Please visit:

Academic Appointments

Program Affiliations

  • Symbolic Systems Program

2023-24 Courses

Stanford Advisees

  • Doctoral Dissertation Advisor (AC)
    Xubo Cao
  • Doctoral (Program)
    Jun Lin, Josephine Tan

All Publications

  • Facial recognition technology and human raters can predict political orientation from images of expressionless faces even when controlling for demographics and self-presentation. The American psychologist Kosinski, M., Khambatta, P., Wang, Y. 2024


    Carefully standardized facial images of 591 participants were taken in the laboratory while controlling for self-presentation, facial expression, head orientation, and image properties. They were presented to human raters and a facial recognition algorithm: both humans (r = .21) and the algorithm (r = .22) could predict participants' scores on a political orientation scale (Cronbach's α = .94) decorrelated with age, gender, and ethnicity. These effects are on par with how well job interviews predict job success, or alcohol drives aggressiveness. The algorithm's predictive accuracy was even higher (r = .31) when it leveraged information on participants' age, gender, and ethnicity. Moreover, the associations between facial appearance and political orientation seem to generalize beyond our sample: The predictive model derived from standardized images (while controlling for age, gender, and ethnicity) could predict political orientation (r ≈ .13) from naturalistic images of 3,401 politicians from the United States, the United Kingdom, and Canada. The analysis of facial features associated with political orientation revealed that conservatives tended to have larger lower faces. The predictability of political orientation from standardized images has critical implications for privacy, the regulation of facial recognition technology, and understanding the origins and consequences of political orientation. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

    View details for DOI 10.1037/amp0001295

    View details for PubMedID 38512164

  • Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT NATURE COMPUTATIONAL SCIENCE Hagendorff, T., Fabi, S., Kosinski, M. 2023
  • Large language models know how the personality of public figures is perceived by the general public. Scientific reports Cao, X., Kosinski, M. 2024; 14 (1): 6735


    We show that people's perceptions of public figures' personalities can be accurately predicted from their names' location in GPT-3's semantic space. We collected Big Five personality perceptions of 226 public figures from 600 human raters. Cross-validated linear regression was used to predict human perceptions from public figures' name embeddings extracted from GPT-3. The models' accuracy ranged from r = .78 to .88 without controls and from r = .53 to .70 when controlling for public figures' likability and demographics, after correcting for attenuation. Prediction models showed high face validity as revealed by the personality-descriptive adjectives occupying their extremes. Our findings reveal that GPT-3 word embeddings capture signals pertaining to individual differences and intimate traits.

    View details for DOI 10.1038/s41598-024-57271-z

    View details for PubMedID 38509191

    View details for PubMedCentralID 4313801

  • Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT. Nature computational science Hagendorff, T., Fabi, S., Kosinski, M. 2023; 3 (10): 833-838


    We design a battery of semantic illusions and cognitive reflection tests, aimed to elicit intuitive yet erroneous responses. We administer these tasks, traditionally used to study reasoning and decision-making in humans, to OpenAI's generative pre-trained transformer model family. The results show that as the models expand in size and linguistic proficiency they increasingly display human-like intuitive system 1 thinking and associated cognitive errors. This pattern shifts notably with the introduction of ChatGPT models, which tend to respond correctly, avoiding the traps embedded in the tasks. Both ChatGPT-3.5 and 4 utilize the input-output context window to engage in chain-of-thought reasoning, reminiscent of how people use notepads to support their system 2 thinking. Yet, they remain accurate even when prevented from engaging in chain-of-thought reasoning, indicating that their system-1-like next-word generation processes are more accurate than those of older models. Our findings highlight the value of applying psychological methodologies to study large language models, as this can uncover previously undetected emergent characteristics.

    View details for DOI 10.1038/s43588-023-00527-x

    View details for PubMedID 38177754

  • Political Attitudes and Disease Threat: Regional Pathogen Stress Is Associated With Conservative Ideology Only for Older Individuals. Personality & social psychology bulletin Brown, G. D., Walasek, L., Mullett, T. L., Quispe-Torreblanca, E. G., Fincher, C. L., Kosinski, M., Stillwell, D. 2023: 1461672231183199


    What environmental factors are associated with individual differences in political ideology, and do such associations change over time? We examine whether reductions in pathogen prevalence in U.S. states over the past 60 years are associated with reduced associations between parasite stress and conservatism. We report a positive association between infection levels and conservative ideology in the United States during the 1960s and 1970s. However, this correlation reduces from the 1980s onwards. These results suggest that the ecological influence of infectious diseases may be larger for older people who grew up (or whose parents grew up) during earlier time periods. We test this hypothesis by analyzing the political affiliation of 45,000 Facebook users, and find a positive association between self-reported political affiliation and regional pathogen stress for older (>40 years) but not younger individuals. It is concluded that the influence of environmental pathogen stress on ideology may have reduced over time.

