A new study shows that while ChatGPT can boost student grades, students need explicit critical-thinking training to produce original ideas. This research points to a major challenge: standard school grading systems often reward polished, AI-generated work over genuine, creative reasoning.
What Happened
Researchers at Bocconi University and OpenAI Economic Research conducted a randomized controlled trial with 1,053 first-year undergraduate students. The students, split into four groups, tackled a real-world business case. One group used ChatGPT Edu (GPT-4o). A second group received training in causal reasoning. A third group received both, while a control group had neither.
According to our coverage of the study's primary findings, students who used ChatGPT scored 0.86 points higher on a five-point scale than the control group's average of 2.09. Their recommendations also closely matched the advice of industry experts.
However, students trained in causal reasoning, who learned to analyze cause and effect and identify potential failures, showed different progress. Their final answers contained highly diverse and unconventional ideas. Because traditional grading rubrics prioritize standard solutions and polished presentation, these innovative students actually received lower scores.
The Bigger Picture
This discrepancy reveals a flaw in how schools evaluate academic success. Traditional assignments are vulnerable to AI because standard rubrics often reward conformity. A study in Frontiers in Education found that educators who switched to "AI-aware" rubrics saw significant increases in critical thinking and deep learning.
Banning AI or using basic "AI scales" to limit tool usage does not work. A post on Kate Lindsay Blogs warns that restriction-first policies lead to "performance theatre" and unfair, inconsistent grading.
Instead, schools must redesign assignments. The Journal of Instructional Design and Technology suggests frameworks like the "UnBlooms™ Metacognitive Reflection Loop" to ensure tasks require human thought. Without structured approaches, students risk developing a "cognitive dependency" on automated systems, according to a review in the Muallim Journal of Social Sciences and Humanities.
What This Means for Families
For parents, a high grade on a homework assignment no longer guarantees that a child understands the material. Students who rely solely on AI to generate answers miss the thinking processes required for future success.
Instead of banning ChatGPT, parents and teachers should focus on comprehensive AI literacy. Common Sense Education notes that true literacy means understanding how AI works and its limitations, rather than just learning to use the tools. Because technology changes rapidly, teaching children how to write prompts is not enough. We must teach them to treat AI as a "cognitive partner," according to AI in Education.
Used correctly, AI tools can help students study independently. For example, EdTool allows students to generate personalized practice quizzes and get instant feedback. This builds the confidence needed to stay in school, which can raise first-year retention rates by five to eight percent, according to the Presidents Forum.
What You Can Do
- Ask your child to explain the mechanism behind their solution and why an alternative might fail when helping them with homework. This replicates the causal reasoning training used in the Bocconi study.
- Talk to teachers about how their grading systems reward originality, logic, and self-reflection instead of polished, easily generated text.
- Have your child use AI tools to quiz them on core concepts and find knowledge gaps, rather than using the tools to write their papers.