The AI-fallibility warning did not significantly change error rate or time spent per problem-solving step¶

Claim · All claims
Evidence · 1 study · causal r2 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? The Mistake and NoMistake groups did not differ significantly on error rate or average time spent per problem-solving step. → Tomohiro Nagashima 2026

Evidence¶

Tomohiro Nagashima 2026¶

Tomohiro Nagashima, Mirella Hladký, and Vera Rief. (2026). Warning About AI Fallibility Increases Help-Seeking in an Intelligent Tutoring System. https://arxiv.org/abs/2606.03822 doi:10.48550/arxiv.2606.03822

q3 · i? · causal · r2

Same classroom regression analysis of problem-solving log data: "The groups did not differ significantly on the error rate and average time spent" (error rate β=−0.02, p=.61; time spent β=−0.23, p=.39). Table 1 reports Mistake means of 0.49 error rate and 1.60 time per step versus NoMistake 0.46 and 1.36.

"The groups did not differ significantly on the error rate and average time spent (error rate:β=−0.02,t(235) =−0.51,p=.61; time spent:β=−0.23,t(235) =−0.86,p=.39)."

Discussion¶

Related Claims¶

• A warning message about potential AI mistakes increases seventh-graders' hint requests in a math intelligent tutoring system — related
• Users alter expectations and behavior based on perceived AI capabilities even when actual AI performance is unchanged (the placebo effect of AI) — a broader claim this one bears on
• Adaptive students request significantly fewer on-demand hints and spend longer on a step before requesting help than Control students — related
• Warning labels alone did not improve performance or reduce reliance on GenAI advice, and students sometimes adopted incorrect AI answers — possibly the same claim (merge candidate)
• Error rates per completed problem-solving step did not significantly differ between conditions — possibly the same claim (merge candidate)