Summary

The investor-relevant warning is that a useful scientific or AI model does not necessarily reveal reality’s underlying architecture. Chirimuuta accepts abstraction as unavoidable for “finite knowers,” but rejects the Platonic leap from tractability to truth. A successful computational neuroscience program may justify temporary “tunnel vision”; it does not establish that “the brain is a computer.”
AI’s search for clean rules and geometric inductive priors may erase variations that matter outside the immediate inquiry. Responding to François’s “kaleidoscope hypothesis,” Chirimuuta argues that dividing data into signal and noise is a scientist’s decision: discarded irregularity could govern how a natural system behaves elsewhere, while denoising may partly create the pattern it appears to discover.
Reflex theory is a cautionary precedent for simple frameworks that can be pursued too far despite failing to explain the full data. Around the late 19th and early 20th centuries, researchers pursued the “simple reflex” even while Charles Sherrington admitted it probably did not exist in reality. Chirimuuta says the program might have continued longer until computational theory supplied another neat framework during the Second World War era—not necessarily because simplification itself had been cured.
Embodiment remains an unresolved constraint on claims that artificial networks can reproduce biological cognition. Tim proposes a lottery-ticket analogy: after training, a dense network can lose 90% of its connections, so perhaps a billion-plus years of evolution left embodiment, autopoiesis and agency as removable residue. Chirimuuta doubts biology could afford such waste, pointing to the brain’s energy economy and the possible continuity between neuronal signaling, metabolism and the rest of the living body.
Behavioral equivalence is insufficient evidence for consciousness or human-like understanding. Inputs and outputs can support practical classification, but Chirimuuta calls it a “philosophical leap” to treat mechanism and subjectivity as irrelevant. An embodied robot might move closer to understanding, yet genuine meaning may be connected to being alive—“always in a situation that is problematic to you”—rather than merely manipulating language.
The same fantasy of disembodiment shapes both AI narratives and digital infrastructure economics. The cloud appears weightless while depending on computers, scarce resources and energy; social media feels like another world while operating through physical devices and human imaginative investment. Chirimuuta’s sharpest social-risk call concerns children spending less time with faces: “We’re running a big experiment on the next generation,” and “we don’t know” whether they will be socialized for happy lives.

Deep dive

1. Abstraction earns traction, not privileged access to reality

The Brain Abstracted grew from Chirimuuta’s neuroscience training and computational-modeling work, philosophy papers beginning around 2014, and writing formally started around 2018 before publication in 2024. Her motivating question was why brain simulations are alleged not merely to model neurons, but to “duplicate the function” of cells in an organ treated as a computer itself.
Abstraction omits known details, as Newtonian exercises omit friction; idealization assigns knowingly false properties, as genetics assumes infinite populations. Idealizations often make calculations tractable, while both approaches simplify representations; idealization especially presents something “cleaner and better than the thing in real life.”
Tim’s AI specimen is François’s “kaleidoscope hypothesis”: beneath messy appearances lies mathematical code whose rules can be decomposed. Chirimuuta’s diagnosis is blunt—“That’s Plato,” reproducing the ancient split between stable forms and the “messy, flowing, complicated reality” of becoming.
Even the moderate claim that denoising reveals “real patterns” embeds judgment. What one scientist calls irrelevant irregularity could matter to another inquiry or to the system’s natural operation; the cleaning procedure may also be “creating pattern through the very denoising process.”

2. Reflex theory shows how a useful fiction can become a scientific cul-de-sac

The late-19th- and early-20th-century reflex program sought to explain every brain function through conditioned sensory-motor arcs. Charles Sherrington acknowledged that the “simple reflex” probably did not exist in real life, yet researchers treated it as the key for decomposing nervous-system complexity.
Chirimuuta’s historical reading: Occam’s razor began as a reasonable heuristic, then scientists “ran with it way too far,” never explaining as much data as promised. The framework might have persisted longer had computational theory not arrived around the Second World War with its own neat “idealization toolbox.”
The same black-box move remains relevant to AI. Treating minds as black boxes and tracking inputs and outputs can be legitimate in some contexts, but inferring consciousness from behavioral similarity alone ignores mechanism and first-person subjectivity; “it’s much too quick to just go behaviorist.”

