This fly is not alive. But when something rushes at it, the decision to bolt is made by brain cells copied out of a real fly — the same cells, joined up the same way scientists found them under a microscope. Move your cursor at it.
This is the fly above, thinking, twenty times every second. Move your cursor at it and watch the red group light up — those are the cells that spot something coming. Watch getting tired creep up the longer it stays in the air, until it gives in and lands.
Anyone can write “real fly brain” on a web page. So here is the test that settles it. Take the same brain cells. Keep the same number of connections, and the same strengths. Then join them up at random instead. If the fly behaves exactly the same either way, the real wiring was never doing anything and I should stop talking. Press the button — your own computer runs it, right now, in front of you.
wired as found in a real brain
same cells, joined up at random
The connections in the panic circuit were measured from a real fly brain and published by scientists. I did not invent them.
Every cell has a name and an ID number. You can look any of them up yourself and find the same cell in the real dataset.
The random-rewiring test above was run at exactly the same strength setting as the real one, so neither side was tuned to win.
The brain thinks at a steady twenty times a second, no matter how fast your screen draws.
This is not a whole fly brain. A real one has about 140,000 cells. This has a few hundred, and only for one reflex — flinching. Everything else it does, I wrote by hand.
These are simplified cells. No electrical spikes, no chemistry, and the fly does not learn anything.
A wiring map tells you what connects to what — and nothing else. How fast each cell reacts, and how hard it pushes, are numbers I had to choose.
It looks like a housefly. The brain cells come from a fruit fly. Different animals — the body is just a costume.
Connectome: Dorkenwald et al., Nature 634 (2024), and Schlegel et al., Nature 634 (2024) · FlyWire, Princeton University and the FlyWire Consortium · data reused under CC BY 4.0.
Circuit sources: Klapoetke et al. 2017 (LPLC2, radial expansion) · Ache et al. 2019 (LC4, angular speed) · von Reyn et al. 2014 & 2017 (giant-fibre escape modes) · Seelig & Jayaraman 2015 and Hulse et al. 2021 (heading ring) · Aso et al. 2014 (mushroom-body valence) · Donlea et al. 2014 (sleep drive) · Seeds et al. 2014 (grooming hierarchy).
Control-experiment rationale after the 2026 digital sphinx result (Brunton lab): a connectome model that behaves correctly is not by itself evidence that the connectome is doing the work.