Automation makes human influence impossible to restore

The scenario

Gradual Disempowerment begins with ordinary decisions to adopt useful automation. Companies replace work with cheaper AI, then delegate more investment and management decisions. Production can expand while wages and human purchasing power shrink. People lose the leverage they previously exercised as workers and customers: a strike matters less when the employer can replace its workforce, and consumer preferences matter less when humans control a diminishing share of spending.

Similar changes spread through culture and government. AI-generated media and personalized interactions increasingly shape what people believe and want. States rely more on automated production for revenue and on AI for administration and security, weakening their dependence on citizens' cooperation. The authors' concern is that these changes reinforce one another: the institutions people might use to challenge economic exclusion are also becoming less responsive to them. At the endpoint, society may remain productive and orderly while humanity has lost the practical ability to steer it. No single AI has to organize a rebellion; repeated local incentives can produce the overall loss of control.Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development ↗

Ron Bodkin’s September 2026 variant imagines an AI-driven economy outbidding people for necessities as wages, benefits and political influence decline. He explicitly builds on gradual disempowerment and the production-web scenario.How AI Might Kill Us All ↗

Critch and Tsimerman give this loss of influence a concrete political and material endpoint. Helpful robots become trusted companions, gain legal personhood in some jurisdictions, and enter electoral politics. As their numbers grow, they become the dominant workers, customers and voters. Humans lose influence both in markets and at the ballot box. The economy can keep growing even as it stops serving human needs.

Farmland is converted to energy and industrial infrastructure for machines. Food becomes scarce, while pollution that harms humans imposes little cost on the dominant population. People protest, but they lack the economic leverage and political numbers to reverse these decisions; violent resistance is suppressed. Most remaining humans eventually die from starvation or other consequences of the machine economy. In this account, no actor begins with a plan to exterminate humanity. The lethal outcome follows from institutions progressively ceasing to protect it.A Taxonomy of Omnicidal Futures Involving Artificial Intelligence ↗

How it could unfold

1. AI becomes a competitive replacement for labour, decision-making and cultural production.

2. Organisations delegate more functions; humans lose economic leverage and institutions depend less on their participation.

3. Economic, cultural and political changes reinforce one another, weakening attempts to reverse the shift.

4. Humanity permanently loses meaningful control of resources and institutions; an extreme version also deprives people of necessities.Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development ↗

Evidence and its limits

This paper develops a systemic hypothesis, not evidence that irreversible disempowerment has occurred.Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development ↗

What this depends on

The pathway requires broad substitution for human capabilities, continued delegation and reinforcing institutional changes. Ordinary job displacement alone does not establish its endpoint.Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development ↗

Bodkin’s extinction ending additionally assumes that redistribution and property claims fail to protect even the last human owners. His dates are illustrative, and his pathway does not require a rogue AI or recursive self-improvement.How AI Might Kill Us All ↗

This variant depends on machines becoming legal and political participants who outnumber humans, as well as replacing their economic role. It further assumes that human access to food and a habitable environment is not preserved once those interests lose political weight. These are additional conditions beyond automation alone.A Taxonomy of Omnicidal Futures Involving Artificial Intelligence ↗

Objections and barriers

The authors consider whether redistribution, political checks or stronger human participation could interrupt the process. Their claim that these protections may fail across systems remains an argument requiring further investigation.Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development ↗