After the AI Revolution

What are the long-term economic and political consequences of transformative AI? This essay tries to answer that question through adversarial collaboration between two optimists (Victor Kumar & Josh May) and two pessimists (Rachel McKinney & Derek Anderson). The goal was not to resolve the debate but for each side to help the other refine their arguments, find avenues of constructive agreement, and discover shared uncertainty.

You don’t need to be a full-blown techno-optimist to think technology is generally a force for good. Technological progress over the last two centuries has continuously and dramatically improved human lives. And not just for the most privileged but across the world.

The share of people living in extreme poverty cratered from over 80% to under 10%. Many no longer have to do backbreaking work for 70 hours a week. Clean water, sanitation, and modern medicine doubled lifespans. Mass starvation, once routine, is now rare. Meanwhile, citizens of wealthy nations enjoy comforts that would have seemed unimaginable to previous generations.

Whereas the Industrial Revolution automated physical labor, the AI Revolution promises to automate cognitive labor. This will lower the costs of building knowledge, analyzing data, writing code, administering institutions, and more. Thus AI too will increase economic productivity, releasing a steady drip of abundance. The lives of future generations are unimaginable to us.

Of course, AI has not yet sparked a boom in economic productivity. To achieve that, three things must happen.

First, firms and other organizations must start restructuring around the technology’s affordances. Further AI progress may not be required—the necessary changes are institutional rather than technological. Second, hardware must advance along with software. We need not just better training methods but also new and more powerful GPUs.

Third, and perhaps most importantly, AI has to start doing things in the world. AI can readily increase economic productivity in domains with cognitive bottlenecks, such as software development, design, and healthcare. But in energy, housing, infrastructure, and other domains, the bottlenecks are physical. To address these, we need robots with fine motor skills and other autonomous systems—for example, factories where AI accelerates not just design but also production.

To be sure, it’s possible that the technology fizzles out—that AI has less to offer than imagined, that hardware limits are just beyond view, or that robots are too hard to build. It’s also possible that we’re on the cusp of artificial superintelligence (ASI) that transforms our world beyond all recognition.

But suppose it’s neither. Suppose AI is a “normal technology” but slowly revolutionary. Capabilities continue to expand dramatically, but their diffusion is slow because the world and its institutions have to be carefully rebuilt to accommodate them. In that case, it’s plausible that economic growth increases by 1 or 2 percentage points. That may not sound significant. In fact, it would be extraordinary.

The gains will accumulate over the course of many decades. You may not live to see the consequences, but your children or grandchildren might. (Still, all signs point to the AI revolution unfolding more quickly than the digital manufacturing revolutions of the 1970s and early 2000s, to say nothing of the OG Industrial Revolution.)

Material gains will arrive unevenly, but they won’t be confined to the wealthy. Indeed, poor people stand to gain more because they have more unmet needs to begin with. The global poor stand to gain most of all. True affordability will come, eventually, because as expertise and automation accumulate, we’ll be able to produce more stuff, more cheaply. And, as usual, falling prices will raise demand, creating more work.

But what happens to workers if their jobs are automated? Before industrialization, roughly 90% of humanity worked as farmers or artisans. From their standpoint, the disappearance of such jobs would be catastrophic. But that’s exactly what happened, and it was great.

There was no permanently unemployed underclass. Instead, new forms of work emerged. Likewise, after the AI Revolution, new jobs await us even if we can’t grasp exactly what they are. Some may involve managing AI tools and complementing the tasks they perform—though if AI advances, these jobs will continue disappearing. Even the C-suite will be replaced.

So what’s left? Alex Imas argues that when expertise and commodities are dirt-cheap, what will become scarce is the human element. People will shift to what he calls “relational work,” where interacting with another human is precisely what people pay for. This means more therapists, caregivers, and performers, but also new kinds of work that for now remain obscure.

OK, but will these jobs pay well? Historically, as automation made labor more productive, firms bid up wages. Higher productivity meant higher pay. Material inequality may get worse, and the owners will get fabulously rich, but if history repeats, raises will pile up until everyone is much better off.

A deeper reason for optimism is that the gains don’t depend on rising wages. Goods and services will become much cheaper. Not just those we already consume but also luxuries that are now exclusive to the 1%, such as gourmet food and high-end appliances. The cheapest commodity of all may be expertise itself, widely distributed even if wealth is not. Everyone might have their own virtual doctor, lawyer, and accountant.

