Artificial Artificial Artificial Intelligence: Crowd Workers Widely Use Large Language Models for Text Production Tasks
Open accessAbstract
Large language models (LLMs) are remarkable data annotators. They can be used to generate high-fidelity supervised training data, as well as survey and experimental data. With the widespread adoption of LLMs, human gold--standard annotations are key to understanding the capabilities of LLMs and the validity of their results. However, crowdsourcing, an important, inexpensive way to obtain human annotations, may itself be impacted by LLMs, as crowd workers have financial incentives to use LLMs to increase their productivity and income. To investigate this concern, we conducted a case study on the prevalence of LLM usage by crowd workers. We reran an abstract summarization task from the literature on Amazon Mechanical Turk and, through a combination of keystroke detection and synthetic text classification, estimate that 33-46% of crowd workers used LLMs when completing the task. Although generalization to other, less LLM-friendly tasks is unclear, our results call for platforms, researchers, and crowd workers to find new ways to ensure that human data remain human, perhaps using the methodology proposed here as a stepping stone. Code/data: https://github.com/epfl-dlab/GPTurk
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
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?
Past work suggests that online survey respondents frequently use AI tools to generate answers.
“Given that past work suggests that online survey respondents frequently use AI tools to generate answers [24], we explicitly instructed participants in the control condition to not use AI in this task.”
From The Role of Human Agency in Human-AI Co-Creativity (2026) · cites this paper as [24]
The Results report that an estimated 33–46% of MTurk crowd workers used LLMs to produce summaries of research abstracts. This supports frequent AI use for online text-production tasks, but the study did not measure whether online survey respondents use AI to generate survey answers; the Discussion explicitly limits its findings to text summarization. (LLM, conf=0.87)
Experimental — accuracy not yet measuredThis phenomenon stems from a classification bias that favours more prevalent classes, a bias that is amplified through successive iterations of learning.
“This phenomenon stems from a classification bias that favours more prevalent classes 164 , a bias that is amplified through successive iterations of learning 165 .”
From Machine culture (2023) · cites this paper as [164]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
ImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333 · cited 2× · see citing works
“What do massive computer vision datasets (Deng et al., 2009) and large-scale psychological experiments (Pennycook et al., 2021) have in common?”
ChatGPT outperforms crowd workers for text-annotation tasks
Gilardi, F., Alizadeh, M., & Kubli, M. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2305016120 · Cited by 1,197 · cited 1× · see citing works
“Further, an ever-increasing body of work illustrates the good performance of LLMs as proxies for human labeling (Wang et al., 2023; Gilardi et al., 2023; Ziems et al., 2023) and for generating high-quality text (which is, however, often factually inaccurate) that has captured the imagination of the…”
Shifting attention to accuracy can reduce misinformation online
Pennycook, G., Epstein, Z., Mosleh, M., Arechar, A. A., Eckles, D., & Rand, D. G. (2021) · Nature · doi:10.1038/s41586-021-03344-2 · Cited by 1,155 · cited 3× · see citing works
“What do massive computer vision datasets (Deng et al., 2009) and large-scale psychological experiments (Pennycook et al., 2021) have in common?”
An MTurk Crisis? Shifts in Data Quality and the Impact on Study Results
Chmielewski, M., & Kucker, S. C. (2020) · Social Psychological and Personality Science · doi:10.1177/1948550619875149 · Cited by 1,027 · cited 1× · see citing works
“…to accomplish a variety of tasks (Draws et al., 2021; Bragg et al., 2013), conducted au-dits on the overall quality of crowdsourced annotations (Chmielewski and Kucker, 2020; Kennedy et al., 2020; Smith et al., 2016), and shed light on the demographics and socioeconomic conditions of workers…”
Out of One, Many: Using Language Models to Simulate Human Samples
Argyle, L. P., Busby, E. C., Fulda, N., Gubler, J. R., Rytting, C., & Wingate, D. (2023) · Political Analysis · doi:10.1017/pan.2023.2 · Cited by 783 · cited 3× · see citing works
“As previously mentioned, LLMs can act as effective proxies for human sub-populations (Argyle et al., 2022), leading to a series of studies using LLMs as “silicon samples” (Ar-gyle et al., 2022; Horton, 2023; Dillion et al., 2023).”
ChatGPT: Jack of all trades, master of none
Kocoń, J., Cichecki, I., Kaszyca, O., Kochanek, M., Szydło, D., Baran, J., Bielaniewicz, J., Gruza, M., Janz, A., Kanclerz, K., et al. (2023) · Information Fusion · doi:10.1016/j.inffus.2023.101861 · Cited by 758 · see citing works
Is ChatGPT a Good NLG Evaluator? A Preliminary StudyPreprint
Wang, J., Liang, Y.-L., Meng, F.-D., Sun, Z., Shi, H., Li, Z.-X., Xu, J.-A., Qu, J.-F., & Zhou, J. (2023) · NEWSUM · doi:10.48550/arxiv.2303.04048 · Cited by 669 · cited 1× · see citing works
“Further, an ever-increasing body of work illustrates the good performance of LLMs as proxies for human labeling (Wang et al., 2023; Gilardi et al., 2023; Ziems et al., 2023) and for generating high-quality text (which is, however, often factually inaccurate) that has captured the imagination of the…”
Utility data annotation with Amazon Mechanical Turk
Sorokin, A. A., & Forsyth, D. A. (2008) · 2009 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops · doi:10.1109/cvprw.2008.4562953 · Cited by 620 · cited 2× · see citing works
“…MTurk spearheaded a paradigm shift in how machine learning datasets were created (Paullada et al., 2021; Marge et al., 2010) and how user studies (Sorokin and Forsyth, 2008), surveys (Buhrmester et al., 2011), and social science experiments (Pennycook et al., 2021; Salganik, 2019) were conducted.”
Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?
