Is AI killing DH?

Michael Piotrowski

Link to RSS Feed

For the field of “DH,” “AI” has been phenomenal:1 OCR, handwriting recognition, visualization, stylometry, network analysis, topic modeling, image similarity analysis, even interpretation: all these hard problems are now essentially solved, thanks to AI! And all this without having to learn how to program—which is sure to delight, in particular, those who have always argued that even using LaTeX constitutes a form of exclusion (see, for example, Dombrowski 2023).

I don’t agree with Harald Klinke that DH is structurally a partnership between “one subject-matter expert + one methodologist (computer scientist),” although I do agree that—especially with third-party funding, projects often tend to have this structure. This has always been a difficult relationship because, as Abraham Moles remarked, it’s not that easy to be interdisciplinary: you have to adopt the language of a different discipline, so that in practice, “le vœu interdisciplinaire, si séduisant en principe, soit très décevant dans la réalité” (Moles [1990] 1995, 158). And in many, if not most projects, this has eventually lead to imbalances: scientific, methodological, political ones.

Regarding the last point, the humanities side has been traditionally much more dependent on the computing side, or, as Klinke puts it, “methods were expensive.” With AI, the balance of power has been massively upset: when, as a humanities scholar, you can have AI do it for you, why work with computer scientists, who are often perceived as arrogant, uncultured nerds. When you look at some of the things people are vibe-coding, I’ve little doubt that this is totally feasible for much of what counts as research in DH. And since most of the software is intended only for research, and often specifically for a single project, the thorny issue of maintenance of AI-generated code doesn’t even arise.

In the first issue of Computers and the Humanities, published in 1966, Louis Milic jubilantly asserted that “the consolidation phase of computer-aided study in the humanities is beginning to provide us with all the good things we have been lacking for so long” (Milic 1966, 3). Sixty years later, are we entering the “age of AI-driven superabundance,” as the CEOs of Big Tech are promising?

Well, so far, AI may “empower”2 individuals, but on a larger scale, everything seems to be going to the dogs. I’d like to leave aside everything else (such as the environmental impact) here and just look at academia. If nothing else changed at the same time, the arrival of this technology would indeed be phenomenal. But if you look at the history of technology, not only is every technology created at a specific time, in a specific place, in a specific society, but the context continues to evolve—both in response to the new technology, but also because of developments that have already been underway.

This means, for example, that AI doesn’t give you an edge when everybody else is using it, too. Indeed, conferences and journals are already drowning in slop; peer-review is on the verge of breaking down, or also being delegated to AI. Students are using AI and faculty is using AI; as Mark Carrigan summarized the argument of his book (Carrigan 2026):

The result is a multidimensional crisis of trust. Staff don’t trust students. Students don’t trust staff. Students don’t trust students. Staff don’t trust staff. Trust breaks down slowly at first, then all at once.

Most importantly, perhaps, universities everywhere are facing massive and ever-increasing budget cuts, while the vendors on which they made themselves dependent are hiking up the prices. And even though the AI technology is certainly here to stay, the economic bubble will inevitably burst, with a heavy, if not unprecedented impact on public funds—which means that there will be even fewer jobs for early-career researchers.

This is the context we need to consider when we talk about DH and AI, in research, as well as in teaching. I agree with Klinke: if you can quickly vibe-code your tools, this devalues the skills necessary for building them yourself, and for teaching this means: why teach programming and computational thinking? DH students are likely to see even less of a need for acquiring these competencies than they do already today. The same goes even more for statistics, which some colleagues argue to be absolutely essential, as “AI is just statistics.” But again, this overlooks the larger context: AI—as we’re encountering it today—is not just a technology, but a cultural, social, epistemological, economic, and political phenomenon. If the humanities don’t consider these dimensions, and take up the position of “it’s just a tool”—who needs them anymore? And this goes twice for DH.

Under the current conditions, one may wonder whether a retreat to hermeneutics as “the core of the humanities” is a winning move. Klinke argues that AI makes the humanities lose “their technical–formalizable foundation,” which is true (in the specific sense considered here), but AI can also produce plausible interpretations (i.e., texts that read as plausible interpretations). The risk, then, for the humanities—and again, for DH in particular, is not just being “reduced to supervising AI-generated analyses,” and becoming “a supporting discipline of computer science,” but rather completely abandoned. We have already been seeing whole departments in the humanities getting closed for many years now, especially in the UK and the US, and this trend is very likely to continue in the face of budget cuts. However, Klinke specifically mentions Schleiermacher; and Schleiermacher’s—rather than Gadamer’s—approach could indeed be a lead, and one that suggests that DH needs to urgently reflect on its goals and methods. Neither the uncritical adoption of AI, nor a merely “critical” stance are likely to ensure the continued relevance of DH, and of the humanities more generally.

References

Discussion on Mastodon

  1. I put both “DH” and “AI” in quotes to make clear that I’m using these terms “as currently understood in a wide sense,” rather than how I’d define them. For my take on DH (or rather computational humanities) see, e.g., Piotrowski (2024). ↩

  2. If you discount all externalities, I do think that LLMs can certainly be useful for certain tasks, but if you look at corporate ads for AI, they’re quite honest as to what they understand by “empowering.” ↩