Trust and Accuracy in AI-Assisted Visual Problem Solving
Author
Term
4. term
Education
Publication year
2024
Submitted on
2024-06-14
Pages
10
Abstract
Human-AI collaboration has been on the rise in recent years due to Artificial Intelligence (AI) advancements. They have shown great promise in assisting humans with various tasks but pose chal- lenges that are non-trivial to overcome. In this paper, we investigate human-AI collaborative problem-solving in a visual task, where the AI acts as a decision support system, giving feedback mid-way through the task. We manipulate the AI assistant to purposefully be incorrect and change the level of confidence in the message it conveys. Through a within-subject study (N = 25), we investigate how the AI feedback impacts participants of varying skill level in terms of performance, among other metrics. We found that people with lower skill levels benefit greatly in performance, achieving greater results than without feedback, and that people prefer to use the assistant as an inspiration for patterns. Given that, we con- clude that an assistant might benefit from serving as a support type, giving ideas for patterns that the human collaborators might be missing.
Keywords
Documents