Research map / Large language models / AI and society

Explainability matters: The effect of liability rules on the healthcare sector

NeurIPS 2026 · Large language models › AI and society · approach: Human-centered trust and agency · 1 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

A legal-game-theoretic perspective arguing that AI explainability shapes practitioner and manufacturer liability in healthcare, comparing an opaque "Oracle" with an explainable "AI Colleague" and linking this to defensive medicine.

Problem

How the level of automation and explainability of AI systems affects the allocation of medical liability between practitioners and manufacturers, and whether this incentivizes defensive medicine in healthcare.

Approach

Models two extreme AI archetypes (opaque Oracle vs. explainable AI Colleague) with simple expected-payoff games over practitioner agreement and independent assessment, deriving an indifference diligence threshold k' that falls toward 0.5 as explainability rises, and argues explainability should be a certification criterion alongside accuracy.

Key results

• Under the Oracle model, agreement is dominant for any accuracy above 50% when shared liability is unfavourable, with indifference accuracy p' = X/(X-L) approaching 0.5 as X approaches -L.
• For the AI Colleague model, the indifference diligence threshold k = (1-pE)/((1-p)E+1-pE) approaches 0.5 as explainability E approaches 1.
• Figure analysis fixes accuracy at p=0.7 to show that lower explainability requires higher practitioner diligence.

Surprising findings

• Even a less diligent practitioner (k near 0.5) is better off making independent assessments when AIS is highly explainable.
• Under the Oracle setting, a system with accuracy below 50% could still induce agreement as a liability-minimizing strategy.

Limitations

• Payoffs are hypothetical and simplified for illustration, not calibrated to real legal cases.
• Analysis focuses on the practitioner's liability and treats accuracy p as fixed by certification.
• Legal discussion is largely jurisdiction-general (EU, US, Italy, France, Germany examples) and does not empirically test defensive medicine effects.

Experimental setup

Metrics: expected payoff indifference accuracy threshold diligence threshold k'

Builds on: defensive medicine literature product liability doctrine

Closest papers by meaning

• Evaluating the Effectiveness of Explainable Artificial Intelligence Approaches (Student Abstract) (AAAI 2024 · similarity 0.82)
• Position: AI's growing due process problem (ICML 2025 · similarity 0.81)
• Position: Explain to Question not to Justify (ICML 2024 · similarity 0.81)
• SafetyAnalyst: Interpretable, Transparent, and Steerable Safety Moderation for AI Behavior (ICML 2025 · similarity 0.80)
• Position: Explainable AI Cannot Advance Without Better User Studies (ICML 2025 · similarity 0.79)
• Regulating AI: Applying Insights from Behavioural Economics and Psychology to the Application of Article 5 of the EU AI Act (AAAI 2024 · similarity 0.78)