Charles Isbell: Computing, Interactive AI, and Race in America | Lex Fridman Podcast #135

Isbell’s home-automation experiments showed that with simple statistical models, you can predict the next button a person will press on a remote with ~93% accuracy, and cluster their behaviors to reach even higher accuracy on action sets. This reveals that much of daily human activity is repetitive and structurally similar across people, despite our self-image as unique and spontaneous.

Isbell argues that machines can map people’s behavioral ‘distributions’ and find overlaps—commonalities in experiences, actions, or preferences—then use those overlaps as safe entry points for dialogue. The goal isn’t to make everyone agree, but to nudge people to ask, “How are we more alike than different?” and to see each other as fully human rather than abstract opponents.

Isbell views real intelligence as something that emerges in interaction with others over time. He criticizes much of machine learning for being over-fitted to narrow tasks and short horizons, and argues that progress requires deploying systems into messy, uncontrolled environments for months or years so they can adapt to changing users and contexts—the essence of lifelong learning.

For Isbell, computing is distinct from math, science, or traditional engineering because it treats models, languages, and machines as interchangeable representations of executable processes. This mindset forces practitioners to surface their assumptions explicitly (through code and parameters) and to think in terms of dynamic systems, which he believes every educated person will need, much like basic scientific literacy.

Using faculty pipeline data, Isbell shows that top departments overwhelmingly hire from a small set of elite PhD programs, not because that’s where all the best talent necessarily is, but because it feels ‘safe.’ This risk-averse behavior minimizes the chance of hiring a perceived mistake but greatly narrows opportunity, calcifies historical inequities, and discourages unconventional talent.