R.P. Raghupathi
Bio:
Dr. R.P. Raghupathi, drawing on a law background, is collaborating with Prof. Saharia and Prof. Ren on a trilateral model of responsible AI that includes design, ethics, and regulation. In addition, Dr. Raghupathi conducts empirical work on reproducibility of AI and explainability in deep learning in health care and law domains. Dr. Raghupathi’s long-term research has focused on the application of AI and deep learning in legal reasoning. Dr. Raghupathi also applies machine learning text analytics to the analysis of legal cases in AI litigation to gain insight into the key issues surrounding responsible AI.
Abstract:
Explainability in deep learning for healthcare is often portrayed as a cure-all for the “black-box” problem. However, universal transparency can create confusion, bias, and cognitive overload. This paper asks: Is explainability required for all AI in healthcare? Integrating systemic concepts, the analysis argues that explainability is a context-dependent systemic property, essential where AI intersects with clinical reasoning, ethics, or accountability, but unnecessary for routine or axiomatic applications such as scheduling, signal normalization, or resource optimization. Through the metaphors of panacea and Pandora’s box, the paper shows that explainability becomes a panacea when proportionate and embedded within socio-technical feedback loops, but a Pandora’s box when imposed universally or superficially. Limitations of current explainable-AI techniques—Saliency Maps, LIME, and SHAP—are examined. Explainability is reframed as a risk-proportionate systemic capability: adaptive, contextual, and required only where human judgment and patient safety converge.