Commentary column: The map of connections of 166,700 neurons in the fruit fly is already used to build models in Minecraft and other games — but this is a partial simulation, not a living or conscious copy of a brain
Credit: FlyEM Project Team, HHMI Janelia; Cambridge Connectomics Group; MRC Laboratory of
Editor's note: MaleCNS he is Connectum — a wiring map of the central nervous system — and not a fully functional copy of brain The fly. In the NeuroCraft Fly project, neural dynamics are simulated, and manually selected outputs trigger scripted behavioral programs. There is currently no evidence that the model feels pain, fear, or desire, or has consciousnessThe column below also discusses the possible implications of future developments in the field.
Today's story is one of the strangest, funniest, and scariest I've ever come across. And it may also suggest that we all, ultimately, live insimulation A big one. I mean, in a computer game, or in "The Matrix" that someone else is running. Maybe.
The wiring diagram of the fly
It all started on September 3, when a team of researchers from HHMI Janelia, the University of Cambridge, the Medical Research Council's Laboratory of Molecular Biology in the UK, and Google Research Publish MaleCNS: Mapping the central nervous system of an adult male fruit fly, which includes the brain and the ventral nerve cord. This is a project that has lasted years, during which the researchers painstakingly identified each nerve cell and tried to understand which other cells it communicates with. In total, they mapped 166,700 nerve cells and about 25.6 million connections between them. A historic achievement by all accounts.
Then they uploaded the connection map to the computer.
And why not, actually? Once we have a detailed map of the cells and the connections between them, we can begin to build a computational model of their activity. Developers NeuroCraft Fly They report that their model stores the network of 166,700 neurons on a computer and connects their activity to a game world. This requires about twenty gigabytes of memory and storage, and the system is designed to run on a powerful Mac computer.
And that's it. You can start running a model on a computer based on the map of the connections of the fly's brain.
To be clear: This is not a perfect simulation of the brain. It does not include, for example, all the processes in the supporting cells that are important for the proper functioning of the nervous system. It also does not accurately reproduce the way each cell works, and it does not learn and update over time as a living fly would. It can respond to input with simulated dynamics and pre-selected outputs, which trigger behaviors in the game. It is difficult to seriously say that this model has “consciousness.”
Does the model have feelings?
But maybe he already has feelings? Immediate pain? Temporary discomfort? That possibility is harder to examine.
When activity in units labeled as part of pathways associated with a noxious stimulus triggers an escape program, does anything there feel pain? We can see that the computer character moves away from the source of the stimulus. But what's going on inside, inside the model?
We don't know.
When units that were associated in the original fly with courtship behavior are activated, and the model's output accordingly activates scripted behavior — does this mean that the model feels "desire"?
We don't know.
And while we can’t say for sure, it’s impossible to completely ignore the question of whether future, much more complete and accurate models might experience flashes of feeling: pain, desire, discomfort, or fear. The current model is a very partial reconstruction of a real network, not a replica of a living brain. But the better the reconstructions get, the harder the question will be.
All of this means that we need to be careful and ethical about these simulations until we better understand whether they can have a real internal experience. This is all the more true when we realize that what can be done today with the map of connections of a fly, it may be possible in the future to do the same with a cat, a dog, and even a human. We need to be careful. We need to be responsible.
Then the data and tools of the connection map were published online, and the world started to go crazy.
How to abuse a fly
Anyone who expected the internet community to act responsibly and ethically was far too optimistic. Shortly after the model was published, someone decided that they simply had to let the fly live in the world of מיינקראפט. He connected what happens in the game world to inputs in the computer network, and between outputs selected from it and action plans. This way, the character can fly, groom itself, eat, escape threats, or perform other actions based on activity in the model. This does not mean that the fly “chooses” in the human sense of the word, but activity in the network does affect what happens in the game.
It is worth noting that the project developer said that he used GPT-6 Astra for the benefit of the cause. Because this is what happens when we have artificial intelligence that can help any person do things that previously required a high level of expertise. Everyone starts playing with extraordinary tools, and the results are accordingly.
And do you think it ends there?
Four days after connecting the model to Minecraft, someone else had it fight in Doom. The next day, they connected it to Beat Saber, a virtual reality game in which participants have to hit boxes with swords to the sound of music. Then they decided to connect it to the stock market. The fly’s network activity was used to make decisions about whether to buy or sell cryptocurrencies like Bitcoin, and units related to reward pathways were activated when the model “made a profit.” These were programmed demonstrations, not evidence that the model had learned the games or understood trading.
