Engineers run Minecraft, Doom and Mario 64 on Google's fly brain map

Google Research and the Howard Hughes Medical Institute's Janelia Research Campus released a complete map of an adult male fruit fly's brain and central nervous system on September 3, 2026, and software engineers had it running video games within three days. By September 6, a simulated fly brain was being trained to play the original Doom, with Super Mario 64, Minecraft and Beat Saber following. Polygon frames the reaction as the internet deciding that a fruit fly's consciousness is now trapped inside the games, a reading that runs ahead of what the dataset actually is.

A connectome is a wiring diagram. It records which neurons connect to which and captures anatomy, not activity. The fly map catalogs more than 166,000 neurons and roughly 125 million synaptic connections, which makes it the largest brain map by neuron count published to date. Google and Janelia describe it as a fundamental resource for using fruit flies as a model organism for studying how brains work. The scale comparison offered in coverage is the estimated 86 billion neurons in a human brain. The project itself ran about a decade.

Researchers sectioned the nervous system into millions of ultra-thin slices and imaged each at high resolution. AI combined the resulting millions of 2D images into 3D neural shapes, reconstructing a record-breaking 166,000-plus neurons, and human proofreaders corrected the automated reconstruction afterward. Google says the map covers the central nervous system as well as the brain, which is why the released dataset, MaleCNS v1.0, can drive movement rather than just visual input.

Alex Wormuth, a software engineer at Coinbase, posted on September 6 that the fly brain was being trained to play Doom using the full MaleCNS v1.0 connectome. Each Doom frame stimulates sensory neurons, he wrote, and neural activity maps to game controls; taking damage triggers a stimulus to two PPL101 dopamine cells as reinforcement. The run is livestreamed, showing both Doom's first-person view and a third-person render of the fly moving through the maps. Jessica Paquette published code for a Mario 64 version on GitHub and described it as 100% vibe coded with GPT Astra. A separate user reported running all 166,700 neurons of the retained connectome inside Minecraft, with simulated neural activity driving a fly's movement.

Other demos pushed the same dataset further from the lab. One user used replays of a song in Beat Saber plus reinforcement learning to teach the simulated fly where to hit, and acknowledged that the fly is not reacting to notes in real time yet; the stated goal is to reduce reliance on external signals. Another experiment trained the brain to fire Y-M-C-A poses in response to four tones. A third had it watch the Bad Apple!! animation. The pattern is consistent: a public, well-documented dataset goes live and hobbyist programmers adapt it within days, faster than peer-reviewed follow-ups can arrive.

Google and the researchers attach a caveat the game demos elide. A connectome is a static anatomical snapshot rather than a simulation of everything happening inside a living brain, and the fly-brain game projects make significant approximations when they convert wiring into activity. There is no evidence the digital fly has consciousness or subjective experience, so the Black Mirror framing in the headline conflicts with the scientists' own description of what they published. Whether the fly learns anything, as Wormuth's post asks, stays open.

The demonstrations establish something narrower than they look: a possible interface between connectome-scale data and real-time control, not a fly reproduced as a digital mind. What remains unknown is whether any in-game behavior reflects the fly's actual neural computation or the approximations layered on top of the wiring. Google has not validated the hobbyist projects, and because the datasets are public, more adaptations are likely before any of them are tested rigorously.