arXiv Science⌕ Search

arXiv · 2611.21376

A Crowd-Audited Atlas of Implicit Bias in LLM-Powered Agents

Abstract

Prior work has shown that large language models (LLMs) express implicit biases in high-publicity, one-off decision scenarios such as hiring and intelligence assessment. However, as people increasingly employ autonomous, LLM-powered AI agents that make decisions on their behalf in doing day-to-day tasks, it is of critical importance that we understand how implicit biases, based on sociodemographic assumptions, emerge across a broad milieu of workaday decision scenarios. We used the 17 domains of daily activity in the American Time Use Survey (ATUS) to generate an atlas of 506 decision scenarios in which LLMs --- endowed with personas representing different races, gender identities, and political ideologies --- expressed significant implicit bias. Next, in a crowdsourced audit, 500 participants made 8,010 binary choices about which implicitly biased decision scenarios most deserved further review. We found that participants prioritized decision scenarios that had broad relevance (e.g., about food choices), were high stakes (e.g., about health), and where all members of a simulated sociodemographic group made unanimous decisions (because it was a clear indication of stereotyping). They were less swayed by the magnitude of an observed disparity or the specific ATUS domain the decision scenario represented. When examining the rationale behind their prioritization choices, we found that participants often interpreted disparities as traits of demographic groups and judged whether those traits seemed plausible, often based on their own internalized stereotypes. We discuss implications for building personalized AI agents, digital twins, social simulations, and for using crowdsourced bias audits.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Jason Alexander, Yuxuan Li, Michael Licamele, Wesley Hanwen Deng, Sauvik Das. 2026-11-16. A Crowd-Audited Atlas of Implicit Bias in LLM-Powered Agents. https://arxiv.science/abs/2611.21376

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Asymmetric Discourse Homogenization on Reddit Was a Text-Length Artifact: A Reanalysis

In the first version of this paper I reported that conservative Reddit users began writing more alike after ChatGPT was released while progressive users did not. That result came from a similarity measure that rises and falls with comment length. I redo every analysis of the original paper on all 6,056,593 comments with a sentence encoder and a dispersion measure computed within length strata, and the conservative break goes away. The two camps show no reliable difference in how within-group similarity changed after late 2022, and the ChatGPT date behaves like any of the more than two thousand other cutoffs I test. The matched flair comparison, the cumulative AI index, the stayer panel and the entropy signature each lose their support once length is held fixed. The one change that survives is older. Left and right comments drew closer to each other between 2019 and 2022, and that convergence was complete before ChatGPT arrived. I end with what this implies for studies that measure homogenization with embeddings.

cs.CY↗

Predicting Student Attrition in Competitive Programming: A Large-Scale Study Integrating Survey Insights and Global Behavioral Logs

We compared the recorded behavior of competitive programmers who became inactive on Codeforces with that of active users of similar skill. From 6,927 accounts collected via the public Codeforces API in June 2026, we excluded 888 accounts with incomplete submission histories and restricted both cohorts to a peak rating below 1500. This yielded 3,421 active accounts and 908 accounts with no submissions for at least three months, labeled as attrition. Behavioral differences between the two groups were small. Attrition accounts accepted smaller shares of their attempts overall and on mathematics and data-structure problems (rank-biserial r from -0.16 to -0.12), and solved fewer problems per rated contest (median 5.95 vs. 7.80; r = -0.13). Struggle time, upsolving, and recovery after rating drops did not differ significantly after correcting for multiple testing. Two pre-specified models trained on eight observation-window-independent features achieved cross-validated AUROCs of 0.596 (logistic regression) and 0.611 (random forest), yielding predicted probabilities barely above the base rate. Across random samples, stricter filtering rules, alternative definitions of inactivity, and tenure-matched cohorts, AUROC ranged from 0.58 to 0.66. Restricting features to activity from the 6- or 12-month windows ending three months prior to each user's final submission produced AUROCs of 0.59 to 0.65, indicating that minor behavioral differences precede inactivity but remain too weak to reliably identify individual at-risk users. All data and code are publicly archived, with a single command reproducing every reported result.

cs.CY↗

Can Power Draw Constrain Covert Compute? Limits of Analogue Verification for AI Governance

Frontier AI treaties or agreements on limiting computation require external verification; an external auditor must be able to confirm how much computation actually ran and that parties are adhering to the agreement. Analogue, off-chip measurements such as power draw provide an information channel for verification. It is unknown how well these analogue channels can constrain the amount of computation against an adversary who actively tries to subvert the audit. We derive a closed form for $β$, the largest hidden computation a power trace cannot exclude, as a fraction of the declared machine capacity. Measurements on NVIDIA A100 GPUs give $β= 1.16$ in the worst case, while an adversarial matched-energy schedule hides at least $0.41$. Analogue power measurements alone therefore constrain compute weakly. Additional restrictions granted by the threat model, such as the ability of the verifier to re-execute the declared work at an observed operating point, let the verifier push $β$ down to $0.059$ in the maximally restricted case. This gives a quantitative estimate of what analogue measurements can contribute to compute verification.

cs.CY↗