Feedforward versus feedbackward: An ethological alternative to the law of effect

Abstract The view that learning is governed by positive and negative consequences has dominated theory and application throughout this century. In some systems stimulus-response connections are stamped in or stamped out by the consequences of action, in others, cognitive expectancies are formed by experience with past consequences. The evidence from early experiments with rats and pigeons and the feedback principles of early servomechanisms seemed to offer both hard evidence and a plausible model for the law of effect in either its behaviorist or its cognitive form. A large body of evidence demonstrates that the results of operant conditioning appear regardless of and often in spite of response-contingent consequences. Experiments designed to measure a residual effect of consequences exhibit an inevitable ex post facto error that vitiates all possible versions of this experimental design. Experiments designed to measure the effect of predictive contingency in Pavlovian conditioning exhibit a corresponding error. There appears to be a fundamental logical defect in all contingency models of the learning process. Meanwhile, modern developments in ethology and computer science provide a unified feedforward model of the learning of adaptive and maladaptive behavior under both laboratory and field conditions. Because the feedforward model is more parsimonious, it is also more compatible with Darwinian principles of biological economy. Research on teaching new and challenging tasks to freeliving, well-fed subjects such as children and cross-fostered chimpanzees illustrates the wide applicability and practical effectiveness of feedforward.