从显性威慑到隐性内化：AI监管和黑暗三联征人格对诚实行为的影响

doi:10.3724/SP.J.1041.2026.0381

摘要/Abstract

摘要：

随着生成式人工智能逐步演变为具备自主影响力的社会代理, 其对人类道德决策的影响愈加显著。然而, 传统伦理监管模式基于理性人假设, 忽视了人格特质对道德选择的调节作用, 导致监管效果因人而异, 难以实现最优效率。为突破这一局限, 本研究聚焦于道德研究中的诚实维度, 构建了监管类型×人格特质交互模型, 系统探讨显性监管、隐性激励与道德反馈三类AI驱动的干预策略对自恋、马基雅维利主义和精神病态人格的差异化影响。结果显示：(1)显性监管有效提升诚实行为, 尤其对高马基雅维利主义与高精神病态个体具有显著约束力; (2)监管即使隐藏也能规范诚实行为, 其不确定性加速了决策反应; (3)隐性激励促进诚实行为并具内化效应, 但高马基雅维利主义者在奖励撤除后的诚实率显著降低, 揭示了该策略在此类人群中的脆弱性; (4)道德反馈对整体诚实率提升有限, 仅对高马基雅维利主义者具有一定影响, 而高自恋者表现出持续不诚实倾向。本研究创新性地揭示了人格特质(黑暗三联征)在AI伦理监管中的关键作用, 并为传统的理性人监管范式提供了一个基于人格差异化的视角, 为未来设计情境化与个性化AI伦理干预策略提供了重要的理论与实践依据。

关键词: AI监管, 诚实行为, 黑暗三联征, 人格特质, 情境化干预

Abstract:

As generative artificial intelligence (GenAI) evolves into social agents with autonomous influence, its impact on human moral decision-making is becoming increasingly significant. Current regulatory models are often grounded in the “rational person hypothesis,” which assumes uniform responses to ethical constraints. This perspective, however, overlooks the profound moderating role of personality traits in moral choices, leading to divergent regulatory effects and a loss of efficiency. The Dark Triad of personality (narcissism, Machiavellianism, and psychopathy) is a robust predictor of moral deviation. To address this gap, we constructed a “Regulation Type × Personality Trait” interaction model. We hypothesized that the effectiveness of different AI-driven intervention strategies—namely explicit regulation, implicit incentives, and moral feedback—would be significantly moderated by individuals' Dark Triad traits when making decisions about honesty.

A series of experiments were conducted to test our hypotheses. The study utilized a modified coin-flip task where participants privately guessed and reported outcomes, a paradigm designed to create opportunities for dishonest behavior for personal gain. Participants' honesty rates and reaction times were recorded as the primary dependent variables. Across the experiments, we manipulated the AI-driven intervention strategies. These strategies included: (1) explicit (visible) versus implicit (invisible) AI surveillance which involved potential penalties for dishonesty; (2) implicit monetary incentives which rewarded consistent honesty; and (3) moral feedback which provided textual messages in response to honest or dishonest reports. Prior to the behavioral tasks, participants' personality traits were measured using the validated Short Dark Triad (SD3) scale.

The results supported our hypotheses, demonstrating significant interactions between intervention types and personality traits. In Experiment 1, explicit AI surveillance significantly increased honest reporting (t(45) = 4.59, p < 0.001), particularly among individuals with high levels of Machiavellianism (t(25) = 4.60, p = 0.005) and psychopathy (t(28) = 4.44, p < 0.001). In Experiment 2, invisible AI surveillance also enhanced honesty but was less effective than visible AI surveillance, F(2, 90) = 18.10, p < 0.001. Notably, invisible surveillance resulted in the shortest reaction time (RT = 0.49), F(2, 90) = 34.10, p < 0.001. High Machiavellian participants displayed greater honesty under visible surveillance (OR = 0.70, p = 0.013) but were more dishonest without or under invisible AI surveillance. In Experiment 3, potential financial rewards increased reaction time (F(2, 118) = 58.59, p < 0.001), while high Machiavellian individuals showed reduced honesty during the internalization stage, t(57.98) = −2.04, p = 0.044. In Experiment 3a and 3b, financial incentives promoted honesty more effectively than moral messaging during the reward stage (t(120) = 3.07, p = 0.003) and maintained this effect into the internalization stage (t(120) = 2.06, p = 0.041), demonstrating the robustness of monetary influence. High Machiavellian participants sustained higher honesty levels in the internalization stage (OR = 1.96, p < 0.001). In contrast, narcissistic participants showed resistance to moral messaging, especially during the reward stage, t(49.95) = −2.55, p = 0.013.

