Conjunction fallacies (CF) have not only been a major obstacle in justifying the rationality of a Bayesian theory of belief update; they have also inspired a variety of theories on probability judgment and logical predication. Here we provide an overview of Bayesian logic (BL) as rational formulation of a pattern-based class of conjunction fallacies. BL is described here as a generalization of Bayesian Occam’s razor. BL captures the idea that probabilities are sometimes used not extensionally but intensionally, determining the probabilistic adequacy of ideal logical patterns. It is emphasized that BL is a class of models that depend on representations and the mea- nings of logical connectives. We discuss open questions and limits of BL. We also briefly discuss whether other theories of the CF may be good supplementary theories of CFs (and predication) as well, if linked to functional explanations.

Social dilemmas specify situations in which (local) egoistic utility optimization prevents achieving the (global) common good of a group. Tragically, in such dilemmas local optimization also reduces the payoff for the individual optimizer. Although social dilemmas essentially reflect inter-individual contexts (conflicting interests, moral attitudes, etc.), innerindividual dilemmas apparently share at least some structural aspects with them: individual behavior can concern more conflicting levels of optimization. For example, starting additional academic projects with potentially positive ‘payoff’ may assume ‘more is more’. However, exogenous effects may arise from optimizing local goals; further contributions may incrementally reduce the quality of other contributions and yield ‘more is less’. In three experiments we explore a oneperson investment game about building hotels, reflecting a social dilemma. The payoffs involve different optima for local and global optimization. Results show that people can be influenced by a default-strategy of ‘more is more’, even if it is irrational.

This paper analyzes probability judgments about properties that can take multiple values (i.e., monadic polytomous events). It extends previous work on pattern-based deviations from standard (extensional) probabilities. Pattern-probabilities are formalized in Bayesian Logic (BL) and should be applicable when testing the overall adequacy of alternative logical hypotheses while allowing for exceptions. BL systematically predicts ‘conjunction fallacies’ (CFs) and, more generally, ‘inclusion fallacies’ (IFs), when a subset is deemed more probable than its superset. BL formalizes a fit between data and explanatory noisy-logical patterns and was supported in previous experiments on dyadic logical connectives with two dichotomous events. Here BL is extended to monadic prediction with several subclasses. BL may for instance predict Ppattern(A) > Ppattern(non-A) even though f(A) < f(non-A), given that non-A has more subclasses than A. Two experiments using material from the Linda paradigm corroborate a pattern approach and rule out confirmation as an alternative explanation.