# AI & the Future of Work | The Weekly Show with Jon Stewart

[Watch on tuber](https://tuber.guzus.xyz/video/RB_WmoH5nQ4/summary) · [YouTube](https://youtu.be/RB_WmoH5nQ4)

**Score: 72/100 · recommended**

Jon Stewart talks with MIT economists Daron Acemoglu (Nobel laureate) and David Autor about AI and work. Their main point is that the risk is less about running out of jobs and more about losing the need for skilled, well-paid human work, possibly very fast and across many fields at once. They argue that where AI goes is a choice, not something fixed, and they offer specific policies: wage insurance, taxing capital and labor more evenly, giving people a stake in capital ("universal basic capital"), and paying people when their data is used.

- **0:00-2:43** — **Cold open and intro.** The episode opens with a clip from late in the conversation, where Acemoglu calls the "ideology of meritocracy" pernicious and says it helps explain Trump and public anger. Stewart frames the episode, recorded the day before Earth Day, around AI as a "third existential threat" next to Iran and climate change. He introduces Acemoglu (Institute Professor) and Autor (Rubinfeld Professor), and Autor jokes that at MIT it's more distinguished *not* to have a Nobel.

- **2:43-5:00** — **Timeline and who is unprepared.** Acemoglu says the timeline is too uncertain to predict, but he worries about the consequences. Neither workers nor students are ready, and students are his biggest concern: with no guardrails, he doubts they will "learn how to learn" or build real expertise when AI hands them answers. Autor says people still need mastery, because if humans don't need to know anything, they aren't needed as workers. He says he'll speak more to the potential and Acemoglu to the risk.

- **5:00-7:48** — **Lessons from earlier disruptions.** Autor says the real worry is not the number of jobs but whether expert human labor is still needed. Everyone carrying boxes from UPS trucks is a very different economy from everyone doing medical care. He cites "Engels' Pause," about 60 years early in the Industrial Revolution when productivity rose but working-class wages didn't. Skilled weavers were wiped out and replaced by unmarried women and indentured children doing unskilled mill work, and specialized skills only came back in the late 1800s. His key point is that adults rarely switch careers mid-life, so adjustment happens across generations. When change is fast, as with the China trade shock, places eventually recover but individual workers mostly don't.

- **7:48-10:30** — **Will AI hit white-collar work the same way?** Acemoglu says yes, but the timing is unclear. The China shock (especially after 2000) and industrial robots both raised productivity in steel, electronics and cars while costing blue-collar jobs and pushing communities into recession. So far there are no mass layoffs from AI, maybe a hiring slowdown. White-collar jobs are less geographically concentrated than textiles or steel, but there are many more of them. Despite big advances "over the last 8 months," models can't yet do whole occupations. In his view, uncertainty is "a very bad reason to be complacent."

- **10:30-15:49** — **"Labor as a tax," derived demand, and truckers vs. call centers.** Stewart quotes an AI executive telling investors AI offers "productivity without the tax of human labor." Autor explains that labor demand is *derived demand*: firms want output, not workers. Firms have always dreamed of full automation and never achieved it; numerically controlled machines, for example, ended up turning machinists into programmers. This time may be different, because older computers could only follow explicit rules, while AI learns from unstructured data and infers rules itself.
  - Autor contrasts two groups of about 3.5 million workers each. Long-haul trucking will change slowly even if self-driving trucks work, because fleets and infrastructure take decades to replace, and change of a few percent a year is absorbed by retirements. Call-center jobs, held mostly by lower-paid women, can disappear fast.
  - The jobs that survive will be fewer, better paid and more expert: the "last 20 people standing" who take calls when the AI gives up. He expects the same in translation, and in software, which will split into a small elite and "infinity vibe coders" paid like Uber drivers. Purely cognitive work is the most exposed; robotics for physical work is "some ways off."

- **17:53-22:50** — **Reduction first, eradication later; bubble or real?** Stewart describes AI as strip-mining 10,000 years of human expertise. Acemoglu says shrinking employment comes first and possible elimination later, with stagnant or falling wages along the way. Past shocks hit a few occupations at a time: weavers lost about two-thirds of their earnings but could still become factory operatives. If Dario Amodei and others are right that many sectors will be hit at once, there is nowhere for displaced workers to go, which he calls "Armageddon." He doesn't expect that soon, but says it's prudent to prepare.
  - On the bubble question, his middle view is that the capabilities will arrive, but more slowly than the trillions in investment assume, so a lot of money will be lost. He can't rule out either extreme (all bubble, or everything within five years). Nobody can test these forecasts because they rest on unreleased models; "we're looking backwards."