This study was the first to systematically reveal the critical moderating role of the Dark Triad personality traits in AI ethical regulation. The findings challenge the traditional 'rational person' paradigm by empirically demonstrating the significant personality-based heterogeneity of regulatory effects. The core contribution of this research is the proposal of an innovative concept: 'personality-regulated regulation.' This framework provides a vital theoretical and practical foundation for designing future AI ethical intervention strategies that are contextualized and personalized. Such an approach allows for the optimization of regulatory resource allocation and enhances overall regulatory efficacy, moving beyond one-size-fits-all models.

Key words: AI regulation, honest behavior, Dark Triad, personality traits, contextualized intervention

中图分类号:

B849: C91

王健树, 姜啸威, 陈亚楠, 王明辉, 杜峰. (2026). 从显性威慑到隐性内化：AI监管和黑暗三联征人格对诚实行为的影响. 心理学报, 58(3), 381-398.

WANG Jianshu, JIANG Xiaowei, CHEN Yanan, WANG Minghui, DU Feng. (2026). From overt deterrence to covert internalization: Moral effects of AI regulation and the moderating role of personality traits. Acta Psychologica Sinica, 58(3), 381-398.

图/表 13

参考文献 42

[1]	Abeler J., Nosenzo D., & Raymond C. (2019). Preferences for truth-telling. Econometrica, 87(4), 1115-1153. doi: 10.3982/ECTA14673
[2]	Aquino K., & Reed A. (2002). The self-importance of moral identity. Journal of Personality and Social Psychology, 83(6), 1423-1440. pmid: 12500822
[3]	Bartlett M. S. (1951). The effect of standardization on a χ² approximation in factor analysis. Biometrika, 38(3/4), 337-344.
[4]	Baskin-Sommers A. R., Waller R., Fish A. M., & Hyde L. W. (2015). Callous-unemotional traits trajectories interact with earlier conduct problems and executive control to predict violence and substance use among high risk male adolescents. Journal of Abnormal Child Psychology, 43(8), 1529-1541. doi: 10.1007/s10802-015-0041-8 pmid: 26081013
[5]	Batson C. D., Thompson E. R., & Chen H. (2002). Moral hypocrisy: Addressing some alternatives. Journal of Personality and Social Psychology, 83(2), 330-339. pmid: 12150231
[6]	Blair R. J. R. (2013). The neurobiology of psychopathic traits in youths. Nature Reviews Neuroscience, 14(11), 786-799. doi: 10.1038/nrn3577 pmid: 24105343
[7]	Borau S. (2025). Deception, discrimination, and objectification: Ethical issues of female AI agents. Journal of Business Ethics, 198(1), 1-19. doi: 10.1007/s10551-024-05754-4
[8]	Campbell W. K., Rudich E. A., & Sedikides C. (2002). Narcissism, self-esteem, and the positivity of self-views: Two portraits of self-love. Personality and Social Psychology Bulletin, 28(3), 358-368. doi: 10.1177/0146167202286007 URL
[9]	Cialdini R. B., Reno R. R., & Kallgren C. A. (1990). A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology, 58(6), 1015-1026. doi: 10.1037/0022-3514.58.6.1015 URL
[10]	Cohen M. K., Kolt N., Bengio Y., Hadfield G. K., & Russell S. (2024). Regulating advanced artificial agents. Science, 384(6691), 36-38. doi: 10.1126/science.adl0625 pmid: 38574134
