But its world isn’t a laboratory. It’s Minecraft.

And what makes it fly isn’t a game script, but a complete connectome of a real male fruit fly’s central nervous system. 166,700 neurons fire inside the simulation, with neural activity directly determining every turn the virtual fly makes.

The brain map, known as MaleCNS v1.0, is the product of a decade-long effort by HHMI Janelia, the University of Cambridge, and Google Research. The project alone required 44 person-years of manual proofreading. The findings were published in the prestigious journal Cell on September 3. Just two days later, the decade of research had been dropped into a video game.

The person who made it happen wasn’t from a frontier research lab. He was Evan Smith, a graduate student at Georgia Tech.

His only assistant?

GPT-6 Astra.

A decade of painstaking research was turned into a simulation mod for a sandbox game in less than a week. From the connectome becoming publicly available to getting the code running, it took him just two days.

An Open-Source Brain

MaleCNS is the first complete connectome of the male fruit fly’s central nervous system. It covers the central brain and optic lobes, as well as the ventral nerve cord, the functional equivalent of a spinal cord in the fly.

Most importantly, the neural connections through the neck have been preserved in full. For the first time, the pathway showing how the brain controls the body can be traced end to end on a single map.

Containing 125 million synaptic connections, the connectome is the largest brain map to date by neuron count.

Researchers first sliced a fruit fly into millions of ultrathin sections and scanned them one by one using electron microscopy. Google then deployed its flood-filling networks and PATHFINDER system to reconstruct the enormous collection of 2D slices into 3D neuronal structures. Janelia experts then carried out extensive manual proofreading and annotation.

After a decade of work and contributions from three leading institutions, what was ultimately delivered to the public was surprisingly accessible: the entire dataset is fully open source under a CC BY license.

All it takes is registering for a token, running a single pip install command, and writing a few lines of code to pull the upstream and downstream connections of any neuron onto a local machine.

A brain containing 125 million synapses has effectively become an open-source file that anyone can download.

Researchers also identified 262 male-specific cell types and 114 cell types with sex-dependent morphological differences. These differences are concentrated in higher-order brain regions. They not only help explain sex-specific behaviors such as courtship and aggression, but also provide evidence that relatively small changes in neural circuits can produce effects across the entire brain.

AI Lowers the Final Barrier

Turning a static wiring diagram into a virtual fly that can actually fly requires three additional engineering layers: neuron dynamics, sensory inputs, and motor outputs.

Using GPT-6 Astra, Evan Smith reduced the engineering workload to something a graduate student could tackle in their spare time.

The significance becomes clearer when you look at the history of whole-brain simulation.

The most famous example, OpenWorm, involves a nematode with just 302 neurons. Scientists and engineers spent more than a decade building an open-source community around the project before finally getting a virtual worm to move inside a simulator.

This time, the fruit fly has more than 500 times as many neurons as the worm.

And the setup?

One graduate student, one AI, and one computer capable of running Minecraft.

Ten years of scientific research. Two days to turn it into something that actually works.

The wall between frontier science and independent experimentation has rarely been thinner.

Evan Smith left three short notes in his original post: V1 is still under development, GPT-6 Astra was used to build it, and the code and mod are coming soon.

The comments exploded.

Some called it “The Matrix, but for fruit flies.” Others joked that simulation theory might finally be real. And some felt sorry for the virtual fly: Imagine being a simulated fruit fly that can’t smell anything.

Pietro Schirano, a former OpenAI designer, left a comment adapted from the famous passage by Zhuangzi:

“Now I don’t know whether I once dreamed I was a fly, or whether I am now a fly dreaming that I am a human.”

Perhaps the biggest impact of frontier AI on scientific research isn’t that it can directly discover the truth for scientists. It’s that it is dramatically lowering the barrier to who gets to test the frontier in the first place.

The little fly in its blocky Minecraft world may have successfully executed the physical instructions of a nervous system.

Whether it can dream, however, remains unknown.

But at least this time, anyone who wants to keep searching for the answer may no longer need the backing of a world-class laboratory.