    View details for DOI 10.1177/01461672231183199

    View details for PubMedID 37424438

  • Overlap in meaning is a stronger predictor of semantic activation in GPT-3 than in humans. Scientific reports Digutsch, J., Kosinski, M. 2023; 13 (1): 5035


    Modern large language models generate texts that are virtually indistinguishable from those written by humans and achieve near-human performance in comprehension and reasoning tests. Yet, their complexity makes it difficult to explain and predict their functioning. We examined a state-of-the-art language model (GPT-3) using lexical decision tasks widely used to study the structure of semantic memory in humans. The results of four analyses showed that GPT-3's patterns of semantic activation are broadly similar to those observed in humans, showing significantly higher semantic activation in related (e.g., "lime-lemon") word pairs than in other-related (e.g., "sour-lemon") or unrelated (e.g., "tourist-lemon") word pairs. However, there are also significant differences between GPT-3 and humans. GPT-3's semantic activation is better predicted by similarity in words' meaning (i.e., semantic similarity) rather than their co-occurrence in the language (i.e., associative similarity). This suggests that GPT-3's semantic network is organized around word meaning rather than their co-occurrence in text.

    View details for DOI 10.1038/s41598-023-32248-6

    View details for PubMedID 36977744

    View details for PubMedCentralID 5863044

  • Author Correction: Facial recognition technology can expose political orientation from naturalistic facial images. Scientific reports Kosinski, M. 2021; 11 (1): 23228

    View details for DOI 10.1038/s41598-021-02785-z

    View details for PubMedID 34824377

  • Regional personality assessment through social media language. Journal of personality Giorgi, S., Nguyen, K. L., Eichstaedt, J. C., Kern, M. L., Yaden, D. B., Kosinski, M., Seligman, M. E., Ungar, L. H., Andrew Schwartz, H., Park, G. 2021


    OBJECTIVE: We explore the personality of counties as assessed through linguistic patterns on social media. Such studies were previously limited by the cost and feasibility of large-scale surveys; however, language-based computational models applied to large social media datasets now allow for large-scale personality assessment.METHOD: We applied a language-based assessment of the five factor model of personality to 6,064,267 U.S. Twitter users. We aggregated the Twitter-based personality scores to 2,041 counties and compared to political, economic, social, and health outcomes measured through surveys and by government agencies.RESULTS: There was significant personality variation across counties. Openness to experience was higher on the coasts, conscientiousness was uniformly spread, extraversion was higher in southern states, agreeableness was higher in western states, and emotional stability was highest in the south. Across 13 outcomes, language-based personality estimates replicated patterns that have been observed in individual-level and geographic studies. This includes higher Republican vote share in less agreeable counties and increased life satisfaction in more conscientious counties.CONCLUSIONS: Results suggest that regions vary in their personality and that these differences can be studied through computational linguistic analysis of social media. Furthermore, these methods may be used to explore other psychological constructs across geographies.

    View details for DOI 10.1111/jopy.12674

    View details for PubMedID 34536229

  • Facial recognition technology can expose political orientation from naturalistic facial images. Scientific reports Kosinski, M. 2021; 11 (1): 100


    Ubiquitous facial recognition technology can expose individuals' political orientation, as faces of liberals and conservatives consistently differ. A facial recognition algorithm was applied to naturalistic images of 1,085,795 individuals to predict their political orientation by comparing their similarity to faces of liberal and conservative others. Political orientation was correctly classified in 72% of liberal-conservative face pairs, remarkably better than chance (50%), human accuracy (55%), or one afforded by a 100-item personality questionnaire (66%). Accuracy was similar across countries (the U.S., Canada, and the UK), environments (Facebook and dating websites), and when comparing faces across samples. Accuracy remained high (69%) even when controlling for age, gender, and ethnicity. Given the widespread use of facial recognition, our findings have critical implications for the protection of privacy and civil liberties.

