301 | Tina Eliassi-Rad on Al, Networks, and Epistemic Instability

What It Covers

Tina Eliassi-Rad explores how AI systems and humans coevolve through feedback loops, examining network analysis techniques, epistemic instability in democracy, algorithmic bias, and the dangers of treating AI as objective authorities rather than tools with specific limitations.

Key Questions Answered

• •Network Relational Dependencies: Graph machine learning identifies two dominant social network formation patterns: triangle closure where friends of friends become friends, and preferential attachment where people connect to popular nodes. Deviations from these patterns reveal significant relationships like romantic partners who bridge multiple social groups.
• •Prediction Accuracy Limitations: The paradox of big data means vast datasets exist but predicting individual outcomes remains difficult. Recommendation systems exploit popular content rather than explore diverse options, creating filter bubbles based on location and demographics rather than serving individual preferences through personalized exploration.
• •Labor Data Predicts Mortality: Analysis of six million Danish citizens revealed labor sector data predicts death between ages thirty-five to sixty-five with seventy-eight percent accuracy, outperforming health data. Male gender and working as electricians versus office workers emerged as stronger mortality indicators than inconsistent healthcare records.
• •Benchmark Hacking Problem: Machine learning researchers optimize for leaderboard rankings on standardized datasets rather than understanding underlying phenomena. This creates a finite competitive toolbox where models chase one percent improvements without measuring uncertainty, questioning data sources, or documenting technical limitations and assumptions.
• •Epistemic Instability Threat: AI systems introduce instability by eliminating shared trusted information sources like Walter Cronkite represented in the nineteen sixties. Generative AI never asks clarifying questions or admits uncertainty to maintain utility, enabling manipulation through jailbreaking and undermining the shared reality democracy requires.