Research Article
The Effect of Outcome Probability on Generalization in Predictive Learning Hadar Ram1 , Dieter Struyf2, Bram Vervliet2,3, Gal Menahem1, and Nira Liberman1 1
School of Psychological Sciences, Tel Aviv University, Israel
2
Centre for the Psychology of Learning and Experimental Psychopathology, Leuven University, Belgium
3
Harvard Medical School, Massachusetts General Hospital, Boston, MA, USA
Abstract: People apply what they learn from experience not only to the experienced stimuli, but also to novel stimuli. But what determines how widely people generalize what they have learned? Using a predictive learning paradigm, we examined the hypothesis that a low (vs. high) probability of an outcome following a predicting stimulus would widen generalization. In three experiments, participants learned which stimulus predicted an outcome (S+) and which stimulus did not (S ) and then indicated how much they expected the outcome after each of eight novel stimuli ranging in perceptual similarity to S+ and S . The stimuli were rings of different sizes and the outcome was a picture of a lightning bolt. As hypothesized, a lower probability of the outcome widened generalization. That is, novel stimuli that were similar to S+ (but not to S ) produced expectations for the outcome that were as high as those associated with S+. Keywords: generalization, predictive learning, partial reinforcement, learning from experience
The purpose of learning from experience is to enable prediction (De Houwer & Beckers, 2002b; Hohwy, 2013; Liberman, Trope, & Rim, 2011; Suddendorf & Corballis, 2007). When people repeatedly experience event A (e.g., gray clouds/a member of a specific social group) that is followed by outcome B (e.g., rain/the group member offers help), they learn to predict outcome B from event A (e.g., expect rain when seeing gray clouds/expect help from another member of the same group). They might learn that another event, C (e.g., white clouds/a member of a different social group), does not predict the same outcome. Prediction is useful because it enables preparing for the future (e.g., take an umbrella/choose whom to ask for help). Importantly, experience has to be generalized in order to apply to a new situation. That is, outcome predictions should be made for events that are similar but not identical to the original event A. An interesting question is that of generalization breadth, namely what determines the range of stimuli (e.g., the range of grayness/the range of group members) for which outcome B would be predicted.
Ó 2019 Hogrefe Publishing
The factors that affect generalization breadth have been studied mainly within the psychology of learning, but are clearly related also to central topics in social psychology, such as attitudes and stereotypes. For example, when we hear a great talk at a conference, we may not only form a positive attitude toward the speaker, but also generalize this attitude toward the speaker’s laboratory members, his/her discipline or even his/her national group, thereby changing, strengthening, or helping to create social stereotypes. In the present article, we examine one factor that may affect generalization breadth, namely the probability that an outcome appears following a predicting stimulus. Specifically, we examine the hypothesis that low (vs. high) outcome probability (i.e., reinforcement) after a cue (S+) widens generalization. In the learning literature, this factor has been termed reinforcement rate. In what follows, we first present classic and contemporary models of learning which addressed the question of generalization as a function of reinforcement rate while also reviewing central relevant findings. We then turn to discuss why effects of reinforcement rate/outcome probability on generalization are important and elaborate why we believe that they are of interest for social psychology. Thereafter, we describe the predictive learning paradigm we use in our experiments and state our hypotheses in the more concrete terms of that paradigm.
Experimental Psychology (2019), 66(1), 23–39 https://doi.org/10.1027/1618-3169/a000429
