The Mirror
Bio Mirror
Human biology read as an engineering specification. Each entry names a structure, states what that structure makes possible, and holds it against its closest analogue in AI and in robotics — including the place the analogy breaks, which is usually the part worth arguing about.
Structure dictates function.
5 levels of organization · 16 structures
Level 1
Cellular
The Cellular Machinery
The nervous system is built from cells whose physical geometry — dendritic branching, axon diameter, myelin thickness, the molecular architecture of a synaptic cleft — is the computation, not a container for it. This is the level at which the artificial neuron abstraction is thinnest, and therefore the level where the differences from artificial networks are most informative.
A neuron computes with its shape. Change the branching and you change the function.
Level 2
Circuit
Circuits and Plasticity
Above the single cell, function comes from how cells are wired together: excitation balanced against inhibition, recurrent loops that sustain activity, and neuromodulatory systems that reconfigure the whole assembly according to context. This is the level at which reinforcement learning was discovered twice — once in psychology and once in the midbrain.
Connectivity is the program. Learning edits the wiring, and the wiring constrains what can be learned.
Level 3
Regional
Cortical Architecture
The neocortex is strikingly uniform: six layers, the same cell types, the same local connectivity motif, across visual, auditory, motor, and prefrontal areas. Regions differ mainly in their inputs and outputs. That uniformity is the strongest structural argument that a single learning principle might underlie very different capabilities — and it is why a stacked, homogeneous architecture like a transformer is a less absurd model of cortex than it first appears.
The cortex repeats one circuit motif everywhere. What differs is what it is wired to.
Level 4
Systemic
Memory, Motor Control, and Embodiment
This is where the biology speaks most directly to robotics. The cerebellum builds internal models of the body and the world; proprioception supplies the state estimate those models need; the hippocampus binds experience into episodes that can be replayed. Together they solve, in wet tissue, the problems the embodied-control track solves with a Jacobian.
The brain does not compute a plan and then execute it. It predicts, compares, and corrects, continuously.
Level 5
Behavioral
Failure, Narrative, and the Limits of Analogy
This is the section that keeps the rest of Bio Mirror honest. Human neurology and psychiatry offer a century of carefully described failure modes, and several of them resemble what language models do closely enough to be worth naming. The resemblance is a source of hypotheses and vocabulary. It is not evidence, and the entries here say so explicitly.
Two systems can fail in the same shape for different reasons. The shape is a hypothesis; the reason has to be earned.
A standing caution
Every mirror here is a source of hypotheses, not a source of evidence. A resemblance between a brain and a model tells you what to test next. It never tells you what is true of the model — only an intervention on the model can do that.
Open Kiru Lab