Professor at the Federal University of Rio de Janeiro, André Barkaoui conducted an experiment that showed that using ChatGPT in learning can undermine long-term memory and critical thinking. The study involved 120 students, some using ChatGPT to prepare presentations and others using traditional abstracts and databases. After 45 days, participants were tested for understanding the material. The AI user group answered 57.5% of questions correctly, and the traditional one 68.5%.

The author explains the effect through the principle of "desirable difficulties": the brain forms strong neural connections only when it encounters productive resistance - the need to recall, analyze, and build connections between ideas independently. AI eliminates this friction by producing ready answers. Students who used the chat bot spent less time preparing (3.2 hours compared to 5.8 hours with traditional training).

The researcher drew three conclusions: AI creates an "illusion of competence" by not giving real understanding; critical thinking can atrophy in systematic cognitive unpacking; AI should be a second pilot, not an autopilot.