Medical artificial intelligence for clinicians: the lost cognitive perspective.

Area of Science:

• Medical Informatics
• Radiology
• Cognitive Science

Background:

• The rapid development and commercialization of artificial intelligence (AI) in medical decision support systems, particularly in radiology, outpaces the understanding of their true value to clinicians.
• Current AI models often lack the contextual understanding and cognitive flexibility inherent in human clinical decision-making processes.

Purpose of the Study:

• To characterize diagnostic decision-making in radiologists using ecologically bounded reasoning.
• To compare clinician decision-making with AI model decision-making to identify challenges in AI integration.
• To propose future research directions for enhancing clinician-AI interaction, safety, and usability.

Main Methods:

• Analysis of clinician decision-making through the lens of ecologically bounded reasoning.
• Comparative review of cognitive processes in human clinicians versus AI decision-making models.
• Identification of capability misalignments between clinicians and AI.

Main Results:

• Clinicians are contextually motivated, resourceful decision-makers, while AI models are contextually stripped, correlational decision-makers.
• Significant challenges exist in integrating AI into radiology due to fundamental differences in decision-making capabilities.
• Misconceptions about clinician-AI interaction arise from this misalignment.

Conclusions:

• Addressing the cognitive and contextual differences is essential for effective clinician-AI collaboration.
• Future research must focus on the cognitive aspects of decision-making to improve AI safety and usability in high-risk medical contexts.
• Enhanced understanding of these differences will facilitate the responsible implementation of AI in clinical practice.