Reber, P. J.
Northwestern University
Separate lines of research within cognitive psychology, developmental linguistics and artificial intelligence have all converged on the importance of automatically extracting statistical structure from experience. Implicit learning is found across a range of tasks including motor sequence learning, categorization and pseudo-linguistic paradigms. In each, improved performance emerges from experience in the absence of awareness of what was learned. Statistical learning has been as automatically extracting temporally contingent structure from perceptual (linguistic) input. Recent advances in artificial intelligence have shown the power of machine learning mechanisms provided with millions of examples and the computational power to represent high-order statistical information. Each domain appears to reflect a similar underlying inherent ability for neural networks to reorganize from experience to improve efficiency (Reber 2013). This broadly descriptive framework provides points of connection across fields, but does not identify specific mechanisms underlying human implicit, statistical learning. Using a simple model paradigm which produces robust and reliable implicit learning, a series of experiments will be presented reflecting an attempt to reverse engineer the brain?s implicit/statistical/machine learning algorithm. By searching for constraints, limitations and boundary conditions on implicit learning, we may be able to uncover the underlying operating mechanism of this fundamental learning ability.