliers that shorten the distance between intention and result and give us superhuman powers; the key question for any worker is whether technology will make their work more expert or instead displace the valuable expertise they hold.
“our technologies are complementary to our expertise, and they are amplifiers, they're force multipliers for expertise because they... shorten the distance between intention and result... how do you get to be on the right side of this equation”
Geoffrey Hinton famously predicted about a decade ago that we would need no more radiologists because AI would do it better, but the opposite happened: radiologists now use and love AI, there are not fewer of them, and they do more—because much of their work is communication with patients and other caregivers, not just reading scans, illustrating the hubris of assuming superhuman technology means expertise is dead.
“Geoffrey Hinton... famously predicted about a decade ago that we would need no more radiologists... Now, AI is now used by radiologists, they love AI, but they're not fewer radiologists... a lot of their work is not just looking at scans... it's all of the communication with the patients”
The AI shock will differ from the China shock in three key ways: it will not be regionally concentrated, it will affect occupations and tasks rather than wiping out entire industries, and firms will often experience it as productivity-enhancing rather than as a pure negative competitive shock.
“I think there are three very important differences. Uh one is again that regional concentration... Two, the trade trade shock really impacted the viability of industries in themselves... the third thing is um the China trade shock was experienced by US firms as a pure negative competitive shock... AI will be experienced by many firms as productivity increasing.”
The US is already arguably the wealthiest society in history with no real scarcity, yet has many poor people without healthcare, safe housing, or good schools—not because resources don't exist but because we lack a system willing to distribute to people who don't earn it through the labor market—so more wealth alone won't solve distribution, and the US is becoming less generous even as it grows wealthier.
“we are already an incredibly wealthy society... we don't really have any real scarcity here and yet we have a lot of people who are quite poor... It's that we don't have a system where... we really want to distribute that much to people who don't somehow earn it... the US is... not getting more generous as a society even as it's getting wealthier we seem to be getting less generous”
Quality of life depends not just on low prices and material goods but on economic security and reduced uncertainty—whether rent will rise, whether a cancer diagnosis will cause medical bankruptcy—and these uncertainty problems aren't solved by lower prices; inequality also matters less for the gap itself than for whether it produces 'dynastism' where the next generation starts so far ahead that there is no fair chance.
“those aren't really sort of price issues, those are more uncertainty issues... inequality can lead to dynastism where... the next generation... starts so far ahead of the others that... there's no longer... anything close to fair chance in the next... round... those things are harder to deliver just through low prices”
MOOCs (massive open online courseware) largely failed because they were simulated classrooms—'everything you hated about a classroom and worse'—lacking the social engagement and immersion that make education effective; adults learn transitions much better experientially than by returning to classrooms.
“remember... MOOCs, massively online open courseware... they really weren't very successful... they were like... simulated classrooms... it's like everything you hated about a classroom and worse... At least a classroom is a social environment, whereas watching a video is not... adults making transitions... learn much more successfully kind of experientially than they do going back to classrooms”
The China trade shock caused a rapid, regionally concentrated loss of more than a million labor-intensive manufacturing jobs (furniture, textiles, clothing) between 2001 and 2007, devastating specialized towns even though a million jobs spread evenly across a 150-million-worker labor market would be barely noticeable.
“this in a very short order caused a loss of, you know, more than a million manufacturing jobs... that's not the way manufacturing works. It's very regionally concentrated.”
The US is not running out of jobs but running out of workers, due to low population growth, low birth rates, and heavily restricted immigration, which creates a financing challenge because a large retired population needs workers to support its standard of living.
“we're not running out of jobs. Uh we're much more running out of workers. And, uh, this is true in most industrialized countries... it also means you're going to have a large retired population that's expecting has earned a decent... standard of living in retirement, and you need workers to support that”
AI is not reliable for tasks you don't understand—it will misunderstand and lead you astray, and may fabricate plausible falsehoods you can't detect—so a key skill is knowing when you know enough to judge its output; AI works best as a collaboration tool in domains where you have enough expertise to filter and interpret what it produces.
