HiTZ@Antidote: Argumentation-driven Explainable Artificial Intelligence for Digital Medicine

Explanations require multiple layers of sophistication: Effective AI explanations in medicine must balance specificity with accessibility, consider what the audience already knows, point to specific decision-making factors, incorporate expert knowledge beyond the AI's training data, and present evidence against alternative diagnoses—not just support for the predicted outcome.

Argumentation is as important as algorithms: The paper argues that combining deep learning with human argumentation techniques creates better explanations than technical analysis alone, suggesting that how AI systems communicate their reasoning may be as critical as the reasoning itself for clinical decision-making.

Context matters for clinical AI: The research emphasizes that explanations must be tailored to the deliberation needs of medical professionals, implying that one-size-fits-all explanations are insufficient for responsible deployment of AI in healthcare settings where clinicians need to understand and potentially challenge AI recommendations.

New resources enable better research: The team released the Antidote CasiMedicos dataset to support future work on explainable medical AI, recognizing that advancing responsible AI in healthcare requires shared datasets and benchmarks for the research community.

Explainability is multidisciplinary: The project's success depends on combining expertise from deep learning, argumentation theory, and medicine, suggesting that responsible AI development requires collaboration across technical and domain-specific fields rather than technical solutions alone.