Whole-brain emulation · Lobeworks/17

Whole-brain emulation is the reproduction of a particular brain in a computer at the level of its causal parts, so faithfully that the copy responds as the original would to any input; it is a goal, and no nervous system has been emulated yet.

Whole-brain emulation. Whole-brain emulation is the reproduction of a particular brain in a computer at the level of its causal parts, so faithfully that the copy responds as the original would to any input; it is a goal, and no nervous system has been emulated yet.

The word sits at the top of a ladder whose rungs are often confused. A model is a description that leaves things out on purpose. A connectome-constrained model takes its architecture from a measured connectome. A simulation runs a model forward in time. A validated simulation has had its predictions checked against measurements, and for the fly this has happened circuit by circuit (feeding, grooming, motion vision), never for a whole brain. An emulation reproduces the system for any input, and an upload is an emulation of one individual, with what that individual had learned.

The fly has the first four rungs. Its whole brain runs as a leaky integrate-and-fire model on its real wiring, with uniform neurons, no neuromodulation, no plasticity and no gap junctions.

What is missing is mostly state and parameters. A fixed connectome records structure; the strengths, time constants and chemical state that an emulation needs have to be measured in living animals or inferred.

The virtual flies of 2026 were integrations. In March a company joined the fly brain model to a simulated body whose legs were run by their own controllers and called it an upload; its critics called it the first closed loop of connectome-constrained brain and body models, which is what it was.

Questions: Is a simulated connectome an emulation? It is not. A simulated connectome is a connectome-constrained model run forward in time; an emulation would respond as the original brain does to any input, which needs the parameters and state a fixed connectome does not record. The fly has validated simulations of particular circuits, such as feeding and motion vision, and nothing yet that behaves like a fly in general. What did the virtual fly of March 2026 actually run? The FlyWire brain as a leaky integrate-and-fire model with uniform neurons, whose descending neurons were read out and mapped onto commands for walking, turning, grooming and feeding, executed by the controllers of NeuroMechFly, a fly body in the MuJoCo physics engine. The brain chose the behaviour; the body model produced the movement, because the brain map stops at the neck and the nerve cord that drives the legs was not in it. Why did the virtual fly of 2026 borrow its legs from a body model? Because the brain map it ran stops at the neck. FlyWire cuts 1,303 descending neurons there, and the nerve cord that would turn their commands into steps is not in it, so the simulation read the descending neurons and handed their choice to controllers written or trained for a body model. The male connectome of 2026 includes the nerve cord, which makes a model that walks by its own wiring possible to attempt.