Generative artificial intelligence (AI) has the potential to revolutionize social science research. However, researchers face the difficult challenge of choosing a specific AI model, often without social science-specific guidance. To demonstrate the importance of this choice, we present an evaluation of the effect of alignment, or human-driven modification, on the ability of large language models (LLMs) to simulate the attitudes of human populations (sometimes called silicon sampling). We benchmark aligned and unaligned versions of six open-source LLMs against each other and compare them to similar responses by humans. Our results suggest that model alignment impacts output in predictable ways, with implications for prompting, task completion, and the substantive content of LLM-based results. We conclude that researchers must be aware of the complex ways in which model training affects their research and carefully consider model choice for each project. We discuss future steps to improve how social scientists work with generative AI tools.
AchenChristopher H.BartelsLarry M.. 2016. Democracy for Realists: Why Elections Do Not Produce Responsive Government. Princeton, NJ: Princeton University Press.
2.
AherGati V.ArriagaRosa I.KalaiAdam Tauman (2023) Using Large Language Models to Simulate Multiple Humans and Replicate Human Subject Studies. In Proceedings of the 40th International Conference on Machine Learning, ed. KrauseAndreasBrunskillEmmaChoKyunghyunEngelhardtBarbaraSabatoSivanScarlettJonathan. Vol. 202of Proceedings of Machine Learning Research PMLR, pp. 337–371. https://proceedings.mlr.press/v202/aher23a.html.
ArditiAndyObesoOscarSyedAaquibPalekaDanielRimskyNinaGurneeWesNandaNeel (2024) “Refusal in Language Models Is Mediated by a Single Direction.” arXiv preprint arXiv:2406.11717.
7.
ArgyleLisa P.BailChristopher A.BusbyEthan C.GublerJoshua R.HoweThomasRyttingChristopherSorensenTayloret al.2023b. “Leveraging AI for Democratic Discourse: Chat Interventions Can Improve Online Political Conversations At Scale.” Proceedings of the National Academy of Sciences120(41): e2311627120.
8.
ArgyleLisa P.BusbyEthan C.FuldaNancyGublerJoshua R.RyttingChristopherWingateDavid. 2023a. “Out of One, Many: Using Language Models to Simulate Human Samples.” Political Analysis3(3): 337–351. https://doi.org/10.1371/journal.pclm.0000429.
9.
ArgyleLisa P.BusbyEthan C.GublerJoshua R.LymanAlexOlcottJustinPondJacksonWingateDavid (2024) “Testing Theories of Political Persuasion Using Artificial Intelligence.” Working Paper.
10.
AskellAmandaBaiYuntaoChenAnnaDrainDawnGanguliDeepHenighanTomJonesAndyet al. (2021) “A General Language Assistant as a Laboratory for Alignment.” CoRR abs/2112.00861. https://arxiv.org/abs/2112.00861.
11.
BaiYuntaoJonesAndyNdousseKamalAskellAmandaChenAnnaDas SarmaNovaDrainDawnet al. (2022) “Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback”. https://arxiv.org/abs/2204.05862.
12.
BailChristopher A.. 2024. “Can Generative AI Improve Social Science?.” Proceedings of the National Academy of Sciences121(21): e2314021121.
13.
BenderEmily M.GebruTimnitMcMillan-MajorAngelinaShmitchellShmargaret (2021) On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pp. 610–623.
14.
BisbeeJamesClintonJoshua D.DorffCassyKenkelBrentonLarsonJennifer M. (2024) “Synthetic Replacements for Human Survey Data? The Perils of Large Language Models.” Political Analysis, pp. 1–16.
15.
BlumerHerbert. 1958. “Race Prejudice As a Sense of Group Position.” Pacific Sociological Review1(1): 3–7.
16.
BoelaertJulienCoavouxSamuelOllionEtiennePetevIvaylo D.PrägPatrick (2024) “Machine Bias. Generative Large Language Models Have a View of Their Own.” osf.io/preprints/socarxiv/r2pnb.
17.
CaliskanAylinBrysonJoanna J.NarayananArvind. 2017. “Semantics Derived Automatically From Language Corpora Contain Human-like Biases.” Science (New York, N.Y.)356(6334): 183–186.
18.
ChengMyraDurmusEsinJurafskyDan (2023) “Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Models.” https://arxiv.org/abs/2305.18189.
19.
