363 | Chandra Sripada on How LLMs and Humans are Cognitive Cousins
Summary
Sean Carroll interviews philosopher and cognitive scientist Chandra Sripada to explore whether Large Language Models think like humans or represent an alien intelligence. Sripada argues that LLMs exhibit core human cognitive principles, such as dual-process theory, through mechanisms like in-context learning and chain-of-thought reasoning. He presents evidence from psycholinguistics and memory studies showing that LLMs replicate subtle human cognitive biases and processing strategies, suggesting they have rediscovered human-like thinking methodologies rather than inventing a wholly new form…
Topics discussed
Intro and sponsors: Wayfair, Azure Standard
Introduction: The Turing Test and AI consciousness debate
How LLMs work: Transformers, weights, and next-word prediction
Sponsors: Babbel and Incogni
LLM behavior vs. Human cognition: Psycholinguistics and memory effects
Sponsors: Vital Proteins and Fig
Dual Process Theory: System 1 vs. System 2 in LLMs and humans
Sponsor: Fig scrubs testimonial
Production Systems and ResNets: Computational parallels
Predictive Processing and Contravariance in cognition
Consciousness, sentience, and moral patienthood of AI
AI in education: Tutoring, stratification, and societal impact
Conclusion: LLMs as cognitive cousins and future of cognitive science
Listen ad-free on Castria