New Lex Fridman Insight: Noam Chomsky: Language, Cognition, and Deep Learning

Key Insights

• Chomsky argues that language is a unique cognitive capacity linked to human nature, not merely a communication tool.
• Deep learning, according to Chomsky, identifies patterns but lacks the capability to understand language structure or contribute to scientific insights.
• Chomsky suggests that human intelligence has genetic limits, similar to other biological organisms, which constrain what can be achieved.

How the conversation moved

The conversation begins with Chomsky framing language as a unique cognitive capacity that is deeply intertwined with human nature. He argues that language is not merely a communication tool but a fundamental aspect of human cognition, allowing for the construction and expression of complex thoughts. This sets the stage for a broader discussion on the nature of human intelligence and its limitations, which Chomsky compares to the genetic constraints found in other biological organisms.

Chomsky presents the main argument that human intelligence, like that of other organisms, has inherent limits determined by genetic endowment. He supports this by referencing Marvin Minsky's experiment, which suggests that any higher intelligence would have at least basic arithmetic, implying a commonality in cognitive capabilities. Chomsky asserts that these limits are intrinsic to the human condition, setting boundaries on what can be achieved through human cognition and intelligence.

Lex doesn't challenge Chomsky's framing of language and cognition, though the obvious counter-position could be the view that language evolved primarily for communication rather than cognitive structuring. The real tension arises in Chomsky's critique of deep learning. He argues that while deep learning can identify patterns from large datasets, it fails to advance scientific understanding of language, as it does not address theoretical questions about language structure.

The conversation concludes with Chomsky exploring how human nature can lead to various types of institutions, shaped by historical contingencies rather than being essential to our nature. This pivot highlights the adaptability and complexity of human social structures, leaving open questions about the future role of AI and deep learning in understanding and replicating human cognitive processes. The episode ends without resolving the tension between the capabilities of deep learning and the theoretical understanding of language.

Surprising moments

In-depth

Language and Cognition

• Language is linked to human cognitive nature, not just a communication tool.
• Structure dependence in language is a universal property across all languages.

Deep Learning

• Deep learning finds patterns in data but doesn't contribute to scientific language understanding.
• Chomsky critiques deep learning for not addressing theoretical language questions.

Limits of Intelligence

• Human intelligence has genetic limits, similar to biological organisms.
• These limits constrain what can be achieved by human cognition.

Still open

• Chomsky questions whether deep learning can ever address theoretical questions about language structure.