Scenarios of Social Explainable AI in Practice

Abstract

Abstract A key goal of explainable AI (XAI) is to ensure the trustworthiness of AI systems when they interact with humans in real-world settings. These interactions involve individuals with diverse backgrounds, varying levels of knowledge, and different abilities to comprehend explanations. This chapter presents scenarios that illustrate the challenges and requirements that arise for XAI methods in such real-life contexts. Specifically, we highlight the importance of adapting explanations to both the context and the explainee(s), which may involve using appropriate and multiple modalities. Based on these scenarios, we identify three key requirements for effective XAI systems: multimodality (the ability to use different explanation formats), incrementality (the ability to refine explanations over time), and patternedness (the ability to present explanations in a structured and recognizable manner).

Ähnliche Arbeiten