Towards a Flexible Bayesian and Deontic Logic of Testing Descriptive and Prescriptive Rules

Abstract

Research on the Wason selection task (WST) has raised fundamental doubts about the rationality of human hypothesis testing and added to the development of both domain-specific and domain-general theories of reasoning. This work proposes a rational but domain-specific synthesis aimed at integrating converging lines of research in the WST debate. For this synthesis two realms are distinguished, that of testing descriptive rules (hypotheses) and that of testing prescriptive rules (prescripts). For both realms, accounts are proposed that have normative aspects, but also domain-specific aspects. For the testing of descriptive hypotheses, a flexible Bayesian logic is developed, which is opposed to the falsificationist research program and builds on previous Bayesian accounts (on Oaksford and Chater, 1994, 2003, in particular). However, instead of advocating a universal Bayesian model a knowledge-based account is pursued which may explain the negative results of previous experiments. Additionally, the Bayesian treatment of conditionals is extended to other logical connectors. For the testing of conditional prescripts, a flexible deontic logic is proposed which draws more fully on the deontic logic of prohibitions, obligations and permissions than previous accounts. Moreover, this is combined with a goal-based, but systematic, mechanism of cooperator and cheater detection. Twelve experiments largely support the predictions of the proposed account. In two experiments the different strategies for testing descriptive or prescriptive rules were investigated. In seven experiments, positive support for the Bayesian account was obtained by actively introducing the preconditions of the models. Additionally, different Bayesian models of a conditional were distinguished and first evidence for a Bayesian logic of different connectors was obtained. In three experiments, the deontic logic of checking prescripts and its interaction with the goals of cheater or cooperator detection (including double foci) was investigated. The results cannot be explained by other current theories of the selection task, such as mental model theory, social contract theory, or pragmatic resoning theory, but do at least necessitate substantial extensions of these theories. In contrast, the results support the flexible Bayesian and deontic logic of testing descriptive and prescriptive rules.