Fly plays.
A tiny brain. An unfamiliar arcade.
A full fruit fly connectome, two games, and a controller learning from scratch.
Meet the brain: MaleCNS
Let’s see what happens.
Checking brain data…
Your browser does the work. Your progress stays here.
The lights come on when you hit Start.Then you can watch the signals move.This is what the eyes get.
Trying: Waiting
Rewards help it choose its next move.
Each finished game adds a dot here.
It might get better. It might just get very good at losing.
Your browser saves what it learns when it can. Switch games for a fresh round; each game remembers its own practice.
Someone mapped a fly. So I gave it games.
First, the really big science project.
Google Research teamed up with HHMI Janelia, Cambridge, and MRC LMB to map a male fruit fly’s brain and nerve cord. The result is called MaleCNS. It’s a huge wiring diagram showing which nerve cells connect to which. Scientists call that a connectome.
Google’s AI helped turn flat electron-microscope images into 3D shapes of nerve cells. Human experts then checked and corrected the map. The goal is to understand how a brain turns things it sees, smells, or hears into something it does.
Read Google’s side of the story ↗Then, this slightly less serious idea.
I took that wiring and hooked it up to an arcade. Your browser runs simple rules for how signals move through the cells, then a small controller turns those signals into button presses. It starts without knowing how to play.
The wiring comes from a real fly. The signal rules, screen-to-eye mapping, and game controls are things I added. So this is a computer experiment built around real anatomy, with plenty of guesswork about how it should behave.
The loop is simple: see the game, try a move, get feedback, try again.
A very tiny game screen.
The game gets shrunk to a 64 by 48 pixel picture. That picture feeds 6,006 simulated eye cells. The scores and buttons you see around the game stay out of that picture. The little vision panel shows exactly what goes in.
The wiring does its thing.
Signals travel through all 166,700 cells and 25.6 million connections in this version of the map. A cell builds up a signal, fires, and passes it to its neighbors. The glowing dots show a small sample of that activity at real cell locations.
Practice, with feedback.
A small controller listens to 1,316 cells that carry signals away from the brain. Ten times a second, it tries a move. Good or bad results nudge its future choices. Only this controller learns; the fly’s wiring stays fixed.
A few things you might be wondering +
What counts as a good move?
Space Invaders gives rewards for hitting enemies and finishing waves, and a penalty for losing a life. Tetris rewards cleared lines and penalizes piling blocks all the way to the top. It also gives small nudges for keeping the stack low and leaving fewer holes. The controller only gets the resulting reward and brain signals, so it can’t peek at the board or enemy positions.
Will it actually get good?
Maybe. The first short test runs didn’t show a clear win over random button presses, and none cleared a Tetris line. It’s an experiment you can watch, pause, and come back to. Learning here means its choices can change with feedback, not that a high score is around the corner.
Is this a living fly in my browser?
The real part is the wiring map. We don’t know enough to recreate everything a living fly’s brain does. The simulated cells use simplified rules and a little background activity. This doesn’t show that a fly is thinking about the game or experiencing it.
What am I looking at in the brain panel?
The dots show 4,096 measured cell locations. A bright dot means that simulated cell just fired. All the other cells still run; showing every one would make the picture hard to read. Signal rate is the average firing rate of the cells the controller listens to. Hz means signals per second.
What happens when I leave?
Your browser saves each game’s practice separately, when it has room. Reloading starts a new round with that saved learning. Restart keeps the practice; Reset Learning wipes it for just that game. Nothing syncs between devices, and clearing browser storage clears the saves.
Why might it run slowly?
That’s a lot of brain to fit in a browser. On a slower device, the brain and game slow down together so every step still happens. The speed readout shows how fast it’s really going. Switching tabs pauses the experiment until you hit Resume.
Inspired by ornata / fly ↗
Built with MaleCNS v1.0 data ↗ · CC BY 4.0
Data: FlyEM / HHMI Janelia, Cambridge,
MRC LMB and Google Research. Modified for this experiment.