Filippas, A., Horton, J., & Manning, B. S. (2023) · ACM Conference on Economics and Computation · doi:10.1145/3670865.3673513 · Cited by 602
Can Large Language Models Transform Computational Social Science?
Ziems, C., Held, W., Shaikh, O., Chen, J., Zhang, Z., & Yang, D. (2024) · Computational Linguistics · doi:10.1162/coli_a_00502 · Cited by 585 · cited 3× · see citing works
“…an ever-increasing body of work illustrates the good performance of LLMs as proxies for human labeling (Wang et al., 2023; Gilardi et al., 2023; Ziems et al., 2023) and for generating high-quality text (which is, however, often factually inaccurate) that has captured the imagination of the…”
Data and its (dis)contents: A survey of dataset development and use in machine learning research
Paullada, A., Raji, I. D., Bender, E. M., Denton, E., & Hanna, A. (2021) · Patterns · doi:10.1016/j.patter.2021.100336 · Cited by 548 · cited 1× · see citing works
“…et al., 2010) to image annotation (Sorokin and Forsyth, 2008), MTurk spearheaded a paradigm shift in how machine learning datasets were created (Paullada et al., 2021; Marge et al., 2010) and how user studies (Sorokin and Forsyth, 2008), surveys (Buhrmester et al., 2011), and social science…”
The shape of and solutions to the MTurk quality crisis
Kennedy, R., Clifford, S., Burleigh, T., Waggoner, P. D., Jewell, R., & Winter, N. J. G. (2020) · Political Science Research and Methods · doi:10.1017/psrm.2020.6 · Cited by 506 · cited 1× · see citing works
“…tasks (Draws et al., 2021; Bragg et al., 2013), conducted au-dits on the overall quality of crowdsourced annotations (Chmielewski and Kucker, 2020; Kennedy et al., 2020; Smith et al., 2016), and shed light on the demographics and socioeconomic conditions of workers using crowdsourcing platforms…”
Dale, R. (2020) · Natural Language Engineering · doi:10.1017/s1351324920000601 · Cited by 461 · cited 1× · see citing works
“…the good performance of LLMs as proxies for human labeling (Wang et al., 2023; Gilardi et al., 2023; Ziems et al., 2023) and for generating high-quality text (which is, however, often factually inaccurate) that has captured the imagination of the general public and the media (Dale, 2021).”
(2023) · Journal of Applied Learning & Teaching · doi:10.37074/jalt.2023.6.1.23 · Cited by 421 · cited 2× · see citing works
“…popularity of LLMs such as ChatGPT, there is widespread concern about their usage in areas such as social media (Pan et al., 2023) (where they could be used to generate spam or misinformation) or higher education (Rudolph et al., 2023) (where they could be used to cheat assignments and exams).”
Smith, S. M., Roster, C. A., Golden, L. L., & Albaum, G. (2016) · Journal of Business Research · doi:10.1016/j.jbusres.2015.12.002 · Cited by 415 · cited 1× · see citing works
“…2021; Bragg et al., 2013), conducted au-dits on the overall quality of crowdsourced annotations (Chmielewski and Kucker, 2020; Kennedy et al., 2020; Smith et al., 2016), and shed light on the demographics and socioeconomic conditions of workers using crowdsourcing platforms (Burnham et al., 2018;…”
Can AI language models replace human participants?
Dillion, D., Tandon, N., Gu, Y., & Gray, K. (2023) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2023.04.008 · Cited by 405 · cited 2× · see citing works
“As previously mentioned, LLMs can act as effective proxies for human sub-populations (Argyle et al., 2022), leading to a series of studies using LLMs as “silicon samples” (Ar-gyle et al., 2022; Horton, 2023; Dillion et al., 2023).”
Human heuristics for AI-generated language are flawed
Jakesch, M., Hancock, J. T., & Naaman, M. (2023) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2208839120 · Cited by 342 · cited 1× · see citing works
“Distinguishing LLM-from human-generated text is difficult for both machine learning models and humans alike (Sadasivan et al., 2023; Jakesch et al., 2023).”
A Survey of Controllable Text Generation Using Transformer-based Pre-trained Language Models
Zhang, H., Song, H., Li, S., Zhou, M., & Song, D. (2024) · ACM Computing Surveys · doi:10.1145/3617680 · Cited by 317 · cited 1× · see citing works
“Typically, these analyses have been done through a variant of controlled text generation (see Zhang et al. (2022) for a comprehensive review).”
Using the Amazon Mechanical Turk for transcription of spoken language
Marge, M., Banerjee, S., & Rudnicky, A. I. (2010) · IEEE International Conference on Acoustics, Speech, and Signal Processing · doi:10.1109/icassp.2010.5494979 · Cited by 272 · cited 2× · see citing works
“…annotation (Sorokin and Forsyth, 2008), MTurk spearheaded a paradigm shift in how machine learning datasets were created (Paullada et al., 2021; Marge et al., 2010) and how user studies (Sorokin and Forsyth, 2008), surveys (Buhrmester et al., 2011), and social science experiments (Pennycook et…”
On the Risk of Misinformation Pollution with Large Language ModelsPreprint
Pan, Y., Pan, L., Chen, W.-H., Nakov, P., Kan, M.-Y., & Wang, W. (2023) · Conference on Empirical Methods in Natural Language Processing · doi:10.48550/arxiv.2305.13661 · Cited by 238 · cited 2× · see citing works
“…in the context of the explosive popularity of LLMs such as ChatGPT, there is widespread concern about their usage in areas such as social media (Pan et al., 2023) (where they could be used to generate spam or misinformation) or higher education (Rudolph et al., 2023) (where they could be used…”
Who Are the Turkers? A Characterization of MTurk Workers Using the Personality Assessment Inventory
McCredie, M. N., & Morey, L. C. (2018) · Assessment · doi:10.1177/1073191118760709 · Cited by 235 · cited 1× · see citing works
“…(Chmielewski and Kucker, 2020; Kennedy et al., 2020; Smith et al., 2016), and shed light on the demographics and socioeconomic conditions of workers using crowdsourcing platforms (Burnham et al., 2018; McCredie and Morey, 2019; Ogletree and Katz, 2021; Ipeirotis, 2010; Gray and Suri, 2019).”
