Google Mapped A Fruit Fly Brain, So Engineers Taught It To Play Doom

by Aaron Leong — Thursday, September 10, 2026, 10:22 AM EDT

A male fruit fly - Image: Philip Veater via Unsplash

When Google released a complete 3D map of an adult male fruit fly’s central nervous system, scientists celebrated it for its impact on neuroscience. Software engineers, however, had other ideas by immediately putting the dataset of the fly to greater use: training it play video games like Doom, Super Mario 64, and Beat Saber.

A quick word about the underlying achievement, though: the decade-long project between Google Research and HHMI’s Janelia Research Campus has produced the largest complete brain map by neuron count of a fly to date. Designated as MaleCNS v1.0 connectome, the digital map consists of over 166,000 neurons and roughly 125 million synaptic connections across the fly’s brain and central nervous system (beating out prior achievements by quite a margin).

Researchers achieved this by slicing the insect’s nervous system into millions of ultra-thin sections, scanning them under high-powered electron microscopes, and relying on AI algorithms alongside years of manual verification to trace every micro-pathway in 3D space. The goal was to provide an open-access blueprint for understanding how animal brains process sensory input, control movement, and handle neural damage.

Yet within days of Google publishing the open-source dataset, enthusiasts and developers recognized that a fully mapped neural network is, in essence, a complex computational engine ready to be programmed.

Frontal section of Google Research's digital map of a fly - Image: Google

Software engineer Alex Wormuth captured the internet's attention by converting the fly’s connectome into an active neural simulator to play Doom (1993 version, of course). By mapping each frame of the game directly into the fly’s virtual sensory neurons, Wormuth configured the resulting neural activity to trigger movement controls. Whenever the digital fly took damage in the game, the simulation sent an electrical pulse to two specific dopamine-releasing cells (PPL101) to act as a biological reinforcement learning loop, rewarding survival and punishing mistakes.

Other developers quickly joined the party. Programmer Jessica Paquette built a framework using the 166,700-neuron model to pilot Mario in Super Mario 64, even sharing footage of the Nintendo mascot hopping and running directly into walls as the simulated insect brain attempted to parse 3D platforming space. Meanwhile, other coders ported the connectome into rhythm games like Beat Saber and Minecraft (to drive a virtual fly's movement).

Now, if scientists can program flies not to land on my food, that'd be great - Image: Google

These demos might sound like fun and games, and they certainly are, but by showing that a mapped biological brain can process arbitrary sensory data and output functional decisions, programmers are validating the fly brain map's real-world utility. Google is already eyeing much larger targets, including fish and mice.