Why AI Has a Plato Problem: Mazviita Chirimuuta on the Philosophy of Neuroscience

The Plato Problem in AI: Mathematical Idealization vs. Biological Reality

Modern AI research often operates on a "Platonic" assumption: the belief that the universe is governed by neat, decomposable mathematical rules and that the messy physical world is merely a "world of appearance." Professor Mazviita Chirimuuta argues that this perspective—similar to Plato's distinction between the world of forms and the world of becoming—leads researchers to believe that if they can simply decode the underlying mathematical patterns, they can replicate human intelligence.

This approach relies on idealization (attributing properties to a system that are known to be false to make calculations tractable) and abstraction (ignoring specific details). While these are useful tools for scientific progress, Chirimuuta warns that they can become philosophical traps. When scientists treat the "signal" as the only truth and dismiss the "noise" as irrelevant, they are making a subjective decision. In biological systems, what a researcher labels as "noise" may actually be critical to how the system functions in the real world.

The Danger of Oversimplification: Lessons from Reflex Theory

History provides cautionary tales of how elegant simplifications can lead scientific fields astray. Chirimuuta cites the reflex theory of the late 19th and early 20th centuries, which attempted to explain all brain functions as simple sensory-motor loops (reflex arcs).

While prestigious physiologists like Charles Sherrington admitted that the "simple reflex" was an idealization that likely didn't exist in real life, the theory remained dominant because it provided a parsimonious explanation. It was only the arrival of the computational theory of mind during the World War II era that provided a new explanatory framework. This demonstrates that scientific "truth" is often a matter of which idealization toolbox is currently most useful, rather than a linear progression toward a single, final truth.

Haptic Realism: Knowledge Through Interaction

Contrary to the "spectator theory of knowledge"—the idea that we can passively absorb information from a distance (like a God's eye view)—Chirimuuta proposes haptic realism. This view emphasizes that knowledge is acquired through active, physical engagement with the world.

• Interaction as Epistemology: Knowledge is not about reading the "source code of the universe" but is the product of an interactive process between an agent and their environment.
• The Haptic Metaphor: Just as the hand is both a sensory organ and a tool for manipulation, scientific models are both means of acquiring knowledge and means of changing the world.
• Nature as Protean: Chirimuuta describes nature as "Protean" (after the shape-shifting sea god Proteus), meaning it is inexhaustibly complex. While we can "pin it down" to get a specific answer, it continues to shift and offer different patterns depending on how we interact with it. Therefore, a single, final "Theory of Everything" is unlikely in the biological sciences.

Why the Brain is Not Just a Computer

The dominant metaphor of the brain as a computer allows researchers to ignore the "messy" biological details—biochemistry, vasculature, and the immune system—by focusing solely on computational properties. Chirimuuta argues that this is a category error known as ontologization: assuming that because we can model the brain as a computer, the brain is a computer.

The Role of Embodiment and Causal Powers

Drawing on the work of John Searle, Chirimuuta argues that computation is a mathematical formalism and does not possess causal powers. Causal powers belong to concrete physical systems.

• Biological Integration: Human cognition is not a set of discrete modules (like a language faculty) that can be detached and replicated in a Large Language Model (LLM). Instead, language is deeply entangled with sensory-motor engagement and biological existence.
• Distal vs. Proximal Sensitivity: A key differentiator of intelligence is the ability to be responsive to "distal" causes—events far away in time or space (e.g., childhood memories)—rather than being purely driven by "proximal" immediate inputs. Non-living physical systems are typically constrained by what is immediately proximal to them.

Technology and Human Finitude

Referencing Martin Heidegger, Chirimuuta discusses the concept of human finitude—the fact that we are bounded, situated knowers. The drive to create a disembodied, universal absorber of facts (like an LLM) reflects a desire to transcend these biological boundaries.

This aspiration to move into a "spiritual world of pure information" ignores the reality that human knowledge is grounded in discrete sensory experiences. Chirimuuta expresses concern that as we increasingly mediate our lives through digital interfaces, we may be truncating the essential social and sensory interactions—such as children looking at human faces—that are necessary for normal psychological development. This "massive experiment" on the next generation may fundamentally alter how humans relate to one another and the world.