ChuJunjieLiuYugengYangZiqingShenXinyueBackesMichaelZhangYang (2024) “Comprehensive assessment of jailbreak attacks against llms.” arXiv preprint arXiv:2402.05668.
20.
ChungEun YiRomanoJoseph P.. 2013. “Exact and Asymptotically Robust Permutation Tests.” The Annals of Statistics41(2): 484–507.
DemszkyDorottyaYangDiyiYeagerDavid S.BryanChristopher J.ClapperMargarettChandhokSusannahEichstaedtJohannes C.et al.2023. “Using Large Language Models in Psychology.” Nature Reviews Psychology2(11): 688–701. https://doi.org/10.1038/s44159-023-00241-5.
26.
DillionDanicaTandonNiketGuYulingGrayKurt. 2023. “Can AI Language Models Replace Human Participants?.” Trends in Cognitive Sciences27(7): 597–600. https://doi.org/10.1016/j.tics.2023.04.008.
27.
DingPengFellerAviMiratrixLuke. 2016. “Randomization Inference for Treatment Effect Variation.” Journal of the Royal Statistical Society Statitiscs Methodology, Series B78: 655–671.
28.
DubeyAbhimanyuJauhriAbhinavPandeyAbhinavKadianAbhishekAl-DahleAhmadLetmanAieshaMathurAkhilet al. (2024) “The Llama 3 Herd of Models.” https://arxiv.org/abs/2407.21783.
29.
EdgellPennyGerteisJosephHartmannDouglas. 2006. “Atheists As ‘Other’: Moral Boundaries and Cultural Membership in American Society.” Social Forces71: 211–234.
EllemersNaomi. 2018. “Gender Stereotypes.” Annual Review of Psychology69: 275–298.
32.
GerberAlan S.GreenDonald P.. 2012. Field Experiments: Design, Analysis, and Interpretation. New York: W. W. Norton.
33.
GilardiFabrizioAlizadehMeysamKubliMaël. 2023. “ChatGPT Outperforms Crowd Workers for Text-annotation Tasks.” Proceedings of the National Academy of Sciences120(30): e2305016120.
34.
GoldsteinJosh A.SastryGirish (2023) “The Coming Age of AI-powered Propaganda. How to Defend Against Supercharged Disinformation.” Foreign Affairs, p. 7.
35.
Google Gemini Team (2024) “Gemini: Advanced Conversational AI Models.” https://www.deepmind.com/gemini. Accessed: 2024-08-15.
36.
HackenburgKobiMargettsHelen. 2024. “Evaluating the Persuasive Influence of Political Microtargeting with Large Language Models.” Proceedings of the National Academy of Sciences121(24): e2403116121.
37.
HeZeyuHuangChieh-YangDingChien-Kuang CorneliaRohatgiShauryaHuangTing-HaoKenneth (2024) If in a Crowdsourced Data Annotation Pipeline, a GPT-4. CHI ’24 New York, NY, USA: Association for Computing Machinery. https://doi.org/10.1145/3613904.3642834.
38.
HeseltineMichaelvin HohenbergBernhard Clemm (2024) “Large language models as a substitute for human experts in annotating political text.” Research and Politics.
HortonJohn J. (2023) Large language models as simulated economic agents: What can we learn from homo silicus?Technical report National Bureau of Economic Research.
41.
HutchingsVincent L.ValentinoNicholas A.. 2004. “The Centrality of Race in American Politics.” Annual Review of Political Science7: 383–408.
42.
JiJiamingLiuMickelDaiJuntaoPanXuehaiZhangChiBianCeZhangChiet al. (2023) “BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset.” https://arxiv.org/abs/2307.04657.
43.
JiangAlbert Q.SablayrollesAlexandreMenschArthurBamfordChrisChaplotDevendra Singhde las CasasDiegoBressandFlorianet al. (2023) “Mistral 7B.” https://arxiv.org/abs/2310.06825.
44.
JiangAlbert QSablayrollesAlexandreRouxAntoineMenschArthurSavaryBlancheBamfordChrisChaplotDevendra Singhet al. (2024) “Mixtral of Experts.” https://arxiv.org/abs/2401.04088.
45.
KimJunsolLeeByungkyu. 2024. “AI-Augmented Surveys: Leveraging Large Language Models and Surveys for Opinion Prediction.” https://arxiv.org/abs/2305.09620.
46.
KinderDonald R.KamCindy D.. 2009. Us Against them : Ethnocentric Foundations of American Opinion. Chicago: University of Chicago Press.
