Enriched conversational XAI methods for healthcare

Despite the rich set of eXplainable Artificial Intelligence (XAI) methods that have been proposed to justify the outcome of Machine Learning (ML) models in healthcare applications, many open challenges remain: most of these efforts are focused on algorithm developers rather than healthcare professionals, who often have little or no knowledge of ML models. The majority of the XAI systems also rely on complex visual representations of models and their features. In addition, the vast majority of the applications of XAI are focused on providing low-level "narrow" explanations of how an individual decision was reached and provide insufficient insight into the reasoning of models and the explanatory depth that people require to accept and trust the decision-making of the model. In this project, we will research the following objectives to address these challenges: first, we will research how conversational explanation methods able to provide insight into model beliefs and reasoning with natural language can be used as opposed to complex visual representations. Second, we will research how these explanations can be enriched to provide more general, broader, explanations with sufficient meaning. Third, we will research how such interfaces can be adapted on the fly to different personal characteristics. The overall objective is to come up with a generic framework for enriched conversational explanation methods that meet the communication needs of healthcare professionals.