The delusion of increased AI productivity

Is AI a bubble? We are told that it depends on whether massive investments in the technology can massively boost productivity. Optimists are convinced that future productivity gains could be colossal, even if a market correction is necessary to draw some hot air out of the current boom. Pessimists, from Goldman Sachs’ Jim Covello to MIT’s Daron Acemoglu, have their doubts. Optimists and pessimists alike are in the grip of what could be called AI productivity delusion.

What both sides miss is that AI could devastate capitalism even if it boosts productivity beyond our imaginations — and regardless of whether it delivers the promised economy-wide automation of everyday corporate workflows. The reason why we tend to miss this point is that we look at today’s world through the lens of another world that cloud capital, especially now that it is AI-enhanced, has rendered obsolete.

Until recently, it was a foregone conclusion that every technological revolution represents a moment of creative destruction. Whole sectors, professions, and skill sets were destroyed by innovations which ushered in new sectors, new professions, and new skill sets. Automobiles ruined saddlers, farriers and knackers but begat auto workers, road builders, mechanics and service-station personnel. The transition from coal to oil condemned whole mining communities and trade unions to oblivion. In net terms, more was created than was destroyed. Creative destruction became synonymous with economic growth.

Faced with the AI juggernaut, it is natural to focus on the technology’s net effects. Could it create more than it destroys? The real question is, “More of what?” To avoid the AI productivity delusion, we must rethink how we define productivity.

If by productivity we mean tasks accomplished per unit of human labor, there is no doubt that AI could prove an immense success — in fact, it already has. If we continue to measure productivity in terms of monetary value produced per unit of labor, the chances are that AI is going to prove an unmitigated disaster.

Earlier technological revolutions, from the steam engine and the automobile to the personal computer and the smartphone, expanded the domain of capitalist markets. Despite the wreckage of whole professions and communities they left in their wake, they resulted in simultaneously more material output and a higher aggregate monetary value of that output.

Can AI do likewise? Maybe in China, where AI is coupled with robotics, material production could rise and, at least in the short run, so would net revenues. In the West, AI agents are deployed in the service sector — from Amazon.com to back offices, law firms, hospitals and more. In those domains, AI-enhanced cloud capital could shrink markets in two distinct ways.

The first way, the one everyone worries about, is that AI is going to kill off entire professions: paralegals, coders, translators, accountants, even architects, mathematicians and school teachers, while creating far fewer new jobs filled by the prompters, overseers, engineers and operators of the new machinery. The second way is far less visible but perhaps even more worrying: the conversion of markets into cloud fiefs and of profits into rents.

As I argued long ago, Amazon might look like a market, with countless buyers and sellers within its digital walls, but it lacks the central feature of a market: decentralization. It is more centralized than the Soviet planned economy ever was, with an inscrutable algorithm deciding which consumer is matched to which vendor on the basis of a probabilistic model optimized to maximize the likelihood that Jeff Bezos collects 40 percent of the maximum price every customer is willing to pay.

As AI becomes more deeply embedded in Bezos’ algorithm, interfaces like Alexa would communicate with us not as mere shopping assistants but as an all-knowing friend, companion and even therapist. Soon, capitalist vendors’ profits would end up in some tech lord’s account in the form of cloud rent, which he would most certainly invest in more AI-enhanced cloud capital that produces more power to extract more cloud rents.

So, as Apple’s Siri, Google’s Gemini and an army of such agentic assistants develop the same capacities for their tech lords as Amazon has for Bezos, the capitalist market domain is going to continue to shrink. While AI might have proven exceedingly productive in terms of economic activities performed per unit of human labor, the profit rates of capitalist enterprises would collapse, following society’s aggregate wages to the bottom. At that point, none of the tools of the liberal state — no amount of quantitative easing, no politically feasible levels of universal basic income, no wealth tax that the tech lords might find hard to evade — would be able to prevent a systemic collapse.

The question of whether AI is going to generate massive productivity gains is easy to answer: Yes. AI is already proving capable of exponentially increasing the amount of work that is being done per human worker. However, as long as it is used to eliminate wage labor while siphoning off profits from actual markets into our tech lords’ cloud rents, the economic value produced per worker is going to tank.

Unless this distinction is made, the AI productivity delusion is going to persist. And as long as it does, we are unable to hope for a post-capitalist future of abundance where we, along with our AI helpers, produce according to ability and distribute according to need.

Yanis Varoufakis, a former finance minister of Greece, is leader of the MeRA25 party, professor of economics at the University of Athens, and a senior research fellow at Fudan University.