Humanities scholars issue 8 provocations for generative AI

Eight claims. That is what a group of humanities researchers put forward as their "provocations" for how generative AI is built and studied, in a paper on arXiv by Lauren Klein, Meredith Martin, André Brock, Maria Antoniak, Melanie Walsh, Jessica Marie Johnson, Lauren Tilton and David Mimno.

The authors' framing is blunt. "The effects of generative AI are experienced by a broad range of constituencies, but the disciplinary inputs to its development have been surprisingly narrow," they write, describing humanities researchers as "currently underrepresented in AI development."

The eight claims are stated flatly: "Models make words, but people make meaning"; "Generative AI requires an expanded definition of culture"; "Generative AI can never be representative"; "Bigger models are not always better models"; "Not all training data is equivalent"; "Openness is not an easy fix"; "Limited access to compute enables corporate capture"; and "AI universalism creates narrow human subjects."

The paper also provides "a working definition of humanities research" and applies its theories and methods to the current AI landscape. It ends on a sharper note than most position papers, with what the abstract calls "a discussion of the importance of resisting the extraction of humanities research by computer science and related fields." Two of the researchers on our radar shared the link recently, and the co-author list itself carries the argument: it reaches from literary studies through Black digital studies to NLP.