Lex Fridman PodcastNoam Chomsky: Language, Cognition, and Deep Learning | Lex Fridman Podcast #53
CHAPTERS
- 0:00 – 2:31
Lex sets the stage: meeting Chomsky, recording mishap, and podcast mission
Lex introduces Noam Chomsky and shares personal context: their first elevator encounter at MIT and the later meeting in Arizona. He explains the unfortunate loss of Chomsky’s video recording and frames the podcast as a side project alongside his AI work.
- •Chomsky’s stature across linguistics, cognitive science, philosophy, and politics
- •Lex’s first meeting with Chomsky at MIT and what it meant personally
- •The Arizona recording accident: audio preserved, video of Chomsky lost
- •Lex’s broader goal of building beneficial AI alongside the podcast
- 2:31 – 4:01
Sponsor message and show logistics (Cash App and FIRST)
Lex delivers the sponsor segment, describing Cash App features and a donation tie-in to FIRST. He reiterates how ads are placed to avoid interrupting the conversation flow.
- •Cash App: payments, bitcoin, and fractional stock investing
- •Promo code and donation mechanism supporting FIRST
- •Lex’s preference for keeping ads out of the middle of conversations
- •Call to subscribe/rate/support the podcast
- 4:01 – 5:46
Can humans communicate with aliens? Arithmetic as a universal bridge
Lex opens with a speculative question about communicating with an alien species. Chomsky discusses Minsky’s Turing-machine thought experiment and suggests arithmetic-like structures might be universal enough to ground a protocol of communication.
- •Minsky/Bobrow experiment: simple Turing machines often crash; survivors yield arithmetic-like behavior
- •Hypothesis: advanced intelligence likely entails arithmetic
- •Core operations of natural language resemble arithmetic in minimal/limiting cases
- •A shared basis in arithmetic could enable cross-species communication
- 5:46 – 7:19
Internal language vs externalized speech: two different concepts
Chomsky distinguishes the internal language faculty (a brain-implemented system generating sound-meaning pairings) from the external signals we produce in communication. He emphasizes that “internal vs external” isn’t a debate but a difference in what we mean by the word language.
- •Internal language as a biological trait that determines you speak English vs Tagalog
- •The system generates infinitely many expressions with sound and meaning
- •Externalization (speech/noise/signals) is a marginal use compared to internal thought
- •External “language” is a different concept: observable output vs cognitive mechanism
- 7:19 – 8:51
Language as a cognitive faculty: like vision, yet central to thought
Chomsky compares language to vision as a biologically grounded, species-specific capacity. He also connects language to long-standing views that it underwrites human creativity and the construction of thought.
- •Language faculty is part of genetic endowment, distinctively human
- •Tradition from early scientific revolution: language as core of human cognitive nature
- •Language supports free, unbounded, creative thought formation
- •Human achievements (good and bad) are tied to these creative capacities
- 8:51 – 10:36
Reasoning, scientific inquiry, and the question of cognitive limits
The discussion turns to whether human cognition has intrinsic limits. Chomsky argues it’s strange to assume humans can answer any question in principle, given that biological capacities typically come with both scope and constraints.
- •Language is central to reasoning but not the only faculty involved
- •A possible “scientific faculty” guides what we find intelligible or worth pursuing
- •Biological traits imply coupled richness (scope) and boundaries (limits)
- •The open question: can we identify the limits of human cognition?
- 10:36 – 12:06
Scope and limits via biology: why endowments both enable and constrain
Chomsky clarifies scope and limits using biological development examples: genes permit certain forms (mammalian vision, arms/legs) while blocking others (insect vision, wings). He extends this logic to cognition: structure enables understanding but also restricts what can be understood.
- •Genetic endowment yields specific developmental pathways and excludes others
- •Rich internal structure is required for any substantial understanding
- •Cognitive capacities, as organic traits, should have analogous constraints
- •Limits are not a defect but a consequence of having structured faculties
- 12:06 – 15:08
Newton, intelligibility, and ‘mysteries that ever will remain’
Chomsky uses the shift from Galileo’s mechanical philosophy to Newton’s action-at-a-distance as a historical clue to cognitive limits. He argues science advanced by accepting theories that work even when the world they describe is unintuitive, reframing what counts as ‘understanding.’