- **22:50-27:00** — **Autor's case for upside, and their one disagreement.** Autor says AI investments can pay off through real productivity gains in healthcare, transport and education, not only by cutting jobs. He notes that even struggling regions today are far better off than pre-industrial England, though he admits transition costs are huge and the winners aren't the people who lose. He sees potential in science, climate adaptation, energy, medicine, and in spreading expertise to the developing world, much as mobile phones did. Both men back "pro-worker AI." Their disagreement: Acemoglu thinks it requires a complete change in the industry's direction, because the current path is focused on automation. He describes Autor as more hopeful that good outcomes can emerge anyway (Autor jokingly disputes this). Acemoglu wants the public debate to move past doom versus hype to the question "are we using it for the right thing?"

- **27:00-33:00** — **What "pro-worker" means, and the AGI problem.** Stewart notes that society-wide benefits like curing diseases are "pro-human" but not necessarily pro-worker, and that globalization showed "capital travels and labor doesn't." Autor cites their paper with Simon Johnson. Pro-worker AI means tools that extend human expertise and create new work. His example is the roughly 250,000 US data scientists earning about $120K at the median, a job that didn't exist 20 years ago; others are solar electricians, pediatric oncologists and fitness coaches.
  - Automation keeps devaluing expertise while new ideas create new kinds of it. What gets automated is easy to predict; new work, and who will get it, is not. Most new work in the last 40 years went to college graduates, and only about 40% of US adults have a degree. So the goal is AI that lets people without elite credentials do more valuable work in medical care, law, programming, contracting and skilled repair.
  - Autor thinks most tech leaders "are not evil" and simply don't know how to steer this. Acemoglu replies that they aren't interested, because the industry is locked into AGI, where status depends on getting closer to it. True AGI combined with flexible robots would break Ricardo's comparative-advantage logic, because cheap models would beat humans at everything.

- **34:20-41:47** — **Ideology, control and authoritarian uses.** Stewart argues that unlike the companies that laid fiber or electrical lines, today's AI builders are ideological and want to own society's "operating system." Acemoglu agrees: centralizing all information in a few companies works against using AI to make novices more expert. Autor warns against trusting economists on AI forecasts, or Amodei on economic forecasts. He adds that even AGI wouldn't end all work, because many problems are political or interpersonal; Acemoglu answers that this holds only if no one has dictatorial power.
  - Stewart uses the splitting-the-atom analogy (we built the bomb first) and points to Palantir supplying data on undocumented people to the government. Acemoglu says technology's direction is highly malleable, and authoritarian or wartime conditions push it toward the worse uses. Autor agrees AI is "God's gift to authoritarians," pointing to China's export of surveillance and censorship and to similar work privatized in the US. Acemoglu says the current targets are mass automation, a surveillance state, and a merger of the security apparatus with tech companies.

- **41:47-47:57** — **AI as an "enclosure" of the internet, and data rights.** Autor cites Oxford economist Max Kasy's book *The Means of Prediction*. AI is enclosing the internet the way lords enclosed common land: taking shared material (photos, writing, film) and monetizing it so it now competes with the people who made it. He calls this an under-discussed, one-sided transfer of property rights. Acemoglu adds the deeper problem: pro-worker tools, such as one that helps novice electricians do expert work, need high-quality data from top experts, which only gets produced if there are data property rights and markets. Extraction destroys that incentive.
  - Stewart calls AI a "human expertise laundering machine" with no royalties. Autor compares this to the Napster era, solved only through licensing like Spotify. Acemoglu notes a key difference: this time it's the most powerful corporations doing the copying, not consumers. Autor says fair use "never envisioned this," the law should change to pay people continuously, and Google engineers told him ongoing attribution is technically feasible but they have no incentive to build it.
  - Acemoglu says Silicon Valley's "beat China to AGI" framing creates artificial wartime conditions that justify anything. Autor recalls that Oppenheimer later opposed the H-bomb and lost his security clearance.

