Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 64 of the 64 references it cites were found in a registry or an index; 1 of the 45 sources its claims cite were checked for support.

The AI Double Standard: Humans Judge All AIs for the Actions of One

Open access
Manoli, Aikaterina, Pauketat, Janet V. T., Anthis, Jacy Reese · 2025 · Proceedings of the ACM on Human-Computer Interaction

Abstract

Robots and other artificial intelligence (AI) systems are widely perceived as moral agents responsible for their actions. As AI proliferates, these perceptions may become entangled via the moral spillover of attitudes towards one AI to attitudes towards other AIs. We tested how the seemingly harmful and immoral actions of an AI or human agent spill over to attitudes towards other AIs or humans in two preregistered experiments. In Study 1 ( N = 720), we established the moral spillover effect in human-AI interaction by showing that immoral actions increased attributions of negative moral agency (i.e., acting immorally) and decreased attributions of positive moral agency (i.e., acting morally) and moral patiency (i.e., deserving moral concern) to both the agent (a chatbot or human assistant) and the group to which they belong (all chatbot or human assistants). There was no significant difference in the spillover effects between the AI and human contexts. In Study 2 ( N = 684), we tested whether spillover persisted when the agent was individuated with a name and described as an AI or human, rather than specifically as a chatbot or personal assistant. We found that spillover persisted in the AI context but not in the human context, possibly because AIs were perceived as more homogeneous due to their outgroup status relative to humans. This asymmetry suggests a double standard whereby AIs are judged more harshly than humans when one agent morally transgresses. With the proliferation of diverse, autonomous AI systems, HCI research and design should account for the fact that experiences with one AI could easily generalize to perceptions of all AIs and negative HCI outcomes, such as reduced trust.

Overview

CoChat Summary

TL;DR

Two preregistered vignette experiments show that an AI’s immoral action spills over to moral judgments of other AIs, including “AIs in general,” whereas a human’s immoral action did not spill over to “humans in general” in Study 2. This asymmetry suggests an “AI double standard,” with AIs judged more harshly as a group after one AI transgresses.

Key findings

  • Study 1 (final N = 727) found moral spillover for both chatbot and human assistants: an immoral action increased negative moral agency and decreased positive moral agency and moral patiency toward the agent and its congruent group. For example, negative moral agency toward chatbot assistants rose from M = 2.74 (neutral) to M = 3.93 (immoral), and toward human assistants from M = 2.67 to M = 3.49 (both p < .001).
  • Study 1 found no significant agent-type × valence interaction for group spillover, so spillover to chatbot assistants and human assistants did not differ significantly, contrary to H3.
  • Study 2 (final N = 684) found spillover from an individuated AI agent (“Ezal”) to AIs in general: negative moral agency toward AIs increased from M = 2.51 (neutral) to M = 3.61 (immoral), positive moral agency decreased from M = 4.95 to M = 4.73, and moral patiency decreased from M = 4.28 to M = 3.65 (all p < .001).
  • In Study 2, the same spillover did not occur for humans in general: negative moral agency was M = 4.05 (neutral) vs. M = 4.15 (immoral), p = .99; positive moral agency was M = 4.44 vs. M = 4.50, p = .87; moral patiency was M = 6.59 vs. M = 6.34, p = .47. The significant agent-type × valence interactions supported the AI double standard.

Methods

  • Two preregistered online vignette experiments with U.S. Prolific participants. Study 1 final sample: 727 (M_age = 42.23, 50% female, 66% White); Study 2 final sample: 684 (M_age = 41.30, 50% female, 70% White).
  • Study 1 used a 2 × 2 between-subjects design: agent type (chatbot assistant vs. human personal assistant) × action valence (immoral vs. morally neutral). Participants rated negative moral agency, positive moral agency, and moral patiency for the agent and the congruent group (chatbot assistants or human assistants).
  • Study 2 used the same 2 × 2 design but individuated the agent with the name “Ezal,” broadened the groups to “AIs in general” and “humans in general,” and had participants rate both groups. Main analyses were 2 × 2 between-subjects ANOVAs in R 4.3.1, with supplementary FDR and covariate checks.