Does that sound like abusing flies on a computer? There is also a positive side: On September 10, one of the networkers decided it was time to let the model relax a bit, and ran it in a "fly paradise," with fruit trees all around and no predators, virtual monsters, or swords.
Oh, and the next day they changed the model so that it would respond to both male and female cues. This was described online as the fly becoming “bisexual.” In this case, at least, it’s a demonstration that resembles a scientifically feasible experiment, even though it’s not a peer-reviewed study. The person who did it relied on previous research that found that certain neurons that control the fly’s courtship response are activated in response to chemicals released by the female. He changed the model so that the courtship network would also be activated in response to cues attributed to males.
And it worked within the simulation. Thus, according to the amusing description online, a model of a fly was created that was attracted to both males and females.
From network demos to experiments
This last topic is particularly interesting, because it shows us a direction for the immediate future: a world in which people who are not professors and researchers in university labs are able to contribute to the world of science by playing with complex simulations. And, a world in which ordinary people are given the power to run simulations that raise questions about pain, fear, and… sexual arousal in computerized minds.
How long will it take before we can also run models of brains from more advanced creatures in this way? For example, dogs or even humans? And will we treat those models with the same casual approach that we currently see taken towards models based on the nervous system of flies?
And all of this brings us to simulation theory.
Simulation theory
According to simulation theory, we all live in a computerized world. That is, we are all—each and every one of us—essentially a simulation of a brain running on a computer. This does not mean that we do not have inner experience or self-awareness. Quite the opposite: we do, because the simulation is so powerful and comprehensive that it gives us these qualities.
Who created this simulation? Well, according to The simulation argument Published by philosopher Nick Bostrom in 2003, sufficiently advanced civilizations could easily create such simulations. Just as easily as we as humans can, for example, create games like Age of Empires, in which thousands of units fight each other. And even more easily, so that any child could set up such a simulation at home.
Why would advanced civilizations even develop such simulations? Maybe they want to explore their past. Maybe they want to examine how people will behave in certain situations. Maybe they want to do an election poll, so they run a simulation of the entire world to figure out what ballot people will cast in October 2026.
Does it sound crazy that all the enormous computing power required for such a task would be in the hands of a child?
Sure. But the possibility that every person would have a computer that could run a model based on the entire neural network of a fly would have been considered a complete fantasy just thirty years ago. And now we have people connecting such models to virtual gardens of paradise. So you can see for yourself that what we think is inconceivable and impossible today may seem obvious and easy to do to our children or grandchildren.
The simulation theory has traditionally been met with a smile and a sneer. I suspect that the better we can improve our simulations of minds—both flies and more advanced creatures—the more seriously we will take this theory.
In a world where we are entering deeper into the realms of science fiction every year, no idea can be dismissed easily.
So this is what happened on the internet in the last ten days.
How are you?
Questions and Answers
What is a connectom?
A connectome is a map of nerve cells and the connections between them. It describes the structure of the neural network, but does not by itself reproduce all the electrical, chemical, and biological processes that occur in a living brain.
Is NeuroCraft Fly a complete fly brain?
No. The model uses the MaleCNS network of connections, simulated computational dynamics, and hand-picked outputs to trigger behavioral programs in the game world. Its developers themselves emphasize that it does not replicate the physiology or natural behavior of a fly.
Is there evidence that the model felt pain or desire?
No. It is possible to program a response of distancing oneself from a stimulus or to activate pathways related to courtship, but an external response does not prove the existence of an internal experience or consciousness.
What is the scientific value of such demonstrations?
They can help test hypotheses about the relationship between network structure and activity and behavior, and make complex tools accessible to developers and a wide audience. To turn a demonstration into scientific research requires experimental design, appropriate reviews, documentation, and peer-reviewed analysis.
The scientific article
More on the subject on the science website
- The monkey whose brain was copied into a computer — and is undermining the global balance of power
- When can I upload my brain to the computer?
- We may never know if artificial intelligence is conscious, says Cambridge philosopher
- A million inhabitants on Mars, digital consciousness and the singularity: Science puts limits on Silicon Valley's dreams
- What can the fruit fly smell?
For the original publication: [The Fly That Came to Life in Minecraft — The Guide to the Future, Dr. Roy Tsezana](