[11]	Faul F., Erdfelder E., Lang A.-G., & Buchner A. (2009). Statistical power analyses using GPower 3.1: Tests for correlation and regression analyses. Behavior Research Methods*, 41(4), 1149-1160. doi: 10.3758/BRM.41.4.1149 pmid: 19897823
[12]	Fowles R. (2018). Rational econometric man. Eastern Economic Journal, 44(3), 489-490. doi: 10.1057/eej.2015.28 URL
[13]	Frey B. S., & Jegen R. (2001). Motivation crowding theory. Journal of Economic Surveys, 15(5), 589-611. doi: 10.1111/joes.2001.15.issue-5 URL
[14]	Gao Q. (2022). Replace or coexist—Cronbach’s alpha and McDonald’s omega. Psychology of China, 4(8), 954-961. doi: 10.35534/pc.0408114 URL
	[高清辉. (2022). 取代还是共存——信度估计系数阿尔法与欧米伽. 中国心理学前沿, 4(8), 954-961.]
[15]	Gerlach P., Teodorescu K., & Hertwig R. (2019). The truth about lies: A meta-analysis on dishonest behavior. Psychological Bulletin, 145(1), 1-44. doi: 10.1037/bul0000174 pmid: 30596431
[16]	Giroux M., Kim J., Lee J. C., & Park J. (2022). Artificial intelligence and declined guilt: Retailing morality comparison between human and AI. Journal of Business Ethics, 178(4), 1027-1041. doi: 10.1007/s10551-022-05056-7
[17]	Glenn A. L., Koleva S., Iyer R., Graham J., & Ditto P. H. (2010). Moral identity in psychopathy. Judgment and Decision Making, 5(7), 497-505. doi: 10.1017/S1930297500001662 URL
[18]	Goodlad L. M. E. (2023). Review of the book Atlas of AI: Power, politics, and the planetary costs of artificial intelligence, by K. Crawford. Critical Inquiry, 49(2), 284-286. doi: 10.1086/722404 URL
[19]	Hart W., Richardson K., & Tortoriello G. K. (2021). Revisiting the interactive effect of narcissism and self-esteem on responses to ego threat: Distinguishing between assertiveness and intent to harm. Journal of Interpersonal Violence, 36(7-8), 3662-3687. doi: 10.1177/0886260518777551 URL
[20]	Hovland C. I., & Weiss W. (1951). The influence of source credibility on communication effectiveness. Public Opinion Quarterly, 15(4), 635-650. doi: 10.1086/266350 URL
[21]	Huys Q. J. M., Browning M., Paulus M. P., & Frank M. J. (2021). Advances in the computational understanding of mental illness. Neuropsychopharmacology, 46(1), 3-19. doi: 10.1038/s41386-020-0746-4
[22]	Jones D. N., & Paulhus D. L. (2014). Introducing the short dark triad (SD3): A brief measure of dark personality traits. Assessment, 21(1), 28-41. doi: 10.1177/1073191113514105 pmid: 24322012
[23]	Jones D. N., & Paulhus D. L. (2017). Duplicity among the Dark Triad: Three faces of deceit. Journal of Personality and Social Psychology, 113(2), 329-342. doi: 10.1037/pspp0000139 pmid: 28253006
[24]	Kaiser H. F. (1970). A second generation Little Jiffy. Psychometrika, 35(4), 401-415. doi: 10.1007/BF02291817 URL
[25]	Kasabova A. (2017). From shame to shaming: Towards an analysis of shame narratives. Open Cultural Studies, 1(1), 99-112. doi: 10.1515/culture-2017-0010 URL
[26]	Khan S. A., Sheikhi G., Opdahl A. L., Rabbi F., Stoppel S., Trattner C., & Dang-Nguyen D.-T. (2023). Visual user- generated content verification in journalism: An overview. IEEE Access, 11, 6748-6769. doi: 10.1109/ACCESS.2023.3236993 URL
[27]	Köbis N., Bonnefon J.-F., & Rahwan I. (2021). Bad machines corrupt good morals. Nature Human Behaviour, 5(6), 679-685. doi: 10.1038/s41562-021-01128-2 pmid: 34083752
