The Fly Brain Test: What 166,700 Neurons Teach About Small AI and AI Hype
The male fruit fly connectome (FlyEM male CNS v1.0, released 8 June 2026, Cell paper 3 September 2026, CC-BY 4.0) maps about 166,700 neurons in 11,710 cell types across the brain and nerve cord, from HHMI Janelia, the University of Cambridge, the MRC Laboratory of Molecular Biology and Google Research. The article explains what a connectome is, what a small specialised brain suggests about efficient AI (Lappalainen et al. 2024 predicting neural activity from wiring, the Shiu et al. 2024 whole-brain model, a whole fly brain on 12 Intel Loihi 2 chips in a 2025 Sandia preprint), and walks through the giant fibre escape circuit: 126 LC4 and 185 LPLC2 visual neurons making 11,224 synapses onto the two giant fibres (DNp01), which trigger the jump motor neuron TTMn and the flight muscles through PSI. A live simulation built on the real synapse counts shows that silencing LC4 removes escapes from fast looms, silencing LPLC2 delays escapes on slow looms, shuffling or equalising synapse counts within a cell type changes almost nothing, and random input partners abolish the reflex, with stated caveats about model gain and the giant-fibre-independent escape. It reviews connectome demos including Doomfly, Stonkfly, NeuroTerrarium and Eon Systems' embodied fly using their own published results, and turns neuroscience controls into a five-question check for AI claims: what is measured versus assumed, switch-off tests, scrambled or naive baselines, held-out data, and published failures.
Frequently Asked Questions
- What is a connectome?
- A connectome is a wiring diagram of a nervous system: every neuron and every synapse between them, reconstructed from electron microscope images. It records which cells connect and how many synapses each connection has. It does not record how fast each neuron responds, how strong each synapse is, or electrical connections between cells, so any simulation built on it adds assumptions.
- What is the male fruit fly connectome?
- It is the first complete wiring diagram of a male Drosophila central nervous system, covering the brain and the ventral nerve cord in one animal. Version 1.0, released on 8 June 2026 by HHMI Janelia's FlyEM team with the University of Cambridge, the MRC Laboratory of Molecular Biology and Google Research, contains about 166,700 neurons in 11,710 cell types. The paper appeared in Cell on 3 September 2026 and the data is free to use under a CC-BY 4.0 licence.
- Can you simulate a whole fly brain?
- Yes, in simplified form. Shiu et al. published a leaky integrate-and-fire model of the whole female fly brain in Nature in 2024, built on the FlyWire connectome, and used it to study the circuits behind feeding and grooming. A 2025 preprint from Sandia ran that whole-brain connectome, about 140,000 neurons and 50 million synapses, on 12 Intel Loihi 2 neuromorphic chips. These models use real wiring with simple neuron models, so they are useful for generating hypotheses and are not complete replicas of a living brain.
- Is a fly connectome an AI model?
- No. A connectome is measured anatomy. It becomes a model only when someone adds neuron dynamics, sensory inputs and motor outputs, and it becomes an AI system only when someone connects it to a task. Most of the behaviour in public connectome demos comes from those added choices, which is why it matters to ask what is real data and what is engineered.
- Has anyone uploaded a fly brain into a computer?
- Eon Systems published an embodied fly simulation in March 2026 that connected a whole-brain model based on the FlyWire connectome to the NeuroMechFly body in the MuJoCo physics engine. The simulated fly groomed, fed and moved towards food, with brain and body synchronised every 15 ms. Eon lists the limits itself: simplified neurons, no learning or internal state, and hand-mapped links between brain and body. It is a simulation built on connectome data, so it inherits those limits.
- Can a fly connectome play Doom or trade crypto?
- People have wired fly connectome models to Doom, Minecraft, Super Mario 64 and paper trading accounts, and several of those projects publish their negative results. Doomfly's README says learned survival has not been demonstrated and its latest candidate failed its validation gates. Stonkfly's README says profitable learning has not been demonstrated. The wiring is real in these projects; the game controls, rewards and readouts are engineering choices.
- What is the giant fibre escape circuit?
- It is the fly's fastest escape reflex. Visual neurons called LC4 signal how fast a shape is looming and LPLC2 neurons signal how big it looks. Both feed the giant fibre, a large descending neuron (DNp01) that triggers the jump muscle motor neuron (TTMn) and the flight muscles through the PSI neuron. In the male CNS connectome, 126 LC4 and 185 LPLC2 cells make 11,224 synapses onto the two giant fibres. Flies also have a slower escape that does not depend on the giant fibre.
- How accurate is the escape demo in this article?
- The wiring is real: each of the 311 visual cells uses its measured synapse count onto the right giant fibre. The rest is simplified and chosen by us: the looming tuning curves, Poisson spiking, a single leaky integrate-and-fire giant fibre and the gain. Its results agree in direction with published experiments, such as LC4 carrying the fast-loom signal, but the exact percentages depend on the gain and should not be read as fly measurements.
- What does the fly connectome mean for businesses?
- There is nothing to buy today. The practical lessons are indirect: narrow, specialised networks do real work with very few units, sparse and event-driven computation is where efficiency comes from, and the controls neuroscientists use (switch a part off, compare with scrambled wiring, publish failures) are a good way to test any AI claim a vendor makes.
Our Services
Contact Cloud First Consulting
Email: info@cloudfirstconsulting.com
Location: London, United Kingdom
Hours: Monday-Friday, 9:00 AM - 6:00 PM GMT
Book a Free 30-min Discovery Call