It is a time-honored approach in machine intelligence to study the way humans manage intellectual tasks. In this two-page essay, Smith notes that the way machines learn still differs significantly from how humans and animals learn. He calls attention to a staged machine learning approach called “shaping” that originated with B.F. Skinner and was later used with some success in language learning and robot vision. Shaping introduces exercises of gradually increasing complexity, from basic to advanced.

I can imagine two reactions to this reminder from the machine learning community. One is to say, “Okay, this is worth a try for our current project.” The other is to say, “Well, this was tried in 1993 and 1996, and it works for some problems, but it is of limited application in our area.” Perhaps this essay is most valuable as an educational note to newer members of the community who may have been unaware of the provenance of shaping or of its continuing potential relevance for their endeavors.