Generative artificial intelligence (AI) is rapidly transforming numerous aspects of our lives, and education is no exception. While generative AI tools like GPT-4 show incredible promise in boosting productivity and providing immediate assistance, a new study from The Wharton School raises a critical concern: Generative Ai Can Harm Learning. This research delves into the nuanced impact of these powerful tools on skill acquisition and long-term educational outcomes, highlighting potential pitfalls alongside the widely acknowledged benefits.
The Double-Edged Sword of AI in Education: Performance vs. Learning
The study, detailed in a research paper from The Wharton School, investigates the effect of generative AI on learning in a real-world educational setting. Researchers Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakcı, and Rei Mariman conducted a field experiment involving nearly a thousand high school students. They introduced two GPT-4 based tutors into math classes, representing about 15% of the curriculum. One tutor, “GPT Base,” mirrored a standard ChatGPT interface, while the other, “GPT Tutor,” incorporated prompts designed to protect and encourage learning.
The initial results were promising, echoing previous findings on AI’s positive impact on performance. Students using GPT-4 tutors showed significant improvement: 48% with GPT Base and an impressive 127% with GPT Tutor. This immediate performance boost underscores the potential of AI to assist students and enhance their problem-solving capabilities in the short term.
The Crutch Effect: Generative AI’s Negative Impact on Long-Term Learning
However, the study uncovered a concerning downside. When the researchers removed access to the GPT-4 tutors, students who had previously used GPT Base performed worse than students who had never used AI assistance. This group experienced a 17% reduction in performance, indicating that while generative AI improved immediate results, it hindered actual learning and skill development. This phenomenon suggests that students may become overly reliant on generative AI, using it as a “crutch” during practice. While this reliance helps them solve problems with AI assistance, it prevents them from developing the independent problem-solving skills necessary for long-term success.
This negative learning effect was significantly mitigated by the GPT Tutor, which incorporated safeguards to encourage active learning. This highlights the crucial role of thoughtful design and implementation when integrating generative AI into education.
Why AI Assistance Can Hinder Skill Acquisition
The research suggests that the ease of access to AI solutions can inadvertently discourage students from engaging in the deeper cognitive processes required for true learning. When students readily obtain answers or solutions from AI, they may bypass the struggle and critical thinking that are essential for internalizing concepts and developing problem-solving abilities. This reliance can lead to a superficial understanding and a decline in independent skills over time.
Safeguarding Learning in the Age of Generative AI
The findings of this study carry significant implications for educators and policymakers. While generative AI offers exciting opportunities to enhance education, it’s crucial to be aware of its potential to impede learning if not implemented carefully. The success of GPT Tutor in mitigating negative effects suggests that incorporating specific pedagogical strategies into AI tools can help maximize their benefits while minimizing the risks. These strategies might include prompts that encourage students to explain their reasoning, break down problems step-by-step, and reflect on the solutions provided by AI, rather than simply accepting them at face value.
Conclusion: Proceed with Caution and Thoughtful Integration
The Wharton School research provides compelling evidence that generative AI can harm learning if used without careful consideration of its impact on skill development. While the immediate performance gains are undeniable, educators must prioritize long-term learning outcomes. Moving forward, the focus should be on designing and implementing AI tools in ways that foster genuine understanding and independent problem-solving skills, rather than simply providing shortcuts to answers. This balanced approach will ensure that generative AI serves as a powerful tool to enhance, not hinder, the educational journey.
Reference:
Bastani, Hamsa and Bastani, Osbert and Sungu, Alp and Ge, Haosen and Kabakcı, Özge and Mariman, Rei, Generative AI Can Harm Learning (July 15, 2024). The Wharton School Research Paper, Available at SSRN: https://ssrn.com/abstract=4895486
[