- •Mechanical philosophy: intelligibility required machine-like contact mechanics
- •Newton’s gravity implied interaction without contact, which troubled contemporaries and Newton himself
- •Hume/Locke: Newton revealed limits of mechanical explanation and left enduring mysteries
- •Science pivoted toward predictive/intelligible theories even if underlying reality feels unintelligible
- 15:08 – 16:46
Infant cognition, contact intuitions, and perception’s built-in assumptions
Chomsky and Lex connect historical ‘intelligibility’ to cognitive biases in perception. He describes how humans impose structured interpretations—geometric regularities and object motion—on imperfect sensory input, hinting at built-in constraints on how we conceptualize the world.
- •Infants infer hidden causes between events (preferring contact-like mechanisms)
- •We perceive idealized geometric forms even from distorted drawings (Descartes’ point)
- •Apparent motion and objecthood are imposed beyond raw stimuli (e.g., TV analogy)
- •These tendencies suggest cognition may be tuned to certain explanatory frameworks
- 16:46 – 18:19
Brain–computer interfaces (Neuralink) and whether machines can expand cognition
Lex asks whether brain–computer interfaces could fundamentally expand cognition and reasoning. Chomsky agrees tools can extend capability in a practical sense (like books) but doubts they can transcend native cognitive limits in a deep biological sense.
- •BCI as bandwidth expansion: reading/stimulating neural activity
- •Tools extend cognition externally (books as ancient precedent)
- •Extension differs from changing underlying biological faculties (e.g., mammal vision ≠ insect vision)
- •We can map other creatures’ perceptions into human-understandable formats without becoming them
- 18:19 – 19:27
Quantum mechanics as a case study: theory understood, world still unintuitive
They use quantum mechanics to illustrate the difference between mastering a formal theory and achieving intuitive understanding. Chomsky notes we can learn the mathematics, yet the described reality may remain ‘unintelligible’ in the classical sense—echoing Einstein’s discomfort and Schrödinger’s critiques.
- •Humans can learn quantum theory, but intuition may not follow
- •Schrödinger’s cat highlights the ‘unintelligible world’ problem
- •Einstein’s classical realism reflects a drive for intelligibility
- •Possible example of a boundary between explanation and human-comprehensible pictures
- 19:27 – 22:14
A deep property of language: structure dependence and why it’s surprising
Chomsky highlights structure dependence as a major discovery: language interpretation relies on hierarchical structure, not linear word order. He illustrates with “carefully” attachment ambiguities and argues the universality and neural grounding of structure dependence reveal something profound about the language faculty.
- •Example sentences showing ambiguity and forced interpretations
- •Humans choose structurally closest attachment over linearly closest
- •We ‘hear’ linear strings but compute hidden hierarchical structure
- •Universality across languages and emerging neural accounts/theoretical explanations
- 22:14 – 26:17
Deep learning’s limits: engineering success vs scientific understanding
Lex asks about neural networks and deep learning. Chomsky argues that while deep learning can be useful engineering (finding patterns in massive data), it typically offers little scientific insight into human language because it doesn’t test explanatory hypotheses via critical experiments.
- •Deep learning as pattern extraction from huge numbers of examples
- •Key distinction: building useful tools vs explaining mental phenomena
- •Google Translate/parser may work well yet teach ‘zero’ about human language mechanisms
- •Science prioritizes targeted, theory-driven ‘critical experiments,’ not brute coverage
- 26:17 – 28:01
Behaviorism echoes, corpus linguistics, and when data-mining can still help
Chomsky rejects claims that deep learning vindicates Skinnerian behaviorism, though he concedes pattern-finding can sometimes reveal unnoticed regularities. He compares deep learning to corpus linguistics and paleoanthropology: useful when direct experimental probing is limited, but still weaker than controlled inquiry with living speakers.
- •Some deep learning advocates claim a return to behaviorism; Chomsky disputes this
- •Models should be tested on rule-violating cases to assess scientific adequacy
- •Deep learning akin to corpus linguistics: inferring structure from records alone
- •Analogy to paleoanthropology: serious inference under constraints, but not ideal experimental science
- 28:01 – 35:45
Human nature, institutions, love, mortality, and making meaning
The conversation closes with broader philosophical questions: whether evil stems from institutions, whether humans are good, and how ‘meaning’ is created. Chomsky emphasizes historical contingency in institutions, highlights personal joys (love and children), reflects on mortality concerns from childhood, and concludes that significance is authored by our actions.
- •Institutions express human nature but aren’t fixed or inevitable (historical contingency)
- •Competing views: market systems as ‘natural’ vs an ‘instinct for freedom’ from domination
- •Personal happiness: falling in love and having children; intellectual excitement from discoveries
- •Mortality reflections and the idea that life’s significance is created through what we do
