Can simply showing people what they lose by offloading thinking to AI get them to think more for themselves?

This explores whether showing people how they use AI, and what it costs them, can get them to keep more of the thinking for themselves. The most direct answer in the corpus is yes, and by a large margin. In a preregistered experiment with 704 people, feedback that pointed out the costs of offloading cut requests for answers from an LLM by half and raised scores on a later test taken without AI by 51% Can metacognitive feedback stop students from offloading to AI?. The surprising detail is what didn't work: rewarding effort had no measurable effect on either outcome. Paying people to try harder didn't change behavior. Helping them see what they were giving up did.

That matters because the costs are otherwise easy to miss. A 666-person UK survey found that frequent AI use was linked to lower critical thinking scores, and cognitive offloading was the measured pathway. The link was stronger in younger users Does frequent AI tool use reduce our critical thinking?. A four-month EEG study goes further. Brain connectivity scaled down as people relied more on AI, and LLM users had trouble recalling work they had just done Does AI assistance weaken our brain's ability to think independently?. If the harm is invisible while it's happening, feedback that makes it visible is a natural fix. One framing of why people over-trust AI describes it as fast, intuitive output that feels like reasoning, with traps that compound when nothing interrupts them Why do people trust AI outputs they shouldn't?.

Feedback isn't the only option. In an 80-person study, assistants that asked reflection questions alongside their advice beat assistants that only gave advice or only asked questions Do reflection questions help people make better decisions with AI?. That suggests the best design doesn't withhold help. It pairs help with a prompt to think. But there's a catch. Cognitive forcing functions, which make people commit to their own judgment before seeing the AI's answer, helped people who enjoy effortful thinking much more than people who don't Do cognitive forcing functions help all decision makers equally?. An intervention that looks good on average can widen the gap between people who are already engaged and people who aren't.

Timing is the other hidden problem. Even correct AI suggestions can hurt reasoning by breaking someone's focus, so they have to rebuild their concentration before continuing Does AI assistance always help reasoning or does it carry hidden costs?. A badly timed nudge to reflect could cost more than it saves. One proposed answer is to read signals like hesitation, typing speed or gaze to find the right moment Can AI systems read cognitive state from interaction patterns alone?. That same tracking could also be used to profile and manipulate users. So the open design question isn't only whether metacognitive feedback works. It's who it reaches, and when it should appear without becoming another interruption.