Earlier this week Art Sherman let me know about a New York Times article, “Is There a Smarter Path to Artificial Intelligence? Some Experts Hope So“. In addition to discussing the limitations of Deep Learning that are becoming clearer as the approach fails to produce the kind of flexibility that is needed for many AI applications, the NT Times article discusses three other approaches that aim to allow AI to “learn” from far smaller data sets—approaches that aim to varying degrees at what we mean by “common sense.” While these approaches do not reject the benefits of Deep Learning, they look beyond that approach in search of something much more akin to human learning, where good decisions can be made on the basis of far less systematically identified data.
Here’s a short video from the Allen Institute for Artificial Intelligence that states the problem in clear terms.
The New York Times article is certainly worth reading, and for further information you can visit the websites of the three innovative organizations it discusses: