The promise of artificial intelligence to revolutionize education is running into a surprisingly human problem: the value of struggle. New research from the Wharton School at the University of Pennsylvania suggests that while AI can significantly improve learning outcomes, its effectiveness hinges on how it’s deployed. Specifically, learners benefit more from structured, system-controlled guidance than from unrestricted, on-demand access to AI assistance.
The study, published on Tuesday, , examined skill development in a cohort of over 200 chess learners over a three-month period. Participants were divided into two groups. One received AI guidance at predetermined intervals, dictated by the system. The other could request AI help whenever they encountered difficulty. Crucially, both groups had access to the same total amount of AI support; the difference lay solely in the timing of its delivery.
The results were striking. Learners receiving the structured guidance improved their performance by approximately 64% during the training period. Those with on-demand access saw an improvement of around 30%. Follow-up testing conducted weeks later revealed that the advantage conferred by structured support persisted, indicating a deeper, more durable understanding of the game rather than a temporary boost in performance.
Researchers attribute this disparity to a phenomenon known as “productive struggle” – the cognitive benefit derived from grappling with challenging problems before receiving assistance. When learners were forced to independently confront difficult positions and received guidance only at carefully calibrated moments, they were more likely to internalize strategies and refine their decision-making processes. The study suggests that the act of working through challenges, even if initially frustrating, is a critical component of effective learning.
The contrast with the on-demand group is telling. Participants with immediate access to AI assistance often resolved problems more quickly in the short term. However, they demonstrated less engagement with the underlying reasoning, and principles. Interviews revealed a common awareness among these learners that frequent assistance might hinder their long-term development, yet they still frequently opted for immediate help when faced with obstacles. This suggests that the ease of access to AI can undermine self-regulation and discourage the kind of deliberate practice necessary for mastery.
The implications of this research extend far beyond the realm of chess. As AI-powered tutors and “copilots” become increasingly integrated into classrooms, professional training programs, and corporate learning platforms, the findings offer valuable guidance for developers and educators. Systems designed to deliver calibrated prompts, staggered hints, and limited access to solutions may be more effective at fostering durable learning than those offering unrestricted, immediate support. The challenge lies in striking a balance between leveraging AI’s adaptive capabilities and preserving the cognitive effort essential for genuine understanding.
The need to understand how people interact with AI is also highlighted by Anthropic’s AI Fluency Index, which measures how individuals are developing proficiency in using AI tools. The index tracks behaviors such as iterative refinement, questioning the AI’s reasoning, and clearly defining goals – all indicators of effective collaboration with AI systems. Early findings suggest that the most productive interactions involve users building upon previous exchanges with the AI, rather than simply accepting the initial response at face value. This reinforces the idea that AI is most effective when used as a tool to augment human intelligence, rather than replace it.
The rise of AI-powered learning platforms is already transforming the landscape of professional development. According to a report from 360Learning, an estimated 1.4 billion people will need to reskill within the next three years to adapt to the changing demands of the job market. This underscores the urgency of developing effective learning solutions that can equip individuals with the skills needed to thrive in an AI-driven world. The challenge, as the Wharton study demonstrates, is not simply about providing access to AI tools, but about designing learning experiences that maximize their potential while preserving the cognitive benefits of productive struggle.
Microsoft is also actively integrating AI into its learning platforms, with features like “Ask Learn” – an AI assistant designed to provide personalized support and troubleshooting assistance. The company is also introducing new “Applied Skills” scenarios, including training on how to accelerate app development using GitHub Copilot and create AI agents using Microsoft Copilot Studio. These initiatives reflect a broader trend towards leveraging AI to enhance skill acquisition and empower individuals to navigate the evolving technological landscape.
The findings from Wharton, coupled with the growing investment in AI-powered learning solutions, suggest a pivotal moment in the evolution of education and training. The key takeaway is that the mere presence of AI is not enough. The design of the learning experience – specifically, the balance between guidance and independent effort – will ultimately determine whether AI serves as a catalyst for genuine skill development or simply a shortcut to superficial understanding.
