Michal Kosinski: Living in a post-privacy world

Episode Summary

Executive Summary: The episode examines how digital footprints—from Facebook likes to browsing, location, and search data—can be used to infer personality, preferences, well-being, and even sensitive traits with surprising accuracy. Stanford’s Michal Kosinski argues that such inference is both powerful and increasingly unavoidable, creating major opportunities for personalization, research, and intervention, but also profound privacy and ethical risks in a post-privacy world.

Main Topics: Personality prediction from digital footprints (Priority: 5/5): Kosinski explains that stable personality traits can be modeled using the Big Five and inferred from online behavior, showing that computers can predict psychological profiles from seemingly ordinary data. Facebook likes and algorithmic accuracy (Priority: 5/5): The discussion highlights research showing that a small number of Facebook likes can outperform acquaintances in predicting personality, and that more likes dramatically improve accuracy. Beyond Facebook: the real scale of data collection (Priority: 5/5): The guest argues that Facebook is only one relatively limited source of data; much more revealing data comes from search logs, browsing history, credit cards, and smartphone location records. Cambridge Analytica and political microtargeting (Priority: 4/5): The conversation addresses Cambridge Analytica as a symptom of a much broader industry using behavioral data for persuasion, emphasizing that the company was relatively unsophisticated compared with larger actors. Privacy, ethics, and the post-privacy world (Priority: 5/5): Kosinski asserts that privacy is effectively gone and that policy should focus on making the post-privacy environment safe rather than assuming data can be fully sealed off. Using prediction to help, not just manipulate (Priority: 4/5): The episode argues that behavioral prediction can support beneficial interventions—career advice, diagnosis, product matching, and well-being improvements—if used responsibly. Inference of sensitive traits from images and social media (Priority: 5/5): Kosinski discusses work showing that facial images and other indirect signals can reveal traits such as sexual orientation and well-being, raising major concerns about consent and surveillance.

Key Arguments: Personality is not as fuzzy as many assume; it can be represented by stable dimensions that are predictive across cultures. Computers can detect subtle patterns in online behavior that humans cannot, enabling highly accurate predictions about personality and behavior. A small set of Facebook likes can predict personality better than a work colleague, and enough likes can rival a spouse’s judgment. Facebook is not the most revealing source of personal data; search, browsing, location, and purchase records are often far more invasive and predictive. Cambridge Analytica was important mainly because it exposed a broader, much larger ecosystem of behavioral targeting, not because it was uniquely sophisticated. Predicting behavior is often a precursor to changing it, and such interventions can be beneficial when used for counseling, health, or commerce. Society should stop assuming privacy can be restored fully and instead design systems and laws for a world where data leakage is endemic. Sensitive traits such as sexual orientation and well-being can be inferred from seemingly unrelated digital traces, even without explicit disclosure.

Data Points: Big Five personality dimensions: 5 - Framework used to describe personality: extroversion, conscientiousness, neuroticism, openness, agreeableness Facebook likes needed to beat a work colleague: 10 random likes - A computer algorithm predicted personality questionnaire responses better than a colleague with only 10 likes Facebook likes needed to beat a spouse: 200-250 likes - With roughly 200 to 250 likes, the algorithm outperformed a spouse in predicting personality-related responses Accuracy benchmark for sexual orientation prediction: Comparable to state-of-the-art Parkinson’s diagnosis or mammogram accuracy - Kosinski described image-based inference of sexual orientation as reaching medical-diagnostic-level performance Cambridge Analytica questionnaire scale: 30,000-60,000 participants - Subcontractor questionnaires plus friends’ data produced large-scale data extraction Average Facebook friends per person: 500-600 friends - Used to explain how participant data could scale to millions of profiles Potentially exposed profiles: Millions - Estimated reach of Cambridge Analytica-style collection through friends’ data Relative spending on psychological targeting: Hillary Clinton spent 3 times more than Donald Trump - Kosinski cited this as evidence that behavioral targeting was widespread beyond Cambridge Analytica

Pivotal Quotes: "You can predict some of my personality dimensions as well as my wife can." — Russ Altman: Introduces the surprising claim that algorithms can match intimate human judgment "We are living in a post-privacy world." — Michal Kosinski: Central claim about how society should frame data governance and policy "If you're not paying for something, it means that you are the product." — Michal Kosinski: Explains the economics of free digital services and user data

Implications: Listeners should assume most digital activity is inferentially rich and not truly private. The future lies in regulating and auditing data use, not pretending it can be eliminated, while ensuring predictive systems are used to help people rather than manipulate them.