Doom has found another bizarre platform: a simulated 166,000 neuron fruit fly brain

Google Research and a group of scientific partners have mapped the brain and central nervous system of an adult male fruit fly in remarkable detail, producing a digital model containing more than 166,000 neurons. Software developers wasted little time finding something unusual to do with it. They connected versions of the reconstructed neural network to Doom, Super Mario 64 and Beat Saber. Yes, Doom again. The game has spent decades appearing on calculators, printers, cameras, tractors and almost anything else capable of running code. A simulated fruit fly nervous system is a rather different proposition. Here, developers aren’t squeezing Doom onto strange hardware. They’re feeding game information into a computer model built from biological connections and turning its neural activity into game controls. It’s less about whether a fly can frag a demon and more about what happens when a huge neuroscience dataset lands in the hands of programmers.

A fruit fly nervous system, mapped neuron by neuron

The dataset is called MaleCNS v1.0. It covers the brain and central nervous system of an adult male Drosophila melanogaster, with more than 166,000 reconstructed neurons packed into the model.

Researchers produced it from millions of microscopic image slices. Computer vision systems helped identify biological structures across those images, with the resulting data assembled into a three-dimensional map of neural connections.

This type of map is called a connectome. Think of it as a wiring diagram for a nervous system, showing which neurons connect with others and where signals have possible routes through the network.

MaleCNS goes beyond the fly’s brain. It includes the ventral nerve cord, which carries signals between the brain and much of the insect’s body. For researchers studying movement and sensory processing, this wider coverage gives the model far more material to work with. For software developers, it gives them something else: a very strange controller.

Doom turns neural activity into movement

One experiment links the reconstructed nervous system with the original Doom. Frames from the game can be converted into sensory input and fed into parts of the simulated neural network. Activity generated inside the model can be interpreted by software and mapped onto Doom controls.

Taking damage can trigger stimulation in dopamine-related cells, creating a basic reinforcement signal. The system receives information, activity moves through its simulated connections, and software translates part of the response into an action inside the game. Don’t expect a tiny digital fly speedrunning E1M1.

The simulation isn’t sitting there appreciating level design, searching for keycards or deciding which shotgun shell to pick up. Developers are using biological wiring as part of a computational control system. Doom provides a convenient test bed since its inputs are simple, its behaviour is predictable and programmers have decades of tools available for experimenting with it. Still, seeing a neural reconstruction produce activity capable of controlling a game makes the research far easier to grasp than staring at a huge database of neuron connections.

Mario 64 gives the fly another job

Doom wasn’t the end of the experiment. Similar work has brought the simulated nervous system into Super Mario 64. Mario presents a different problem. Movement happens in three dimensions, the environment changes continuously and useful actions require more varied control inputs. Connecting a biologically derived network to such a game gives developers another way to test how simulated sensory information can produce usable outputs.

Again, nobody is claiming a fruit fly understands Mario. What matters is the pipeline. A game supplies input, the simulated neural system processes it, and software converts resulting activity into movement. Games make the process visible in seconds. Instead of watching graphs change, you can watch Mario move.

Beat Saber pushes the experiment further

Beat Saber has joined the list too, with recorded gameplay information used alongside reinforcement learning to train a simulated neural system to respond to incoming targets. Current experiments still rely on extra training information. They aren’t examples of a fully independent fly connectome seeing Beat Saber blocks and immediately working out how to hit them.

There’s plenty of software sitting between the biological map and the game. For researchers and developers, reducing reliance on training assistance is one obvious technical target. More behaviour could eventually come from sensory input passing through simulated biological connections, rather than actions copied or inferred from existing gameplay recordings.

The interesting part isn’t the high score

Nobody needs a fruit fly brain to play Doom. Conventional AI systems can handle videogame tasks far more directly. The interesting part is what programmers can now do with biological datasets. A connectome once existed mainly as research material to inspect, compare and analyse. MaleCNS can become an active component inside software. Developers can stimulate parts of the model, observe neural activity and connect its outputs to other programs.

Games happen to make excellent demonstrations. They’re interactive, easy to observe and packed with clean inputs and measurable responses. There’s still a huge distance between a digital wiring map and a living nervous system. A real brain involves electrical activity, chemical signalling, changing biological states and processes a connection diagram alone can’t reproduce.

MaleCNS doesn’t recreate a fruit fly in software. It does give programmers a 166,000-neuron biological network to experiment with, though. And naturally, one of the first questions someone asked was whether it could run Doom.

Spread the love