How AI is using your data to influence you

Episode Summary

Executive Summary: The episode examines psychological targeting: how digital traces from social media, phones, purchases, and emerging neurotech can be used to infer personality and influence behavior. Sandra Matz argues this is now widely accessible via AI, effective in some contexts, ethically fraught, and increasingly important to regulate through privacy-preserving technologies like federated learning.

Main Topics: Psychological profiling from digital footprints (Priority: 5/5): Matz explains how online traces can reveal Big Five traits and other psychological dimensions, even from years-old posts, enabling inference about who people are and what motivates them. Behavioral targeting and persuasion (Priority: 5/5): The conversation explores how matching ads and messages to personality can increase effectiveness, using examples from beauty retail and savings interventions. The Cambridge Analytica era and limits of proof (Priority: 4/5): The hosts revisit the 2018 scandal, noting that while the targeting was real, there was little direct evidence proving large-scale electoral persuasion. Privacy, surveillance, and the data economy (Priority: 5/5): Matz details the breadth of data collected through phones, apps, payments, cameras, and location signals, arguing current transparency-and-control models are insufficient. Changing legal and ethical landscape (Priority: 4/5): The episode compares U.S. state-level privacy rules with the EU, while stressing that regulation often shifts burden onto users rather than preventing risky collection. Future of data: AI, federated learning, and neurotech (Priority: 5/5): The discussion looks ahead to brain-computer interfaces, body sensors, and privacy-preserving machine learning that could deliver personalization without centralizing raw data. Practical user defenses (Priority: 3/5): The show closes with listener advice: manage app permissions, use privacy-respecting tools, and be mindful that phone and location data are difficult to fully hide.

Key Arguments: Digital traces can be enough for accurate psychological inference; personality traits remain relatively stable, so older data can still predict current tendencies. Psychological targeting is effective when the message aligns with a person’s motivations; Matz cites large lifts in buying and saving behavior when ads match traits. AI has democratized profiling: off-the-shelf models can infer traits from public posts without custom training data. The biggest risks are not just commercial persuasion but opaque political, health, and social influence with weak user consent. 'Nothing to hide' is not a reliable defense because data can become dangerous when laws, leadership, or social conditions change. Transparency and control are necessary but insufficient; users cannot realistically manage every data permission or term of service. Privacy-preserving methods like federated learning can preserve convenience while reducing centralized data collection, and companies may benefit from lower breach risk. Future neurotech could make the problem more intimate by reading signals from inside the body and brain rather than inferring from outside behavior.

Data Points: Beauty ad targeting lift: about 50% higher likelihood of purchase - Extrovert-matched ads for a UK beauty retailer produced higher buying rates when message fit personality. Savings intervention lift: 60% more people hit the goal - A study with households having less than $100 in savings used personality-based messaging to help people save an additional $100 over four weeks. Facebook records in Cambridge Analytica scandal: upwards of 50 million plus Facebook records - Referenced as the scale of data access in the 2018 scandal. Saved amount target in study: $100 over 4 weeks - Participants with under $100 in savings were encouraged to save more using tailored messages. Personality model used: Big Five traits - Openness, conscientiousness, extroversion, agreeableness, and neuroticism were inferred from social media traces.

Pivotal Quotes: "You don't need to be a big corporation to profile people these days." — Sandra Matz: Explains how AI now allows ordinary users to infer personality from public posts. "Your phone is essentially a stalker 24-7 because it is someone in your pocket who knows exactly where you are at any given point in time." — Sandra Matz: Describing how ubiquitous mobile and location data collection is. "It's absolutely necessary. But it's nowhere near sufficient." — Sandra Matz: On transparency and control as privacy principles in regulation.

Implications: Listeners should assume far more data is being collected than they realize and use app permissions, privacy tools, and cautious sharing. For industry, federated learning and similar methods offer a path to personalization with less surveillance.