Lex Fridman PodcastCharles Isbell: Computing, Interactive AI, and Race in America | Lex Fridman Podcast #135
CHAPTERS
- 0:00 – 2:32
Charles Isbell’s background, charisma, and why this conversation matters
Lex introduces Charles Isbell as Georgia Tech’s Dean of Computing and a long-time AI/ML researcher with broad interests. He frames the episode as touching not only AI, but also empathy, online discourse, and race in America.
- •Charles Isbell’s roles: dean, researcher, educator in AI and computing
- •Lex’s motivation for the conversation and the cultural moment (pre-election context)
- •Episode themes preview: AI, music, race, empathy in communication
- •Sponsor mentions and housekeeping
- 2:32 – 8:45
Top movies: screwball comedy, world-building action, and a perfect ending
Charles answers Lex’s “top 3 movies” question by “cheating” with genres and exemplars. He highlights dialogue-driven storytelling, immersive worlds that don’t over-explain, and the power of a final scene to summarize an entire narrative.
- •His Girl Friday as a dialogue/banter-driven classic
- •Crouching Tiger, Hidden Dragon and John Wick as similar ‘drop you into a world’ storytelling
- •Favorite Jackie Chan pick: Drunken Master 2 and what makes its fights special
- •Nothing But a Man and The Godfather’s ending as a compact narrative masterpiece
- 8:45 – 14:28
Quantifying daily life: smart homes, two days of data, and human predictability
Lex asks about Charles’s self-tracking and home instrumentation experiments. Charles explains how simple behavioral traces (like IR signals from remotes) can predict actions with startling accuracy, revealing how routine-driven humans are.
- •Wiring a home and capturing infrared remote-control activity
- •Why ‘the right two days’ (often a weekend) can reveal most routine behavior
- •Surprisingly high prediction accuracy with very simple statistical methods
- •Clustering people and actions: discovering structure in button presses and habits
- 14:28 – 26:12
Breaking out of social media silos: language, empathy, and finding overlap
The conversation moves from predictability to polarization. Charles argues the hardest part of bridging bubbles isn’t exposure—it’s mismatched language and implicit assumptions—and suggests AI could help by identifying points of genuine commonality.
- •Why groups in ‘silos’ often talk past each other (the “clock on the wall” story)
- •Empathy vs sympathy: separable concepts with different social effects
- •AI as a tool to find overlap between people (similarity mappings)
- •The key challenge: motivating people to ask “How are we alike?”
- •Online othering: why it’s easier to troll people who don’t feel ‘real’
- 26:12 – 32:46
What ‘Interactive AI’ means: intelligence as social, adaptive, and long-horizon
Charles defines Interactive AI as aiming at true intelligence, not just task performance. He argues intelligence is fundamentally about interacting with others over time—communicating, adapting, and learning amid changing people and contexts.
- •Charles’s identity: ML researcher motivated by the larger AI goal
- •AI vs ML: overlapping fields with different aims (engineering vs intelligence ideals)
- •Why interaction matters: intelligence is meaningful in relation to observers/others
- •Adaptation over time: people change across minutes, years, and contexts
- •Communication gaps and projection as core problems in human and machine interaction
- 32:46 – 41:14
Lifelong learning and the limits of task-focused ML (plus GPT-3-style scaling)
Lex probes lifelong learning as an underexplored frontier. Charles critiques the community’s task fixation and argues progress requires systems that live in messy real environments for months, while also questioning whether ‘scale alone’ can reach general intelligence.
- •Why modern ML often optimizes narrowly and overfits to benchmark tasks
- •Lifelong learning requires long deployments in uncontrolled, chaotic environments
- •Incentives in academia/industry discourage multi-year, high-uncertainty bets
- •Scaling argument (GPT-3/self-supervision): useful, but doesn’t “solve it” automatically
- •How technology changes the problem definition (Google search as minimizing false positives)
- 41:14 – 48:56
Faculty hiring as search: rankings, risk aversion, and minimizing false positives
Charles connects the “false positive minimization” lens to academic hiring. He presents data suggesting elite departments disproportionately hire from the same elite PhD programs and explains why institutional incentives reward safety over discovering unconventional talent.
- •Hiring as a search problem shaped by risk and long-term commitment (tenure timelines)
- •Striking concentration: top departments hiring heavily from top-4/top-10 PhD origins
- •Why the system self-reinforces (career risk, reputational hedging)
- •How to find ‘the 13th page’ candidate: enlarge pools, adjust incentives/loss functions
- •Graduate admissions as a pipeline bottleneck that pushes decisions earlier (age ~17)
- 48:56 – 56:10
University rankings and reputation: what they measure—and what they distort
Lex challenges the legitimacy of rankings; Charles explains CS rankings are largely reputation-based. They discuss how narratives become “objective” in practice, how institutions game incentives, and how ‘objectivity’ itself can be shaped by power and design decisions.
- •US News CS rankings: essentially reputation surveys, not a formula of outputs
- •How Georgia Tech climbed: leadership during inflection points (college, undergrad reform, online masters)
- •Rankings as feedback loops: attracting talent reinforces the narrative
- •‘Narrative is objective’ as an objective measure of collective opinion
- •Example of ‘engineered objectivity’: camera/film historically optimized for lighter skin tones until market pressures shifted trade-offs
- 56:10 – 1:03:52
Science communicators, tenure as training, and academia’s relationship with fame
The discussion turns to public intellectuals and why academia often discourages bold, popular communication. Charles argues tenure is less a switch and more a conditioning process, and that great universities need visible engagement with the wider world.
