AI Aversion: Do People Trust and Accept Artificial Intelligence Risk Calculator Recommendations?

Area of Science:

• Health Informatics
• Human-Computer Interaction
• Medical Decision Making

Background:

• Medical risk calculators are widely used, but user aversion to their recommendations is a concern.
• The increasing integration of artificial intelligence (AI) in healthcare may exacerbate this aversion.
• Explainable AI (XAI) is proposed as a method to increase transparency and mitigate AI aversion.

Purpose of the Study:

• To investigate the impact of calculator type (statistical vs. AI) and model explanation on public trust, acceptance, and comfort.
• To examine the role of a risk reference table in evaluating AI-based risk calculator outputs.
• To understand user perceptions of AI-driven medical risk prediction tools.

Main Methods:

• A randomized experimental design with a 2x2x2 factorial structure was employed.
• Participants were exposed to hypothetical scenarios with risk calculator outputs.
• Measures included trust, comfort, and acceptance of the calculator's recommendation.

Main Results:

• AI-based calculator outputs led to lower trust and comfort compared to statistical models.
• Providing an explanation of the underlying model significantly increased trust and acceptance.
• No significant effect of XAI on trust, acceptance, and comfort was observed, nor were there significant interactions.

Conclusions:

• Findings support the AI aversion hypothesis in the context of medical risk calculators.
• Explanations of AI models can foster greater trust and acceptance.
• More research is required to determine optimal methods for explaining AI-based calculators to enhance user trust and comfort.