What the game is—and is not

A CONNECTOME, NOT A MIND.

Human vs Brain uses a computational model constrained by the published MaleCNS connectome: a mapped adult male Drosophila central nervous system. The game is a narrow, binary stimulus experiment. It is not a model of understanding, and it does not show that a biological fly understands this game.

How one round works

1. The same LEFT or RIGHT target shown to you is selected for the model.
2. The game directly stimulates the selected same-side LC10a visual-projection-neuron population.
3. Activity propagates through the MaleCNS-derived connectome simulator.
4. Activity in the left and right DNa02 descending neurons is measured.
5. The side with higher DNa02 output activity becomes the model's LEFT or RIGHT action.

No AI model, LLM, trained policy, learned readout, or hidden heuristic chooses the fly's move. The action comes only from LC10a stimulus → MaleCNS simulation → bilateral DNa02 activity.

What is assumed

The published connectivity is transformed into a lightweight point-neuron simulation. It uses homogeneous leaky integrate-and-fire dynamics, normalized synaptic weights, and predicted neurotransmitter signs. The game directly injects LC10a rather than simulating optics, retina, lamina, object-motion computation, body movement, or closed-loop steering.

The public challenge uses a fixed LC10a stimulus strength of 0.27 model voltage units per simulation step, weaker than the 0.80 scientific reference configuration. This creates a non-trivial discrimination problem inside the same full MaleCNS simulation and unchanged DNa02 decoder. The value was selected by computational game calibration; it is not a biologically calibrated stimulus magnitude. Validation found a modest residual LEFT advantage, which is a model result rather than a corrected game mechanic.

These choices make a reproducible computational probe, not a validated whole-animal model. The displayed DNa02 rates are model outputs, not claims about a fly's absolute firing rate in this game.

Why these populations

LC10a is a published visual-projection-neuron type implicated in target tracking and ipsiversive turning. DNa02 is a bilateral descending-neuron type associated with steering. Those findings motivate the input and readout; they do not validate every modeling assumption here.

Sources and attribution

• MaleCNS v1.0 official download — data licensed CC BY 4.0.
• Berg et al., “Sexual dimorphism in the complete Drosophila male central nervous system connectome,” Cell (2026).
• Ribeiro et al., LC10 visual projection neurons and directed courtship, Cell (2018).
• Hindmarsh Sten et al., Lateralized LC10a visual responses, Nature (2021).
• Yang et al., Fine-grained DNa02 steering control, Cell (2024).
• Rayshubskiy et al., Bilateral DNa02 activity and steering, eLife (2025).

MaleCNS attribution: FlyEM project at HHMI Janelia, University of Cambridge, MRC Laboratory of Molecular Biology, Google Research, and the MaleCNS authors. Human vs Brain does not redistribute the dataset.