“AI isn't really useful for things that you don't understand. Uh because it's not that reliable... if you're using AI to do something you really don't get you're kind of out over your skis... it might very well make up some Roman currency you've never heard of and you wouldn't know... learning the instinct of when you know enough to know if it's doing something useful”
A consensus in the learning research literature is that AI tutors should not simply tell people what to do, because that isn't how we learn; instead, presenting contrasting options ('take A and this would happen, or B and this would happen') gives learners information to reason about, which is far more instructive than a directive answer.
“a consensus in the literature is what they should not do is to sit around telling us what to do... instead... the AI can say... you could take A and this would happen or B and I predict this would happen... the contrast between those things actually very instructive for learning... you need to give them information that supports that choice and enables them to reason about it”
Contrary to the narrative that America is in decline, US productivity has risen about 30% relative to Europe over the last 20 years, US labor markets work well, and the US has an exceptional culture of innovation—so Americans underestimate both their strengths and their good fortune.
“people think that America's been in decline, but you know, our productivity has risen 30% relative to Europe over the last 20 years... Our labor markets actually work really well... the US is an incredibly innovative country”
22% of all US manufacturing employment was lost between 1999 and 2007, and cumulatively about a third once the Great Recession is included.
“22% of all US manufacturing employment was lost between 1999 and 2007. And then I cumulatively about a third uh once we go into the Great Recession.”
Automation can either increase the expertise of work by eliminating supporting tasks so you focus on what you're good at, or deskill work by automating the expert parts and leaving only a low-skill 'last mile'; it also often creates new work requiring new expertise, but usually performed by different people.
“an automation... can either increase the expertise of your work by eliminating the supporting tasks and allowing you to focus on what you're really good at... Or it can deskill your work by automating the expert parts and just leaving you with a sort of last mile... it also often creates new work that requires new forms of expertise... it's usually different people who are doing that work”
Machines acquire skills far more rapidly than people—once one machine does something, you quickly have many machines doing it—so AI-driven displacement of specific roles (language translators, illustrators, workman-level software coders) can happen abruptly; there is already a notable decline in employment among people who write programs, though not necessarily software engineering broadly.
“machines can acquire skills much more rapidly than people can... If you're a language translator... you're under threat. Uh, if you're an illustrator, you're under threat... we do see a big decline in employment in computer coding right now in software development... in the people who write programs”
US economic insecurity for low-wage workers is a policy choice, not a necessary fact of life: someone working at a McDonald's in Norway, Denmark, or France gets vacation, healthcare, and sick leave, showing that security need not be reserved for the middle and upper class.
“people in the United States have much greater level of economic insecurity than... other people in... less affluent economies working the same jobs... If you work in a McDonald's in... Norway or Denmark or even in France, you're going to have vacation... health care... sick leave... that's a choice that we make, but we tend to think it's inevitable, and it's not.”
Healthcare and education—about 20% of US GDP, much of it public money—are where AI offers the greatest opportunity, and investment should focus not just on rare-disease treatments but on making healthcare more available to everyone for longer, higher-quality lives.
“health care and education, two activities... in the United States, that's 20% GDP, a lot of it's public money... this is where there's such great opportunity, where AI could be a tool that could be so helpful... investing doesn't just mean... more treatments for rare diseases, it means things that make health care more available to everyone”
Work that pays well is high-stakes decision-making work requiring discretion (how to land this plane, care for this patient, remodel this kitchen) where simple rules don't exist—because if simple rules existed the work would already be automated.
“work that pays well is decision-making work... where the stakes are high, it's a one-off choice... And there aren't simple rules. If there were simple rules, they'd already be automated”
Much valuable work in advanced economies is monopolized by elites (professors, lawyers, doctors, financiers) protected by high barriers to entry, which keeps healthcare, education, and legal services expensive; enabling more people to compete in these domains via supporting roles would let people move up rather than be pushed down into low-paid services.
“a lot of valuable work in advanced economies is monopolized by elites... we don't face that much competition... there's huge barriers to entry... health care is expensive, education is expensive, legal services are expensive... if we could enable more people to compete in those domains... it would be great if we could use this technology to enable more people to move up”
Your belief about whether AI turns out well is not really a belief about AI but about what humanity will do with the opportunity; the future is a design exercise, not a forecasting exercise, so breaking our way depends on making good collective choices—which is feasible but extremely hard.
