Explaining in Time

Explainability has emerged as a critical AI research objective, but the breadth of proposed methods and application domains suggest that criteria for explanation vary greatly. In particular, what counts good explanation, kinds are computationally feasible, become trickier in light oqaque “black box” systems such deep neural networks. Explanation cases drifted from many philosophers stipulated having to involve deductive causal principles mere “interpretation,” which approximates happened target system varying degrees. However, post hoc constructed rationalizations highly problematic social robots operate interactively spaces shared with humans. For contexts, explanations behavior, and, justifications violations expected should make reference socially accepted norms. this article, we show how robot’s actions can face explanatory demands it came act on its decision, goals, tasks, or purposes design had those pursue norms constraints recognizes course action. As result, argue will need be accurate representations system’s operation along causal, purposive, justificatory lines. These generate appropriate references norms—explanations based “interpretability” ultimately fail connect behaviors determinants. We then lay out foundations cognitive robotic architecture HRI, together particular component algorithms, generating engaging dialogues human interactants. Such track actual decision-making themselves determined by normative robot describe use justifications.