A Watermark for Large Language Models
Kirchenbauer, J., Geiping, J., Wen, Y., Katz, J., Miers, I., & Goldstein, T. (2023) · arXiv · doi:10.48550/arxiv.2301.10226 · Cited by 194 · cited 2× · see citing works
“These concerns have led to work on watermarking LLM output (Kirchenbauer et al., 2023) by slightly altering the odds of each token during sampling, and to further work on methods for improving the detection of synthetically generated text (Yu et al., 2023; Verma et al., 2023; Diwan et al., 2021).”
Who is Mturk? Personal characteristics and sample consistency of these online workers
Burnham, M. J., Le, Y. K., & Piedmont, R. L. (2018) · Mental Health Religion & Culture · doi:10.1080/13674676.2018.1486394 · Cited by 175 · cited 1× · see citing works
“…(Chmielewski and Kucker, 2020; Kennedy et al., 2020; Smith et al., 2016), and shed light on the demographics and socioeconomic conditions of workers using crowdsourcing platforms (Burnham et al., 2018; McCredie and Morey, 2019; Ogletree and Katz, 2021; Ipeirotis, 2010; Gray and Suri, 2019).”
Can AI-Generated Text be Reliably Detected?
Sadasivan, V. S., Kumar, A., Balasubramanian, S., Wang, W., & Feizi, S. (2023) · arXiv · doi:10.48550/arxiv.2303.11156 · Cited by 161 · cited 1× · see citing works
“Distinguishing LLM-from human-generated text is difficult for both machine learning models and humans alike (Sadasivan et al., 2023; Jakesch et al., 2023).”
Törnberg, P. (2023) · arXiv · doi:10.48550/arxiv.2304.06588 · Cited by 152 · see citing works
The Curse of Recursion: Training on Generated Data Makes Models Forget
Shumailov, I., Shumaylov, Z., Zhao, Y., Gal, Y., Papernot, N., & Anderson, R. (2023) · arXiv · doi:10.48550/arxiv.2305.17493 · Cited by 151 · cited 2× · see citing works
“This may degrade the performance of “recursive” LLMs, i.e., those trained on synthetically generated data (Shumailov et al., 2023), and magnify the impact of the values and ideologies encoded by these models (Santurkar et al., 2023; Liu et al., 2022).”
Crowdsourcing Multi-Label Classification for Taxonomy Creation
Bragg, J., Mausam, & Weld, D. S. (2013) · AAAI Conference on Human Computation & Crowdsourcing · doi:10.1609/hcomp.v1i1.13091 · Cited by 145 · cited 1×
“Previous work has studied how to efficiently use crowdsourcing, with all its limitations, to accomplish a variety of tasks (Draws et al., 2021; Bragg et al., 2013), conducted au-dits on the overall quality of crowdsourced annotations (Chmielewski and Kucker, 2020; Kennedy et al., 2020; Smith et…”
Text Embeddings by Weakly-Supervised Contrastive Pre-training
Wang, L., Yang, N., Huang, X., Jiao, B., Yang, L., Jiang, D., Majumder, R., & Wei, F. (2022) · arXiv · doi:10.48550/arxiv.2212.03533 · Cited by 135 · cited 2× · see citing works
“We used an e5-base pretrained model as our main architecture (Wang et al., 2022).”
WANLI: Worker and AI Collaboration for Natural Language Inference Dataset Creation
Liu, A., Swayamdipta, S., Smith, N. A., & Choi, Y. (2022) · Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) · doi:10.18653/v1/2022.findings-emnlp.508 · Cited by 131 · cited 3× · see citing works
“This may degrade the performance of “recursive” LLMs, i.e., those trained on synthetically generated data (Shumailov et al., 2023), and magnify the impact of the values and ideologies encoded by these models (Santurkar et al., 2023; Liu et al., 2022).”
Whose Opinions Do Language Models Reflect?
Santurkar, S., Durmus, E., Ladhak, F., Lee, C., Liang, P., & Hashimoto, T. (2023) · arXiv · doi:10.48550/arxiv.2303.17548 · Cited by 114 · cited 2× · see citing works
“This may degrade the performance of “recursive” LLMs, i.e., those trained on synthetically generated data (Shumailov et al., 2023), and magnify the impact of the values and ideologies encoded by these models (Santurkar et al., 2023; Liu et al., 2022).”
A Checklist to Combat Cognitive Biases in Crowdsourcing
Draws, T., Rieger, A., Inel, O., Gadiraju, U., & Tintarev, N. (2021) · Proceedings of the AAAI Conference on Human Computation and Crowdsourcing · doi:10.1609/hcomp.v9i1.18939 · Cited by 86 · cited 2× · see citing works
“Previous work has studied how to efficiently use crowdsourcing, with all its limitations, to accomplish a variety of tasks (Draws et al., 2021; Bragg et al., 2013), conducted au-dits on the overall quality of crowdsourced annotations (Chmielewski and Kucker, 2020; Kennedy et al., 2020; Smith et…”
How Do Older Adults Recruited Using MTurk Differ From Those in a National Probability Sample?
Ogletree, A. M., & Katz, B. (2020) · The International Journal of Aging and Human Development · doi:10.1177/0091415020940197 · Cited by 72 · cited 1× · see citing works
“…(Chmielewski and Kucker, 2020; Kennedy et al., 2020; Smith et al., 2016), and shed light on the demographics and socioeconomic conditions of workers using crowdsourcing platforms (Burnham et al., 2018; McCredie and Morey, 2019; Ogletree and Katz, 2021; Ipeirotis, 2010; Gray and Suri, 2019).”
