EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Revealing implicit social bias through EEG-informed generative AI

Aotong Li, Tuukka Ruotsalo, Christian Montag, Michiel Spapé

Computers in Human Behavior Reports · 2026-11 · 101443

DOI: 10.1016/j.chbr.2026.101443 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computers in Human Behavior Reports, 2026-11. Verfasst von Aotong Li und 3 weiteren Autor:innen. Die Arbeit nennt 114 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: University of Macau Research Services and Knowledge Transfer Office, Fund for the Development of Science and Technology.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Our first impression of others is commonly defined by our emotions and attitudes towards perceived group identities, which can influence how we judge their competence and trustworthiness. Investigating this remains challenging as social desirability and lack of awareness obscure attitudinal measures of first impressions. Although behavioural, computerised tests can provide an alternative, indirect means for assessing bias, such measures are not easily interpretable and may not be individually diagnostic. Rather than relying on questionnaires or reaction times, we suggest EEG evoked by faces can be used to infer how social bias affects first impressions. Participants (N = 38) viewed AI-generated faces while their EEG was recorded. Brain responses to these faces were then used to train classifiers to predict which candidates each individual would later vote for in a fictional leadership contest. These predictions were used as implicit “BrainVotes” to recalibrate the generative neural network to produce new images that aligned with individual leadership perceptions. Two months later, participants from the original sample took part in a mock election: BrainVote-generated images received 80 % of the vote and were 2.67 times more likely to win than matched controls. Visual analysis of the BrainVote images gave evidence for a common senior, masculine leader stereotype, especially for those participants who scored high in gender-career implicit association tests. Thus, using brain-responses in interaction with generative AI can project individual impressions of what constitutes leadership, and in so doing, reveal our implicit social biases.

Förderung

  • University of Macau Research Services and Knowledge Transfer Office
  • Fund for the Development of Science and Technology

Zitieren

APA

Li, A., Ruotsalo, T., Montag, C., & Spapé, M. (2026). Revealing implicit social bias through EEG-informed generative AI. Computers in Human Behavior Reports, 101443. https://doi.org/10.1016/j.chbr.2026.101443

Vancouver

Li A, Ruotsalo T, Montag C, Spapé M. Revealing implicit social bias through EEG-informed generative AI. Computers in Human Behavior Reports. 2026:101443. doi:10.1016/j.chbr.2026.101443

BibTeX
@article{Li2026,
  author = {Li, Aotong and Ruotsalo, Tuukka and Montag, Christian and Spapé, Michiel},
  title = {Revealing implicit social bias through EEG-informed generative AI},
  journal = {Computers in Human Behavior Reports},
  year = {2026},
  pages = {101443},
  doi = {10.1016/j.chbr.2026.101443}
}
Kostenlos starten →