Over the past year, I’ve had many disquieting realizations — that the American voter is more forgiving of insurrections than inflation, that being 37 means having one bad knee (and another not particularly good one), and that the New York Mets made a pact with Satan in the fall of 1986, for which subsequent generations of fans would eternally pay.
But my most unsettling epiphany may have been this: Robots can now do much of my job better than I can.
For a decade, I have made a living partly by synthesizing news articles and academic studies into accessible explainers, which white-collar workers could skim while pretending to do their jobs. The skills that this vocation requires — the capacity to rapidly digest text, craft clean copy, and sound like I know more about any given subject than I actually do — are ones that ChatGPT possesses in spades.
Worse, it can exercise these abilities far more efficiently than I can. ChatGPT requires five seconds and a speck of electricity to explain how President Donald Trump’s tariffs have impacted the dollar. To perform the same task, I need about 48 hours, six meals, 37 Twitter breaks, and at least three medications. And unlike myself, the chatbot can instantly rewrite its analysis in the voice of a pirate.
But journalists are hardly alone in fearing our impending economic obsolescence (which is, admittedly, a longstanding pastime within the field). Knowledge-workers of all kinds — from software engineers to financial analysts to filmmakers — are sweating their robo-competition. And not without reason.
In investor calls, executives are touting plans to satisfy their labor needs with artificial intelligence at a historically high rate. Goldman Sachs is slowing hiring and speeding AI deployment. The fintech firm Klarna has used AI to slash its payroll by 40 percent. Salesforce says machines can now perform 50 percent of the firm’s work.
AI-generated mass unemployment is still the stuff of science fiction. But the technology does appear to be slowing hiring in heavily exposed fields, particularly for junior positions. And labor market data indicates that something weird is afoot: Over the past 35 years, recent college graduates have almost always had a lower unemployment rate than American workers writ large. Now, a young college grad is more likely to be involuntarily jobless than the typical US worker.
Meanwhile, the major AI labs are promising to deliver an artificial general intelligence (AGI) — which is to say, a machine that can outperform humans at all cognitive tasks — in the near future. Anthropic CEO Dario Amodei says AI is likely to wipe out half of all entry-level white-collar jobs by 2030. Investors seem to be betting a lot of money on such outcomes. This year alone, companies are poised to sink $375 billion into AI infrastructure.
Among anxious laptop workers, these trends have fed half-ironic chatter about the coming of a “permanent underclass”: Once AI renders virtually all human labor commercially useless, most people will be condemned to eternal subjugation and precarity. No company will pay you for work that a robot can do better. And no market economy will let you climb the income ladder if your labor has no value. In Silicon Valley, such reasoning has generated a peculiarly dystopian variant of hustle culture: Make your fortune in the next five years, and you’ll claim a place in the perpetual aristocracy of AI owners — fail, and you’ll forever be at their mercy.
All these claims are wildly speculative. It’s not certain that today’s AI labs have functioning business models, much less the wherewithal to develop omnicompetent robots. Yet of all the nightmare scenarios spun by fatalistic futurists, AGI-induced neofeudalism strikes me as among the most plausible.
AGI may or may not decide to liquidate the human race. But it will tank the value of human labor, more or less by definition. We don’t know how ordinary people will fare in a world where the wealthy can do without their talents and exertions. But it is reasonable to worry that the answer is “not too well.” After all, there are already societies in which workers enjoy relatively little economic leverage over elites. And they typically aren’t nice places to be an ordinary person.
It’s therefore worth examining precisely how AI could generate an immutable oligarchy —and what can be done to prevent that from happening.
You will own nothing and be starving
Automation is the mother of all prosperity.
Without labor-saving machines, most of us would still spend our days coaxing calories from the Earth, forever a few unfriendly weather events shy of starvation. Robots are why most Americans can go to bed with a full stomach in a climate-controlled home, facing a flat screen that supplies infinite audiovisual entertainment on demand — all without ever needing to risk a sunburn or broken back.
Yet automation has also routinely ruined people’s lives. True, robots tend to benefit most workers in the long run. But in the meantime, some displaced laborers die poor.
Nevertheless, in every past wave of automation, it was at least theoretically possible for the disemployed to secure better-paying jobs by updating their skills. Even as machines rendered humans uncompetitive at certain tasks, they facilitated the discovery of new economically useful endeavors. When the steel plow ran field hands off the farm, they could find new work in factories. When industrial robots shouldered manufacturing workers off the assembly line, they could — hypothetically, as the saying goes — “learn to code.” Far from exhausting a finite supply of occupations, technological progress generated more and better ones.
But there is no reason why this must always be the case. No physical law forbids silicon from facilitating reasoning as well as gray matter. Nor does any property of the universe guarantee that human hands will always be better at folding towels (or any other fine-grained activity) than robotic ones. And the number of tasks that machines can’t do has, in fact, been rapidly shrinking.
Six years ago, the world’s most sophisticated AI models struggled to write a snippet of Python or a coherent paragraph. Today, they can get an A+ in constitutional law and outperform 99.5 percent of humans on certain coding problems.
