Ought Implies Can, Omission and Probabilistic Deliberative STIT

Roberto Ciuni (RUB) gives a talk at the MCMP Colloquium (24 May, 2012) titled "Ought Implies Can, Omission and Probabilistic Deliberative STIT". Abstract: STIT logics are prominent formal settings for modelling the interaction among agents. Remarkably, they endorse some basic game-theoretical notion, most notably alpha-effectivity: the moves of one agent are independent from the ones of any other agent. This is in turn reflected in the semantics of the deliberative stit operator [a : X], which reads 'agent a ensures - in the next moment'. In the last ten years, STIT logics have been applied to a number of issues at the cross between deontic logic, ethics and legal philosophy, such as the notion of responsibility. The Principle 'Ought implies Can' (OC) is a widespread principle in ethics, which is taken to provide a criterion of fairness for establishing 'oughts'. Responsibility for omission is that kind of responsibility which holds of an agent if she did not ensure something she had the obligation of ensuring. Among the cases of omission, we have omission w.r.t. other agents' wrong-doing (not preventing others from accomplishing violations of norms). In the paper, I show that responsibility for this kind of omission (RNP) is impossible in a deontic STIT logic with the above operator and which endorses OC (the latter being a consequence of employing [a:X]). Since there are reasons for keeping STIT, OC and RNP, I explore a way to fix the problem which does not renounce to any of them: I introduce a deliberative STIT logic which is able to deal with the notions of 'attempting' and 'maximizing the chance that another agent fails'. Interestingly, this is done without renouncing to the 'Independence of Agents', which is indeed the characteristic principle of STIT logics and the responsible of traditional STIT's failure in modeling RNP.