Exploring the artificial intelligence “Trust paradox”: Evidence from a survey experiment in the United States
Open accessThe DOI matches its Crossref record; 23 of the 23 references it cites were found in a registry or an index.
Abstract
Advances in Artificial Intelligence (AI) are poised to transform society, national defense, and the economy by increasing efficiency, precision, and safety. Yet, widespread adoption within society depends on public trust and willingness to use AI-enabled technologies. In this study, we propose the possibility of an AI “trust paradox,” in which individuals’ willingness to use AI-enabled technologies exceeds their level of trust in these capabilities. We conduct a two-part study to explore the trust paradox. First, we conduct a conjoint analysis, varying different attributes of AI-enabled technologies in different domains—including armed drones, general surgery, police surveillance, self-driving cars, and social media content moderation—to evaluate whether and under what conditions a trust paradox may exist. Second, we use causal mediation analysis in the context of a second survey experiment to help explain why individuals use AI-enabled technologies that they do not trust. We find strong support for the trust paradox, particularly in the area of AI-enabled police surveillance, where the levels of support for its use are both higher than other domains but also significantly exceed trust. We unpack these findings to show that several underlying beliefs help account for public attitudes of support, including the fear of missing out, optimism that future versions of the technology will be more trustworthy, a belief that the benefits of AI-enabled technologies outweigh the risks, and calculation that AI-enabled technologies yield efficiency gains. Our findings have important implications for the integration of AI-enabled technologies in multiple settings.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Baron, R. M., & Kenny, D. A. (1986) · Journal of Personality and Social Psychology · doi:10.1037/0022-3514.51.6.1173 · Cited by 91,532 · cited 2×
“Though useful, the Baron and Kenney [85] method has no significance test.”
MethodPeeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI)
Adadi, A., & Berrada, M. (2018) · IEEE Access · doi:10.1109/access.2018.2870052 · Cited by 6,294
A general approach to causal mediation analysis.
Imai, K., Keele, L., & Tingley, D. (2010) · Psychological Methods · doi:10.1037/a0020761 · Cited by 4,143 · cited 1×
“These studies adopt what Imai et al. refer to as the “sequential ignorability assumption,” whereby both possible pretreatment confounders and treatment assignment are assumed to be statistically independent from the potential outcomes and mediators [67].”
Robots and Jobs: Evidence from US Labor Markets
Acemoglu, D., & Restrepo, P. (2020) · Journal of Political Economy · doi:10.1086/705716 · Cited by 4,088
Automation and New Tasks: How Technology Displaces and Reinstates Labor
Acemoglu, D., & Restrepo, P. (2019) · Journal of Economic Perspectives · doi:10.1257/jep.33.2.3 · Cited by 2,612
Hainmueller, J., Hopkins, D. J., & Yamamoto, T. (2014) · Political Analysis · doi:10.1093/pan/mpt024 · Cited by 1,990 · cited 1×
“In adopting this approach, we ensure to fulfill a key assumption for the orthogonality of attributes in expectation, which enables us to differentiate complex treatment effects into their constituent parts [39, 40].”
MethodIMAI, K., KEELE, L., TINGLEY, D., & YAMAMOTO, T. (2011) · American Political Science Review · doi:10.1017/s0003055411000414 · Cited by 1,537
Validating vignette and conjoint survey experiments against real-world behavior
Hainmueller, J., Hangartner, D., & Yamamoto, T. (2015) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1416587112 · Cited by 1,344 · cited 1×
“In adopting this approach, we ensure to fulfill a key assumption for the orthogonality of attributes in expectation, which enables us to differentiate complex treatment effects into their constituent parts [39, 40].”
MethodGorwa, R., Binns, R., & Katzenbach, C. (2020) · Big Data & Society · doi:10.1177/2053951719897945 · Cited by 1,097 · cited 1×
“This finding is consistent with previous studies showing that algorithmic content moderation online remains opaque and difficult to audit or understand [83].”
ResultCoppock, A., & McClellan, O. A. (2019) · Research & Politics · doi:10.1177/2053168018822174 · Cited by 987
Artificial intelligence, bias and clinical safety
Challen, R., Denny, J., Pitt, M., Gompels, L., Edwards, T., & Tsaneva-Atanasova, K. (2019) · BMJ Quality & Safety · doi:10.1136/bmjqs-2018-008370 · Cited by 956 · cited 1×
“Further, others could introduce the role of bias and a spectrum of consequences to flesh out public tolerance for the range of unintended outcomes of these technologies [100].”
BackgroundPelau, C., Dabija, D.-C., & Ene, I. (2021) · Computers in Human Behavior · doi:10.1016/j.chb.2021.106855 · Cited by 889 · cited 1×
“Indeed, people may rely on AI-enabled technologies not so much because they trust these capabilities but because they perceive that the anticipated benefits [20] of adopting AI-enabled technologies will exceed the expected costs [21] of doing so.”
