EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Large language models show human-like content biases in transmission chain experiments

Alberto Acerbi iD, Joseph M Stubbersfield iD

Proceedings of the National Academy of Sciences · 31.10.2023 · 120 (44)

DOI: 10.1073/pnas.2313790120 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the National Academy of Sciences, online erschienen am 26.10.2023. Verfasst von Alberto Acerbi und 1 weiteren Autor:innen. Beteiligte Einrichtung: Department of Sociology and Social Research, University of Trento, Trento 38122, Italy. Hauptthemen laut MEDLINE: Language, Cultural Evolution. Die Arbeit wurde bisher 142-mal zitiert (Crossref) und nennt 27 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

As the use of large language models (LLMs) grows, it is important to examine whether they exhibit biases in their output. Research in cultural evolution, using transmission chain experiments, demonstrates that humans have biases to attend to, remember, and transmit some types of content over others. Here, in five preregistered experiments using material from previous studies with human participants, we use the same, transmission chain-like methodology, and find that the LLM ChatGPT-3 shows biases analogous to humans for content that is gender-stereotype-consistent, social, negative, threat-related, and biologically counterintuitive, over other content. The presence of these biases in LLM output suggests that such content is widespread in its training data and could have consequential downstream effects, by magnifying preexisting human tendencies for cognitively appealing and not necessarily informative, or valuable, content.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

LanguageCultural EvolutionHumansMental RecallBiasEthical Theory

Stichwörter der Autor:innen

ChatGPTcontent biasescultural evolutionlarge language modelstransmission chains

Autor:innen und Institutionen

  • Department of Sociology and Social Research, University of Trento, Trento 38122, Italy.
  • Department of Psychology, University of Winchester, Winchester SO22 4NR, United Kingdom.

ORCID-Profile: Alberto Acerbi, Joseph M Stubbersfield

Zitieren

APA

Acerbi, A., & Stubbersfield, J. M. (2023). Large language models show human-like content biases in transmission chain experiments. Proceedings of the National Academy of Sciences, 120(44). https://doi.org/10.1073/pnas.2313790120

Vancouver

Acerbi A, Stubbersfield JM. Large language models show human-like content biases in transmission chain experiments. Proceedings of the National Academy of Sciences. 2023;120(44). doi:10.1073/pnas.2313790120

BibTeX
@article{Acerbi2023,
  author = {Acerbi, Alberto and Stubbersfield, Joseph M},
  title = {Large language models show human-like content biases in transmission chain experiments},
  journal = {Proceedings of the National Academy of Sciences},
  year = {2023},
  volume = {120},
  number = {44},
  doi = {10.1073/pnas.2313790120}
}
Kostenlos starten →