Toulmin Argumentation for Explainable Medical AI Diagnosis
Researchers propose structuring ML diagnostic claims (e.g., retinal disease) using Toulmin's argumentation model: claim, grounds, warrant, qualifier, rebuttal, and backing.
A biomarker-extraction model provides grounds (evidence); a MedGemma agent with medical knowledge validates the warrant (reasoning link); qualifiers and rebuttals capture confidence and counterarguments.
Trades raw XAI outputs for structured, human-auditable argument chains—useful when clinicians need to understand not just what the model says, but why.