The Quality of Test Context and Contra-evidence as a Moderating Factor in the Belief Revision Process

The Quality of Test Context and Contra-evidence as a Moderating Factor in the Belief Revision Process Kristien Dieussaert (kristien.dieussaert@psy.kuleuven.ac.be) University of Leuven, Department of Psychology Tiensestraat 102, Leuven, Belgium Walter Schaeken (walter.schaeken@psy.kuleuven.ac.be) University of Leuven, Department of Psychology Tiensestraat 102, Leuven, Belgium Gery d’ydewalle (gery.dydewalle@psy.kuleuven.ac.be) University of Leuven, Department of Psychology Tiensestraat 102, Leuven, Belgium Abstract In this study, we describe the influences of qualitative changes to the reasoning problem on the reasoning process. The first manipulation is the quality of the test context: A rule is learned in a certain context and contradicted in another. The belief in the rule is then measured in the learning context, the contradictory context, and a new context. The second manipulation is the quality of contradiction: The contradictory rule can neutralize or inverse the learned rule. Both qualitative changes influence the belief revision process. Introduction Research in Artificial Intelligence is often conducted to develop systems that think/act rationally (or like humans, depending on the approach). Minsky (1975) was one of the first to point out the problem with deductive systems, and from then on, several researchers developed non-monotonic reasoning systems (for an overview, see Brewka, Dix, & Konolidge, 1997). While these systems can be very interesting from the viewpoint of an engineer, Rips (1994) mentions two main reasons why these nonmonotonic logics are less than ideal for cognitive purposes: They do not lend themselves to simple implementations (higher order logics are incomplete) and they do not seem to reflect the deliberations that actually underlie human reasoning with defaults. He (1994, p. 299) stated: “For purposes of philosophy (and perhaps AI), we need normative rules about how to change our minds in the face of conflicting evidence; for purposes of psychology, we also need descriptive information on how people deal with the same sort of conflict.” Elio and Pelletier (1997) wrote a pioneering article on this topic. In their experiments, they first presented participants with a conditional premise and a categorical premise that affirmed the antecedent [or denied the consequent]. Then, they added a third piece of information, which conflicted with the conditional and the categorical premise. This piece of information was a categorical premise denying the consequent [or affirming the antecedent]. The three pieces of information together are in contradiction with the valid Modus Ponens [Modus Tollens] argument: If A then B, A, thus B [If A then B, not-B thus not-A]. Participants were asked to resolve the contradiction by rejecting one of the two first pieces of information. Elio and Pelletier (1997) observed that participants chose to disbelieve the conditional premise rather than the categorical one when resolving the contradiction. Their results were refined by Dieussaert, Schaeken, De Neys, and d’Ydewalle (2000) and by Politzer and Carles (2001). They found that the initial belief in the conditional premise influenced the belief revision choice that participants prefer to make. When participants had a strong belief in the conditional premise, the preferred to reject (doubt) the categorical premise, when the conflicting information is added. When participants had a weak belief in the conditional premise, the results of Elio and Pelletier (1997) were confirmed: In this case, participants prefer to reject (doubt) the conditional premise. This shows that it is important in belief revision research to be aware of the belief state participants hold before conflicting information is presented. Therefore, we conducted a measure of the initial belief state in the following experiments, before adding conflicting information. The following experiments were inspired by research in the field of conditioning. Among others, Bouton (e.g., 1988, 1994) showed that the extinction of behavior does not necessarily means the rejection of a learned rule. One of the phenomena that confirm this hypothesis is ‘renewal’: When a behavior is learned in context A, and extinguished in context B, the behavior might show up again with a new test in context A. Bouton explains this phenomenon as follows: Individuals learn a dominant rule and exceptions to this rule in certain contexts (see also: Holyoak, Koh, &