3. Knowledge is made through contact with a shape-shifting nature

Chirimuuta’s constructivism does not make theories unconstrained social inventions. Scientists enter the world with agendas, questions and technological goals; nature pushes back experimentally, so knowledge is always the product of interaction between human framing and phenomena.
Her “haptic realism” rejects John Dewey’s “spectator theory of knowledge,” the conceit that observers passively absorb a God’s-eye view. Hands capture science’s double face: they sense by touching, but also manipulate and change what is being investigated.
Contingency creates paths not taken. Hasok Chang’s pluralism recommends exploring many paths to maximize knowledge; Chirimuuta is more guarded because narrowing inquiry can make sense and resources are not unlimited, though every chosen path carries opportunity costs.
Her governing metaphor is Proteus, the shape-shifter who tells the truth only when pinned down. Nature can yield true answers under a particular experimental grip, but upon release “it will carry on shape-shifting”; biology’s inexhaustible particularity therefore makes one convergent final theory less plausible than it may appear from physics.

4. The brain-computer metaphor feeds itself by engineering its own lens

From Descartes’s mechanistic body through reflex theory and cybernetics, researchers repeatedly built devices inspired by biology, then used those devices to reinterpret biology. McCulloch and Pitts’s 1943 treatment of neurons as logic gates exemplifies the loop that helped originate neural networks.
The comparison licenses computational neuroscience to bracket biochemistry, vasculature, immune interaction and other properties absent from non-living machines. Chirimuuta accepts that scientists need such “tunnel vision”; her objection is its “ontologization”—success in one inquiry does not prove that the brain literally is a computer.
Tim’s pushback is functional: connectionists need not claim identical mechanisms, only sufficient equivalence, perhaps through more biologically plausible autonomy, diversity and agency. Chirimuuta replies that neuronal signaling may be an outgrowth of metabolic signaling throughout the body, making equivalent functionality in a non-living machine “more of a stretch.”

5. Cognition may require distal agency, while computation supplies no causal power

Tim’s lottery-ticket analogy sharpens the disagreement: networks need density for stochastic-gradient-descent training, then can be pruned by 90% without losing performance; perhaps evolution’s billion-plus-year process likewise left embodiment and autopoiesis vestigial. Chirimuuta answers that biological cognition’s tight energy budget suggests evolution already performed substantial pruning.
Their positive account of agency centers on sensitivity to what is distal. A childhood event or imagined future can be as relevant to present behavior as what is happening in the room, whereas non-living systems are more tightly governed by proximal causes—“the distal is always screened off by the proximal.”
Against Daniel Dennett’s physical, design and intentional stances, Chirimuuta withholds ontological priority from the physical stance. If representation and intentionality do explanatory work in science, she asks why their reality must first be validated through a non-intentional, lower-level causal story.
Putnam’s rock problem exposes indiscriminate computational mappings: if mapping physical dynamics onto formalism suffices, rocks, sofas and stomachs compute too. Chirimuuta’s categorical distinction is that “computation itself is mathematical formalism”; concrete systems have causal powers. She takes Searle’s challenge seriously: cognitive science still owes an account of how non-causal computation explains physical cognition.

6. Understanding and digital life expose the costs of denying finitude

Chirimuuta finds human-like LLM understanding implausible because language, perception and sensory-motor engagement are mutually shaping, not detachable modules. A language faculty copied without embodiment or worldly action lacks the integrated cognition through which humans understand.
Tim’s robot reply moves closer: add sensory-motor affordances and physical engagement. Chirimuuta says that this is more along the lines of the kind of thing that could have understanding, and that robots placed in increasingly precarious, lifelike situations might develop something closer; she explicitly says it is “not beyond the realms of possibility.”
Heidegger’s useful provocation is that cybernetics—and now AI—culminates a philosophical aspiration to transcend material finitude. The “disembodied absorber of facts” appears in the LLM image; Tim extends a related disembodiment fantasy to the weightless “cloud,” concealing computers, resource shortages and energy consumption behind an immaterial consumer image.
Tim argues that virtual experiences can still be real; Chirimuuta locates the issue in attention and opportunity cost. Her concrete concern is developmental: young children now see fewer faces despite being predisposed toward gaze and social interaction. Recalling 1950s experiments depriving monkeys of maternal contact, she asks, “Should we be doing that experiment on children?”

Will ChatGPT Ads Change OpenAI? + Amanda Askell Explains Claude's New Constitution Why Every Brain Metaphor in History Has Been Wrong [SPECIAL EDITION]