Noam Chomsky: Language, Cognition, and Deep Learning

Episode Summary

Executive Summary: Noam Chomsky argues that language is an internal biological faculty shaping thought, and that human cognition has real limits set by our nature. He critiques deep learning as useful engineering but weak science, discusses how structure dependence reveals hidden principles of language, and suggests meaning in life comes from our own actions rather than fixed answers.

Main Topics: Language as an internal faculty: Chomsky distinguishes internal language in the mind/brain from externalized speech, arguing that language is a biological system that generates thought and only secondarily becomes spoken language. Language, thought, and reasoning: He claims language is the mechanism by which we reason and structure thought, though other faculties also contribute to scientific and conceptual understanding. Limits of human cognition: Chomsky argues that, like vision, cognition likely has built-in bounds. He rejects the idea that humans can understand anything in principle and points to historical and perceptual examples of limits. Deep learning and machine learning: He says neural networks can be useful for engineering, but they do not by themselves explain language or science. Their success on corpora is not evidence that they capture linguistic theory. Structure dependence in language: He highlights structure dependence as a deep linguistic property: humans interpret sentences by grammatical structure, not simple linear proximity, revealing non-obvious cognitive operations. Human nature, institutions, and freedom: Chomsky argues that institutions like markets are historically contingent, not fixed by nature, while human nature may support both domination and freedom depending on circumstances. Mortality and meaning: He reflects on mortality, childhood existential fear, and concludes that life’s meaning is not given externally but created through our own activities.

Key Arguments: Language is not just external speech; it is an internal mental system that determines what language a person speaks and enables thought. The human language faculty is a unique biological endowment, analogous to vision, with both richness and limits. Reasoning uses language, but also involves other faculties such as an undefined scientific capacity. Human cognition likely has limits; the pursuit of a fully intelligible mechanistic world may be impossible, as shown historically by Newtonian and quantum physics. Deep learning is valuable as engineering, but it does not constitute a scientific explanation of language because it succeeds on corpora without testing theoretical structure. A corpus-based model can perform well while failing on structurally important cases, so high accuracy is not proof of understanding. Structure dependence is a key insight: speakers rely on syntactic hierarchy rather than linear word order, even when that seems less obvious. Human institutions are not simply “natural”; they are products of history and contingency, though they arise from human nature in some broad sense. Meaning in life is self-authored: significance comes from what we do, not from an external universal answer.

Data Points: MIT tenure: over 60 years - Introduced as Chomsky’s long academic career at MIT University of Arizona meeting: recently joined - The conversation took place after Chomsky joined the University of Arizona First meeting with Lex Fridman: about four and a half years ago - Fridman recalls first seeing Chomsky in an elevator at MIT FIRST nonprofit reach: hundreds of thousands of students - Mentioned in the Cash App sponsorship read about STEM education FIRST global presence: over 110 countries - Describing the nonprofit supported by Cash App Charity Navigator rating: perfect rating - Used to emphasize efficient use of donated money Cash App bonus: $10 - Users get $10 for using code LexPodcast, and Cash App donates $10 to FIRST Language corpora example: Wall Street Journal corpus - Used to explain how parser models are evaluated on large text datasets Historical period: 1950s - Referenced as the era when modern linguistics began challenging externalist views of language Scientific revolution reference: 17th century - Used as the period when modern science developed mechanical philosophy

Pivotal Quotes: "There is something about you that determines that you're talking English and not Tagalog, let's say." — Noam Chomsky: Explaining the internal language faculty versus external speech "What deep learning is doing is taking huge numbers of examples and finding some patterns." — Noam Chomsky: Describing why neural networks are useful but limited scientifically "The significance of your life is something you create." — Noam Chomsky: Answering the question about meaning and mortality

Implications: Listeners should expect AI to improve tools, not replace scientific understanding of mind and language. Chomsky’s view encourages skepticism toward surface-level model performance and emphasizes human limits, structure, and self-created meaning.