Software Engineers Wire a Fully Mapped Fruit Fly Brain to Play Doom and Super Mario 64

The most detailed map of a biological nervous system ever created is currently trying to survive the first level of a 1993 first-person shooter. On September 3, 2026, Google Research and the HHMI Janelia Research Campus released MaleCNS v1.0, a complete structural map of an adult male fruit fly's brain containing 166,700 neurons and 125 million synaptic connections. Within three days, independent software engineers had wired that biological blueprint into custom software wrappers, forcing the digital connectome to play Doom, Super Mario 64, and Beat Saber.[2][3][5]

The result is a bizarre, high-speed collision between computational neuroscience and gaming culture that proves how quickly open-source biological data can be transformed into a software testbed. The experiment relies on translating game data into simulated biological stimuli. In the Doom setup built by Coinbase software engineer Alex Wormuth, every frame rendered by the game engine is converted into 3,335 brightness signals and 811 color signals.[1][3][7]

These visual inputs are fed directly into the digital model's artificial photoreceptors. From there, the software tracks how the simulated electrical signal cascades through the fly's 11,710 distinct cell types, tracing the route from visual processing to motor response. Specific motor neurons are mapped directly to keyboard inputs.[7]

When the simulated signal reaches the DNp20 motor neuron, the software translates that spike into a command to rotate the player's view. When the DNpe017 neuron fires, the character moves forward and shoots simultaneously. "Each Doom frame stimulates sensory neurons. Neural activity is mapped to game controls. Damage triggers a stimulus to two PPL101 dopamine cells as reinforcement," Wormuth wrote when publishing the open-source repository on September 6.[1][2][3][7]

That dopamine trigger is the crucial variable in the experiment. The connectome itself is just a static map—a wiring diagram of a brain, not a living, thinking entity. To make it "play," the software has to simulate a learning loop. Whenever the Doom marine takes damage from an enemy, Wormuth's code delivers a 200-millisecond negative pulse to the PPL101 cells, which handle punishment and avoidance in a living fruit fly.[4][6][7]

The open question is whether this simulated feedback loop can actually alter the digital pathways enough to teach the model to avoid incoming fire. Early results have been predictably chaotic. A live web interface tracking the Doom experiment showed the digital fly struggling through its first 6,300 rounds, frequently spinning in circles or firing into walls as the simulated network attempted to make sense of the sensory flood.[1][7]

Doom is not the only testing ground. Developer Jessica Paquette wired the same 166,700-neuron dataset into Nintendo's Super Mario 64, generating the integration code entirely through the AI assistant GPT Astra. Early footage showed Mario repeatedly jumping into a wall, highlighting the messy early stages of simulated neural training.[1][2][3]

Meanwhile, a developer known as Lyra Bubbles took a different approach with the rhythm game Beat Saber. Instead of dropping the connectome into a live environment, they fed the network replays of a completed song, using reinforcement learning to teach the model where and when to strike the incoming blocks based on historical data.[1]

These gaming experiments are running on top of a monumental scientific achievement. Mapping the Drosophila melanogaster central nervous system required slicing a physical brain into millions of ultra-thin sections, imaging them with electron microscopes, and using AI to trace the tangled web of synapses. It took a decade of work by Google and HHMI Janelia to produce a dataset that researchers hope will unlock new understandings of neurological conditions.[2][4]

Because mapping the estimated 86 billion neurons in a human brain remains technologically impossible, scientists rely on smaller organisms to understand how neural circuits influence behavior. But a digital map is not a living mind, and the gaming community's experiments come with severe biological caveats.[2][5][6]

Google explicitly notes that MaleCNS v1.0 is a structural snapshot, lacking the complex chemical soup of neurotransmitters and hormones that dictate real biological behavior. The Doom and Mario experiments simulate electrical spikes traveling along known paths, but they rely on heavy approximations. A digital fly taking damage in a video game does not feel pain or consciously understand its environment; it merely processes a simulated electrical penalty according to the rules written by a programmer.[4][6]

Still, the speed at which developers turned a landmark biological dataset into a playable software model marks a shift in how neuroscience intersects with computer science. By dropping a mapped biological network into a controlled, high-stakes digital environment, engineers are creating a new kind of sandbox.[3]

The fruit fly connectome may not be self-aware, but as it navigates the pixelated corridors of a 33-year-old game, it is providing a real-time visualization of how biological wiring processes a hostile world. The next milestone will be seeing if the digital fly can finally clear the first level without getting stuck in a corner.[1][3]