Should you show the AI's confidence?
Design it like this
Show confidence where it varies meaningfully case to case, and visually flag low-confidence answers.
What the research says
Showing a confidence score helped people trust the AI more when it was right and less when it was wrong. But it didn't automatically improve final decisions. The goal is trust that matches how reliable the AI really is, not maximum trust.
“we show that confidence score can help calibrate people’s trust in an AI model, but trust calibration alone is not sufficient to improve AI-assisted decision making”
Zhang et al., 2020
Sources
- Effect of Confidence and Explanation on Accuracy and Trust Calibration in AI-Assisted Decision Making2020 · FAT* 2020 · Yunfeng Zhang, Q. Vera Liao, Rachel K. E. Bellamy
“we show that confidence score can help calibrate people’s trust in an AI model, but trust calibration alone is not sufficient to improve AI-assisted decision making”
Abstract - Trust in Automation: Designing for Appropriate Reliance2004 · Human Factors · John D. Lee, Katrina A. See