“your belief about what's going to happen is not really a belief about AI, it's a belief about what humanity will do with this opportunity... the future is... not a forecasting exercise, it's a design exercise... breaking our way is not just a matter of luck, it's a matter of making good collective choices”
A better world would be one in which more people without a four-year college degree can do software development, legal work, medical technical work, or kitchen design, because AI could let non-elites enter the high-value decision-making domains currently monopolized by the credentialed.
“if a world in which more people who don't have a four-year college degree can do software development, can do some legal work, can do medical technical work, can do kitchen design... That's a better world in my opinion.”
The good AI scenario is one where AI supports people in doing more valuable decision-making work—both using existing expertise more effectively and acquiring it faster—and the great challenge of the era is to build AIs that help people use expertise better and learn faster, which is hard because over-reliance leads people not to bother learning.
“The good scenario would be one where we were able to use AI to support people to do more valuable decision-making work, both to use their expertise more effectively and to acquire it more efficiently... If you rely too much on technology, you kind of won't bother.”
Contrasting the 'Wall-E' scenario (leisure dystopia where no one works) with the 'Mad Max Fury Road' scenario (warlords controlling scarce resources), Autor argues a very wealthy world can still leave most people with nothing, and the Wall-E outcome is actually the good one because the more likely failure mode is violent inequality.
“I like to compare... the the Wall-E and the Mad Max scenario... it's a future where basically people... watch... holographic TV drinking big gulps... But I view that as the good scenario. Uh because the more likely scenario to me looks much more like Mad Max Fury Road where everybody's competing over... a few remaining resources... controlled by... some warlord”
Automation tools (automatic transmissions, elevators, toll takers) fully encode specialized knowledge in machinery and successfully replace jobs, whereas most tools are collaboration tools that require the user to bring expertise—they let you take knowledge you already have and do it faster, further, or better.
“I like to distinguish between what I... call automation tools versus collaboration tools... automation tools are tools... like... the automatic transmission... the elevator... the toll taker... All of these specialized knowledge that was required is now fully encoded in machinery... Most tools require you to bring some expertise to the table to use them”
A central open research question is whether AI tools built to support experts merely yield better results or also help people acquire judgment faster—judgment being the slowly-developed, experiential part of expertise (in law, medicine, carpentry, actuarial work) that makes professionals valuable beyond book knowledge.
“to ask whether tools that are built to support experts... do they help them acquire judgment faster... if you're a lawyer... a doctor... a carpenter... You develop judgment over time. That's part of the expertise that makes you so valuable... You learn how to make the right decisions at the right time”
Collaboration differs from automation in that when two experts disagree (answers A and B) and put their heads together, they may arrive at answer C—something neither originally considered—an emergent outcome that cannot happen when a machine simply tells you what to do; this makes collaborative design essential for tools that help people learn.
“collaboration is, you know, like if two people, two experts, look at the same problem and reach different answers... the answer they come up with may not be the average of A or B, it may be C... which is not something that can happen when a machine tells you what to do”
China's growth from the 1980s through the 1990s under Deng Xiaoping involved adopting Western techniques, allowing foreign direct investment, and moving hundreds of millions from unproductive rural agriculture into export processing zones, and WTO accession in 2001 plus permanent normal trade relations made China more competitive by forcing reform and opening.
“China just has this incredible productivity growth... They allow foreign direct investment. They allow hundreds of millions of people to move from unproductive rural agriculture into export processing zones... in 2001, they... become a member of the World Trade Organization... China actually became more competitive as a result of having to reform and open because of the WTO.”
Learning when to turn to AI is itself a skill that improves with practice, analogous to how people developed the instinct to Google things; the speaker keeps a chat window open to bounce ideas and uses AI for quick tasks like turning a photographed table into a bar chart in minutes.
“Learning to use AI well, actually, is a... is an important skill in itself... knowing actually when to turn to AI for things is also like not obvious. It's something you get better at... I always have a chat window open and I use it to sort of... bounce ideas or look things up”
On a 7-week cross-country trip after college, the speaker heard on NPR about a computer learning center opening at a Black Methodist church in San Francisco called Computers and You, volunteered there, and became director of education for three years, which sparked his interest in technology, work, and inequality.
“they talked about this computer learning center opening up at a Black Methodist Church in San Francisco uh called Computers and You... I ended up as the director of education for 3 years and that kind of got me interested in technology and work and inequality.”