[28]	Köbis N., Rahwan Z., Rilla R., Supriyatno B. I., Bersch C., Ajaj T., Bonnefon J.-F., & Rahwan I. (2025). Delegation to artificial intelligence can increase dishonest behaviour. Nature, 646, 126-134. doi: 10.1038/s41586-025-09505-x
[29]	Konrath S., Ho M. H., & Zarins S. (2016). The strategic helper: Narcissism and prosocial motives and behaviors. Current Psychology, 35(2), 182-194. doi: 10.1007/s12144-016-9417-3 URL
[30]	Liu Y., Zhang K., Li Y., Yan Z., Gao C., Chen R., … Sun L. (2024). Sora: A review on background, technology, limitations, and opportunities of large vision models (arXiv:2402.17177). arXiv.
[31]	Malär L., & Giuffredi-Kähr A. (2024). The dark triad of brand personality: Scale development and validation. Psychology & Marketing, 41(11), 2728-2740. doi: 10.1002/mar.v41.11 URL
[32]	Marbut A., Harms P. D., & Credé M. (2025). In the service of the prince: A meta-analytic review of Machiavellian leadership. Journal of Organizational Behavior, 46(8), 939-969. doi: 10.1002/job.v46.6 URL
[33]	Nass C., & Moon Y. (2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56(1), 81-103. doi: 10.1111/josi.2000.56.issue-1 URL
[34]	Ojanen T., & Findley-Van Nostrand D. (2019). Affective- interpersonal and impulsive-antisocial psychopathy: Links to social goals and forms of aggression in youth and adults. Psychology of Violence, 9(1), 56-66. doi: 10.1037/vio0000160
[35]	Peirce J., Gray J. R., Simpson S., MacAskill M., Höchenberger R., Sogo H., Kastman E., & Lindeløv J. K. (2019). PsychoPy2: Experiments in behavior made easy. Behavior Research Methods, 51(1), 195-203. doi: 10.3758/s13428-018-01193-y pmid: 30734206
[36]	Rijnders R. J. P., Terburg D., Bos P. A., Kempes M. M., & van Honk J. (2021). Unzipping empathy in psychopathy: Empathy and facial affect processing in psychopaths. Neuroscience & Biobehavioral Reviews, 131, 1116-1126. doi: 10.1016/j.neubiorev.2021.10.020 URL
[37]	Ryan R. M., & Deci E. L. (2020). Intrinsic and extrinsic motivation from a self-determination theory perspective: Definitions, theory, practices, and future directions. Contemporary Educational Psychology, 61, Article 101860.
[38]	Schüz, B., Cooke R., Schüz N., & van Koningsbruggen G. M. (2017). Self-affirmation interventions to change health behaviors. In L. Little, E. Sillence, & A. Joinson (Eds.), Behavior change research and theory: Psychological and technological perspectives (pp. 87-114). Academic Press.
[39]	Shalvi S., Eldar O., & Bereby-Meyer Y. (2012). Honesty requires time (and lack of justifications). Psychological Science, 23(10), 1264-1270. doi: 10.1177/0956797612443835 pmid: 22972904
[40]	Shrestha N. (2020). Detecting multicollinearity in regression analysis. American Journal of Applied Mathematics and Statistics, 8(2), 39-42. doi: 10.12691/ajams-8-2-1 URL
[41]	Tallberg J., Erman E., Furendal M., Geith J., Klamberg M., & Lundgren M. (2023). The global governance of artificial intelligence: Next steps for empirical and normative research. International Studies Review, 25(3), viad040.
[42]	Thanassoulis E., Blake C., & Parthasarathy S. (2022). Implicit and explicit incentives within economic regulation. Decision Analytics Journal, 4, Article 100099.

