AI and productive struggle
A recent paper by researchers at Stanford highlights the "evaluation paradox" at the heart of AI-assisted learning: AI systems that users rate as most helpful don't necessarily lead to better learning outcomes. Their research shows that students often prefer systems that provide immediate solutions but miss the valuable learning that comes from productive struggle.
As it stands the widespread availability of AI tools like ChatGPT risks eroding the skills needed to foster deep understanding by eliminating the struggle necessary for authentic learning. In one experiment, participants who received AI training they rated as less helpful actually demonstrated better skill development and task performance than those who received "preferred" AI assistance.
The authors argue that AI's true value lies not in eliminating challenges but in enhancing it. They developed "Tutor CoPilot" – an AI system designed to help teachers foster productive struggle rather than providing solutions.
Image: The researchers found that students working with tutors that had access to their Tutor CoPilot software were 4 percentage points more likely to pass their math lesson tests.
The need for productive struggle extends beyond education to the workplace, where meaningful skills development and real world expertise increasingly depend on balancing technological tools with opportunities for on-the-job problem solving and growth through doing.