Accountability in artificial intelligence: what it is and how it works

Abstract Accountability is a cornerstone of the governance artificial intelligence (AI). However, it often defined too imprecisely because its multifaceted nature and sociotechnical structure AI systems imply variety values, practices, measures to which accountability in can refer. We address this lack clarity by defining terms answerability, identifying three conditions possibility (authority recognition, interrogation, limitation power), an architecture seven features (context, range, agent, forum, standards, process, implications). analyze through four goals (compliance, report, oversight, enforcement). argue that these are complementary policy-makers emphasize or prioritize some over others depending on proactive reactive use missions governance.