Gary Marcus is a psychology professor at New York University and founder of a company called Geometric Intelligence. His aim is to develop artificial intelligence (AI) that is more sophisticated and efficient than currently available AI. Driving his work is the belief that creating good AI entails understanding how young children learn new concepts and generalize. As a trained scholar of human cognition, as well as father to an inquisitive two-year-old boy, Marcus is well versed in the underpinnings of his model.
Commonly, AI algorithms have been developed with a “deep learning” approach, modeled on the way that neurons and synapses in the brain change in response to new information after encountering many examples. Such technology has enabled pattern detection abilities, such as face recognition and word identification. Marcus recognizes that, unlike deep learning technology, toddlers are able to abstract conceptual generalizations from relatively little data. The brain stores and manipulates learned rules so that it can arrive at useful conclusions from few examples. For example, children quickly grasp and apply grammatical rules, while learning the exceptions by rote. Marcus and his colleagues have been coupling this understanding with probabilistic algorithms to take artificial intelligence in new directions.
This is a good reminder that as much as we may recognize cognitive challenges that can benefit from technological assistance, the brain is an amazingly sophisticated tool that technology is only beginning to emulate.
Image credit: www.flickr.com/photos/cblue98/