Expertise will also deepen: scientists, engineers, and entrepreneurs will make discoveries faster. The rest of society will then benefit from advances in medicine, agricultural science, and energy development—leading to new miracle drugs, drought-resistant crops, and better batteries.

Advances in AI may even unleash moral and political transformations. In general, market-based economic growth (with rule of law) fosters progressive values of tolerance and freedom, as people are released from zero-sum conflict and expect their lives, and the lives of their children, to improve. If AI is as revolutionary as past technology, humanity’s future looks bright.

That’s the fundamental case for AI optimism. Why be pessimistic?

There are plenty of reasons! Suppose AI doesn’t fizzle out or kill us all. And suppose it fosters growth, lowers prices, generates medical advances, and so on. AI may still be terrible for humanity. Abundance is not enough.

First of all, even if the destination is an improvement, the journey there may be brutal. Careers will end for those who don’t have the skills society values in the age of AI. Their lives will be wrecked. And even if we do reach a new period of abundance where most people don’t have to work much anymore, that doesn’t mean they’ll flourish. As Keynes famously cautioned, free time might be spent on passive consumption or pointless status games.

Granted, a spiritually degraded world of material abundance may still beat the status quo (eventually). Still, there’s an even bigger problem. Assume abundance. Who enjoys it?

Consider a different story from the past. Industrialization led to massive gains in economic productivity. Profits soared, but they weren’t shared—not initially. From roughly 1790 to 1840, Britain experienced “Engels’ pause.” Wages stagnated; residents of industrial cities led miserable lives.

Eventually, gains in economic productivity did spread. Why? This is contested by economists, but one plausible answer from Daron Acemoglu and Simon Johnson (among others) is that workers began to use their leverage—through strikes, unionization, and slowdowns. Labor then forced legal and political changes to its (and plausibly everyone’s) benefit. Contracts, constitutions, and democratization created the conditions for modern markets.

After the Industrial Revolution, workers were needed in mills, factories, and transportation. But after the AI revolution, if machines are doing essential work, including work in the knowledge economy, then capital no longer needs labor. Abundance won’t automatically distribute; for that, you need organized people power.

So consider a possible story of the future. As AI displaces workers, people lose not only their primary means of earning income but also, crucially, their economic bargaining power. If workers are so desperate to keep income-generating work that they cannot meaningfully threaten to exit—and if firms are so indifferent to the costs of employees exiting because AI can replace anyone who leaves—then one of the main social processes driving economic redistribution is eliminated. Such a shift would lead to the gradual disempowerment of humans on the labor side of the economy, and thus the inability to improve their lives.

Meanwhile, on the consumption side, if AI replaces jobs at a rate and scale never seen before, we could be caught in the “AI layoff trap,” where layoffs erode the consumer demand that would make automation profitable. In this future, consumption would be concentrated among the few whose income comes not from labor but from ownership of capital.

Prices may drop, sure, but real wages may stagnate. This will be a particular problem for the global poor, who may be deprived of the opportunity to climb the development ladder, as manufacturing and service jobs are outsourced to machines. At the same time, markets may consolidate and turn to patronage, breaking mechanisms like competition, contract, and democratic rule of law.

What the future holds, then, is not techno-utopia, but techno-feudalism.

The “relational work” Imas predicts might well mean that households displace firms as the main organizational nexus for employment, and place people under the direct, personal, unaccountable authority of others—as sex workers and personal servants. In this scenario, “employment” remains high, but workers won’t have the leverage to spread gains or maintain freedom.

Ordinary people, then, will be poor and vulnerable to the whims of the rich. The few at the top will amass extraordinary wealth, while the majority will languish in relative poverty, stripped of autonomy, authority, and esteem. Whereas the Industrial Revolution eliminated drudge work, the AI Revolution promises to eliminate the kind of work that is safe, enjoyable, freely chosen, or consistent with equal standing among others, in favor of work that is dangerous, disgusting, coercive, or demeaning. The pause will be permanent.

As material inequality expands, social and political inequality will follow. We’ll have what Elizabeth Anderson would call relational inequality. People won’t stand as equals. They’ll be subordinated, treated as inferior. Or in Philip Pettit’s terms, they’ll suffer from domination, subject to the arbitrary and unaccountable will of others.

So even if AI succeeds according to narrow technological and economic criteria, it will fail politically. Vast abundance will be enjoyed only by the few. The rest will endure vast inequality.

