How a serious study of a fly's brain turned into an internet meme — its brain model was taught to play Doom and trade cryptocurrency

In October 2024, researchers from Princeton University, together with the FlyWire consortium, presented in the journal Nature a connectome of the female Drosophila melanogaster brain, which contained nearly 140,000 neurons and over 50 million synapses.

In September 2026, a team from HHMI Janelia Research Campus, the University of Cambridge, and Google Research published in Cell a complete connectome of a male specimen. This model, named MaleCNS, included not only the central brain but also the ventral nerve cord — an analogue of the vertebrate spinal cord. The database contains information on 166,700 neurons and about 125 million synaptic connections.

The fruit fly brain measures about 750 by 350 by 250 microns, which is smaller than a poppy seed, yet it contains about 150 meters of neural connections with a thickness of 50 to 100 nanometers. To create its precise digital map, scientists used the method of serial electron microscopy. The biological sample was turned into a solid block and cut with diamond knives into more than 7,000 ultrathin layers, each about 40 nanometers thick. Each slice was scanned, resulting in researchers obtaining 21 million images with a total volume exceeding 40 trillion pixels.

To combine two-dimensional images into a three-dimensional model, artificial intelligence algorithms were used, specifically Google's Flood-Filling Networks neural network technology.

The program automatically defined cell boundaries, but human involvement was required to verify and correct computer recognition errors. The collective efforts of thousands of volunteers and scientists amounted to approximately 33 person-years of manual verification. According to researchers' estimates, without the use of artificial intelligence, such a modeling process would have taken about 50,000 person-years.

A distinctive feature of the MaleCNS model was precisely the integration of the ventral nerve cord, which for the first time allowed tracing pathways from sensory receptors to motor neurons that directly control the insect's movement. Researchers also discovered and mapped specific sexually dimorphic neurons. These cells differ in males and females, responsible for different behavioral patterns during mating and sensory signal processing.

How to run a nervous system in a software environment

A connectome is not a full-fledged simulation of a living mind. It is a static database, a precise "wiring diagram" that only shows which neurons are connected to each other and how many synapses form these connections.

The digital map lacks information about the current electrical activity of cells, the concentration of chemical substances, the presence of neuromodulators (such as dopamine or serotonin), the state of glia, and the general hormonal background.

For this static scheme to start functioning in a software environment, each neuron is transformed into a simplified mathematical model. Most often, the leaky integrate-and-fire model is used for this. A virtual neuron accumulates incoming signals from neighboring cells, and upon reaching a certain threshold value, generates its own impulse, transmitting it further along the chain, after which it returns to its initial state.

To interact the simulation with the external world — a video game, a trading terminal, or a text editor — developers create special software interfaces. External data (for example, pixels on the screen) are converted into artificial signals and fed to the fly's virtual sensory neurons (usually to visual or olfactory receptors).

Further, impulses propagate through the network strictly according to the biological map of synaptic connections. At the final stage, another program code records the activity of the output (motor) neurons, which in a real fly are responsible for the movement of wings, legs, or proboscis, and translates this activity into program commands: pressing a key on the keyboard, moving in a game, or selecting a trading operation.

The model itself does not understand the context and, unlike a living brain, does not possess the ability for natural neuroplasticity and learning.

From Minecraft to cryptocurrency exchange trading

The availability of an open interactive dataset allowed programmers to start their own experiments. Already on September 7, YouTuber Ro0oney published a video where the simulated network controlled a character in the game Minecraft.

The next day, another blogger released a video showing the model playing the rhythm game Beat Saber. It garnered 22 million views and spawned an internet meme where the virtual fly was assigned various human tasks.

the fly brain can play beat saber /AqKMqD6gzP— lyra bubbles (@_lyraaaa_) September 9, 2026

In the following days, dozens of new projects appeared. One demonstrated how the simulation simultaneously parks a car in a driving simulator. In a project called DOOM-x-Fly, visual data from the cult shooter was streamed to the insect's vision model, and the program's output layer analyzed the activity of thousands of neurons to press game keys.

the fly has completed* the ultimate test: parallel parking /KqdDg8dvo6— Breg Grockman (@alright_mark) September 10, 2026

An engineer under the nickname SpikeCalls went further and connected the simulation to the physical world: he connected the digital brain to a camera and four motors of a real drone, where visual signals directly controlled the thrust of the propellers.

Coinbase crypto exchange engineer Alex Warmuth developed the Stonkfly project: he turned a digital copy of the fly into a day trader for trading the BTC-USDC currency pair.

I gave the fly brain $100 to trade bitcoin.Dopamine neurons are stimulated when the fly makes profit. Neuron activity controls buy/sell decisions and makes trades on the fly get rich? /XVWSFattGu— Alex Wormuth (@wormuth) September 10, 2026

The program converts a standard candlestick chart of historical quotes into a 320x180 pixel color image and transmits these RGB values directly to the virtual brain's photoreceptors, distributing the image between the model's left and right "eyes."

Within the simulation, the fly has three actions available: buy, sell, or hold the asset. The basic decision-making interval ("thinking") is 500 milliseconds, and market data is updated every 60 seconds. For one operation, the algorithm can stake up to $10, but no more than 24 times a day.

The author implemented a biological reinforcement system: portfolio value growth simulates a dopamine release that activates 15 specific cells. Conversely, financial losses or trading fee deductions activate two disgust cells.

However, the algorithm is deprived of the ability to use leverage or short sell.

The Stonkfly project code is available in open access on GitHub for macOS and Linux operating systems. To run your own "neuro-trader," 16 GB of RAM, Python 3.11, and a C++ 17 compatible compiler are required. By default, the simulation runs in test mode with a starting virtual balance of $100, but the developer has left instructions for connecting the algorithm to a real account.

Ethical questions

The widespread use of the digital model for entertainment purposes sparked ethical discussions on platforms like Reddit. Users began to express concern about the possible digital suffering of the simulated creature.

Questions arose as to whether the virtual fly experiences stress or pain when forced to continuously process data from video games or financial markets. The situation is even compared to dystopias like the "Black Mirror" series and the concept of "The Matrix."

Neurobiologists deny the possibility of consciousness or the ability to feel within the current technology.

A connectome is solely a static map of synaptic connections. Digital simulations created based on it use simplified mathematical models of neurons that do not reflect the full complexity of a biological organism.

In the software environment, physiological processes that ensure vital activity and perception are absent: metabolism, glial cells, hormonal background, and the ability for neuroplasticity.

Despite living fruit flies having pain receptors and reacting to negative stimuli, the virtual model is deprived of the ability to feel pain or fatigue.

The "dopamine release" in projects like Stonkfly is only a mathematical variable in the program code, not a real chemical substance.

The real scientific value of the FlyWire and MaleCNS projects is not related to internet trends. Since fruit flies share about 75% of genes associated with human diseases, a precise map of their brain will allow researchers to study the mechanisms of diseases such as Parkinson's or Alzheimer's at the level of individual neural circuits.

The connectome structure will also aid in the development of more efficient architectures for artificial intelligence. As for the digital modeling of the human brain, due to its scale (about 86 billion neurons compared to 166 thousand in a fly), this task remains technologically unreachable for now.