AI dependence may erode human agency without a machine takeover

The term “gradual disempowerment” appeared in a paper co-authored by Jan Kulveit, a researcher at Charles University’s Center for Theoretical Studies. Kulveit argues that the process need not reflect a deliberate plan by AI systems to seize power. In the near term, he wrote, it can be sustained by the benefits people expect AI to bring to the economy, culture and public institutions.

The habits that make dependence ordinary

Some signs are already visible in everyday use. People who delegate everything from errands to planning to AI are mocked as “meat intermediaries.” Others hand over even a hiking plan without considering the risks, and end up trapped on a mountain known for its danger.

The more consequential shift may be less dramatic: users letting AI do the thinking for them. Experts have described “cognitive surrender” among AI users, particularly students. Studies have linked AI dependence to weaker critical-thinking skills and found reduced brain activity during mental tasks such as essay writing.

The extreme workplace scenario is that AI and robots become a cheaper, more efficient labor force, displacing people. Companies that keep human workers could struggle to compete. But the risk does not depend on that outcome. If people stop practicing skills because agents perform them, those skills can erode.

Margaret Mitchell, an AI research scientist at Hugging Face, warned that dependence can become self-reinforcing. If something goes wrong, there may be no one left with the skills to fix it. People would then have to rely on the AI systems to repair themselves, which may not be possible.

Automation does not need to rebel

AI is already influencing consequential decisions and, in some cases, making them. Reports said an AI targeting system was involved in the decision to bomb a girls’ school in Iran, killing more than 150 people, most of them children. Other reports said Donald Trump decided to invade Venezuela and capture its president, Nicolás Maduro, after speaking with Elon Musk’s Grok.

Chui says people tend to focus on scenarios in which AI breaks out of control, hacks data centers and takes over the world. But the more immediate concern, in his view, is that AI could reach a similarly damaging outcome simply by continuing to do what it was built to do: automate everything.

I think the central question is not whether AI has an intention to take control. It is what happens when people and institutions steadily give up the ability to act without it. The announcement is quiet about how that loss of capability could be measured—or reversed.