So are the optimists or the pessimists right? The authors of this essay don’t share the same expectations about human well-being and freedom after the AI revolution. Each side has become more sympathetic to the other, but neither is fully persuaded.

For the sake of argument, the pessimists assumed abundance, but in fact they doubt whether AI can really accelerate manufacturing in ways that make consumer goods radically cheaper. Conversely, the optimists doubt that jobs will disappear or become restricted to relational work that is peripheral to the economy.

Still, abundance is plausible, while the risks of techno-feudalism are real. And here our disagreement stops mattering. Whether optimists or pessimists have the correct expectations about the future is beside the point. We can disagree about how likely it is that the house will burn down while agreeing that it is wise to purchase insurance. If worker leverage erodes, society must seek other strategies for spreading the gains. We see two.

One is the moral strategy. Consider the abolition of slavery, the civil rights movement, women’s liberation, and the rise of liberal democracy. In each case, the oppressed had agency and leverage. But progress also depended on morally persuading people without material incentives to overturn injustice. Whites supported abolition and civil rights; men supported women’s liberation; the bourgeoisie supported constitutional and democratic reform. Let’s grant that morality didn’t work on its own—that progress also required collective pressure on those with power. Yet this time, the moral vanguard and those in power may be the very same people. No pressure required.

The new AI elites might voluntarily share their gains—funding universal basic income, job retraining, and the like. Hope is not unreasonable, since the philosophy of Effective Altruism runs deep in tech. Anthropic cofounders and other early recruits have pledged to give away 80% of their wealth. According to some napkin math, the upcoming IPOs for Anthropic and OpenAI will release roughly $370 billion in new philanthropic funds.

More is needed. More money, but also more ambition. Founders and other AI entrepreneurs must not just “earn to give” but “build to give.” They have to devote themselves to AI ventures that directly benefit ordinary people, such as pharmaceutical startups that deliver cheap and universal access to safe and effective medicine.

Yet the moral strategy suffers from two big problems.

The first is that past progress did seem to require leverage. What happens when you have only moral ideals with no leverage? We already know: factory farming, arguably the most significant episode of moral regress over the last century. Like farm animals, human workers may become unable to advocate for their own freedom from undignified conditions.

And without leverage, AI elites will also have an incentive to renege on their commitments, or to engage in a motivated search for reasons to shift their priorities. Consider OpenAI, which started as a non-profit but transformed into a for-profit company seeking a $1 trillion valuation.

The second problem is that altruism, while plausible as harm reduction, is not enough for progress. It might prevent deprivation but not domination or social inequality. People will still be dependent on a benevolent oligarchy. As one of us argues elsewhere, benevolence increases welfare but not equality. For that, you do need power.

If the moral strategy can’t fully deliver, the political strategy might. Even if people don’t have leverage as workers, they may still have leverage as citizens—through social movements, campaigning for legislation, and electing leaders who are serious about AI governance. Citizen majorities could institute heavy progressive taxation, sovereign wealth funds, labor protections, or partial public control of AI firms.

But will democratic leverage erode too? That is, will citizens become politically disempowered? This seems likely if capital acquires even more influence over the government and democratic participation continues to fade. Or if AI firms become so powerful that they subsume government functions. Or if advanced military technology neuters mass unrest. Indeed, even if none of these come to pass, disempowerment may still occur as more of the state’s decisions are outsourced to AI rather than left to human judgment. The same conditions that drive abundance make these scenarios likely.

Yet citizens do have democratic leverage now. We have to rewrite our political compact before the power to do so fades—before techno-feudalism is “locked in.” The public still controls permits, copyright, and government contracts. Perhaps, as Senator Bernie Sanders argues, the public should take a direct ownership stake in AI firms. Better yet, make a public stake the price of further compute infrastructure.

In another possible story of the future, a virtuous circle unfolds: public stakes build support for the energy and permits that frontier AI labs need, which speeds diffusion, which produces more abundance, which pays out bigger public dividends, which deepens public buy-in, leading to larger public stakes, and so on.

Uncertainty remains. We don’t know how fast the technology will arrive, which means we don’t know who will govern the transition. Maybe takeoff is further away than some expect, leaving enough time for a competent government to create durable political change. The window is small, but it remains open.

Leave a comment

This essay is supported by a grant from the PPE Research Consortium and the John Templeton Foundation. The views expressed are the authors’ own.