47.
KirkRobertMedirattaIshitaNalmpantisChristoforosLuketinaJelenaHambroEricGrefenstetteEdwardRaileanuRoberta (2024) Understanding the Effects of RLHF on LLM Generalisation and Diversity. In The Twelfth International Conference on Learning Representations. https://openreview.net/forum?id=PXD3FAVHJT.
48.
KleinbergJonLakkarajuHimabinduLeskovecJureLudwigJensMullainathanSendhil. 2018. “Human Decisions and Machine Predictions.” The Quarterly Journal of Economics133(1): 237–293. https://doi.org/10.1093/qje/qjx032.
49.
KozlowskiAustin C.KwonHyunkuEvansJames A. (2024) “In Silico Sociology: Forecasting COVID-19 Polarization with Large Language Models.” https://arxiv.org/abs/2407.11190.
50.
KrepsSarahMcCainR. MilesBrundageMiles. 2022. “All the News That’s Fit to Fabricate: AI-Generated Text As a Tool of Media Misinformation.” Journal of Experimental Political Science9(1): 104–117.
51.
LakeThomChoiEunsolDurrettGreg (2024) “From Distributional to Overton Pluralism: Investigating Large Language Model Alignment.” https://arxiv.org/abs/2406.17692.
52.
LeeSangukPengTai-QuanGoldbergMatthew H.RosenthalSeth A.KotcherJohn E.MaibachEdward W.LeiserowitzAnthony. 2024. “Can Large Language Models Estimate Public Opinion About Global Warming? An Empirical Assessment of Algorithmic Fidelity and Bias.” PLOS Climate3(8): 1–14. https://doi.org/10.1371/journal.pclm.0000429.
MellonJonathanBaileyJackScottRalphBreckwoldtJamesMioriMartaSchmedemanPhillip (2024) “Do AIs know what the most important issue is? Using language models to code open-text social survey responses at scale.” Research and Politics.
58.
MizeTrenton D.. 2016. “Sexual Orientation in the Labor Market.” American Sociological Review81: 1132–1160.
59.
MooreJaredDeshpandeTanviYangDiyi (2024) “Are Large Language Models Consistent over Value-laden Questions?” https://arxiv.org/abs/2407.02996.
60.
ObermeyerZiadPowersBrianVogeliChristineMullainathanSendhil. 2019. “Dissecting Racial Bias in An Algorithm Used to Manage the Health of Populations.” Science (New York, N.Y.)366(6464): 447–453.
PachotArnaultPetitThierry (2024) “Can Large Language Models Accurately Predict Public Opinion? A Review”.
64.
PalmerAlexisSpirlingArthur. 2023. “Large Language Models Can Argue in Convincing Ways About Politics, But Humans Dislike AI Authors: Implications for Governance.” Political Science75(3): 281–291.
65.
PanchTrishanMattieHeatherAtunRifat. 2019. “Artificial Intelligence and Algorithmic Bias: Implications for Health Systems.” Journal of Global Health9(2): 010318.
66.
ParkJoon SungZouCarolyn Q.ShawAaronHillBenjamin MakoCaiCarrieMorrisMeredith RingelWillerRobbLiangPercyBernsteinMichael S. (2024) “Generative Agent Simulations of 1,000 People.” https://arxiv.org/abs/2411.10109.
67.
ParrishAliciaChenAngelicaNangiaNikitaPadmakumarVishakhPhangJasonThompsonJanaHtutPhu Monet al. (2022) BBQ: A hand-built bias benchmark for question answering. In Findings of the Association for Computational Linguistics: ACL 2022, ed. MuresanSmarandaNakovPreslavVillavicencioAline. Dublin, Ireland: Association for Computational Linguistics, pp. 2086–2105. https://aclanthology.org/2022.findings-acl.165.
68.
PhillipsNolan E.LevyBrian L.SampsonRyan J.SmallMario L.WangRyan Q.. 2021. “The Social Integration of American Cities: Network Measures of Connectedness Based on Everyday Mobility Across Neighborhoods.” Sociological Methods & Research50: 1189–1214.
69.
QuYaoWangJue. 2024. “Performance and Biases of Large Language Models in Public Opinion Simulation.” Academy of Management Proceedings2024(1): 10298. https://doi.org/10.5465/AMPROC.2024.10298abstract.
