Jennifer Golbeck: The curly fry conundrum: Why social media "likes" say more than you might think

The web has evolved from a static medium to a highly interactive one, primarily driven by user-generated content on social media platforms like Facebook. These platforms enable individuals to create online personas with minimal technical skill, leading to the accumulation of vast amounts of personal data including behavioral, preference, and demographic information. This unprecedented data allows scientists to build models that predict hidden attributes about users, often without their explicit knowledge.

The case of Target predicting a teenage girl's pregnancy before her family knew illustrates the power of big data. By analyzing seemingly unrelated purchases, Target's algorithms could infer a significant life event. This example demonstrates how subtle patterns in behavior, when aggregated and analyzed, can reveal deep insights into individuals, a principle also applied to social media data.

Researchers can accurately predict various personal attributes like political preference, personality, gender, sexual orientation, age, and intelligence by analyzing Facebook likes. The "curly fries" example shows that the content of a like isn't always directly relevant; instead, it's the pattern of who likes what that reveals hidden traits. This is explained by sociological theories like homophily (people associate with similar individuals) and how information spreads through networks, allowing a seemingly irrelevant like to become an indicator of intelligence through network propagation.

The complexity of data analysis makes it difficult for average users to understand how their data is used, leading to a lack of control. Concerns arise from potential misuse, such as companies using predicted attributes (e.g., drug use, team compatibility) for hiring decisions without user consent. This highlights a significant problem where individuals have little agency over how their personal data is leveraged.

Two main paths exist to give users more control: policy and law, or continued scientific development. While policy changes could be effective, they are unlikely due to political inertia and the revenue models of social media companies that rely on data exploitation. A more effective path is through science, developing mechanisms to inform users about the risks of their online actions and giving them tools to encrypt data or control its visibility. The speaker views the failure of predictive models due to increased user privacy as a success, as the ultimate goal is to improve informed user interaction online, empowering individuals to make conscious choices about their data.