Male Fruit Fly Brain Trained to Play Classic Video Games
Digital fly brain model controls Doom, Beat Saber, and Super Mario 64 through neural mapping.
1Breakthrough in Neural Mapping
Google created a detailed 3D reconstruction of a male fruit fly's brain with 139,255 neurons. This digital model connects sensory inputs to game controls, where visual frames stimulate neurons that map to actions. The project demonstrates how complex behaviors can emerge from relatively simple neural architectures, providing insights for AI development.
2Doom as a Testing Ground
The fly brain was first trained on the original Doom, with damage events triggering dopamine cell stimulation as reinforcement. Engineer Alex Wormuth developed the system where neural activity directly controls movement and shooting. Despite 6,385 rounds, the fly performs poorly—highlighting challenges in translating biological processes to digital environments.
3Expanding to Rhythm Games
Modders adapted the system for Beat Saber, using replays to teach the fly when and where to strike notes via reinforcement learning. Lyra Bubbles' approach shows the brain responding to pre-programmed patterns rather than reacting organically. Current training aims to reduce reliance on external signals for fully reactive gameplay.
4Super Mario 64 Adaptation
Jessica Paquette shared results of running the digital brain through Super Mario 64's 3D environment. The experiment tests spatial navigation in complex worlds, with the fly brain processing platform jumps and enemy avoidance. This expands the model's applications beyond first-person shooters and rhythm games.
5Minecraft Integration
YouTuber Ro0oney implemented the system in Minecraft using Flywire.AI—a Princeton University collaboration analyzing a female fly brain with 160,000+ neurons. While neuron counts differ from Google's male model, both projects demonstrate how biological neural networks can interface with diverse game engines.
6Technical Implementation
The system maps specific dopamine cells (PPL101) to reward signals, mimicking biological reinforcement learning. Visual inputs are processed through the fly's optic lobe simulations, with motor outputs controlling in-game actions. Researchers monitor neural activity in real-time through custom visualization tools.
7Research Implications
These experiments advance understanding of how simple organisms process complex environments. The projects provide scalable models for studying decision-making pathways and could inform more efficient machine learning architectures. They also demonstrate unexpected applications for game engines in neuroscience research.
8Future Directions
Researchers aim to develop fully reactive systems without pre-programmed inputs. Future work may explore cross-species neural comparisons and applications in robotics. The open-source nature of these projects encourages community-driven development at the intersection of biology and artificial intelligence.
Original article: Slashdot — Male Fruit Fly Brain Trained to Play Classic Video Games