Counting the research base behind new AI and data science degrees: a share that the classification decides

Abstract

United States colleges have been launching bachelor's degrees in artificial intelligence and data science at a rapid rate, and a natural question is how many do so without research activity in the field. This data note reports that public data cannot answer that question at the level of a number. Of the 200 institutions that first awarded such a degree between 2020 and 2025 and could be matched to a research identifier, the share with fewer than five AI works in 2018 to 2022 is 5.0 per cent counted with the OpenAlex concept C154945302 and 28.5 per cent counted with the OpenAlex topics subfield 1702, a gap of 23.5 percentage points whose Wilson intervals do not overlap. A stop condition fixed before the data were retrieved fires at 10 points, so neither figure may be reported as the answer. What survives every counting is direction and not level: doctoral R1 and R2 institutions are at 0 of 67 under all four counting rules, while baccalaureate colleges run from 8.9 to 55.6 per cent. The disagreement is one-directional at this threshold. The two schemes differ in documented ways, in status and in classification logic, since OpenAlex concepts are deprecated and frozen while topics are the maintained classification, but the note does not decompose the gap into those causes and does not claim to have explained it. The practical finding is for anyone using OpenAlex to measure institutional research activity in a field: the choice between the frozen and the maintained classification can move a headline share by more than twenty points. All seven pre-registered falsifiers were run; one row was found whose research identifier matches no OpenAlex institution record, and the note reports what that does to the affected figure. Prepared with AI assistance, disclosed in full in the note's AI use statement.