Blocking Requires Uncertainty about Novel Cues

Blocking Requires Uncertainty about Novel Cues Pei-Pei Liu (peiliu@notes.cc.sunysb.edu) Department of Psychology Stony Brook, NY 11729-2500 USA Christian C. Luhmann (christian.luhmann@stonybrook.edu) Department of Psychology Stony Brook, NY 11729-2500 USA Abstract Blocking is a well-studied learning phenomenon in which previous learning inhibits subsequent learning about novel cues. Existing models provide different explanations for blocking and predict different beliefs about novel cues early in the second phase of blocking. Two experiments examined learners’ beliefs when first encountering novel cues. The results suggest that the introduction of the novel cue in the second phase of a blocking paradigm adds uncertainty and that learners entertain the possibility that novel cues are preventative. A novel computational account is proposed to explain people’s beliefs, because existing models cannot fully account for these findings. Keywords: inference learning, blocking, uncertainty, Bayesian Blocking was first reported by Kamin (1969) and is one of the most intensively studied phenomena in the field of learning (Pineno & Miller, 2007). In a blocking paradigm, participants first learn that the presence of a single cue event (cue A) is always followed by a certain outcome (O). Subsequently, cue A is paired with another, novel event (cue X), and this pair is followed by the same outcome (O). Despite the fact that cue X is always followed by the outcome, learners do not learn to associate cue X with the outcome. Learning about cue X is said to be blocked by the previous experience with cue A. Blocking was first observed in conditioning experiments using non-human animals, and thus theories of conditioning have been developed to explain this phenomenon. According to one of the most famous proposals (Rescorla & Wagner, 1972), learning only occurs when outcomes are unexpected. The first phase of the blocking paradigm is designed so that learners come to expect the outcome when they encounter cue A. In the second phase, when cue A is paired with the novel cue X, the Rescorla-Wagner model suggests that learners will expect the same outcome that followed cue A. Given that this is exactly what follows the A-X pair, there is no surprise and thus no learning takes place. Thus, Rescorla-Wagner predicts that nothing is learned about the relationship between cue X and the outcome because there is never an opportunity to do so. Attentional accounts (Mackintosh, 1975a; Kruschke, 2001), on the other hand, argue that blocking occurs because participants learn that cue X does not predict any change in the outcome (beyond cue A itself) and thus come to ignore cue X. Kruschke and Blair (2000) reported evidence consistent with this proposal. They first presented participants with a traditional blocking procedure. Subsequently, another novel cue was added to the A-X pair, which was then followed by a novel outcome. Their results showed that there was less learning about cue X even for the new, novel outcome; a finding inconsistent with Rescorla- Wagner. Such results suggest that the blocking procedure caused learners to ignore cue X, which subsequently prevented learners from learning about the relationship between cue X and the novel outcome. In addition, Mackintosh (1975b) conducted a blocking study in which rats first experienced a light repeatedly paired with a shock. Later, the rats received a single trial on which light and tone were presented together and followed by a shock. When tested, the rats’ behavior indicated that they (weakly) expected the shock to follow the tone presented by itself. That is, the rats learned something about the novel cue after receiving only a single compound A-X trial. This suggests that, contrary to Rescorla-Wagner, learning in a blocking paradigm proceeds normally, at least during the first trial of the second phase, and that at least one compound trial may be required before blocking occurs. Although the blocking paradigm involves a sequence of trial-by-trial presentations, previous studies have largely focused on learners’ expectations at the end of the entire blocking procedure (the Mackintosh, 1975b study is a rare exception). Given that the prominent theories of blocking all make predictions about the trial-by-trial dynamics that underlie blocking, direct measurement of these dynamics seems to be an efficient way to distinguish between the competing theories. In the present study, we investigate blocking by focusing on participants’ beliefs about the novel cue X. According to the Rescorla-Wagner model, participants’ expectations about the outcome the first time they encounter the A-X pair should be identical to their expectations at the end of the first phase when confronted with A alone. In contrast, the attentional accounts predict that, participants must gradually learn to ignore the novel cue X. Thus, the first time participants observe the A-X pair, their expectations about the outcome should be less certain than their expectations about cue A alone. Experiment 1A: The First Phase 2 Trial Method