What an Agent Ought To Do A Review of John Horty’s ‘Agency and Deontic Logic’

John Horty’s book ‘Agency and deontic logic’ appeared at Oxford Press in 2001. It develops deontic logic against the background of a theory of agency in nondeterministic time. Several philosophical reviews of the book appeared since then [1–5]. Our goal is to present the book to a general AI audience that is familiar with action theories developed in AI, classical decision theory [6], or formalizations of temporal reasoning like Computation Tree Logic (CTL) [7, 8]. Therefore, in contrast to the philosophical reviews, we discuss and explain several key examples in the book. We do not explicitly discuss the relevance for AI and law, because the book itself is not concerned with the application of the theory to the legal domain. However, the relevance of deontic logic and normative reasoning for legal reasoning is well established by a number of publications on deontic logic in AI and law, see for example the special issue of this journal on agents and norms (volume 4, 1999). Horty presents a formal account of what individuals and groups of agents ought to do under various conditions and over extended periods of time, using the ‘Seeing To It That’ or STIT framework [9]. He explicitly develops a utilitarian / consequentialist perspective, which means roughly that an act is obligatory if performing it results in an optimal state of affairs. However, the question whether a state of affairs is ‘optimal’ is not a question that is exclusively linked to the deontic point of view. And also, seeing deontic necessity only from the perspective of an agent’s welfare (optimality), might not suffice to model all subtleties involved in the semantics of deontic notions. Therefore, it is easy to be confused by the examples; sometimes it is not immediately clear why they are especially relevant for deontic logic. Horty focusses on the common assumption that what an agent ought to do can be identified with the notion of what it ought to be that the agent does, and argues that this assumption is wrong. The assumption is based on the well-known conceptual distinction in deontic logic that concerns the notions of ought-to-be and ought-to-do. Roughly, ought-to-be deontic statements express a norm about the satisfaction of certain conditions at certain moments. Oughtto-do deontic statements apply to actions, which have been argued to fall in a different ontological category than conditions [10, 11]. The distinction between ought-to-do and ought-to-be forms the starting point for Horty’s journey.