Can People Make Better Decisions with AI?

This recent paper (https://doi.org/10.3390/app14167192) looked at whether people’s decision making in identifying real or fake (artificially generated) images improved when given suggestions from an AI. An accurate and less inaccurate version of AI were tested in the experiment. The paper found that if the AI’s skill level was better than the human participant’s, the participant’s performance improved similarly to how talking over a situation with a more experienced or more skilled colleague would help make a better decision.

This is perhaps unsurprising, but it does mean that people in this experiment were able to sense the AI’s skill level in relation to their own to some extent, and not go along when they knew better. In tasks related to risk or safety hazard identification, if this effect holds up, it could mean that AI helpers could enhance performance and make it less likely that potential risks or hazards would not be identified. Like many of these studies, the sample size was small (N=30), so further investigation is recommended.