BackgroundBarth, S., & de Jong, M. D. T. (2017) · Telematics and Informatics · doi:10.1016/j.tele.2017.04.013 · Cited by 798 · cited 1×
“Indeed, people may rely on AI-enabled technologies not so much because they trust these capabilities but because they perceive that the anticipated benefits [20] of adopting AI-enabled technologies will exceed the expected costs [21] of doing so.”
BackgroundHuman- versus Artificial Intelligence
Korteling, J. E. (., van de Boer-Visschedijk, G. C., Blankendaal, R. A. M., Boonekamp, R. C., & Eikelboom, A. R. (2021) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2021.622364 · Cited by 619 · cited 1×
“One argument suggests that AI-enabled technologies have different cognitive qualities than biological systems— including a set of mental models about ethics and empathy—and should be viewed less as a replacement for human intelligence but rather as a partner to humans [60].”
BackgroundDefending Against Neural Fake News
Zellers, R., Holtzman, A., Rashkin, H., Bisk, Y., Farhadi, A., Roesner, F., & Choi, Y. (2019) · arXiv · doi:10.48550/arxiv.1905.12616 · Cited by 569 · cited 1×
“In a context of natural language models, for example, algorithmic detection of AI-generated text is measured based on the percentage of text examples that the algorithm correctly identified as AI versus human generated [47].”
MethodLoyola-Gonzalez, O. (2019) · IEEE Access · doi:10.1109/access.2019.2949286 · Cited by 559
Information equivalence in survey experiments
Dafoe, A., Zhang, B., & Caughey, D. (2018) · Political Analysis · doi:10.1017/pan.2018.9 · Cited by 356 · cited 1×
“We varied four different attributes, outlined in Table 1, along with their respective levels [41].”
BackgroundPublic acceptance and perception of autonomous vehicles: a comprehensive review
Othman, K. (2021) · AI and Ethics · doi:10.1007/s43681-021-00041-8 · Cited by 347
An Intelligence in Our Image: The Risks of Bias and Errors in Artificial Intelligence
Osoba, O., & Welser, W. (2017) · doi:10.7249/rr1744 · Cited by 326 · cited 1×
“That we do not detect stronger effects is somewhat surprising given the research on the range of biases that AI can have for communities of color, whether in terms of algorithms that discriminate against black patients on transplant lists [70], in policing [71], and marketing [72].”
ResultJakesch, M., French, M., Ma, X., Hancock, J. T., & Naaman, M. (2019) · Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3290605.3300469 · Cited by 190
The AI Index 2022 Annual Report
Zhang, D., Maslej, N., Brynjolfsson, E., Etchemendy, J., Lyons, T., Manyika, J., Ngo, H., Niebles, J. C., Sellitto, M., Sakhaee, E., et al. (2022) · arXiv · doi:10.48550/arxiv.2205.03468 · Cited by 146
International law, military effectiveness, and public support for drone strikes
Kreps, S. E., & Wallace, G. P. R. (2016) · Journal of Peace Research · doi:10.1177/0022343316657405 · Cited by 128 · cited 1×
“In doing so, we replicate an approach adopted by other scholars studying emerging technologies [78, 79].”
ResultSelf-citationThe AI Index 2021 Annual Report
Zhang, D., Mishra, S., Brynjolfsson, E., Etchemendy, J., Ganguli, D., Grosz, B., Lyons, T., Manyika, J., Niebles, J. C., Sellitto, M., et al. (2021) · arXiv · doi:10.48550/arxiv.2103.06312 · Cited by 48
Cited by
Rethinking Plagiarism in the Era of Generative AI
Hutson, J. (2024) · Journal of Intelligent Communication · doi:10.54963/jic.v4i1.220 · Cited by 83
“This lag in awareness usage among older age groups, particularly in educational settings, potentially contributes to a growing mistrust and scepticism towards the application of AI tools in academic contexts by the younger generation [25].”
Wang, F., Li, N., Cheung, A. C. K., & Wong, G. K. W. (2025) · International Journal of Educational Technology in Higher Education · doi:10.1186/s41239-025-00547-9 · Cited by 69
Link, E., & Beckmann, S. (2025) · Communication Research Reports · doi:10.1080/08824096.2024.2427609 · Cited by 17
“As education is suggested to impact a more positive assessment and make the use of AI more likely (e.g., Kreps et al., 2023), the data might have limited between-person variance.”
Metadata
- Authors
- Kreps, S., George, J., Lushenko, P., & Rao, A.
- Venue
- PLOS ONE
- Publisher
- Public Library of Science (PLoS)
- Published
- 2023-07-18
- Volume
- 18
- Issue
- 7
- Pages
- e0288109
- ISSN
- 1932-6203 (ISSN-L)
- OpenAlex
- W4384663237
- Cited by
- 88
- References
- 23
- FWCI
- 21.49
- Type
- journal article
- Language
- en
- Open access
- gold
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.
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.