The formal characterization of self-deception

Awareness is of central interest not only to cognitive scientists but also to computer scientists who are developing models of intelligent, adaptive systems. Self-awareness is a significant aspect of awareness, given that some, at least, of the agents in such systems are capable of representing and reasoning about both the cognitive processes of other agents and also their own. Any comprehensive attempt to analyse and characterize self-awareness will have to consider ways in which it may be constrained, among which, it has often been maintained, self-deception figures prominently, at least in humans. The focus of our ongoing work is the application of logic— specifically modal logics of belief—to the task of formally characterizing a group of types of self-deception. The results so far indicate that it is possible to carry through that task without resorting to paraconsistent logic1 while still making explicit the nature of the apparent inconsistency that self-deception exhibits. They also indicate that the type of belief logic commonly chosen in theoretical artificial intelligence is totally unsuited to this task because it contains strong awareness assumptions as axioms. The latter are usually referred to as the ‘positive introspection axiom’ (the modal schema 4, which in the case of belief says that if an agent believes that p then he believes that he believes that p), and the ‘negative introspection axiom’ (the modal schema 5, which in the case of belief says that if an agent does not believe that p then he believes that he does not believe that p). Consequently, our work focuses on the development of a formal-logical conceptual model, using logics which can formallymodel the critical concepts, so that a small ‘family’ of types of self-deception can be clearly characterized. These include, for instance, the situation in which an agent does not believe that p, but believes that he does believe that p,2 and the situation in which an agent believes that p, but believes that he does not believe that p. The approach taken reflects the methodological position described in a recently published paper,3 in which we argue that—not least in the design of sociotechnical Figure 1. Michel de Montaigne (1533–1592). His remark about selfdeception2 was the point of departure for Jaakko Hintikka’s formallogical analysis in his book ‘Knowledge and Belief.’4 Our work criticizes and develops Hintikka’s approach. (Image credit: c © Georgios Kollidas, Dreamstime.com.)