Via Brian Leiter, here is an analysis of how AI figures in philosophy job ads these days.
Misc. notes:
I disagree with Prof. Leiter's pessimism about the prospects of making good hires here ("This is going to result in a lot of weak appointments..."). It's perfectly true that philosophers don't have long experience with generative AI, and that almost no philosophers have expertise in the full range of issues required to work proficiently on this subject. But nobody has long experience with generative AI, and almost nobody has expertise in that full range of issues. This domain needs philosophical work, and philosophers ought to be doing (at least some of) that work. There is a long tradition of philosophers' providing intellectual leadership amid collective ignorance, and now is a good time to draw on that tradition.
Perhaps I'm misunderstanding Prof. Leiter (could "as huge as you thought it was" be a wry joke?), but I don't read Prof. Lassiter's analysis as finding a "huge" amount of AI jobs. The 2023-2025 increase in AI-related job share is driven not by more AI-related jobs but fewer overall jobs, as Prof. Lassiter's second chart makes clear. The 2026 data point is based on partial data.
I'd like analysis like Lassiter's to be clearer about their techniques. I'm grateful for what information there is, but important questions are not answered. How, exactly, was Sonnet 3.6 used to determine AI relevance? If this analysis rested on keyword matching or regular expressions, what was the role of generative AI? (Was it just for collecting and cleaning the data? One certainly doesn't need generative AI to run job ads through a regular expression.)
If one really does want to use generative AI in this kind of analysis, I strongly recommend something more powerful than Sonnet 3.6. The scale of this analysis is so small that using (e.g.) Opus shouldn't be significantly more expensive. More powerful models tend to work much better for, e.g., asking whether a given job ad is "AI-related."1
If one really wants to know the state of philosophy job ads, I'd recommend hand-labeling the data: there are only a couple hundred data points so far this year. The best use of generative AI in this task might be to set up a system for labeling the data: here as elsewhere, using AI well often involves removing friction from the application of human judgment, not replacing that judgment. (For analyzing thousands of job ads over many years, it makes a lot more sense to set up a system where AI does the first-order work.)
AI is broad enough these days that I'd expect a lot of variety in the ways AI figures in the "AI-relevant" job ads. Many job ads might mention machine learning or algorithmic bias without having much to do with modern generative AI. (And, again, I don't know enough about how regular expressions and other techniques produced this analysis.)
I agree very much with Lassiter's point that we ought to be looking to the history of philosophy right now. I alluded to Socrates above and have been thinking about the 19th century in Europe, but that has a lot to do with my own background and preferences.
...and, re: Lassiter's speculations about philosophers using AI: yes, philosophers can benefit from AI. (Though note that as of this writing, Fable is not publicly available.) I think that philosophical expertise is one of the human competencies most immune to replacement by AI, but it is highly amenable to assistance from AI.
Lassiter's post is dated from only a few weeks ago.↩
