Modeling the cognitive processes of accepting clinical decision support

Area of Science:

• Cognitive Psychology
• Artificial Intelligence Ethics
• Clinical Decision Support

Background:

• Algorithm aversion is the tendency to distrust algorithmic advice, even when superior to human input.
• Understanding the cognitive basis of algorithm aversion is crucial for effective human-AI collaboration in critical fields like medicine.

Purpose of the Study:

• To investigate the cognitive mechanisms driving algorithm aversion in a clinical decision-making context.
• To differentiate between increased caution and reduced perceived reliability as causes of algorithm aversion.

Main Methods:

• Two experiments involving participants evaluating X-rays for bone fractures with advice from either an algorithm or a human.
• Utilized evidence accumulation modeling to analyze decision thresholds and response times.
• Compared decision strategies when human advice was attributed to laypersons versus expert radiologists.

Main Results:

• Participants exhibited longer response times and set higher decision thresholds for algorithmic advice, indicating increased deliberation and caution.
• This heightened caution persisted regardless of whether human advice came from lay participants or expert radiologists.
• No significant differences were found in evidence accumulation rates or prior preferences between algorithmic and human advice sources.

Conclusions:

• Algorithm aversion is characterized by a strategic increase in decision caution rather than a diminished perception of algorithmic reliability.
• Formal cognitive models are valuable for understanding trust in automated systems and optimizing human-algorithm collaboration.
• Identifying response caution as a core mechanism advances the theoretical understanding of algorithm aversion.