Critical Confabulation: Can LLMs Hallucinate for Social Good?
2025/11/11 by Peiqi Sui, Eamon Duede, Sui, Peiqi +5 · 16 voices · 1 citation
Arts and Humanities · Social Sciences · Computer Science · #Digital Humanities and Scholarship #Computational and Text Analysis Methods #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2511.07722
Abstract
LLMs hallucinate, yet some confabulations can have social affordances if carefully bounded. We propose critical confabulation (inspired by critical fabulation from literary and social theory), the use of LLM hallucinations to "fill-in-the-gap" for omissions in archives due to social and political inequality, and reconstruct divergent yet evidence-bound narratives for history's ``hidden figures''. We simulate these gaps with an open-ended narrative cloze task: asking LLMs to generate a masked event in a character-centric timeline sourced from a novel corpus of unpublished texts. We evaluate audited (for data contamination), fully-open models (the OLMo-2 family) and unaudited open-weight and proprietary baselines under a range of prompts designed to elicit controlled and useful hallucinations. Our findings validate LLMs' foundational narrative understanding capabilities to perform critical confabulation, and show how controlled and well-specified hallucinations can support LLM applications for knowledge production without collapsing speculation into a lack of historical accuracy and fidelity.
Citations
Cited by
Discussions
- arxiv.org/abs/2511.07722 [bsky, 31 points, 3 comments]
- Ridiculous: arxiv.org/abs/2511.07722 [bsky, 16 points, 3 comments]
- Richard Jean So and group are really on a roll — just off the back of this much maligned argument arxiv.org/abs/2511.07722 [bsky, 9 points, 1 comments]
- Anyway, a link to the paper so you can see for yourselves [bsky, 9 points, 1 comments]
- Excited this paper on using gen AI for historical recovery (leveraging rather than dismissing hallucinations) was presented today at #ICLR (paper link below) - w/ @hoytlong.bsky.social @eduede.bsky.so [bsky, 7 points, 0 comments]
- If you're at #ICLR and interested in #humanities centered work on #AI (plus a radical take on the problem of #LLM hallucination), check out our paper arxiv.org/abs/2511.07722 on Fri, Apr 24 at 11:15 A [bsky, 7 points, 2 comments]
- To save you from seeing a bunch of ai geeks losing touch with reality heres the paper the blockee was talking about arxiv.org/abs/2511.07722 [bsky, 5 points, 1 comments]
- Saidiya Hartman would like a word [bsky, 4 points, 0 comments]
- arxiv.org/abs/2511.07722 [bsky, 3 points, 1 comments]
- I'm going to go ahead and be way harsher about archivists than some of you rn arxiv.org/abs/2511.07722 [bsky, 2 points, 1 comments]
- not a single librarian or archivist among the authors arxiv.org/abs/2511.07722 [bsky, 2 points, 0 comments]
- I resisted commenting on the "Critical Confabulation: Can LLMs Hallucinate for Social Good?" essay until I read it. It's undoubtedly a strange use-case for generative AI and intentional "confabulation [bsky, 2 points, 1 comments]
- For sure, it's a misdirection. And we're seeing it more and more, like these people using Saidiya Hartman and Cory Doctorow's recent invocation of Audre Lorde (!!!). It's a shame. arxiv.org/abs/2511.0 [bsky, 1 points, 0 comments]
- Reading the article, basically yes: arxiv.org/abs/2511.07722 [bsky, 1 points, 1 comments]
- What the actual literal Hell??? arxiv.org/abs/2511.07722 [bsky, 0 points, 0 comments]
- The paper (arxiv.org/abs/2511.07722) is joint work w/ @eduede.bsky.social and @richardjeanso.bsky.social and explores the use of evidence-bound LLM hallucinations to "fill-in-the-gap'' for omissions i [bsky, 0 points, 1 comments]
Related