Fly Experiment

Protocol

A real connectome is deciding when this fly eats.

It smells. Slop in the city emits a concentration field. The fly's antennae sample it.
Its neurons fire. That signal is injected into real gustatory receptor neurons in an 873-neuron cut of the FlyWire fruit fly connectome, stepping live in your browser.
MN9 decides. The proboscis extends when — and only when — motor neuron MN9 crosses 60 Hz. Nothing in the code checks how close the food is.

The control: bitter slop (🥦 broccoli, 📚 homework) drives a different set of receptors. The connectome refuses to extend for them. Watch the fly land on the broccoli, fail, and give up.

Motor neuron MN9

quiet

0Hz

0proboscis extends above 60 Hz400

proboscis0%

behaviour—

target—

Stimulus at the antennae

sugar 0.00

bitter control 0.00

Spike raster

live

sugar GRN0

proboscis MN0

descending L0

descending R0

Trial log

0 fed · 0 refused

waiting for the first encounter…

Be suspicious of this

real The —-neuron network is a genuine cut of FlyWire FAFB v783 — real root IDs, — connections, — synapses, real neurotransmitter signs. Leaky integrate-and-fire, Shiu et al. 2024 parameters.

subgraph Not the whole brain. That is — neurons and runs 4.4 ms of wall clock per millisecond simulated — impossible in a browser. Cut free of the rest of the brain, this fires hotter than the full network.

scripted Flying, landing, walking and target choice are hand-written. No connectome produces locomotion here. Only the decision to extend is neural.

Full provenance ledger and the failures →

City: Kenney City Kit (Industrial), CC0 · kenney.nl