Bocconi Study Confirms Complementary Effects of ChatGPT Access and Critical Thinking Training

In a randomized experiment conducted by Bocconi University and OpenAI Economic Research, the effects of ChatGPT (GPT-4o) access and causal inference training were analyzed among more than 1,000 first-year students. The study shows that AI tool usage and critical thinking training contribute to student performance in different ways.

Bridging the Expertise Gap with ChatGPT

The group allowed ChatGPT access scored nearly 1 point higher on a 5-point evaluation scale. Their answers were richer in ideas, had higher logical consistency, and showed high similarity to expert recommendations. Students did not simply delegate tasks to the AI, but instead went through a process of framing questions and evaluating responses.

The Unique Value of Critical Thinking Training

Students who received causal inference training explained more clearly why ideas work and when they might fail, but showed no score difference under existing evaluation criteria (achievement of marketing goals). However, automated text analysis results indicated that this group produced more diverse and unique ideas than their peers. This suggests that existing evaluation methods fail to capture the originality they miss.

Complementary Effects and the Need for Changes in Evaluation Methods

Students who received both ChatGPT access and critical thinking training exhibited both effects. AI enhances the completeness and logic of answers, while thinking training fosters original ideas and questioning of assumptions. As AI facilitates the generation of refined answers, educational settings must adapt assignments and evaluation methods to measure other competencies such as originality.