    View details for DOI 10.1038/s41598-020-79310-1

    View details for PubMedID 33431957

  • Spouses' faces are similar but do not become more similar with time. Scientific reports Tea-Makorn, P. P., Kosinski, M. 2020; 10 (1): 17001


    The widely disseminated convergence in physical appearance hypothesis posits that long-term partners' facial appearance converges with time due to their shared environment, emotional mimicry, and synchronized activities. Although plausible, this hypothesis is incompatible with empirical findings pertaining to a wide range of other traits-such as personality, intelligence, attitudes, values, and well-being-in which partners show initial similarity but do not converge over time. We solve this conundrum by reexamining this hypothesis using the facial images of 517 couples taken at the beginning of their marriages and 20 to 69years later. Using two independent methods of estimating their facial similarity (human judgment and a facial recognition algorithm), we show that while spouses' faces tend to be similar at the beginning of marriage, they do not converge over time, bringing facial appearance in line with other personal characteristics.

    View details for DOI 10.1038/s41598-020-73971-8

    View details for PubMedID 33046769

  • Individual-Level Analyses of the Impact of Parasite Stress on Personality: Reduced Openness Only for Older Individuals. Personality & social psychology bulletin Mullett, T. L., Brown, G. D., Fincher, C. L., Kosinski, M., Stillwell, D. 2019: 146167219843918


    The parasite stress hypothesis predicts that individuals living in regions with higher infectious disease rates will show lower openness, agreeableness, and extraversion, but higher conscientiousness. This article, using data from more than 250,000 U.S. Facebook users, reports tests of these predictions at the level of both U.S. states and individuals and evaluates criticisms of previous findings. State-level results for agreeableness and conscientiousness are consistent with previously reported cross-national findings, but others (a significant positive correlation with extraversion and no correlation with openness) are not. However, effects of parasite stress on conscientiousness and agreeableness are not found when analyses account for the data's hierarchical structure and include controls. We find that only openness is robustly related to parasite stress in these analyses, and we also find a significant interaction with age: Older, but not younger, inhabitants of areas of high parasite stress show lower openness. Interpretations of the findings are discussed.

    View details for DOI 10.1177/0146167219843918

    View details for PubMedID 31046588

  • Privacy in the age of psychological targeting. Current opinion in psychology Matz, S. C., Appel, R. E., Kosinski, M. n. 2019; 31: 116–21


    Psychological targeting describes the practice of extracting people's psychological profiles from their digital footprints (e.g. their Facebook Likes, Tweets or credit card records) in order to influence their attitudes, emotions or behaviors through psychologically informed interventions at scale. We discuss how the increasingly blurred lines between public and private information, and the continuation of the outdated practices of notice and consent, challenge traditional conceptualizations of privacy in the context of psychological targeting. Drawing on the theory of contextual integrity, we argue that it is time to rethink privacy and move beyond the questions of who collects what data to how the data are being used. Finally, we suggest that regulations of psychological targeting should be accompanied by a mindset that fosters (1) privacy by design to make it easy for individuals to act in line with their privacy goals, as well as (2) disclosure by choice, to allow individuals to freely decide whether and when they might be willing to forsake their privacy for better service.

    View details for DOI 10.1016/j.copsyc.2019.08.010

    View details for PubMedID 31563799

  • A Computer Adaptive Measure of Delay Discounting ASSESSMENT Mahalingam, V., Palkovics, M., Kosinski, M., Cek, I., Stillwell, D. 2018; 25 (8): 1036–55
  • The Promotion of a Bright Future and the Prevention of a Dark Future: Time Anchored Incitements in News Articles and Facebook's Status Updates. Frontiers in psychology Garcia, D., Drejing, K., Amato, C., Kosinski, M., Sikström, S. 2018; 9: 1623