CHEAT: A Large-Scale Dataset for Detecting CHatGPT-writtEn AbsTracts
Yu, P., Chen, J., Feng, X., & Xia, Z. (2025) · IEEE Transactions on Big Data · doi:10.1109/tbdata.2025.3536929 · Cited by 65 · cited 1× · see citing works
“These concerns have led to work on watermarking LLM output (Kirchenbauer et al., 2023) by slightly altering the odds of each token during sampling, and to further work on methods for improving the detection of synthetically generated text (Yu et al., 2023; Verma et al., 2023; Diwan et al., 2021).”
ChatGPT as a Factual Inconsistency Evaluator for Text Summarization
Luo, Z., Xie, Q., & Ananiadou, S. (2023) · arXiv · doi:10.48550/arxiv.2303.15621 · Cited by 53 · cited 2× · see citing works
“ing easily done with the aid of commercially available LLMs (Luo et al., 2023).”
Veselovsky, V., Ribeiro, M. H., Arora, A., et al. (2023) · arXiv.org · doi:10.48550/arxiv.2305.15041 · Cited by 49 · cited 3× · see citing works
“, 2023) and synthetic data generated by LLMs can still be unfaithful with respect to the real data of interest (Veselovsky et al., 2023).”
Self-citationLarge Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?
Horton, J. J., Filippas, A., & Manning, B. S. (2023) · arXiv · doi:10.48550/arxiv.2301.07543 · Cited by 47 · see citing works
CHEAT: A Large-scale Dataset for Detecting ChatGPT-writtEn AbsTracts
Yu, P., Chen, J., Feng, X., & Xia, Z. (2023) · arXiv · doi:10.48550/arxiv.2304.12008 · Cited by 45 · see citing works
A Survey of Controllable Text Generation using Transformer-based Pre-trained Language ModelsPreprint
Zhang, H., Song, H., Li, S., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2201.05337 · Cited by 28 · see citing works
Message Distortion in Information Cascades
Ribeiro, M. H., Gligorić, K., & West, R. (2019) · International World Wide Web Conference (WWW) · doi:10.1145/3308558.3313531 · Cited by 18 · cited 3× · see citing works
“We modify a prior MTurk task originally devised by Horta Ribeiro et al. (2019), whose goal was to study the so-called “telephone effect,” whereby information is gradually lost or distorted as a message is passed from human to human in an information cascade.”
InfluentialSelf-citationGhostbuster: Detecting Text Ghostwritten by Large Language Models
Verma, V., Fleisig, E., Tomlin, N., & Klein, D. (2023) · arXiv · doi:10.48550/arxiv.2305.15047 · Cited by 16 · cited 1× · see citing works
“These concerns have led to work on watermarking LLM output (Kirchenbauer et al., 2023) by slightly altering the odds of each token during sampling, and to further work on methods for improving the detection of synthetically generated text (Yu et al., 2023; Verma et al., 2023; Diwan et al., 2021).”
string2string: A Modern Python Library for String-to-String AlgorithmsPreprint
Suzgun, M., Shieber, S., & Jurafsky, D. (2023) · Annual Meeting of the Association for Computational Linguistics · doi:10.48550/arxiv.2304.14395 · Cited by 12 · cited 1× · see citing works
“The fraction of overlap with the original abstract was operationalized as the longest common substring present in both the original abstract and the summary (as in Suzgun et al. (2023)).”
Fingerprinting Fine-tuned Language Models in the Wild
Diwan, N., Chakraborty, T., & Shafiq, Z. (2021) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2021.findings-acl.409 · Cited by 5 · cited 1× · see citing works
“These concerns have led to work on watermarking LLM output (Kirchenbauer et al., 2023) by slightly altering the odds of each token during sampling, and to further work on methods for improving the detection of synthetically generated text (Yu et al., 2023; Verma et al., 2023; Diwan et al., 2021).”
WANLI: Worker and AI Collaboration for Natural Language Inference Dataset CreationPreprint
Liu, A., Swayamdipta, S., Smith, N. A., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2201.05955 · Cited by 5 · see citing works
Fingerprinting Fine-tuned Language Models in the WildPreprint
Diwan, N., Chakravorty, T., & Shafiq, Z. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.01703
Can Large Language Models Transform Computational Social Science?
Ziems, C., Held, W., Shaikh, O., Chen, J., Zhang, Z., & Yang, D. (2023) · arXiv · doi:10.48550/arxiv.2305.03514
Flagged first, then by citations
1 of the 46 references couldn’t be matched to a known paper.
Research works printed but not linked: 3
- Michael Buhrmester, Tracy Kwang, and Samuel D Gosling. 2011. Amazon's mechanical turk: A new source of inexpensive, yet high-quality, data? Per- spectives on psychological science, 6(1):3-5.
- Matthew J Salganik. 2019. Bit by bit: Social research in the digital age. Princeton University Press.
- Oscar Schwartz. 2019. Untold history of ai: How ama- zon's mechanical turkers got squeezed inside the ma- chine. IEEE Spectrum.
