Think of it like GPS for your brain. The directions are great while you’re using the app, but take it away and suddenly you can’t navigate your own neighborhood. That’s essentially what researchers at the MIT Media Lab found when they studied how AI assistants affect people’s ability to spot misinformation.

The study, released on June 9, involved 67 participants over four weeks evaluating news headlines paired with images. While using AI chatbots, participants improved their accuracy in detecting false content by 21 percentage points. But here’s the thing: once the AI was removed, their ability to independently identify misinformation didn’t just return to baseline. It got worse. By the end of the four-week period, overall accuracy had dropped by 15.3 percentage points.

The skill erosion problem

The research paper carries a title that doubles as its thesis: “Dialogues with AI Reduce Beliefs in Misinformation but Build No Lasting Discernment Skills.” It was presented at ACM CHI 2026, one of the premier conferences for human-computer interaction research.

The team behind the study, including Anku Rani, Valdemar Danry, Paul Pu Liang, Andrew Lippman, and Pattie Maes, found that the largest accuracy losses appeared specifically in recognizing fake news content. The decline in discernment abilities was more pronounced than any stability observed in identifying real news. People could still generally recognize legitimate content, but their radar for falsehoods had atrophied.

One particularly notable finding: AI systems that provided direct answers, rather than guiding users through reasoning, tended to correlate with greater erosion of independent skills. The researchers drew a direct parallel to how navigation abilities diminish with reliance on GPS technology.

The broader implications for AI-dependent decision making

The study’s finding that direct-answer AI systems produce worse outcomes than more interactive, reasoning-based approaches has practical implications. Tools that explain their reasoning, challenge assumptions, or ask users to evaluate evidence before providing conclusions may carry less risk of skill degradation than those that simply deliver a verdict.

For market participants, the takeaway isn’t to abandon AI tools. The 21 percentage point improvement in accuracy while using them is substantial and real. The takeaway is that these tools should supplement human judgment rather than replace it entirely.