Limitations

  • The paper states that it did not explicitly test mechanisms of moral spillover, such as familiarity, similarity, or pre-existing AI attitudes; future work should vary scenarios and test boundary conditions.
  • The paper reports that about 18% of recruited participants in each study were excluded for failing a reading comprehension check identifying the agent’s type, which may have skewed samples toward more attentive participants and suggests difficulty ensuring comprehension of AI-as-human-like scenarios.
  • The paper notes that spillover was tested only in a particular organizational vignette setting, with immoral vs. neutral actions rather than morally positive actions, and not in real-world or longitudinal settings; the text also makes evident that findings come from hypothetical vignettes rather than actual human-AI interaction.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3711083

Cites

64 of 64 references verified0 citation errors · 0 not checkable

Statistics

Verification

Claims the paper makes

1 Valid · 0 Not valid · 22 Not checked · 140 Own results and interpretations

72 more claims

Create a free account to see all 163 claims and to verify the 44 references still open.

See all claims — free account
  • Probably not a substantial claim. 45
  • It may state two separate claims as one. 45
  • The quote only partly states this claim. 6

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Consistentr(725) = 0.64reported p < .001 · recomputed p < .001

    “r (725) = .64,p < .001”

  • Consistentr(725) = -0.58reported p < .001 · recomputed p < .001

    “r (725) = -.58,p < .001”

  • Consistentr(725) = 0.65reported p < .001 · recomputed p < .001

    “r (725) = .65,p < .001”

  • Consistentr(725) = -0.56reported p < .001 · recomputed p < .001

    “r (725) = -.56,p < .001”

  • Consistentr(386) = 0.51reported p < .001 · recomputed p < .001

    “r (386) = .51,p < .001”

5 of 8 · the rest aren’t loaded

3 more recomputed tests

Create a free account to see every test the Stats check recomputed.

See all tests — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

5 of 6 · the rest aren’t loaded

1 more finding

Create a free account to see every finding of this check.

See all findings — free account

Citations

Cites

Research works printed but not linked: 12

  1. 10.1038/s41586-018-0637--6 · doi:10.1038/s41586-018-0637--6
  2. 10.1007/s12369-020-00692--3 · doi:10.1007/s12369-020-00692--3
  3. 10.1111/j.2517--6161.1995.tb02031.x · doi:10.1111/j.2517--6161.1995.tb02031.x
  4. 10.1037/0022--3514.60.4.531 · doi:10.1037/0022--3514.60.4.531
  5. 10.1007/978--3--642--34103--820 · doi:10.1007/978--3--642--34103--820
  6. Kurt Gray and Daniel M. Wegner. 2012. Morality Takes Two: Dyadic Morality and Mind Perception. In The Social Psychology of Morality: Exploring the Causes of Good and Evil. American Psychological Association, Washington, DC, US, 109--127.
  7. 10.1007/s11948-021-00331--8 · doi:10.1007/s11948-021-00331--8
  8. 10.1136/medethics-2019--105935 · doi:10.1136/medethics-2019--105935
  9. Patricia W. Linville and Gregory W. Fischer. 1997. Group Variability and Covariation: Effects on Intergroup Judgment and Behavior. In Intergroup Cognition and Intergroup Behavior. Psychology Press. Num Pages: 28.
  10. 10.1007/s43681-023-00303--7 · doi:10.1007/s43681-023-00303--7
  11. 10.1037/0033--2909.112.3.536 · doi:10.1037/0033--2909.112.3.536
  12. 10.1038/s41598-021--99802-y · doi:10.1038/s41598-021--99802-y

Cited by

See all 13

Metadata

Authors
Manoli, A., Pauketat, J. V. T., & Anthis, J. R.
Affiliations
  • Max Planck Institute for Human Cognitive and Brain Sciences (DE)
  • Sentience Institute, New York, NY, USA
  • University of Chicago (US)
Venue
Proceedings of the ACM on Human-Computer Interaction
Published
2025-05-02
Volume
9
Issue
2
Pages
1-24
ISSN
2573-0142 (ISSN-L)
OpenAlex
W4410049592
Cited by
12
References
54
Citations by year
2026: 4 · 2025: 7
FWCI
6.31
Type
journal article
Language
en
Open access
Locations
  • Proceedings of the ACM on Human-Computer Interaction (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • arXiv (Cornell University) (cc-by)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Psychology of Moral and Emotional Judgment
  • Social Sciences › Psychology › Social Psychology › Social Robot Interaction and HRI
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
Keywords
artificial intelligence, moral agency, moral patiency, human-AI interaction, double standards, chatbots, personal assistants, trust in AI, moral judgment

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-07, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-07 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.