Cited by
Sorted by citations
A Survey on LLM-Generated Text Detection: Necessity, Methods, and Future Directions
Wu, J., Yang, S., Zhan, R., Yuan, Y., Chao, L. S., & Wong, D. F. (2025) · Computational Linguistics · doi:10.1162/coli_a_00549 · Cited by 212 · see citing works
Kim, S. S. Y., Liao, Q. V., Vorvoreanu, M., Ballard, S., & Vaughan, J. W. (2024) · The 2024 ACM Conference on Fairness Accountability and Transparency · doi:10.1145/3630106.3658941 · Cited by 147 · see citing works
Art or Artifice? Large Language Models and the False Promise of Creativity
Chakrabarty, T., Laban, P., Agarwal, D., Muresan, S., & Wu, C.-S. (2024) · Proceedings of the CHI Conference on Human Factors in Computing Systems · doi:10.1145/3613904.3642731 · Cited by 135 · see citing works
Brinkmann, L., Baumann, F., Bonnefon, J.-F., Derex, M., Müller, T. F., Nussberger, A.-M., Czaplicka, A., Acerbi, A., Griffiths, T. L., Henrich, J., et al. (2023) · Nature Human Behaviour · doi:10.1038/s41562-023-01742-2 · Cited by 132 · see citing works
“In this respect, AlphaGo is in stark contrast to language models, which primarily learn to reproduce human cultural output.”
BackgroundResultHow large language models can reshape collective intelligence
Burton, J. W., Lopez-Lopez, E., Hechtlinger, S., Rahwan, Z., Aeschbach, S., Bakker, M. A., Becker, J. A., Berditchevskaia, A., Berger, J., Brinkmann, L. U. L., et al. (2024) · Nature Human Behaviour · doi:10.1038/s41562-024-01959-9 · Cited by 124 · see citing works
“This may subsequently decrease the production rate of new human-generated material (as opposed to LLM-recycled material) and, in turn, decrease the quantity and quality of collective knowledge that can be shared, remixed and learned from 121,122 .”
Shaping Human-AI Collaboration: Varied Scaffolding Levels in Co-writing with Language Models
Dhillon, P. S., Molaei, S., Li, J., Golub, M., Zheng, S., & Robert, L. P. (2024) · Proceedings of the CHI Conference on Human Factors in Computing Systems · doi:10.1145/3613904.3642134 · Cited by 111 · see citing works
Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
Casper, S., Davies, X., Shi, C., Gilbert, T. K., Scheurer, J., Rando, J., Freedman, R., Korbak, T., Lindner, D., Freire, P., et al. (2023) · arXiv · doi:10.48550/arxiv.2307.15217 · Cited by 101 · see citing works
“It is also becoming increasingly common for human knowledge workers to outsource work to chatbots, defeating the purpose of human oversight [Veselovsky et al., 2023].”
BackgroundNLP Evaluation in trouble: On the Need to Measure LLM Data Contamination for each Benchmark
Sainz, O., Campos, J., García-Ferrero, I., Etxaniz, J., de Lacalle, O. L., & Agirre, E. (2023) · Findings of the Association for Computational Linguistics: EMNLP 2023 · doi:10.18653/v1/2023.findings-emnlp.722 · Cited by 129 · see citing works
A Large Language Model Approach to Educational Survey Feedback Analysis
Parker, M. J., Anderson, C., Stone, C., & Oh, Y. (2024) · International Journal of Artificial Intelligence in Education · doi:10.1007/s40593-024-00414-0 · Cited by 86 · see citing works
“With an eye toward long-term planning, course providers try to identify gaps in course content and formats to maximize learning value and engagement.”
BackgroundAre Large Language Models a Threat to Digital Public Goods? Evidence from Activity on Stack Overflow
del Rio-Chanona, M., Laurentsyeva, N., & Wachs, J. (2023) · arXiv · doi:10.48550/arxiv.2307.07367 · Cited by 62 · see citing works
Who Determines What Is Relevant? Humans or AI? Why Not Both?
Faggioli, G., Dietz, L., Clarke, C. L. A., Demartini, G., Hagen, M., Hauff, C., Kando, N., Kanoulas, E., Potthast, M., Stein, B., et al. (2024) · Communications of the ACM · doi:10.1145/3624730 · Cited by 32 · see citing works
Generative AI Meets Open-Ended Survey Responses: Research Participant Use of AI and Homogenization
Zhang, S., Xu, J., & Alvero, A. (2025) · Sociological Methods & Research · doi:10.1177/00491241251327130 · Cited by 48 · see citing works
CoMPosT: Characterizing and Evaluating Caricature in LLM Simulations
Cheng, M., Piccardi, T., & Yang, D. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.669 · Cited by 60 · see citing works
Large language models reduce public knowledge sharing on online Q&A platforms
del Rio-Chanona, R. M., Laurentsyeva, N., & Wachs, J. (2024) · PNAS Nexus · doi:10.1093/pnasnexus/pgae400 · Cited by 87 · see citing works
“…Overflow and Reddit demonstrate the value of this kind of data for the continued training of LLMs. h The proliferation of LLMs has already impacted other forms of data creation: many Amazon Mechanical Turk workers now generate content (i.e. respond to surveys, evaluate texts) using ChatGPT (43).”
Kieslich, K., Diakopoulos, N. A., & Helberger, N. (2024) · AI and Ethics · doi:10.1007/s43681-024-00497-4 · Cited by 45 · see citing works
“[66]), we disabled copying text into the open text field such that people needed to type the story directly into the text box.”
Start Generating: Harnessing Generative Artificial Intelligence for Sociological Research
Davidson, T. (2024) · Socius Sociological Research for a Dynamic World · doi:10.1177/23780231241259651 · Cited by 51 · see citing works
A Comparative Study on Annotation Quality of Crowdsourcing and LLm Via Label Aggregation
Li, J. (2024) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp48485.2024.10447803 · Cited by 33 · see citing works
“Veselovsky et al. [1] found that the crowd workers on MTurk have been recently using ChatGPT to complete the crowdsourcing tasks.”
