Google Maps Fruit Fly Brain With AI, Engineers Immediately Train It To Play Doom and Mario 64

Google Achieves Neuroscience Breakthrough With Complete Fruit Fly Brain Mapping

Google Research, in collaboration with the Howard Hughes Medical Institute's Janelia Research Campus, announced on September 3 a decade-long achievement: the complete mapping of an adult male fruit fly's brain and central nervous system1

1

2

2

. This connectome represents the largest brain map by number of neurons to date, featuring over 166,000 neurons and 125 million synaptic connections2

2

. The project utilized AI to transform millions of 2D electron microscope images into 3D neural shapes, reconstructing the complete neural architecture of the insect1

1

. Scientists spent years verifying and annotating the results after the initial AI-powered reconstruction2

2

.

Why This Neural Computation Milestone Matters for Brain Research

The complete mapping of an adult male fruit fly's brain fills a critical gap in neuroscience research. While the human brain contains an estimated 86 billion neurons—making complete mapping currently impossible—fruit flies serve as ideal model organisms for understanding fundamental brain function1

1

2

2

. These insects have served research for over a century, particularly in genetics and neuroscience1

1

. The MaleCNS v1.0 connectome provides scientists with a fundamental resource to study how animal nervous systems perceive the world, react to stimuli, and potentially how damaged neural pathways might be repaired2

2

. This wiring diagram showing how neurons connect through the central nervous system accelerates understanding of brain mechanics in ways previously unattainable1

1

.

Software Engineers Train AI-Powered 3D Model to Play Video Games

By Monday, September 6—just three days after the announcement—software engineers began experimenting with the fruit fly brain data in unexpected ways. Alex Wormuth demonstrated training the full MaleCNS v1.0 connectome to play Doom, with each game frame stimulating sensory neurons and neural activity mapped to game controls1

1

2

2

. Damage in the game triggers stimulus to two PPL101 dopamine cells as reinforcement1

1

. The training process runs live online, allowing observers to watch all 166,700 neurons learn in real-time1

1

.

Jessica Paquette, describing herself as a C++ ragebaiteur, created a project training the simulated fruit fly brain to play Super Mario 641

1

2

2

. Her video shows Mario repeatedly jumping and bumping into walls—behavior that, ironically, mirrors actual fly behavior when encountering windows1

1

. The code was developed entirely using GPT Astra1

1

2

2

. Both projects released their code as open source, enabling wider experimentation1

1

2

2

.

Beyond Gaming: Creative Applications Emerge From Connectome Data

The creative applications extended beyond playing Doom and Super Mario 64. One developer successfully ran the entire 166,700-neuron connectome inside Minecraft, using simulated neural activity to drive a fly's movement2

2

. Another engineer fine-tuned the simulated fruit fly brain to make Y-M-C-A poses in response to four different tones, while yet another made the brain watch the viral Bad Apple animation to observe its reactions2

2

. These experiments, while playful, demonstrate the accessibility and potential of the neural data for testing theories about neural computation and learning.

What's Next: Scaling Up to More Complex Organisms

Google is already planning beyond fruit flies. The company confirmed it's working with researchers on connectomes for fish and mice2

2

. These progressively complex organisms will bring scientists closer to understanding mammalian brains, though the leap from 166,000 neurons to the billions found in mice and eventually humans remains substantial. The immediate proliferation of experimental projects using the fruit fly brain data suggests strong community interest in neural modeling. Watch for more sophisticated applications as researchers and developers gain familiarity with the connectome structure and its implications for understanding how neural pathways process information, learn from experience, and generate behavior.