This is an interesting and engaging paper for those who take a serious interest in artificial intelligence (AI), who likely will set the propositions from this paper about the current dominant topics in AI against their own experiences, and for those who have a more general fascination with patterns of behavior change in scientific communities. Published as part of a special track celebrating the centenary of Alan Turing and the 25th anniversary of the journal AI Communications, this paper is an extended version of an invited presentation at the 2012 European Conference on Artificial Intelligence (ECAI).

Using frequency analysis of terms appearing in the titles of papers in the 22 International Joint Conferences on Artificial Intelligence (IJCAI) from 1969 to 2011, and also word clouds of titles from selected years of those conferences, Cristianini illustrates a shift that has occurred in the past 50 years (or so) from knowledge-driven AI to learning-driven AI. The narrative complements the perspective of one of the referenced papers, on some lessons from the success of statistical techniques in language technology research [1].

That the body of published work shows a shift from knowledge-driven to learning-driven AI is, in my view, uncontroversial. I leave the question of whether this shift can be properly conceptualized as a paradigm shift in the Kuhnian sense, or more simply as a drift of emphasis, for others steeped in the philosophies of science to comment on. I do note that philosophers of science have differing perspectives on a contemporary framing of Kuhn’s early work [2]. Regardless, Cristianini’s paper is a worthwhile read, and for some will open a door to other readings in the social context of the practice of science.