What the research found
Researchers at MIT's Media Lab ran a four-week study tracking 67 participants as they used AI assistants to help judge whether news headlines and images were real or fake. The AI tools themselves were genuinely useful — often landing on the accurate answer. But something notable happened to the people using them: participants who leaned on the AI's answers heavily got worse at spotting misinformation on their own over time, not better.
The researchers point to a specific mechanism: the AI tended to hand over a confident answer rather than walk users through how to evaluate the claim themselves. Convenient in the moment, but it meant users weren't practicing the underlying skill — they were outsourcing it. The comparison researchers drew: much like GPS has quietly eroded many people's sense of direction, leaning on AI for judgment calls can erode the judgment itself.
This isn't an isolated finding — a growing body of research this past year has converged on the same pattern across different tasks: AI tools reduce the mental effort a task takes, but that reduced effort sometimes comes at the cost of the reasoning skill the effort was building in the first place.
The practical takeaway
None of this means avoid AI tools — the same research shows they're genuinely useful. The distinction that matters is how you use them:
Outsourcing judgment (accepting an AI's verdict on something and moving on) is where the erosion happens.
Using AI as a sparring partner — asking it to lay out the reasoning, checking that reasoning yourself, or asking it to argue the opposite case — keeps you actively engaged instead of passively accepting an answer.
A simple habit: before accepting an AI's answer on something that matters, ask it "what would make this wrong?" It's a small shift that keeps you in the loop rather than out of it.
This study focused on misinformation-detection tasks over four weeks with 67 participants — a meaningful but specific slice of how people use AI day to day. The broader pattern is echoed elsewhere, but effects likely vary by task and person.
Reference
Rani, A., Danry, V., Liang, P. P., Lippman, A., & Maes, P. "Dialogues with AI Reduce Beliefs in Misinformation but Build No Lasting Discernment Skills." Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3772318.3790656
Top comments (0)