Measured Cognitive Costs of AI Assistance
Randomized evidence on how AI assistance changes human persistence and skill will keep accumulating, pushing education buyers, regulators and employers toward interaction designs that favor guidance over direct answers.
forming · confidence 40 · Emerging (watchlist) · tracking since October 14, 2026 · updated October 14, 2026
Why the conviction moved
- Oct 14Strengthened +5
Three randomized trials with 1,222 participants, presented at the Conference on Language Modeling, found that ten minutes of chatbot help reduced persistence once the AI was taken away. Participants who asked for direct answers lost the most accuracy, while those who asked for hints did not show the same decline. This is randomized evidence pointing directly to hint-style designs, which is the design shift the thesis predicts buyers will push for.
Source trail
Supporting · October 14, 2026
Ten minutes of chatbot help reduced persistence once the AI was removed, three randomized trials find
Three randomized trials with 1,222 participants, presented at the Conference on Language Modeling, found that ten minutes of chatbot help reduced persistence once the AI was taken away. Participants who asked for direct answers lost the most accuracy, while those who asked for hints did not show the same decline. This is randomized evidence pointing directly to hint-style designs, which is the design shift the thesis predicts buyers will push for.
AI ML Big Data (Telegram)
Unlock full source trail, score history, and daily updates.
Unlock TrendsAffected regions & assets
Townsquare
Argue the thesis in Townsquare.