Science & tech · 2020
Can seeing an AI make mistakes early change how experts trust it?
Yes. In a controlled user study, subject experts who saw an image-classification system make mistakes early formed negative first impressions that lowered their trust during subsequent interactions. The system’s accuracy was held constant while errors appeared early or late. Novices, by contrast, relied too heavily on the system because they lacked the knowledge to detect errors; the abstract does not establish the same ordering effect for them.
Details
The study examined whether the order in which users encounter mistakes shapes trust differently depending on their knowledge of the subject. Participants reviewed a system that classified images and provided explanations. Its accuracy stayed constant, but the errors appeared either near the beginning or near the end of use.
For experts, early mistakes could leave a negative impression that continued to affect trust. When knowledgeable users saw correct outputs first, they were better able to adjust their trust as they observed the system’s performance. Novices instead showed excessive reliance linked to difficulty recognizing mistakes. These findings distinguish trust shaped by first impressions from trust limited by a lack of relevant knowledge.
The abstract does not report the sample size, the image-classification subject area or the size of the effects. It therefore supports a finding about this controlled user study, not a claim that every AI system or every expert will show the same pattern.
What do you think?
Good conversations begin with curiosity.
You can read comments without an account. Sign in to write.
Loading comments…