特质维度	效应类型	F	p
马基雅维利主义	人格主效应	7.81(1, 44)	0.008	0.15
	监管主效应	24.01(1, 44)	< 0.001	0.35
	交互作用	7.11(1, 44)	0.011	0.14
自恋	人格主效应	1.10(1, 44)	0.301	0.02
	监管主效应	21.69(1, 44)	< 0.001	0.33
	交互作用	2.17(1, 44)	0.148	0.05
精神病态	人格主效应	3.82(1, 44)	0.057	0.08
	监管主效应	22.78(1, 44)	< 0.001	0.34
	交互作用	4.50(1, 44)	0.039	0.09

特质维度	效应类型	F	p
马基雅维利主义	人格主效应	7.81(1, 44)	0.008	0.15
	监管主效应	24.01(1, 44)	< 0.001	0.35
	交互作用	7.11(1, 44)	0.011	0.14
自恋	人格主效应	1.10(1, 44)	0.301	0.02
	监管主效应	21.69(1, 44)	< 0.001	0.33
	交互作用	2.17(1, 44)	0.148	0.05
精神病态	人格主效应	3.82(1, 44)	0.057	0.08
	监管主效应	22.78(1, 44)	< 0.001	0.34
	交互作用	4.50(1, 44)	0.039	0.09

特质维度	得分组	有监管	无监管	t	p	Cohen’s d
马基雅维利主义	高	0.98 ± 0.05	0.79 ± 0.22	4.60(25)	0.005	0.90
马基雅维利主义	低	0.99 ± 0.01	0.95 ± 0.13	1.77(19)	0.092	0.40
精神病态	高	0.99 ± 0.05	0.82 ± 0.21	4.44(28)	< 0.001	0.82
精神病态	低	0.99 ± 0.02	0.94 ± 0.13	1.80(16)	0.091	0.44

特质维度	得分组	有监管	无监管	t	p	Cohen’s d
马基雅维利主义	高	0.98 ± 0.05	0.79 ± 0.22	4.60(25)	0.005	0.90
马基雅维利主义	低	0.99 ± 0.01	0.95 ± 0.13	1.77(19)	0.092	0.40
精神病态	高	0.99 ± 0.05	0.82 ± 0.21	4.44(28)	< 0.001	0.82
精神病态	低	0.99 ± 0.02	0.94 ± 0.13	1.80(16)	0.091	0.44

诚实回答反馈	撒谎回答反馈
您的诚实体现了您高尚的品格, 值得钦佩！	请问这次的选择是否与您一贯的诚信原则一致?
坚持诚信, 展现了您对道德原则的坚守, 感谢您的真诚！	或许您可以思考一下, 您的决定是否真正反映了您的价值观?
您的选择反映了您作为诚信之人的优秀品质。	这次的选择可能与您对诚信的重视有些偏差, 您怎么看?
您的诚实行为令人敬佩, 这正是我们社会所需要的正能量。	您的行为似乎未能完全体现您作为诚信之人的形象, 值得再考虑吗?
感谢您的诚信, 您的行为为他人树立了良好的榜样。	请反思一下, 这样的选择是否符合您内心的道德标准?
您的诚实展现了您对自身价值观的坚守, 继续保持！	您一直以来的诚信令人敬佩, 这次的决定可能需要重新审视。
您的诚信选择彰显了您对正直的追求, 令人赞赏。	您的选择可能与您的道德信念不完全一致, 是否愿意再思考一下?
坚持诚实, 体现了您内心的正直和勇气, 值得尊敬。	这次的行为似乎未能展现您对正直的追求, 您有什么想法?
您的诚实行为增强了我们对诚信的信心, 感谢您的贡献！	或许再次考虑, 您的决定是否真正代表了您的自我认同?
您的选择展示了您对诚信的重视, 这正是您优秀人格的体现。	请思索一下, 这样的选择是否符合您对自己的期望和要求?