A few days ago I posted a short breaking news summary of a spectacular “Wow!” breakthrough posted by Alex Wissner-Gross: a complete digital copy of a fruit fly brain controlling a physics-based body model.
Eon Systems Founder and CEO Michael Andregg posted an X thread to comment on this breakthrough and its significance. "This is, in our view, a real uploaded animal," he said. "We don't know what its experience is - nobody does. But we take the possibility seriously."
This story has been received with interest and even enthusiastic reactions. “It will be exciting to see what works and doesn't work here (both tell us nontrivial things!),” said Anders Sandberg - who, like Wissner-Gross, is an advisor to Eon Systems.
Popular artificial intelligence (AI) podcaster Matthew Berman is enthusiastic. He says that this is absolutely fascinating to think about and even jumps to the simulation hypothesis. “So what does this actually mean?,” he continues. “Have you heard of simulation theory?... It's basically a theory that says everything that we're experiencing is a simulation. Meaning we're all in a big video game.”
Of course this is not a proof of the simulation hypothesis. It is only a preliminary indication that we could one day simulate complex living beings. But then, who’s to say that we are not being simulated ourselves, whatever that means?
Criticism
Not so fast, say others. “Private advisor... and my most important (unheeded) advice was NOT to declare that you ‘uploaded a fly’ unless you had damn good evidence…” says Kenneth Hayworth., President and Co-Founder of the Brain Preservation Foundation. “I take uploading seriously, so please point me to the publication that offers such good evidence.” Robin Hanson, another advisor to Eon Systems, reposted Hayworth's comment. Other experts that I contacted privately expressed similar sentiments.
Eon Systems acknowledges the limitations of this phase of the project. “We can't trace the actual motor neurons because the body was not scanned,” says Andregg. “However we do know what the brain does when it wants to move in certain ways and that's what we connected to the NeuroMechFly.” In fact, the most evident limitation of this phase of the project is that… the fly doesn’t fly! It only flaps its wings on the ground.
“This is a real limitation of the FlyWire connectome, which is why we plan to scan both the brain and the body,” continues Andregg.
A technical writeup
Eon Systems then posted a more detailed technical update. This “is still very much a work-in-progress,” they say, “and a first step towards showing how an embodied brain would control a virtual body.” While public discourse often frames such milestones as mind uploading, Eon’s technical breakdown clarifies that this is a sophisticated research integration designed to explore how brain structure alone dictates behavior.
The foundation of the project is an integration of several massive datasets and computational frameworks. The brain is based on the FlyWire connectome, which is a map of approximately 140,000 neurons and 50 million synaptic connections. Eon utilizes a Leaky Integrate-and-Fire model that simulates neural signal propagation based on the physical wiring of the fly. This model is further refined by neurotransmitter predictions to determine whether specific connections are excitatory or inhibitory. To give this brain a physical presence, Eon utilizes NeuroMechFly v2, a high-fidelity virtual body model. This digital insect features an anatomically accurate mesh derived from X-ray microtomography and comprises 87 independent joints. The simulation runs on the MuJoCo physics engine, allowing the fly to interact with a virtual environment where gravity, friction, and physical contact are strictly enforced.
The core achievement of the experiment, Eon says, is closing the loop between perception and action through a four-step process. First, virtual stimuli such as sugar or dust activate specific sensory pathways. Second, this activity propagates through the 140,000-neuron connectome. Third, activity in descending neurons, which act as the brain’s output handles, is translated into physical instructions. Finally, the body moves, which in turn alters the sensory input as the fly moves closer to a food source, starting the cycle anew every 15 milliseconds. By piping visual data and taste cues into the connectome, the team demonstrated complex autonomous behaviors. The digital fly can sense invisible taste cues to navigate toward food, recognize when fictive dust has accumulated to trigger a grooming response, and utilize visual motion pathways to navigate its environment.
One of the most technical aspects of the update involves how the brain actually controls the body using a sparse interface of Descending Neurons. Eon uses a car analogy to explain this relationship, suggesting that rather than simulating every internal combustion event of the entire motor hierarchy, they treat specific neurons as control handles for high-level commands like forward velocity, steering, and grooming. These signals are then interpreted by lower-level controllers that manage the complex joint torques required for stable walking. This approach allows the connectome's structure to drive meaningful behavior without requiring a perfect simulation of every muscle fiber.
Eon is careful to manage expectations regarding the upload narrative by highlighting several critical simplifications. The neuron fidelity is currently a point neuron simulation that lacks the dendritic complexity, metabolic states, and biophysical channel diversity of real neurons. Furthermore, the current emulation lacks the software of biology, such as learning, memory, and hormonal changes. While a real fly behaves differently based on hunger or mating states, Eon’s fly is currently a hard-wired machine. The interface is also relatively sparse because a biological fly has over 1,000 Descending Neurons, while Eon’s model currently utilizes only a small subset, meaning the repertoire of movement is still a low-dimensional approximation of true biology.
Despite these limitations, Eon views the project as a qualitative threshold. Historically, researchers have modeled brains without bodies or animated bodies using artificial intelligence without biological brains, but this project unites the two. Eon’s ultimate mission is to scale this technology as they amass data for a mouse brain emulation, which contains roughly 70 million neurons. They argue that by perfecting the structure-to-behavior pipeline in insects, they are laying the groundwork for eventual human-scale emulation. The post concludes by inviting the academic community to help define fidelity frameworks for future emulations, signaling that while the ghost may not be fully in the machine yet, the framework for it is now being built.
Cryonics for uploaders
In reply to a question about his “predictions for the timeline of modeling the full human brain connectome,” Andregg says: “Can't share too much right now but it really depends on funding. I think it's a massive microscope scanning challenge + image processing (stitching, tracing, segmentation, proofreading) but a known engineering problem. The biggest uncertainty is the neuron modeling part.”
I think the next steps could be: first, scanning the full body of the fly and repeating this experiment with full links between the simulated brain and the virtual body. The fly could then fly! Second, repeating the experiment with a physical robotic body.
Andregg clarified that the brain was preserved with aldehyde fixation, linking to a LessWrong post titled “Less Dead” by Aurelia Song (formerly Robert McIntyre). Song is the founder and CEO of Nectome, an advisor to Eon Systems, and a pioneer of the aldehyde fixation method that won the Brain Preservation Foundation's Small Mammal Brain Preservation Prize and Large Mammal Brain Preservation Prize.
I think this fruit fly story adds to the available evidence (without proving, at least not yet) that this brain preservation method works.
“Uploading is still years out,” says Andregg. “But preservation is step one.” He seems right in the sense that, if today’s brain preservation technology can store brains in a way that really preserves all relevant information for future uploading, then step one would one day be followed by step 2: mind uploading. I refer to this brain preservation method as “cryonics for uploaders,” and so does Hayworth.
Hayworth has then posted a longer statement. At the end of a long road, scientists “will at long last make the true statement ‘we’ve uploaded a fruit fly,’” he says.”What EON Systems’ misleading video and claim has done today is to try to steal that future victory and take its valor for their own.”
My reply: Saying that the fly was “uploaded" is overhype - I don't disagree on this point. However, to me the very fact that this experiment has been tried and even produced some very preliminary but essentially encouraging results is significant and worth celebrating.
However, Hayworth reiterated what is, I think, the really important point: “Of course, we can preserve human brains today.” If (an if worth repeating) today’s brain preservation technology can store brains in a way that really preserves all relevant information for future uploading, then the road to mind uploading is already open to all of us - or, more precisely, to those who can afford it.