    Background: Research suggests that humans have the tendency to increase the valence of events when these are imagined to happen in the future, but to decrease the valence when the same events are imagined to happen in the past. This line of research, however, has mostly been conducted by asking participants to value imagined, yet probable, events. Our aim was to re-examine this time-valence asymmetry using real-life data: a Reuter's news and a Facebook status updates corpus. Method: We organized the Reuter news (120,000,000 words) and the Facebook status updates data (41,056,346 words) into contexts grouped in chronological order (i.e., past, present, and future) using verbs and years as time markers. These contexts were used to estimate the valence of each article and status update, respectively, in relation to the time markers using natural language processing tools (i.e., the Latent Semantic Analysis algorithm). Results: Our results using verbs, in both text corpus, showed that valence for the future was significantly higher compared to the past (future > past). Similarly, in the Reuter year condition, valence increased approximately linear from 1994 to 1999 for texts written 1996-1997. In the Facebook year condition, the valence of the future was also significantly higher than past valence. Conclusion: Generally, the analyses of the Reuters data indicated that the past is devaluated relative to both the present and the future, while the analyses of the Facebook data indicated that both the past and the present are devaluated against the future. On this basis, we suggest that people strive to communicate the promotion of a bright future and the prevention of a dark future, which in turn leads to a temporal-valence asymmetrical phenomenon (valence = past < present < future). " I have a dream that one day this nation will rise up and live out the true meaning of its creed: "We hold these truths to be self-evident, that all men are created equal."I have a dream that one day on the red hills of Georgia, the sons of former slaves and the sons of former slave owners will be able to sit down together at the table of brotherhood.I have a dream that one day even the state of Mississippi, a state sweltering with the heat of injustice, sweltering with the heat of oppression, will be transformed into an oasis of freedom and justice.I have a dream that my four little children will one day live in a nation where they will not be judged by the color of their skin but by the content of their character.I have a dream today!"Martin Luther King, Jr., 28th of August 1963, at the Lincoln Memorial, Washington, DC, United States.

    View details for DOI 10.3389/fpsyg.2018.01623

    View details for PubMedID 30271358

    View details for PubMedCentralID PMC6146082

  • Latent human traits in the language of social media: An open-vocabulary approach. PloS one Kulkarni, V., Kern, M. L., Stillwell, D., Kosinski, M., Matz, S., Ungar, L., Skiena, S., Schwartz, H. A. 2018; 13 (11): e0201703


    Over the past century, personality theory and research has successfully identified core sets of characteristics that consistently describe and explain fundamental differences in the way people think, feel and behave. Such characteristics were derived through theory, dictionary analyses, and survey research using explicit self-reports. The availability of social media data spanning millions of users now makes it possible to automatically derive characteristics from behavioral data-language use-at large scale. Taking advantage of linguistic information available through Facebook, we study the process of inferring a new set of potential human traits based on unprompted language use. We subject these new traits to a comprehensive set of evaluations and compare them with a popular five factor model of personality. We find that our language-based trait construct is often more generalizable in that it often predicts non-questionnaire-based outcomes better than questionnaire-based traits (e.g. entities someone likes, income and intelligence quotient), while the factors remain nearly as stable as traditional factors. Our approach suggests a value in new constructs of personality derived from everyday human language use.

    View details for PubMedID 30485276

  • Facial Width-to-Height Ratio Does Not Predict Self-Reported Behavioral Tendencies. Psychological science Kosinski, M. 2017; 28 (11): 1675-1682


    A growing number of studies have linked facial width-to-height ratio (fWHR) with various antisocial or violent behavioral tendencies. However, those studies have predominantly been laboratory based and low powered. This work reexamined the links between fWHR and behavioral tendencies in a large sample of 137,163 participants. Behavioral tendencies were measured using 55 well-established psychometric scales, including self-report scales measuring intelligence, domains and facets of the five-factor model of personality, impulsiveness, sense of fairness, sensational interests, self-monitoring, impression management, and satisfaction with life. The findings revealed that fWHR is not substantially linked with any of these self-reported measures of behavioral tendencies, calling into question whether the links between fWHR and behavior generalize beyond the small samples and specific experimental settings that have been used in past fWHR research.

    View details for DOI 10.1177/0956797617716929

    View details for PubMedID 28976810

  • Frankly, We Do Give a Damn: The Relationship Between Profanity and Honesty. Social psychological and personality science Feldman, G., Lian, H., Kosinski, M., Stillwell, D. 2017; 8 (7): 816-826


    There are two conflicting perspectives regarding the relationship between profanity and dishonesty. These two forms of norm-violating behavior share common causes and are often considered to be positively related. On the other hand, however, profanity is often used to express one's genuine feelings and could therefore be negatively related to dishonesty. In three studies, we explored the relationship between profanity and honesty. We examined profanity and honesty first with profanity behavior and lying on a scale in the lab (Study 1; N = 276), then with a linguistic analysis of real-life social interactions on Facebook (Study 2; N = 73,789), and finally with profanity and integrity indexes for the aggregate level of U.S. states (Study 3; N = 50 states). We found a consistent positive relationship between profanity and honesty; profanity was associated with less lying and deception at the individual level and with higher integrity at the society level.

    View details for DOI 10.1177/1948550616681055

    View details for PubMedID 29187959

    View details for PubMedCentralID PMC5686790