A Fly Walks Out of Math

In March 2026, a small team led by Philip Shiu at Eon Systems let the entire connectome of a fruit fly run inside a simulated body. The body walked. The body groomed. The body foraged. The body did exactly what flies do — except for one strange, telling thing.

What Actually Happened

Eon Systems is a small startup made of researchers with prior tours at Janelia, DeepMind, and Google. The senior scientist on the work, Philip Shiu, is also the lead author on a Nature paper from late 2024 that published a computational model of the entire adult Drosophila melanogaster brain — 125,000 neurons and roughly 50 million synaptic connections, built on top of the FlyWire connectome that Princeton, Janelia, and a small army of citizen-scientists had spent years assembling.

A connectome is not a brain. A connectome is a map of a brain — every neuron, every connection, what kind of synapse it is, which neurotransmitter it uses. By itself, a connectome does nothing. It is anatomy in a database.

What Shiu’s group did, in March 2026, is something different in kind. They wired the connectome up to NeuroMechFly v2, a biomechanically faithful simulation of a fly’s body, and ran the whole thing inside MuJoCo, a physics engine. Sensory input flowed in through simulated eyes and antennae. The connectome integrated the inputs the way a real fly’s brain does. Motor commands flowed out. The simulated legs moved.

And the fly walked. Not in metaphor — in physics simulation, with the actual geometry and dynamics of insect locomotion. The team reports the system reproduces natural fly behavior — walking, grooming, foraging — with about 91% accuracy against the biological reference. Even Elon Musk publicly registered amazement, which is a kind of certification of a particular sort.

This is, by any reasonable accounting, the first whole-brain emulation in history that does the job of being its animal. Earlier landmarks — OpenWorm‘s C. elegans, the Janelia hemibrain — were extraordinary first steps, but they did not yet drive a body through a world. The Eon fly does.

The One Thing the Fly Cannot Do

And now the thing the headlines mostly missed.

The uploaded fly walks. It grooms. It forages. But — as the team is careful to note — it cannot form new memories. The connectome is the wiring captured at one moment. The fly can act, in the sense of running the dynamics that wiring permits. The fly cannot learn, in the sense of permanently updating its wiring based on what just happened.

This sounds like a technical limitation. It is in fact the most philosophically revealing fact in the whole project.

What does it mean that we can build a fly that walks, but not a fly that remembers walking?

It means we have, accidentally, separated two things that were always braided together in biology. Behavior — the moment-to-moment response of a nervous system to its environment — turns out to be tractable from a static map. Self — the carrying-forward of yesterday into today — turns out to require something the static map does not contain: plasticity over time, the live updating of the wiring by the wiring’s own experience.

This is exactly the cleavage the rest of this series is going to walk through. The Eon fly is a body without a biography. It can do today, but it cannot have had a yesterday. Which means it is, in a strict sense, a different fly every microsecond — a perfect actor with no inner continuity.

Why This Matters for Everyone Who Is Not a Fly

The fly is small. The connectome is small. Mapping a human brain — ~86 billion neurons, ~100 trillion synapses — is a project on a wholly different scale, and the people who tell you it is fifteen years away have been telling you it is fifteen years away for fifteen years.

But the principle has now been demonstrated. An animal connectome, when run inside a faithful body and a faithful physics, behaves like the animal it was harvested from. That is not a theory anymore. That is a result.

The remaining engineering challenges — scale, plasticity, the persistence of memory across re-instantiation — are exactly the challenges that “Where Is Memory Stored — Or Why the Question Is Already Wrong” and the next essay in this series, “The Ego Is a Memory Engine,” are about. The fly’s missing memory is not a footnote. It is the headline.

What Eon Did Not Do (the honest list)

Because precision matters more than hype, here is the careful inventory:

• They did not upload a conscious mind. The fly’s connectome lacks the introspective architecture humans have. Even if it had it, we would not yet know how to verify the consciousness of a digital fly.
• They did not create a fly that learns. Synaptic plasticity — the actual mechanism by which biological brains store experience — is not yet running in the emulation. The team has flagged this as the next frontier.
• They did not “copy a mind to silicon” in the popular sense. They re-instantiated the dynamics of a specific anatomy, captured at one instant, inside a different substrate. It is closer to recording an orchestra and replaying it than to teaching a new orchestra the piece.
• They did not solve consciousness. They did, however, narrow the question. After this, no honest person can argue that “behavior” is the hard part. Behavior, it turns out, falls out of the map. The hard part is the part that is left over when behavior is removed: memory, learning, continuity of self.

That residue — the part that does not yet emerge from the connectome alone — is the seat of the self. The next essay names it. The whole rest of this series chases it.