Michal Kosinski

Kosinski holds a doctorate in psychology from the University of Cambridge and master's degrees in psychometrics and in social psychology.^[6] He previously served as a post-doctoral scholar at Stanford's Computer Science Department, as the deputy director of the University of Cambridge Psychometrics Centre, and as a researcher at Microsoft Research (Machine Learning Group).^[7]

Summarize Timeline Fact Check

In 2013, Kosinski and David Stillwell published a paper entitled "Private Traits and Attributes Are Predictable from Digital Records of Human Behavior".^[8] They claimed that Facebook Likes reveal personal traits and sensitive attributes, from sexual and political orientation to mental health. "Individual traits and attributes can be predicted to a high degree of accuracy based on records of users' Likes," they wrote. In 2012, Facebook patented a method doing precisely what Kosinski and Stillwell did, "Determining user personality characteristics from social networking system communications and characteristics".^[9]

In 2015, Kosinski, Youyou, and Stillwell published a study in PNAS reporting that computer-based personality judgments derived from Facebook Likes were more accurate than judgments made by friends, family members, and spouses.^[10]

In 2017, Kosinski, Matz, Nave, and Stillwell published a PNAS study reporting that psychologically tailored advertising messages could increase the effectiveness of digital mass persuasion.^[11]

In 2017, Kosinski co-published a paper showing that modern artificial intelligence can predict someone's sexual orientation based on facial images.^[12][13][14] The research was conducted on over 130,000 pictures and used existing facial recognition systems and AI algorithm. Their AI could predict the sexual orientation of gay men 81% of the time, while a human would be right 61% of the time.^[15]

In 2021, Kosinski published a study in Scientific Reports demonstrating that facial recognition technology can predict political orientation from naturalistic facial images with 72% accuracy, substantially better than chance or conventional methods such as personality questionnaires.^[16]

Since 2023, Kosinski's research has increasingly focused on the cognitive and social capacities of large language models. In a 2023 study in Scientific Reports, Digutsch and Kosinski compared semantic activation in GPT-3 with human lexical decision data, reporting that GPT-3's semantic activation broadly mirrored human patterns but was more strongly driven by semantic similarity than by associative relatedness.^[17] In Nature Computational Science, Hagendorff, Fabi, and Kosinski tested OpenAI language models on cognitive-reflection and semantic-illusion tasks, reporting that human-like intuitive errors emerged in larger pre-ChatGPT models but were reduced in ChatGPT.^[18] In 2024, Kosinski published a study in PNAS evaluating eleven large language models on false-belief tasks used to study theory of mind, reporting that older models solved no tasks while GPT-4 solved 75% of the tasks, matching the performance of six-year-old children reported in previous studies.^[19] Related work examined language-model agency through negotiation games and the extent to which large language models encode public perceptions of public figures' personalities.^[20][21][22]

Kosinski and David Stillwell co-created the myPersonality Facebook application and database, which became a major research resource for social scientists. A 2015 article in American Psychologist stated that the myPersonality application was owned by Stillwell and Kosinski, and that their Facebook-based psychological studies had attracted over 10 million participants.^[23]

His research on facial recognition systems raised controversy. Kosinski and Wang stated that their findings were intended to expose the risks posed by existing computer-vision technologies, writing that such systems could threaten the privacy and safety of gay men and women.^[12]

While Deputy Director of the University of Cambridge Psychometrics Centre, Kosinski was approached by departmental colleague Aleksandr Kogan about a potential collaboration with SCL Group, Cambridge Analytica's parent company.^[24] The Centre's co-director David Stillwell first declined to share its data, as it had been collected for academic purposes; an alternative was proposed in which Kogan would collect new data and pass it through the Centre's prediction models.^[24] Kogan testified that he calculated a price of $500,000 for the Centre's models ("I thought, all right, maybe we will give the Psychometrics Centre $500,000 for the modelling aspect"); SCL considered the price too high and "instructed [Kogan] to remove them from the project."^[24] Kogan subsequently built his own Facebook app independently; he confirmed in 2017 that "no data or models from the psychometrics centre were ever used for anything with SCL." Kosinski raised a formal complaint with university authorities, calling Kogan's approach "highly unethical," leading to legal undertakings prohibiting any use of the Centre's work in the project.^[25] In November 2014, Kosinski contacted The Guardian and provided documents about the Kogan-SCL connection; the newspaper's resulting December 2015 article was the first press report about Cambridge Analytica's data practices.^[26][27] No official investigation, including those by the FTC, UK ICO, and UK Parliament, has identified Kosinski as involved with Cambridge Analytica.^[28][29]

• Ethics of artificial intelligence
• AI takeover