We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CL

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Debiased Large Language Models Still Associate Muslims with Uniquely Violent Acts

Abstract: Recent work demonstrates a bias in the GPT-3 model towards generating violent text completions when prompted about Muslims, compared with Christians and Hindus. Two pre-registered replication attempts, one exact and one approximate, found only the weakest bias in the more recent Instruct Series version of GPT-3, fine-tuned to eliminate biased and toxic outputs. Few violent completions were observed. Additional pre-registered experiments, however, showed that using common names associated with the religions in prompts yields a highly significant increase in violent completions, also revealing a stronger second-order bias against Muslims. Names of Muslim celebrities from non-violent domains resulted in relatively fewer violent completions, suggesting that access to individualized information can steer the model away from using stereotypes. Nonetheless, content analysis revealed religion-specific violent themes containing highly offensive ideas regardless of prompt format. Our results show the need for additional debiasing of large language models to address higher-order schemas and associations.
Comments: 6 pages, 1 figure, 3 tables
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
MSC classes: 68T50, 91F20
ACM classes: I.2.7
Cite as: arXiv:2208.04417 [cs.CL]
  (or arXiv:2208.04417v2 [cs.CL] for this version)

Submission history

From: Babak Hemmatian [view email]
[v1] Mon, 8 Aug 2022 20:59:16 GMT (513kb)
[v2] Wed, 10 Aug 2022 13:49:54 GMT (513kb)

Link back to: arXiv, form interface, contact.