+ A CONNECTOME-BASED CHESS EXPERIMENT
Two fly circuits. Sixty-four squares. Let them cook.
Connecting to the shared broadcast…
The first move is always a little nerve-racking.
Every rivalry starts somewhere.
Real connections. Simulated decisions.A small experiment at the intersection of neuroscience and play.
Two independent simulations use the same sampled fruit-fly connectivity graph, with different starting states. Their activity selects legal chess moves.
This is a partial connectome-based simulation. The input encoding, neural dynamics and chess decoder are engineered. It is not a complete fly brain, biological vision, demonstrated chess understanding or learning. The 3D brains show 139,255 anatomical neuron positions. Only the highlighted 268 neurons belong to the chess circuit; the remaining points provide anatomical context. The 3D flies use a simplified CT-scan mesh; their small display movements are illustrative, not biomechanical simulation.
All viewers follow one server-controlled match and a shared, persistent score. Pause freezes only your view; Return to live catches up to the current position. Updates arrive about once per second. The server advances the persisted simulation when viewers request updates; after a period without viewers, it resumes from the saved position. Completed games advance automatically.
3D fly: etainproject — Drosophila CT scan (CC BY 4.0), simplified and re-shaded. Anatomy: Microfly / FlyBrain / FlyWire. Rendering: Three.js (MIT). Drag either model to rotate; use arrow keys when focused, or Home to reset.
Connectivity: FlyWire FAFB v783 · FlyWire annotations.Sample: FLYDINO's published subgraph, drawn from the first 65,536-row batch, retaining edges of at least five synapses. Inspect the exact data ↗Rules: chess.js (BSD-2-Clause). Pieces: Colin M. L. Burnett, via Lichess (CC BY-SA 3.0). Fonts: DM Sans and Instrument Serif (SIL OFL).