By Nyang’au Araka
A new randomized controlled trial has raised questions about the role of generative AI in higher education.
The trial found that students who relied on ChatGPT during study retained significantly less knowledge over time compared to peers using traditional methods.
The research, conducted at the Federal University of Rio de Janeiro and published in Social Sciences & Humanities Open, tested 120 undergraduate business administration students learning concepts in artificial intelligence and machine learning.
Half were allowed unrestricted use of ChatGPT, while the other half studied using conventional resources such as books and articles.
Forty-five days later, both groups were given a surprise retention test and the results were striking.
Sudents in the AI-assisted group scored an average of 57.5 percent correct, while those in the traditional group achieved 68.5 percent.
“Students who used ChatGPT scored significantly lower on the retention test,” the paper reports.
Lead author André Barcaui explained the findings in terms of cognitive offloading—the tendency to rely on external tools rather than internal memory.
“The ease of obtaining instant, comprehensive answers from ChatGPT could be creating a generation of students who remember where to find information rather than the information itself,” he wrote, echoing earlier concerns about the so-called “Google effect” on memory.
The study also drew on the principle of “desirable difficulties,” which suggests that effortful learning strategies such as retrieval practice, spacing, and generation enhance long-term retention even if they feel harder in the moment.
By removing these productive struggles, AI tools may inadvertently weaken memory consolidation.
“While AI assistance may ease initial learning, it appears to undermine the effortful processes needed for robust learning,” the authors concluded.
The forgetting curve—the rate at which knowledge decays over time—was steeper for the AI-assisted group.
This supports the prediction that bypassing effortful encoding leads to weaker hippocampus-dependent memory traces, which decay faster.
“We expect a steeper forgetting curve in the AI condition because reduced effort during initial learning should impair systems-level consolidation,” the paper explained.
Students worked in naturalistic conditions, preparing presentations on assigned topics with or without AI support, and were tested without prior warning.
This design mirrored authentic student behaviour rather than artificial laboratory tasks.
The findings arrive amid widespread adoption of generative AI in universities worldwide.
Surveys suggest that 60–80 percent of students have experimented with ChatGPT for academic purposes.
Advocates highlight its potential for personalised tutoring and instant feedback, but critics warn of dependency and diminished critical thinking.
Barcaui’s trial adds weight to those concerns by providing rare causal evidence.
“Unrestricted ChatGPT use impaired long-term retention, likely by reducing the cognitive effort that supports durable memory,” the study states.
The researcher opines that while AI can support immediate comprehension, its integration into education must be carefully managed to avoid undermining the very learning outcomes institutions seek to promote.
“These results have important implications for how generative AI tools should be integrated into higher education,” the author said.
Editor’s note: This article is not intended to attack and/ or discredit ChatGPT or any other form of AI but to create awareness on prudent use of the same.