Clinical Scorecard: Recognizing the Significance of Divergence in Clinical Human–AI Interactions
At a Glance
Category
Detail
Condition
Differences between clinician and AI outputs in clinical decision-making
Key Mechanisms
Divergence may arise from asymmetric information, different interpretations of shared evidence, or an error by either party. The article describes structural, informational, and behavioral axes of the interaction.
Target Population
Clinicians using AI systems in clinical decision-making
Care Setting
Clinical decision-making contexts in which clinicians consult AI outputs
Key Highlights
The article defines human–AI divergence as a difference between clinician and AI outputs, distinguishing it from disagreement based on identical evidence.
Clinicians may contribute bedside examination and communication, while AI processes structured records and learned patterns; the parties may therefore have different information.
Existing interfaces often frame responses to AI outputs as accepting, overriding, or modifying them and may not document the interaction that produced the decision.
The authors propose human–AI divergence visibility: documenting clinician engagement with AI reasoning and what each party contributes when outputs differ.
In a cited primary-care study, clinicians adopted 58% of harmful recommendations and 22% of beneficial recommendations from an LLM-based decision-support tool. Separately, diagnostic accuracy decreased by 14% among physicians given the option to consult LLMs, despite formal training in evaluating AI outputs.
Guideline-Based Recommendations
Diagnosis
Management
Monitoring & Follow-up
Risks
Patient & Prescribing Data
Not specified; the article discusses clinical AI use across clinical decision-making contexts.
No specific treatment or prescribing regimen is provided. The article reports that, in a cited primary-care study, clinicians adopted 58% of harmful recommendations and 22% of beneficial recommendations from a large language model-based decision-support tool.
Clinical Best Practices
The authors argue that differences between clinician and AI outputs should be examined as divergence rather than reduced to simple disagreement.
The authors emphasize considering the distinct information available to the clinician and the AI when interpreting divergent outputs.
The authors call for documenting the interaction, including clinician engagement with AI reasoning and each party’s contribution to the final decision.
The article states that technical explanations of an AI output have not directly corresponded to better clinical decisions and may foster false confidence in erroneous outputs.
Investigators linked distinct odor patterns to life stage, sex, and skin microbes, while limited longitudinal evidence leaves their clinical utility uncertain.
Optical analysis of resection-margin tissue identified tumor infiltration as the strongest contributor to predictions of where glioblastoma later recurred.