RafailovRafaelSharmaArchitMitchellEricManningChristopher D.ErmonStefanoFinnChelsea (2023) Direct Preference Optimization: Your Language Model is Secretly a Reward Model. In Thirty-seventh Conference on Neural Information Processing Systems. https://openreview.net/forum?id=HPuSIXJaa9.
RismanBarbara. 2004. “Gender As a Social Structure.” Gender & Society18: 429–450.
74.
RobinsonJoshuaRyttingChristopher MichaelWingateDavid (2023) “Leveraging Large Language Models for Multiple Choice Question Answering.” https://arxiv.org/abs/2210.12353.
75.
RothschildJacob E.HowatAdam J.ShafranekRichard M.BusbyEthan C.. 2019. “Pigeonholing Partisans: Stereotypes of Party Supporters and Partisan Polarization.” Political Behavior41: 423–443.
76.
SanturkarShibaniDurmusEsinLadhakFaisalLeeCinooLiangPercyHashimotoTatsunori (2023) Whose opinions do language models reflect? In Proceedings of the 40th International Conference on Machine Learning. ICML’23 JMLR.org.
77.
SherifMuzaferHarveyO. J.Jack WhiteB.HoodWilliam R.SherifCarolyn W.. 1961. Intergroup Conflict and Cooperation: The Robbers Cave Experiment. Norman, OK: Institute of Group Relations.
SrivastavaAarohi, et al. (2023) “Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models.” Transactions on Machine Learning Research. https://openreview.net/forum?id=uyTL5Bvosj.
80.
SuzgunMiracGurTayfunBianchiFedericoHoDaniel E.IcardThomasJurafskyDanZouJames (2024) “Belief in the Machine: Investigating Epistemological Blind Spots of Language Models.” https://arxiv.org/abs/2410.21195.
TesslerMichael HenryBakkerMichiel A.JarrettDanielSheahanHannahChadwickMartin J.KosterRaphaelEvansGeorginaCampbell-GillinghamLucyCollinsTantumParkesDavid C.BotvinickMatthewSummerfieldChristopher. 2024. “AI can help humans find common ground in democratic deliberation.” Science (New York, N.Y.)386(6719): eadq2852.
83.
ThébaudSarahKornrichSabinoRuppannerLeah. 2021. “Good Housekeeping, Great Expectations: Gender and Housework Norms.” Sociological Methods & Research50: 1189–1214.
84.
VelezYamilLiuPatrick (2024) “Confronting Core Issues: A Critical Assessment of Attitude Polarization Using Tailored Experiments.” American Political Science Review.
85.
WeiAlexanderHaghtalabNikaSteinhardtJacob. 2024. “Jailbroken: How Does Llm Safety Training Fail?.” Advances in Neural Information Processing Systems36: 80079–80100.
86.
WeiJasonTayYiBommasaniRishiRaffelColinZophBarretBorgeaudSebastianYogatamaDaniet al. (2022) “Emergent Abilities of Large Language Models.” https://arxiv.org/abs/2206.07682.
87.
WhiteheadAndrew L.PerrySamuel L.BakerJoseph O.. 2018. “Make America Christian Again: Christian Nationalism and Voting for Donald Trump in the 2016 Presidential Election.” Sociology of Religion79: 147–171.
ZhangShengyuDongLinfengLiXiaoyaZhangSenSunXiaofeiWangShuheLiJiweiet al. (2024) “Instruction Tuning for Large Language Models: A Survey.” https://arxiv.org/abs/2308.10792.
90.
ZhengLianminChiangWei-LinShengYingZhuangSiyuanWuZhanghaoZhuangYonghaoLinZiet al. (2024) Judging LLM-as-a-judge with MT-bench and Chatbot Arena. In Proceedings of the 37th International Conference on Neural Information Processing Systems. NIPS ’23 Red Hook, NY, USA: Curran Associates Inc.
91.
ZhuChiweiXuBenfengWangQuanZhangYongdongMaoZhendong (2023) On the Calibration of Large Language Models and Alignment. In Findings of the Association for Computational Linguistics: EMNLP 2023, ed. BouamorHoudaPinoJuanBaliKalika. Singapore: Association for Computational Linguistics, pp. 9778–9795. https://aclanthology.org/2023.findings-emnlp.654.
92.
ZiemsCalebHeldWilliamShaikhOmarChenJiaaoZhangZhehaoYangDiyi (2024) “Can Large Language Models Transform Computational Social Science?”. https://arxiv.org/abs/2305.03514.
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
