Fruit Fly Connectome Powers Tic-Tac-Toe AI
A biological neural circuit drives game decisions in this experimental AI project.
1Biological circuit as game AI
Researchers extracted a 100-neuron circuit from male fruit fly CNS connectome data to create a tic-tac-toe playing agent. The biological neural network serves as the decision core, though the fly brain has no inherent understanding of the game. The project demonstrates how biological circuits can be repurposed for computational tasks through deliberate engineering conventions.
2Connectome data sources
The neural data comes from male Drosophila melanogaster central nervous system connectome (male-cns:v1.0), hosted on a specialized neuPrint server. The dataset represents collaboration between FlyEM at HHMI Janelia, University of Cambridge, MRC Laboratory of Molecular Biology, and Google Research. All outputs using this data must properly credit these institutions under CC-BY license terms.
3Engineering game conventions
The project maps tic-tac-toe board states onto about 100 real neurons through artificial conventions, as the fly connectome contains no natural board representation. Input/output mappings are entirely engineered rather than biologically inherent. The empty-board opening move is deterministic due to zero initial activation, with tie-breaking handled by highest input-weight output groups.
4Technical implementation
The system uses Python with networkx for graph representation and implements leaky-integrator dynamics for neural simulation. While not biophysically accurate, this provides functional neural behavior. The repository includes modules for connectome extraction, circuit analysis, game simulation, and agent implementation. Users can play against the fly-brain agent or watch two agents compete.
5Gameplay limitations
In 'unbeatable' mode, the system uses minimax algorithms to ensure optimal moves, with the biological circuit only breaking ties among equally good options. Without this safety net, the raw neural circuit can lose games. The project acknowledges these limitations in biological accuracy while demonstrating computational potential of neural circuits.
6Future development paths
The creators suggest implementing spiking neural network simulators like Brian2 as a next step for more biologically realistic dynamics. The current leaky-integrator model serves as an illustrative but simplified approach. The modular design allows swapping different simulation backends while maintaining the same circuit mapping conventions.
7Scientific vs engineering claims
The project carefully distinguishes between biological facts and engineering conventions. It explicitly notes that claims about 'how the fly brain sees the board' should be taken skeptically, as the sensory pathways were deliberately cut during circuit selection. This maintains scientific rigor while demonstrating creative engineering applications.
8Open-source neuroscience
This project showcases how open connectome data enables novel computational experiments. By making all code and methodology available, it invites further exploration of biological neural circuits for artificial intelligence applications. The work bridges neuroscience and computer science through reproducible, curiosity-driven experimentation.
Original article: Hacker News: Show HN — Fruit Fly Connectome Powers Tic-Tac-Toe AI