If AI performs all work, Steve Keen says communism is not a choice but a structural inevitability

Full AI automation does not produce a political choice between economic systems, according to economist Steve Keen. It forecloses the choice. Keen’s position is that if AI reaches the point of performing all work, communism becomes an inevitability, not a preference, not a policy outcome someone might elect, but the only arrangement that can logically follow from the underlying conditions.

The logic Keen is invoking is structural. Under any system where human labor is the primary mechanism by which people earn claims on goods and services, the elimination of human labor from production does not quietly redistribute wealth. It severs the link between contribution and access entirely. People still need goods. They have no remaining mechanism, under a wage-labor framework, to obtain them. What fills that gap is, by Keen’s account, not a reformed capitalism but something that has shed the core premise capitalism depends on: that access to the goods of production flows from participation in the act of producing them.

This is a falsifiable call, not a philosophical musing. It rests on a specific threshold condition: AI performing all work, not most work, not work in particular sectors, but all of it. That threshold matters because the claim collapses if the threshold is not reached. Partial automation, even at very high levels, still leaves space for hybrid arrangements, redistribution schemes, or expanded welfare states that do not require abandoning the wage system in full. Keen’s argument only becomes testable if automation reaches the terminal condition he names.

If we get to the position where AI is doing all the work, communism is an inevitability. Steve Keen

The claim also carries an implicit challenge to those who frame advanced automation as a political opportunity rather than a structural forcing function. A common optimistic view holds that societies can choose, in advance and deliberately, how to organize distribution once machines do most of the work. Keen’s framing suggests that at the endpoint he describes, the choosing is already done. The economic logic of a world in which human beings cannot contribute to production but still require its outputs leaves only one coherent organizational answer. Whether anyone wants that answer, or what they would call it, is a secondary matter.

What Keen does not supply, and what the claim requires before it can be fully evaluated, is any account of timeline or trajectory. Full AI automation of all work is not a near-term horizon by any current measure. The claim is better understood as a conditional: a structural argument about what must follow from a specific endpoint, not a prediction about when that endpoint arrives or how likely it is to do so. Whether the current trajectory of AI development points toward that endpoint at all is a separate question, and one that Keen’s statement does not address.

That restraint in scope is also what keeps the claim intellectually serious. A broader assertion, that significant automation will tend toward communism, or that current AI trends already imply it, would be much easier to contest on empirical grounds. By anchoring the claim to a specific terminal condition, total displacement of human labor, Keen has written a statement that cannot be falsified by pointing to today’s labor markets. It can only be confirmed or disconfirmed as the trajectory of automation becomes clearer over time.

The bet, as stated, will sit unresolved for some time. But it is worth tracking precisely because it reframes a conversation that tends to get stuck on redistribution mechanisms and policy design. Keen is not arguing about the best political response to automation. He is arguing that at a certain level of automation, the system-level response is not a design question at all. Societies do not, in his framing, get to choose the destination once the conditions are met. They only get to choose how they arrive. That is a harder claim to dismiss than most positions in this debate, and a considerably harder one to confirm. It deserves to be held against the evidence as the automation data accumulates over the years ahead.