Srini Narayanan

In the AI Story

Narayanan's 1997 dissertation, KARMA (Knowledge-Based Action Representations for Metaphor and Aspect), developed computational models showing how image schemas and the neural metaphorical mappings Lakoff theorized could be implemented in structured connectionist networks capable of actually performing the cognitive operations the theory described. The work mattered because it demonstrated that the embodied-cognition framework was not merely a philosophical claim about human minds but a specifiable computational hypothesis that could be tested by building systems and observing whether they exhibited the predicted behaviors. The answer, across decades of subsequent work, was qualified: structured connectionist networks implementing image-schematic operations exhibited cognitive capacities that pure symbolic systems lacked, supporting the framework's central claim that embodied grounding matters for the kind of cognition that emerges.

At Google and subsequently DeepMind, Narayanan has worked at the intersection of cognitive science and large-scale AI systems. His position inside the industry gives the argument of The Neural Mind a particular rhetorical force: the claim that deep-learning AI is categorically different from embodied mind is made by a researcher who helps build deep-learning AI and understands its architecture in technical detail. The argument is not an outside critique from a philosopher who does not understand what the systems do; it is an inside analysis from a builder who understands what the systems do and argues that what they do is not what embodied minds do.

The collaboration with Lakoff on The Neural Mind represents the mature joint statement of two decades of convergent work. The book's afterword — "The Neural Mind versus Deep Learning AI" — does not soften the opposition the title presents. It argues that the image schemas structuring human thought are implemented in neural circuits originally evolved for sensorimotor control and repurposed for abstract cognition, and that systems lacking these circuits cannot have the kind of cognition that embodied minds have. The argument is technical in its specifics and philosophical in its implications, and its force depends substantially on Narayanan's credibility as someone who knows the technical specifics from the inside.

Origin

Narayanan completed his Ph.D. in computer science at UC Berkeley in 1997 under Lakoff's supervision in linguistics and Jerome Feldman's in computer science. He held research positions at the International Computer Science Institute and SRI International before joining Google in the 2010s. He has published extensively on computational cognitive science, language understanding, and the neural basis of metaphor.

Key Ideas

KARMA dissertation. Narayanan's 1997 work provided computational models implementing image-schematic reasoning in connectionist networks.

Industry position. His senior role at DeepMind places the embodied-cognition argument inside the AI industry rather than outside it.

Neural theory of language. Narayanan extended Lakoff's framework with specific computational and neural specifications.

The Neural Mind. The 2025 co-authorship with Lakoff represents the mature joint statement applying the framework directly to deep-learning AI.

Rhetorical weight. The argument's force depends substantially on Narayanan's credibility as a senior practitioner who knows the systems from the inside.