Inclusive Explainable AI

Artificial intelligence (AI) and machine learning methods are increasingly relying on complex, black-box models that are not easily understood, especially by everyday users. The goal of research on explainable AI (XAI) is to improve the transparency and understandability of the decisions made by AI systems. However, in current state-of-the-art XAI methods, explanations are typically treated as static one-way interactions without consideration of the expanation partner's understanding. The aim of the Transregional Collaborative Research Center “Constructing Explainability” (TRR 318) is to change this by considering explanations within a social context and constructing them together with the human explainee and the machine explainer.

Our sub-project, “Explaining the multimodal display of stress in clinical explanations” (Project A06), aims to improve clinical explanations of stress by examining signals of stress during an explanatory process and comparing different explanatory strategies and AI-assisted approaches.
In a co-constructive explanation process, a machine explainer should monitor and analyze the explainee’s understanding to tailor the provided explanation to their current level of understanding. However, these signals of understanding are influenced by stress, which is why detecting stress is critical for accurately interpreting signals of understanding. This is particularly important for clinical staff, as they often have to provide explanations to stressed individuals and misinterpretations of understanding can have negative consequences. To develop techniques that train clinical staff to recognize signs of stress in explanatory situations, we first investigate how people express stress in such situations. We then compare both traditional and AI-supported techniques for explaining these signs.

We gratefully acknowledge funding by the Deutsche Forschungsgemeinschaft
(DFG, German Research Foundation): TRR 318/1 2021 – 438445824