MethodČegiň, J., Šimko, J., & Brusilovsky, P. L. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.117 · Cited by 40 · see citing works
Hanley, H. W. A., & Durumeric, Z. (2024) · Proceedings of the International AAAI Conference on Web and Social Media · doi:10.1609/icwsm.v18i1.31333 · Cited by 36 · see citing works
AI’s Epistemic Harm: Reinforcement Learning, Collective Bias, and the New AI Culture War
Humphreys, D. (2025) · Philosophy & Technology · doi:10.1007/s13347-025-00928-y · Cited by 26 · see citing works
Friction: Deciphering Writing Feedback into Writing Revisions through LLM-Assisted Reflection
Zhang, C., Ju, K., Bidoshi, P., Yen, Y., & Rzeszotarski, J. M. (2025) · International Conference on Human Factors in Computing Systems · doi:10.1145/3706598.3714316 · Cited by 46 · see citing works
Adaptive Ensembles of Fine-Tuned Transformers for LLM-Generated Text Detection
Lai, Z., Zhang, X., & Chen, S. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10651296 · Cited by 35 · see citing works
“LLM has penetrated various aspects of daily life and plays a crucial role in many professional workflows such as forecasting and anomaly detection [2], text production [3], reasoning [4], naming [5], promoting tasks [6], vision [7], and across various domains [8]–[13].”
BackgroundUse of large language models as a scalable approach to understanding public health discourse
Espinosa, L., & Salathé, M. (2024) · PLOS Digital Health · doi:10.1371/journal.pdig.0000631 · Cited by 24 · see citing works
Using Large Language Models for Qualitative Analysis can Introduce Serious Bias
Ashwin, J., Chhabra, A., & Rao, V. (2026) · Sociological Methods & Research · doi:10.1177/00491241251338246 · Cited by 22 · see citing works
“, 2023), or even that a substantial proportion of workers on crowd-sourcing platform may be using LLMs in completing tasks (Veselovsky et al., 2023).”
Data Authenticity, Consent, and Provenance for AI Are All Broken: What Will It Take to Fix Them?
Longpre, S., Mahari, R., Obeng-Marnu, N., Brannon, W., South, T., Kabbara, J., & Pentland, A. (2024) · doi:10.21428/e4baedd9.a650f77d · Cited by 23 · see citing works
Large Language Models and the Wisdom of Small Crowds
Trott, S. (2024) · Open Mind · doi:10.1162/opmi_a_00144 · Cited by 22 · see citing works
“Collecting data from human participants is a major bottleneck in terms of both time and money; online data collection tools have ameliorated this somewhat, but recent evidence has raised concerns about the quality of data collected on some platforms (Veselovsky et al., 2023).”
BackgroundMy Future with My Chatbot: A Scenario-Driven, User-Centric Approach to Anticipating AI Impacts
Kieslich, K., Helberger, N., & Diakopoulos, N. A. (2024) · ACM Conference on Fairness, Accountability, and Transparency (FAccT) · doi:10.1145/3630106.3659026 · Cited by 18 · see citing works
“There was a reasonable distribution of the final set of participants across the various sociodemographic variables we measured, including gender, race, and education (See Appendix 5 for full details).”
BackgroundArtificial intelligence chatbots mimic human collective behaviour
He, J. K., Wallis, F. P. S., Gvirtz, A., & Rathje, S. (2024) · British Journal of Psychology · doi:10.1111/bjop.12764 · Cited by 34 · see citing works
“Finally, recruiting genuine human participants is becoming more difficult as workers in online subject pools increasingly use ChatGPT (Veselovsky et al., 2023).”
Bode, I., Huelss, H., Nadibaidze, A., & Watts, T. F. A. (2024) · Big Data & Society · doi:10.1177/20539517241303151 · Cited by 14 · see citing works
Gambetti, A., & Han, Q. (2024) · Proceedings of the International AAAI Conference on Web and Social Media · doi:10.1609/icwsm.v18i1.31437 · Cited by 13 · see citing works
Lessons from complex systems science for AI governance
Kolt, N., Shur‐Ofry, M., & Reuven Cohen, C. (2025) · Patterns · doi:10.1016/j.patter.2025.101341 · Cited by 22 · see citing works
The threat of AI chatbot responses to crowdsourced open-ended survey questions
Traylor, F. (2024) · Energy Research & Social Science · doi:10.1016/j.erss.2024.103857 · Cited by 12 · see citing works
The use of ChatGPT in creative writing assistance
Фіялка, С. Б., Kornieva, Z., & Гончарук, Т. Г. (2024) · XLinguae · doi:10.18355/xl.2024.17.01.01 · Cited by 11 · see citing works
Start Generating: Harnessing Generative Artificial Intelligence for Sociological ResearchPreprint
Davidson, T. (2023) · doi:10.31235/osf.io/u9nft · Cited by 10 · see citing works
Bank Run, Interrupted: Modeling Deposit Withdrawals with Generative AI
Kazinnik, S. (2026) · Management Science · doi:10.1287/mnsc.2024.06807 · Cited by 9 · see citing works
Humans vs. ChatGPT: Evaluating Annotation Methods for Financial Corpora
Kaikaus, J., Li, H., & Brunner, R. J. (2023) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata59044.2023.10386425 · Cited by 8 · see citing works
“Some crowd-workers may in fact utilize LLMs for annotation tasks in order to increase productivity and output [7].”
BackgroundQuality Assured: Rethinking Annotation Strategies in Imaging AI
Rädsch, T., Reinke, A., Weru, V., Tizabi, M. D., Heller, N. E., Isensee, F., Kopp‐Schneider, A., & Maier‐Hein, L. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-73229-4_4 · Cited by 9 · see citing works
Generative AI in Crowdwork for Web and Social Media Research: A Survey of Workers at Three Platforms
Christoforou, E., Demartini, G., & Otterbacher, J. (2024) · Proceedings of the International AAAI Conference on Web and Social Media · doi:10.1609/icwsm.v18i1.31452 · Cited by 8 · see citing works
Reynolds, J., & Kincaid, R. (2024) · Sociological Forum · doi:10.1111/socf.13029 · Cited by 9 · see citing works
Reijnierse, W. G., Brugman, B. C., & Droog, E. (2025) · Journal of Science Communication · doi:10.22323/150520250702095506 · Cited by 6 · see citing works
Network formation and dynamics among multi-LLMs
Papachristou, M., & Yuan, Y. (2025) · PNAS Nexus · doi:10.1093/pnasnexus/pgaf317 · Cited by 9 · see citing works
“Because social networks structure human behavior, preferences, and information diffusion (19 – 23), it is crucial to understand whether LLMs mirror human principles of network formation (18, 24, 25).”
