Information coding and the delta rule

The Rescorla-Wagner model of conditioning has been proposed as a model of human contingency judgment, but the Rescorla-Wagner model cannot account for the phenomenon known as backward blocking. The model fails because it does not allow the strength of a cue to change when that cue is not present. The delta rule network is identical to the Rescorla-Wagner model when presence of a stimulus is coded as 1 and absence is coded as 0. Therefore the delta rule cannot account for backward blocking using 1/0 coding. But Information Coding, proposed in this paper, codes the absence of a stimulus (either cue or target) as less than zero under some circumstances, i.e., when that absence is informative, as defined in this paper, and allows the delta rule to predict backward blocking. Some predictions of the delta rule with Information Coding were tested in the studies reported here. The studies were conducted by presenting college students with trial-by-trial data regarding several medical symptoms (cues) and a fictional disease (the target). Experimental Section 1 tested the prediction that pairing the informative absence of a cue with the presence of a target causes the predictive strength of the cue to decrease. Experimental Section 2 tested the prediction that pairing the presence of a cue with the absence of a target at the beginning of an experiment causes the cue to become an inhibitor. Experimental Section 3 tested the prediction that pairing an informatively absent cue with an absent target increases the predictive strength of the cue. The predictions were all supported. The results of these studies suggest that one factor that subjects take into account in contingency judgments is the correlation between cue and target. It also appears that the absence of a cue can be informative in contingency judgment tasks. The delta rule with Information Coding assumes that a cue\u27s absence can in fact provide information about the correlation between cue and target