This article considers the potential methodological impact of the modern field of machine learning on psychology and neuroscience to improve the predictive potential of neuropsychological models. It discusses the argument advanced by proponents of predictive theories that overcoming the research problems that have led to the reproducibility crisis in psychology can be achieved through the use of large datasets, careful evaluation of model prediction errors, and model simplification by dropping irrelevant coefficients. In this respect, it is noted that small-sample studies criticized by predictive theorists cannot be considered a paragon of explanatory psychological models. Theories in cognitive psychology, cognitive neuroscience, neuropsychology, and psychophysiology are more deeply involved in the business of identifying causal mechanisms of cognition and behavior. Theories of mechanistic explanations and machinelearning-inspired predictive models may be distinguished most importantly by the former’s greater precision and greater potential for manipulating the causal factors of cognition and behavior. The latter, when disconnected from the search for real causal mechanisms, might only capture correlations between events. When making important decisions in the context of educational, biomedical, and legal practices, it is better to rely on knowledge of the causal mechanisms of cognition and behavior. Accordingly, predictive models cannot replace theories that strive for mechanistic explanations of cognition and behavior. Rather than choosing between predictive and explanatory methodologies, it would be more fruitful to search for an optimal balance between them.
psychology; machine learning; prediction; explanation; methodology of science.