Public Tax Disclosures and Fair Share PerceptionsPreprint
Dierynck, B., Jacob, M., Müller, M. A., et al. (2022) · SSRN Electronic Journal · doi:10.2139/ssrn.4173175 · Cited by 6 · see citing works
Informing Machine Perception With Psychophysics [Point of View]
Dulay, J., Poltoratski, S., Hartmann, T. S., Anthony, S. E., & Scheirer, W. J. (2024) · Proceedings of the IEEE · doi:10.1109/jproc.2024.3380905 · Cited by 5 · see citing works
“Current risks include using chatbots to solve tasks [25] and subjects rushing through tasks without trying (evident in identical reaction times or unnatu-rally fast ones).”
BackgroundHomophily in An Artificial Social Network of Agents Powered By Large Language ModelsPreprint
He, J. K., Wallis, F. P. S., & Rathje, S. (2023) · Research Square · doi:10.21203/rs.3.rs-3096289/v1 · Cited by 5 · see citing works
Generative AI in healthcare: challenges to patient agency and ethical implications
Holmes, S. A., Faria, V., & Moulton, E. A. (2025) · Frontiers in Digital Health · doi:10.3389/fdgth.2025.1524553 · Cited by 4 · see citing works
Artificial Intelligence Chatbots Mimic Human Collective Behaviour
He, J., Wallis, F., Gvirtz, A., & Rathje, S. (2024) · doi:10.21203/rs.3.rs-3096289/v2 · Cited by 4 · see citing works
Safeguarding Decentralized Social Media: LLM Agents for Automating Community Rule Compliance
Cava, L. L., & Tagarelli, A. (2025) · Online Social Networks and Media · doi:10.1016/j.osnem.2025.100319 · Cited by 6 · see citing works
Beyond Probabilities: Unveiling the Misalignment in Evaluating Large Language Models
Lyu, C., Wu, M., & Aji, A. F. (2024) · arXiv · doi:10.48550/arxiv.2402.13887 · Cited by 4 · see citing works
Christoforou, E., Demartini, G., & Otterbacher, J. (2025) · Communications of the ACM · doi:10.1145/3706101 · Cited by 3 · see citing works
Establishing Trustworthiness: Rethinking Tasks and Model Evaluation
Litschko, R., Müller-Eberstein, M., van der Goot, R., Weber, L., & Plank, B. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.14 · Cited by 6 · see citing works
Thresh: A Unified, Customizable and Deployable Platform for Fine-Grained Text Evaluation
Heineman, D., Dou, Y., & Xu, W. H. (2023) · Conference on Empirical Methods in Natural Language Processing: System Demonstrations (EMNLP) · doi:10.18653/v1/2023.emnlp-demo.30 · Cited by 3 · see citing works
Evaluating Language Model Agency through NegotiationsPreprint
Davidson, T. R., Veselovsky, V., Josifoski, M., Peyrard, M., Bosselut, A., Kosiński, M., & West, R. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2401.04536 · Cited by 14 · see citing works
“The use of crowdsourcing platforms, traditionally the go-to source for collecting large, human-annotated datasets, has also come under scrutiny due to crowd workers’ increased use of LMs (Veselovsky et al., 2023b;a).”
BackgroundSelf-citationEdossa, F. W., Gassen, J., & Maas, V. S. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4891763 · Cited by 3 · see citing works
Wirth, A., Ludec, C. L., Girard-Chanudet, C., & Cornet, M. (2025) · Digital Society · doi:10.1007/s44206-025-00224-4 · Cited by 2 · see citing works
Synthetic Data in AI-Driven Earth Observation: an Insight Into the SD4EO Project
Fernández, M., Gimeno, J., Gini, R., et al. (2024) · IEEE International Geoscience and Remote Sensing Symposium (IGARSS) · doi:10.1109/igarss53475.2024.10641008 · Cited by 2 · see citing works
“…not only are AI practitioners increasingly resorting to synthetic data for training purposes, but human annotators responsible for providing benchmark annotations for supervised learning tasks are also turning to generative models to increase their productivity and financial returns [32].”
BackgroundArtificial Intelligence Chatbots Mimic Human Collective BehaviourPreprint
He, J. K., Wallis, F. P. S., Gvirtz, A., & Rathje, S. (2023) · doi:10.31234/osf.io/vf5ap · Cited by 2 · see citing works
The Curse of Performative User Studies
Hertzmann, A. (2023) · IEEE Computer Graphics and Applications · doi:10.1109/mcg.2023.3315759 · Cited by 1 · see citing works
“16 MTurk might not be useful for much longer.”
BackgroundUsing Large Language Models to Detect Insufficient Effort Responding in Open-Ended Survey Questions
von Felten, N. (2026) · doi:10.1145/3780045.3780059 · Cited by 2 · see citing works
“Further, the capability of LLMs to generate fluent responses which could mask the lack of genuine effort are a new challenge that warrants further investigation, as it was already shown that the use of LLMs by crowdworkers is on the rise [40].”