- **47:57-51:35** — **Policy 1: wage insurance, and the conversation that comes first.** In a program piloted under Obama, a displaced worker earning $25/hour who takes a $15 job gets half the gap covered (up to about $8K, for up to two years). That brings people back to work faster and acts like an earned income tax credit for returning workers. Unlike unemployment insurance, you get it only when you're working, which makes it politically viable in the US. Autor says the pilot paid for itself through unemployment-insurance savings and extra payroll taxes. Brian Kovak at Carnegie Mellon is trying to launch a multi-state demonstration, and Autor calls it a "no regrets" policy. Acemoglu supports it but says the first step is public discussion: AI is a range of choices, not one thing. The media went from praising the tech barons to "killer robots" talk, and decisions that change civilization shouldn't be left to Amodei, Altman or Thiel.

- **53:02-1:01:30** — **Policy 2: tax neutrality. Policy 3: universal basic capital.** Acemoglu says the US taxes labor heavily and subsidizes capital, a gap he says has grown over 25 years and especially under Trump. That pushes firms toward automation, so the tax treatment should be evened out. Stewart adds that AI profits flow into political power, citing tech CEOs in the front row at the inauguration.
  - Autor proposes **universal basic capital**: an endowment of shares with voting rights given at birth (he notes "Trump accounts" already exist), which can't be spent until 18. It diversifies people away from depending only on their own labor, and spreads ownership the way labor is naturally spread, since everyone owns exactly one worker. The government would buy the shares; he points to Norway's sovereign fund (about 2× GDP) and Alaska's fund.
  - Acemoglu backs this as an addition to work, but warns against UBI-style handouts that let people stop working. Mass non-work would hurt mental health and social peace, and would create a two-tier society of celebrated "creators" and everyone else. Stewart pushes back that we already have a two-tier, makers-vs-takers system that blames poor people for their poverty. He calls such transfers investments, not handouts, and says the public should get "a rake" on the industries it subsidizes. Acemoglu connects this to Michael Sandel's critique of meritocracy (the cold-open clip).

- **1:03:27-1:09:26** — **Political feasibility and regulating Big Tech.** Stewart complains that politics offers platitudes. Autor says wage insurance is being discussed with people in both the Trump administration and Democratic leadership. He also cites Raj Chetty's Opportunity Insights at Harvard, which is working on measurable job training. He warns that "even if it goes well, we're in for a rough ride." Acemoglu says productivity outpacing wages isn't inevitable: from the 1940s to the mid-1970s, wages grew faster than productivity, and non-college workers' wages grew faster than managers'. The US chose a different path around 1980.
  - Acemoglu calculates that each of the seven largest tech firms has annual revenue about twice the British Empire's mid-19th-century GDP. He says China, which he does not endorse, proves AI can be regulated, since Alibaba now serves Party interests; US firms could likewise be aligned with democratic priorities. Autor mocks Altman's claim that paying for intellectual property would kill AI firms, saying it amounts to admitting their product isn't worth paying for. He wants AI steered, not shut down.

- **1:09:26-1:13:04** — **Wrap-up and outro.** Stewart says the guests didn't ease his fears but gave him hope that "the future is not yet written," thanks to concrete policies rather than platitudes, and jokes that he has made AI avatars of them. The final few minutes are a solo outro: he's in a better mood, there's no staff chat because he's on the road, followed by credits. This last part is filler.

**Practical takeaways**
- The key question isn't whether jobs disappear but whether *expert* human work stays valuable. Purely cognitive, standardized roles like call centers, translation and routine coding are most exposed, and change there could be fast.
- Speed matters more than direction. Changes absorbed over a generation (trucking) are manageable, but fast, multi-sector displacement hurts mid-career workers, who rarely retrain successfully.
- Concrete "no regrets" policies to support: wage insurance for workers returning to lower-paid jobs, equal tax treatment of labor and capital, government-funded capital endowments at birth, and ongoing compensation when people's data is used for training.
- Be skeptical of both AGI hype and confident doom. Even experts can't test claims about unreleased models, and uncertainty is a reason to prepare, not to wait.
- AI's direction depends on choices about regulation, data rights and incentives. The guests want public and democratic debate, not tech leaders alone, deciding whether it's built to augment workers or replace and surveil them.

## Audience signal

- No high-signal audience themes available.