At the University of the Basque Country, Héctor Galindo-Dominguez and his team at the ESCUTIC research group have cracked a puzzle that worries educators worldwide: why do some students blindly trust artificial intelligence, and what actually stops them from doing so?
The answer is surprisingly encouraging. A study of 404 students averaging 20 years old, published in Computers in Human Behavior, reveals that self-regulation—a student's ability to organize themselves, make sustained effort, and reflect on their own work—acts as a protective barrier against overconfidence in generative AI. In other words, the very habits that make someone a good learner also make them a skeptical, thoughtful user of ChatGPT and similar tools.
The research uncovered a fascinating paradox. Students with the clearest goals actually tend to trust AI more, not less. "This isn't due to a lack of ability, quite the opposite, in fact; they use AI as a tool to speed up their progress," explains Galindo-Dominguez. That sounds like a problem—but it isn't, not entirely. The same students who have clear objectives also tend to possess other self-regulatory skills: perseverance, the ability to learn from mistakes, and strong decision-making. These qualities function like internal fact-checkers, prompting students to review their work, question AI-generated answers, and refrain from passively accepting whatever the algorithm produces.
The real risk emerges among students lacking these protective habits. Some develop what the researchers call "overconfidence in artificial intelligence"—a tendency to assume that AI-generated answers are correct or adequate without scrutiny. This can lead to dangerous delegation of important decisions and a worrying reduction in personal effort. Yet the study also offers reassuring news: most students don't fall into this trap. The researchers found that the majority use AI sparingly, turning to it mainly for information gathering or answering specific queries. Only a smaller group displays frequent, extensive use that might signal true dependence.
What makes this research particularly valuable is what it doesn't recommend. Galindo-Dominguez and his team explicitly reject the idea of banning or restricting AI tools in education. Instead, they argue for a far more effective approach: teaching students how to use these tools judiciously. This means designing assignments that encourage students to cross-check information, explain their reasoning, and treat AI responses as starting points rather than finished products. It means building reflection into the learning process itself.
"The debate should not be about whether artificial intelligence is good or bad, but about what kind of students use it and how they do so," Galindo-Dominguez emphasizes. When self-regulatory skills are present—when students know how to persevere, how to learn from their mistakes, how to think critically—AI becomes a genuine aid to learning rather than a replacement for it. The message is clear and hopeful: rather than fearing generative AI in classrooms, educators should focus on what actually matters—cultivating the thinking habits that help students use any tool wisely.