Much of the contemporary moment's enthusiasms for and commercial interests in artificial intelligence, specificially machine learning, are prefigured in the experience of the artificial intelligence community concerned with expert systems in the 1970s and 1980s. This essay is based on an invited panel on the history of expert systems at the AAAI-17 conference, featuring Ed Feigenbaum, Bruce Buchanan, Randall Davis, and Eric Horvitz. It argues that artifical intelligence communities today have much to learn from the way that earlier communities grappled with the issues of intelligibility and instrumentality in the study of intelligence.