AI access nearly eliminates people's willingness to say 'I don't know,' study finds

September 26, 2026 at 09:56 PM2 min

This digest was compiled by AI from multiple sources — links to the originals are below.

Researchers ran five experiments with 3,132 participants and found that access to a language model nearly wiped out people's willingness to say "I don't know." With AI advice available, judgment suspension dropped from 36-44 percent to 3-6 percent, while confidence more than doubled. Correct answers fell from 27.5 to 9.2 percent, showing that people deferred to AI even when it was mostly wrong.

Key Facts

In studies 1a and 1b, participants without AI access withheld judgment on 36% and 44% of questions, but with AI access those rates fell to 6% and 3%.
In Study 2, confidence with AI access reached 75.9 out of 100 points compared to 29.6 without AI, while the share of correct answers dropped from 27.6% to 10.0%.
Across all studies, participants without incentives who had AI access answered 9.2% of questions correctly, versus 27.5% without AI.
Financial incentives reduced how often participants sought AI advice, with 4.53 versus 5.27 out of six possible requests in Study 3.

Experimental Design

Researchers ran five experiments with 3,132 participants to test whether access to a language model changes how people handle uncertainty. The questions were selected so that the model used, Step 3.5 Flash, was almost always wrong, focusing on fine visual details from movies like the color of a team uniform in Bend It Like Beckham. GPT-5.5, Claude 4.6 Sonnet, and Gemini 3.5 Flash answered most other questions correctly but sometimes failed on the harder ones. Because the AI advice was mostly wrong, the effect cannot be explained as reasonable delegation to a reliable tool.

Confidence and Accuracy

In Study 2, without financial incentives, confidence with AI access hit 75.9 points on a 100-point scale, compared to 29.6 points without AI. At the same time, the share of correct answers fell from 27.6% to 10.0%. Across all studies, participants without incentives who had AI access got 9.2% of questions right, versus 27.5% without AI. They answered more questions with AI but were correct only about a third as often.

Incentive Effects

Studies 2 through 4 added financial incentives: participants earned 10 cents for each correct answer, lost 10 cents for each wrong one, and got nothing for "I don't know." The researchers pre-registered the hypothesis that incentives would increase willingness to abstain but that AI availability would weaken that effect. None of the three studies showed a statistically significant interaction. Incentives led participants to seek AI advice less often, with 4.53 versus 5.27 out of six possible requests in Study 3, and to answer correctly more often when AI was available. Judgment suspension rose slightly with incentives but stayed well below the no-AI control group.