A field of fruit flies, each one flying on its own spiking neural network —
748 leaky integrate-and-fire neurons named after real Drosophila cell types and
wired in the shape of the pathways measured in the MaleCNS connectome.
Each fly has a mushroom body: its Kenyon cells give every smell its own signature, and
dopamine neurons that fire on a looming threat, a knock, alarm CO2 or a taste of
juice rewrite which smells it is drawn to. Switch it in the Data card.
Nothing in a fly's behaviour is scripted. Light, odour puffs, airflow and leg
contact are injected into sensory neurons; the network runs; and flight is read back out
of the motor neurons. There is no if (nearFruit) goToFruit. Feeding,
landing, escaping the swatter, plume tracking, courtship and egg-laying all fall out of
the wiring — or fail to, which is reported just as plainly.
A full life runs here: flies eat, mate, lay eggs, hatch, hunt for food, get
eaten by spiders and die — each of those drops a marker on the map where it
happened. The clock runs a 300-second day, and night is real to the flies: it dims the
luminance their photoreceptors see, rather than setting any “it is night” flag.
Click any fly to watch its brain fire. Every bar, tag and number in this panel is a
readout of population firing rates, never a driver of them.
It is not the real connectome: 748 neurons against 166,700, with real
cell-type names but invented counts and weights. Nothing here is trained.
What is and isn't real →
