The effects of emotion on human inference: Towards a computational model

A common criticism of AI has been that computational models have not been theoretically grounded and as such lack testability. In this research I attempt to address this criticism by looking at the interaction between emotions and reasoning from both a theoretical and empirical perspective then suggesting a possible role for emotions in both human and machine inference. Three studies examining the role of emotions in three different types of reasoning tasks were conducted: juror reasoning, syllogistic reasoning and narrative reasoning from text. In the first two studies (juror and syllogistic reasoning), happy and sad emotional states were induced by asking subjects to view happy and sad film clips. In the juror reasoning task, subjects were asked to read through a trial transcript to offer a verdict and to answer some other questions about the trial; in the syllogistic reasoning task, subjects were asked to ascertain the validity of ten categorical syllogisms. In the narrative reasoning task no mood inductions were performed, instead subjects were asked to read a short story, and to record simultaneously what emotions they felt. After completing the story they were asked to answer interpretive questions pertaining to the story. The results of all three studies showed that affect plays roles in reasoning that have significant measurable effects--even in a domain as structured as syllogistic reasoning.; What the results of these studies suggested, moreover, is that it is not enough to merely look at the influences of positive and negative mood on social judgment (the method of the large majority of studies thus far performed) but that the influence of particular emotions needs to be examined more carefully. In each study it was not just the effect of positive or negative mood induction, but more specific effects of particular emotions (usually happiness, sadness or anger) that influenced reasoning style. Analyses of the data also suggested that the dynamic event of a change in intensity of emotion may be worth considering as a cause of triggering particular chains of reasoning. The reasoning strategy each emotion triggered was based on a number of variables such as the task domain; the nature of prior search; the juncture reached in the ongoing plan; and the inferencing patterns shown by people in similar situations. A model machine inference that incorporates the effects of emotions in human reasoning that would be useful for programming commonsense reasoning in a multiple-agent, multiple-goal dynamic environment is presented.