- •Why public-facing communication is treated as “vulgar” or unscientific by some
- •Tenure as a training regime shaping behavior and values, not liberation
- •Jealousy and status anxiety around TED-talk-style popularity
- •Need for diverse modes of impact: research excellence plus public engagement
- •Examples: Rod Brooks as a respected ‘public’ figure while doing strong work
- 1:03:52 – 1:15:59
Hip hop, rap, and funk: culture, sampling, and an education plan for newcomers
Lex asks Charles to teach him hip hop and funk. Charles distinguishes hip hop (culture) from rap (subset), recommends a listening path centered on lyrical and political depth, and connects sampling to broader musical traditions like jazz and funk.
- •Hip hop as culture: graffiti/tagging, dance, DJing, and rap
- •Starter canon: Public Enemy (It Takes a Nation of Millions...), Mos Def, EPMD, The Last Poets
- •Sampling as a foundational creative instrument (DJ-centered origins)
- •Funk lineage and revival: George Clinton/Parliament/Funkadelic and beyond
- •Lyrics vs modern trends: concern about drift, but optimism about underground scenes
- 1:15:59 – 1:23:33
What is computing? Mindsets, curriculum as a ‘data structure,’ and human meaning
Charles lays out a philosophy of computing education: disciplines teach mindsets, not just tools. He argues computing’s distinctive mindset is the equivalence of models, languages, and machines—and that the human inside the triangle gives computation its meaning.
- •Curriculum as the fundamental ‘data structure’ of education
- •Disciplines as mindsets (beyond tools and skills)
- •Core computing idea: models ↔ languages ↔ machines are equivalent and executable/dynamic
- •Computing as science + math + engineering through the lens of execution and constraints
- •Why human context matters: data and computation only become meaningful via human interaction
- 1:23:33 – 1:31:58
Computing everywhere: precision, ‘learn to code’ vs ‘learn to think,’ and citizenship
They explore whether computing will ‘eat’ other fields or dissolve into them. Charles argues not everyone must code, but everyone must learn computational precision and the habit of surfacing assumptions—both for work and for navigating systems like social media.
- •Computing as both thriving major and essential service to every discipline
- •Future scholarship (history, psychology, etc.) increasingly depends on data and computation
- •The key educational outcome: precise, executable problem statements and explicit assumptions
- •Coding as a means, not the end; languages as central to how we shape outcomes
- •Computational thinking as civic literacy for understanding algorithmic influence
- 1:31:58 – 1:47:59
Race and identity: Atlanta roots, Georgia Tech and MIT, and navigating stark contrasts
Lex transitions to race; Charles recounts growing up in Atlanta’s Black communities and then encountering predominantly white elite institutions. He contrasts Atlanta with Boston’s social geography at the time and describes feeling institutionally supported while still facing racialized realities in the city.
- •Charles’s identity preference: “Black” (capital B)
- •Cultural shift from predominantly Black schooling to being ‘one of few’ at Georgia Tech/MIT
- •Boston in the early 1990s: lack of cohesive Black middle class and segregated risk maps
- •Institutional support vs broader societal structures: universities vs city systems
- •Empowerment, delusion, and resilience: confidence as useful—but must track reality
- 1:47:59 – 1:57:50
Police encounters and the psychology of fear: the ‘fire’ metaphor and hatred’s cost
Charles tells a story of being pulled over and having a gun drawn on him, and explains the lasting bodily tension that can follow. He frames policing as a necessary but dangerous force—like fire—while warning that hatred is energy-intensive and rarely helpful to carry.
- •Pulled over and confronted with a drawn gun; the fear of impunity and narrative reversal
- •Why police presence can trigger chronic vigilance even decades later
- •‘Fire’ metaphor: necessary institution that must be treated as dangerous
- •How fear can turn into hatred—and why resisting that matters for living well
- •Empathy at the structural level: systems that create clustering, feedback loops, and ‘othering’
- 1:57:50 – 2:05:43
Racial tensions and media visibility: why protests ignite, and why history repeats
They discuss why racial unrest surges in certain periods and how visibility drives change. Charles compares TV’s role in the civil-rights era to social media today, recalls earlier incidents that didn’t “go viral,” and notes the recurring sameness of public reactions across decades.
- •Why ‘now’: generational cycles, demographic change, and increased visibility
- •TV’s role in civil rights (e.g., Birmingham) vs attempts to suppress coverage elsewhere
- •Pre-viral era examples: incidents that vanished from attention vs Rodney King’s spread
- •George Floyd video as a catalyzing visibility moment
- •Repetition of op-eds and arguments across eras: the cost of forgetting history
- 2:05:43 – 2:23:50
MLK vs Malcolm X, the role of struggle, and existential questions about the future
The conversation closes with hard questions: whether violence ‘works,’ whether civilization will survive, and how to live with uncertainty. Charles emphasizes that change requires struggle, expresses cautious optimism about humanity’s future, and reflects on death, time, and meaning.
- •Rejecting a simplistic MLK vs Malcolm X binary; acknowledging violence’s historical efficacy
- •Change as struggle: power rarely yields without pressure
- •Optimism and “This too shall pass” as a coping philosophy
- •Fear of death and the anesthesia story: time discontinuity as a glimpse of oblivion
- •Meaning of life as relational: ‘the universe… is composed entirely of others’