Progress in robotics has been less frenetic. Machines have an easier time acing calculus than assembling Ikea furniture. But the better robots get at processing information, the more comprehensively they’ll be able to outperform humans in the physical world — a point underscored by recent advances in self-driving cars.
“Most of us are currently focused on machines being able to perform all the cognitive labor,” University of Virginia economist Anton Korinek told me. But in his view, one of the big lessons of the past two years is that “if we give our existing robots better brains, they can perform much better.” For this reason, Korinek thinks that machines may “soon” be able to outperform humans at most “cognitive and physical labor.”
None of this means that AGI is two years away, as some of the industry’s boosters (and Cassandras) have been prophesying. But it does offer some cause for thinking that economically omnicompetent machines are possible.
And there are reasons to suspect that their advent may be inevitable over a sufficiently long time horizon.
After all, humans did not evolve to optimally create value in a modern market economy. We cannot perform tasks for any extended period of time without getting a bit hungry, tired, bored, or horny. We struggle to accept orders unless we’re shown a modicum of respect. More fundamentally, barring cyborgification, the efficiency of our brains is bound by our biology, no less than the strength of our arms.
Machines’ capabilities are less inherently constrained. Indeed, given sufficient advances in AI, hardware, and energy technology, it is conceivable that robots could perform a week’s worth of any human’s labor while consuming fewer resources than it would take to keep that person alive.
This has led some in the AI industry to predict that their technology will eventually push human wages below the subsistence level. In that scenario, after briefly freeing people from lives spent tilling soil and chasing squirrels for sustenance, technological progress would have restored most humans to that prehistoric fate.
Or, at least, it would do so — unless elites chose to share the fruits of AGI with the masses. But it’s unclear whether we should expect such redistribution, in a world where an ordinary person’s labor isn’t worth the cost of their calories.
How AI could break the engine of social progress
America today is a profoundly inequitable nation, in which the powerful exploit and neglect the weak in myriad ways.
It is also one of the most egalitarian civilizations to ever exist.
Since the advent of agriculture, most humans have lived in steeply hierarchical societies, where privileged castes hoarded both resources and legal rights. Typically, at the top of these social orders stood nobles, who collected income without working; at the bottom, enslaved people, who worked without pay. In some cases, the former could do most anything to the latter, up to and including killing them.
Ordinary people, meanwhile, boasted limited legal standing, exercised no political influence, and tended to pay higher effective tax rates than the rich, even as they received a piddling share of economic output and few public benefits beyond military protection.
Compared to ancient Rome or China, modern industrial democracies resemble socialistic utopias. Even in the infamously unequal United States, all citizens enjoy formal equality under the law and a minimum living standard within the economy: The working poor are entitled to food assistance and public health insurance, the old or disabled to a guaranteed income, and the young to 13 years of free education. Meanwhile, the typical middle-class family pays a lower tax rate than the typical rich one. Many employers feel economically compelled to provide their workers with paid family leave, healthcare, and other benefits — and all firms are legally required to pay a minimum wage, maintain safe working conditions, and tolerate union organizing
What explains the relative humanity of so many modern societies? Some would point to the egalitarian ideas propagated by universalistic religions, Enlightenment philosophy, and social democratic thought. And such belief systems have surely played a role.
Yet the spread of egalitarian ideas in recent centuries doesn’t merely reflect their intuitive moral appeal but also, their capacity to facilitate mutually beneficial cooperation. The abolition of slavery did not only make America more just; it also rendered the nation more economically productive. Labor markets comprised of autonomous workers are more efficient than those populated by serfs or enslaved people. As a result, the liberation of those bound to the land facilitated an increase in both wages for ordinary people and profits for elites.
Similarly, in a world of robust economic growth — where ordinary workers’ time is becoming steadily more valuable — the wealthy have an interest in securing political stability and labor peace, even at the cost of forfeiting a bit of power. Acquiescing to mass enfranchisement may have loosened elites’ grip on governments. But doing so also helped legitimize orders that provided capital owners with high and rising living standards. Tolerating unions may have forced business owners to accept a higher wage bill, but doing so also insulated them from the threat of strikes. Investing in public education may have required the well-heeled to pay more taxes, but it also cultivated a more productive labor force.
This isn’t to say that material interests mechanically determine history or human behavior. People crave status, meaning, and self-esteem about as fiercely as they do wealth. The pursuit of these other goods can lead economically comfortable people to support progressive reforms, even if they have no crass material or political incentives to do so. In fact, as economic growth satiates many rich people’s financial desires, we would expect them to start putting greater weight on postmaterial concerns — which, in many cases, will mean advocating for greater social and economic equality.
Still, elites’ strong incentive to maintain the economic cooperation and political loyalty of ordinary workers surely made democratization easier. It is not a coincidence that the past two centuries of egalitarian progress has coincided with a sharp increase in the market value of human labor — and thus, the bargaining power of human workers.
AGI threatens to kick all these trends into reverse.
The future is…Congo?
For a preview of fully automated neofeudalism, look to the Democratic Republic of Congo.
