Noam Chomsky on Language, Cognition, Deep Learning, and the Limits of Human Understanding

The conversation with Noam Chomsky delves into the fundamental nature of language, distinguishing between its internal mental system and its external expression. Chomsky posits that language is a core component of human cognitive nature, deeply linked to our capacity for thought and reasoning, and unique among organisms. He explores the possibility of communication with alien species, suggesting a common ground in arithmetic and the core principles of human language, drawing on Marvin Minsky's ideas about the simplest Turing machines yielding arithmetic.

A crucial distinction is made between language as an internal, genetically endowed faculty (determining, for example, speaking English vs. Tagalog) and language as an externalized system of communication. Chomsky also differentiates between engineering and science in the context of AI and deep learning, arguing that while deep learning tools like Google Translator are useful engineering feats, they offer little scientific insight into human language because they focus on pattern recognition in vast datasets rather than critical experiments designed to test theoretical questions about the underlying structure. He highlights the concept of "structure dependence" as a deep, puzzling property of language that deep learning models often fail to capture scientifically.

While not offering direct practical recommendations in the typical sense, Chomsky's insights challenge the common belief in unbounded human intelligence, suggesting that our biological nature imposes inherent "scope and limits" on cognition, much like our genetic endowment dictates physical development. He illustrates this with the historical shift in science from the mechanical philosophy of Galileo and Newton to the acceptance of unintelligible phenomena (like action at a distance), implying a fundamental boundary to human understanding. He also touches on the idea that external tools like books or brain-computer interfaces can expand cognitive capacity in a superficial way, but not fundamentally alter our inherent cognitive limits.

The discussion extends to broader philosophical questions, including the origins of evil in society (attributing it to institutions rather than inherent human nature, though acknowledging human nature's role in creating those institutions) and the meaning of existence, which Chomsky believes is self-created through our actions. His perspective on the limits of science and human understanding has profound implications for fields like AI, suggesting that simply scaling data and computation may not lead to genuine scientific understanding of complex phenomena like language or consciousness, but rather to sophisticated engineering approximations. The conversation underscores the importance of being "willing to be puzzled" by seemingly obvious phenomena, a hallmark of scientific inquiry.

It's a simple fact that there's something about you, a trait of yours, part of the organism you that determines that you're talking English and not Tagalog, let's say. So there is an inner system.

No other organism has anything remotely similar [to the human language faculty].

Language is the sort of the core of human cognitive nature. It's the source, it's the mode for constructing thoughts and expressing them and that is what forms thought and it's got fundamental creative capacities.

If we're biological organisms which are not angels then our capacities ought to have scope and limits which are interrelated.

The very endowment that confers richness and complexity also sets bounds on what can be attained.

Newton to his astonishment and distress proved that there are no machines, that there's interaction without contact.

Science just changed its goals. It abandoned the mysteries. It can't solve it, they'll put it aside. We only look for intelligible theories. Newton's theories were intelligible it's just what they described wasn't.

What deep learning is doing is taking huge numbers of examples and finding some patterns. Okay, that could be interesting and in some areas it is but we have to ask here a certain question. Is it engineering or is it science?

Does it tell you anything about human language? Zero, nothing, and in fact it's very striking. From the very beginning it's just totally remote from science.

The significance of your life is something you create.

It just seemed obvious that that can't be true. Like I said, there's something about you that determines you're talking English not Swahili or something. But that's not really a discovery. That's just an observation of what's transparent.