Video · 2026-01-23 · 54m · 6 moments

Why AI Has a Plato Problem — Mazviita Chirimuuta

✦ AI generated

GUEST Mazviita Chirimuuta HOST Tim Scarfe

timeline · colored by role

Claim◆

Scientific abstraction and mathematical idealization exist not because they reveal a deeper Platonic layer of reality, but because finite human knowers need to simplify an inexhaustibly complex world in order to gain any traction on it.

Chirimuuta contrasts the Platonic view that idealized mathematical models get us closer to an underlying 'true' reality with her own 'down-to-earth' view: abstraction is a coping strategy for cognitively limited beings, not access to a higher truth.

transcript

Mazviita Chirimuuta: instead of thinking that the abstraction gets you like the higher level of reality, just saying that we do abstraction because we're finite knowowers. There's limits to how much complexity any individual person or group of people can actually encompass... it's only by pretending things that are more simple than they actually are that we get some traction.

explains mechanism · 1gives example · 1rebuts · 1

Example

Late 19th/early 20th century reflex theory shows how a scientist can knowingly adopt an idealization (the 'simple reflex') as a useful fiction, only to overextend it into a totalizing explanatory framework that outran the evidence it could actually explain.

Using reflex theory as a case study, Chirimuuta shows physiologist Charles Sherrington admitted the 'simple reflex' was an idealization that likely doesn't exist in nature, yet the field ran with it far past its explanatory limits before computational theory replaced it.

transcript

Mazviita Chirimuuta: a very prestigious and sort of well-regarded physiologist like Charles Sherrington was heavily invested in the reflex theory. But he admitted in his book the integrated action of the nervous system that this notion of a simplex simple reflex is an idealization. It probably doesn't exist in real life. And yet this is the key that's going to kind of unlock neurophysiology.

Definition

Scientific knowledge is fundamentally 'haptic' rather than passively observational — it is acquired through active, touch-like interaction, manipulation, and engagement with the world, not through detached visual absorption of facts.

Chirimuuta introduces her concept of 'haptic realism,' arguing against the 'spectator theory of knowledge' by claiming that acquiring knowledge, especially scientific knowledge, necessarily involves hands-on engagement, manipulation, and interaction rather than passive absorption.

transcript

Mazviita Chirimuuta: if you think that scientific knowledge in particular is more kind of touchlike, you can't ignore the fact that we sort of run into things. We have to pick things up, engage with them, ultimately change them in order for us to acquire knowledge of them. So you cannot discount the fact that we're kind of meddling with things in the process of bringing about our knowledge.

supports · 1

Claim◆

Computation is a mathematical formalism, not a concrete physical system, so it cannot itself possess causal powers — only the physical systems that implement a computation can be causes.

Chirimuuta argues computation and causation belong to different ontological categories: only concrete physical systems have causal powers, so computation per se cannot explain cognition without further justification.

transcript

Mazviita Chirimuuta: computation itself is mathematical formalism, it's like exists, it's uh it is a mathematical structure. Things that have causal powers are concrete physical systems. So I just think they're different kinds of things.

rebuts · 1

Claim◆

An LLM cannot have genuine understanding like humans do because it is a language faculty artificially detached from the rest of embodied cognition, lacking sensorimotor engagement with the world.

Chirimuuta finds it implausible that an LLM could achieve human-like understanding, since human language is inseparably bound up with sensorimotor engagement and embodied experience that LLMs lack.

transcript

Mazviita Chirimuuta: the idea that you could just sort of detach off a language faculty, have it replicated in an LLM that doesn't have the other bits of our cognition and it doesn't have embodiment, doesn't have the capacity to engage with the world, and that it could have understanding in the same way that we do. I find that implausible.

explains mechanism · 1rebuts · 1

Claim

The idea of an LLM as a disembodied, non-situated absorber of all information reveals a failure to acknowledge human finitude — the fact that human knowledge is inherently bounded, culturally situated, and grounded in a limited set of inaccessible sensory experiences.

Drawing on Heidegger's critique of the philosophical tradition's aspiration to transcend embodiment, Chirimuuta argues that treating an LLM as a knower analogous to humans ignores the essential finitude and situatedness of human knowledge.

transcript

Mazviita Chirimuuta: the very idea that a non situated sort of non-embodied absorber of facts like an LLM that kind of just sits there and like sucks in all the information in the world, that that could somehow be a counterpart to how we know things as human beings. I think that's an instantiation of this like lack of acknowledgement of human finitude.

Highlight slides

Abstraction as Coping, Not Transcendence✦ from: Scientific abstraction and mathematical idealization exist not because they reveal a deeper Platonic layer of reality, but because finite human knowers need to simplify an inexhaustibly complex world in order to gain any traction on it.The Down-to-Earth View of Idealization✦ from: Scientific abstraction and mathematical idealization exist not because they reveal a deeper Platonic layer of reality, but because finite human knowers need to simplify an inexhaustibly complex world in order to gain any traction on it.Computation ≠ Causation✦ from: Computation is a mathematical formalism, not a concrete physical system, so it cannot itself possess causal powers — only the physical systems that implement a computation can be causes.Implication for Cognitive Science✦ from: Computation is a mathematical formalism, not a concrete physical system, so it cannot itself possess causal powers — only the physical systems that implement a computation can be causes.LLMs Lack Genuine Understanding✦ from: An LLM cannot have genuine understanding like humans do because it is a language faculty artificially detached from the rest of embodied cognition, lacking sensorimotor engagement with the world.Embodied Cognition Is the Missing Piece✦ from: An LLM cannot have genuine understanding like humans do because it is a language faculty artificially detached from the rest of embodied cognition, lacking sensorimotor engagement with the world.

Related episodes

Article · Simon Willison's Weblog

Scientific abstraction and mathematical idealization exist not because they reveal a deeper Platonic layer of reality, but because finite human knowers need to simplify an inexhaustibly complex world in order to gain any traction on it.

Late 19th/early 20th century reflex theory shows how a scientist can knowingly adopt an idealization (the 'simple reflex') as a useful fiction, only to overextend it into a totalizing explanatory framework that outran the evidence it could actually explain.

Scientific knowledge is fundamentally 'haptic' rather than passively observational — it is acquired through active, touch-like interaction, manipulation, and engagement with the world, not through detached visual absorption of facts.

Computation is a mathematical formalism, not a concrete physical system, so it cannot itself possess causal powers — only the physical systems that implement a computation can be causes.

An LLM cannot have genuine understanding like humans do because it is a language faculty artificially detached from the rest of embodied cognition, lacking sensorimotor engagement with the world.

The idea of an LLM as a disembodied, non-situated absorber of all information reveals a failure to acknowledge human finitude — the fact that human knowledge is inherently bounded, culturally situated, and grounded in a limited set of inaccessible sensory experiences.

llm 0.32rc2

I Built an LLM From Scratch

Ben Horowitz on the Global Race for Tech, Power, and Influence