Incorporating Worker Perspectives into MTurk Annotation Practices for NLP
Huang, O., Fleisig, E., & Klein, D. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-main.64 · Cited by 3 · see citing works
Public Tax Disclosures and Investor PerceptionsPreprint
Dierynck, B., Jacob, M., Müller, M. A., et al. (2020) · SSRN Electronic Journal · doi:10.2139/ssrn.3729938 · Cited by 1 · see citing works
Whose Truth is it Anyway? An Experiment on Annotation Bias in Times of Factual Opinion PolarizationPreprint
van der Velden, M. A. C. G., Reuver, M., Fokkens, A., et al. (2023) · doi:10.31235/osf.io/nd6yr · Cited by 1 · see citing works
Towards a Political Economy of Algorithmic CapitalismPreprint
Egan, M. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4869312 · Cited by 1 · see citing works
promptr: Format and Complete Few-Shot LLM Prompts
Ornstein, J. (2024) · doi:10.32614/cran.package.promptr · Cited by 1 · see citing works
The Mirage of Artificial Intelligence Terms of Use RestrictionsPreprint
Lemley, M. A., & Henderson, P. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5049562 · Cited by 1 · see citing works
The Role of Human Agency in Human-AI Co-Creativity
Wu, S. H., Yang, Y., Lee, A. Y., Liebscher, A., Rapuano, K., Niederhoffer, K., & Hancock, J. T. (2026) · Proceedings of the 2026 Conference on Creativity and Cognition · doi:10.1145/3803784.3816857
“Given that past work suggests that online survey respondents frequently use AI tools to generate answers [24], we explicitly instructed participants in the control condition to not use AI in this task.”
Sudmann, A., Echterhölter, A., Ramsauer, M., Retkowski, F., & Schröter, J. (2023) · AI critique · doi:10.1515/9783839467664-003
Participant use of artificial intelligence in online focus groups: an experiential account.Preprint
Stafford, L., Preston, C., & Pike, A. C. (2024) · doi:10.31234/osf.io/9cpg7
Leveraging edutainment and social networks to foster interethnic harmony
Siddique, A., Vlassopoulos, M., & Zénou, Y. (2024) · doi:10.1920/wp/ifs.2024.2024
Silently Killing your Panelists One Email at a Time: The True Cost of Email SolicitationsPreprint
Ferecatu, A., Bruyn, A. D., & Mukherjee, P. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4804023
The Risks of Using Large Language Models for Text Annotation in Social Science ResearchPreprint
Zhang, Y., & Lin, H. (2024) · doi:10.31235/osf.io/79qu8
Measuring Policy Positioning in U.S. Congressional ElectionsPreprint
Case, C. R., & Porter, R. (2024) · doi:10.31219/osf.io/zhrmv
Political Effects of Exposure to Evidence about Racial Discrimination
Bergan, D. E., Spates, S. A., Teinowitz, L., & Gonzalez, C. (2024) · Political Communication · doi:10.1080/10584609.2024.2397674
“The preregistration stated our intentions to collect data from Amazon’s MTurk platform, but one of the authors experienced a high number of suspicious respondents on that platform in a separate study prior to fielding the current study (see also Veselovsky et al., 2023).”
Hisam, E., Gimeno, J., Miraut, D., Pérez, M., Fernández, M., Gini, R., Rodríguez, R., Meoni, G., & Şeker, D. Z. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5119734
Hisam, E., Gimeno, J., Miraut, D., Pérez, M., Fernández, M., Gini, R., Rodríguez, R., Meoni, G., & Şeker, D. Z. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5129567
New Evidence and Design Considerations for Repeated Measure Experiments in Survey ResearchPreprint
Jordan, D., Ollerenshaw, T., & Trexler, A. (2025) · doi:10.31219/osf.io/q6czp_v2
Sciberas, C., & Fournier, M. A. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5374389
The Ouroboros effect and heterodox domains
Vukov, J. M., Lebkuecher, G., Joseph, T. L., Martinez, E. M., Ramirez, M., & Burns, M. B. (2025) · Ethics and Information Technology · doi:10.1007/s10676-025-09850-5
“Veselovsky et al. ( 2023) estimate that “33– 46% of crowd workers used LLMs” when completing an on “175 billion parameters culled from a large amount of internet text” while GPT-4 was “6 times larger with one tril - lion parameters” (Cohen, 2023 , p. 3; see also Vukov et al., 2023 ); DeepSeek-V3…”
Zablith, F., & Abbas, J. (2026) · Information Systems Frontiers · doi:10.1007/s10796-026-10738-y
Mann, J., & Stumpf-Wollersheim, J. (2026) · Research Integrity and Peer Review · doi:10.1186/s41073-026-00247-6
“Researchers have increasingly investigated the issue of participants’ GenAI use by measuring its frequency, quantifying its effect on research results, and testing different approaches for its prevention and detection [4, 6–10].”
Psychologische Forschung zu Zeiten generativer KI: Eine methodische Positionierung
Carmon, J., Rolfs, M., Steffens, M. C., Kiesel, A., Meiser, T., Schwarzer, G., & Kühn, S. (2026) · Psychologische Rundschau · doi:10.1026/0033-3042/a000784
- Authors
- Veselovsky, V., Ribeiro, M. H., & West, R.
- Affiliations
- École Polytechnique Fédérale de Lausanne (CH)
- Venue
- arXiv
- Publisher
- arXiv
- Published
- 2023
- ISSN
- 2331-8422 (ISSN-L)
- arXiv
- 2306.07899
- OpenAlex
- W4380715494
- Cited by
- 73
- References
- 46
- Citations by year
- 2026: 7 · 2025: 17 · 2024: 34 · 2023: 13 · 2022: 1 · 2020: 1
- Type
- posted content
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (cc-by)
- Topics
- Physical Sciences › Computer Science › Computer Science Applications › Mobile Crowdsensing and Crowdsourcing
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Social Sciences › General Social Sciences › Computational and Text Analysis Methods
- Keywords
- large language models, crowd workers, crowdsourcing, Amazon Mechanical Turk, LLM-assisted annotation, manual annotation, abstract summarization, financial incentives
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