Or at least that’s what the AI entrepreneurs Luke Drago and Rudolf Laine advised earlier this year in an essay titled “Defining the Intelligence Curse.”
As Drago and Laine note, Congo is rich in natural resources. The nation holds substantial oil reserves and about $24 trillion in mineral wealth. Yet Congo is among the five poorest nations on the planet, with more than 70 percent of its people living on less than $2.15 per day.
This dissonant combination of commodity wealth and mass poverty is not unusual. To the contrary, large resource endowments appear to correlate with slower economic growth and higher corruption — a phenomenon that economists and political scientists have dubbed “the resource curse.”
Scholars have attributed this paradox to many different factors. But Drago and Laine emphasize one in particular: Large commodity endowments can reduce elites’ incentives to increase ordinary workers’ productivity.
Vast energy or mineral deposits provide investors with a ready source of profits — and states, with an easy source of revenue. The former don’t have to hazard capital on complex production processes to secure returns, while the latter need not bother with the headaches of forming competent tax collection services, the fiscal costs of developing a skilled labor force, or the political risks of cultivating an educated populace and diverse economy with competing power centers. Oligarchs and public officials can just feather their beds with commodity revenues instead.
According to Drago and Laine, AGI could fuel an even more extreme version of this dynamic. With the aid of superintelligent robots, the theory goes, capitalists won’t need to curry favor with pesky workers in order to turn a profit. And states won’t rely on ordinary people for tax revenue. To the contrary, as machines condemn most workers to perpetual unemployment, governments will have few funding sources beyond the windfall profits of corporations. The typical person’s economic leverage over public and private powers will be kaput. And states and businesses will have little material incentive to invest in their education or well-being.
In autocratic petrostates, ordinary people can still exert some influence over elites through the tacit threat of revolt. According to some political scientists, fears of inciting a popular rebellion constrain the abuses of kleptocracies, compelling some to fund substantial social benefits.
But Drago and Laine argue that, after AGI, mass publics are liable to lose even this timeless check on oppression. In a world of super-intelligent drones and AI-powered surveillance, they worry, governments will need the consent of trivially few humans to maintain an effective monopoly on violence.
Perhaps, the sheer force of enlightenment ideals or humanitarian fellow feeling will lead the powerful to share the wealth in such a society, anyway. But the behavior of many American tech billionaires and public officials today — when they still depend on ordinary workers for their keep — doesn’t inspire much confidence.
We need to AGI-proof American democracy
If we can’t depend on the kindness of oligarchs, what can preserve ordinary people’s power in the face of AI progress?
To economists like Erik Brynjolfsson and Daron Acemoglu, part of the answer is encouraging the development of artificial intelligence that augments labor rather than replacing it. AI tools could plausibly lift wages by making workers more productive. In theory, such innovations could safeguard laborers’ economic leverage and promote a more equitable model of growth. Unfortunately, in this account, the tech industry’s obsession with AGI has inclined it to focus overwhelmingly on labor-automating machines, rather than the labor-augmenting variety.
Brynjolfsson and Acemoglu contend that the government exacerbates this tendency by taxing capital at a lower rate than labor, thereby incentivizing automation. But they also see a role for entrepreneurs to develop the labor-augmenting technologies that large labs are neglecting. This is what Drago and Laine have sought to do with their own startup.
If AGI ever becomes technically feasible, however, the economic incentives to adopt it will be overwhelming; Well-intentioned entrepreneurs are unlikely to beat it back. In that scenario, our best bet for fending off techno-feudalism will be institutional reform.
The more equitably we distribute capital ownership before AGI devalues human labor, the less oligarchic a fully automated economy is liable to be. And the more democratically accountable we make our government, the better ordinary people’s chances of constraining elites through the exercise of political power, even as their economic leverage erodes.
The “resource curse” is a common trap. But it isn’t an inevitable one. Norway is an exceptionally oil-rich country, with fossil fuels generating a quarter of government revenues. Yet it is also among the most prosperous and equal nations on the planet.
The Norwegians’ immunity from the resource curse has a simple explanation. By the time Norway discovered its major oil reserves in 1969, it had already established one of the world’s most egalitarian social orders. When processed through the nation’s social democratic institutions, the emergence of a vast new source of inanimate wealth didn’t yield windfall rents for plutocrats but rather a sovereign wealth fund that bankrolls generous pensions for ordinary Norwegians.
If we can get our political economy in better shape, advanced AI could deliver similarly broad benefits to future Americans. “The very principle of technological progress is that the economy can do more with less,” Korinek said in our interview. “It means you can have more output, more wealth, more goods and services that people can consume. The big question is how will those goods and services be distributed. But the good thing is that, in principle, since we have more, everybody could actually be made better off.”
We can build a highly automated economy in which everyone can flourish — even the obsolete authors of sprawling, self-referential “explainer” journalism. We just need to develop healthy political and economic institutions.
Granted, at the moment, doing that looks about as hard as building a God out of silicon.
The post The most likely AI apocalypse appeared first on Vox.
