It's a Digital Fly, But Not What You Think

A significant amount of excitement has recently circulated, yet tangible evidence remains scarce.

Last week, social media platform X was inundated with posts concerning a purported virtual "embodied fly." These posts gained considerable traction, fueled by AI hype accounts and enthusiastic commentators who appeared to lack a clear understanding of the subject matter.

The videos originated from Eon Systems, a San Francisco-based company articulating a vision for "digital human intelligence." The firm asserts its intention to construct a complete digital emulation of a mouse brain within the next two years—a timeline that is, to phrase it charitably, exceptionally ambitious. Cofounder Alexander Wissner-Gross publicly shared the initial clip, proclaiming it the "world’s first embodiment of a whole-brain emulation that produces multiple behaviors" and alluding to an imminent technological singularity. CEO Michael Andregg released a separate version, characterizing it as a "real uploaded animal."

This constituted the entirety of the presented proof: no comprehensive methodologies, no scientific papers, and no independent verification. What was offered were merely videos depicting what appeared to be a digital fly engaging in activities such as walking, eating, and rubbing its legs together.

One notable post declared: "We've uploaded a fruit fly. We took the @FlyWireNews connectome of the fruit fly brain, applied a simple neuron model (@Philip_ShiuNature 2024) and used it to control a MuJoCo physics-simulated body, closing the loop from neural activation to action. A few things I want to…"

AI-focused accounts across X and Reddit extensively amplified these clips, presenting the accompanying captions as undisputed facts. Predictable endorsements swiftly followed from influential figures such as Elon Musk ("wow"), Bryan Johnson ("this is amazing"), and Peter Diamandis ("this is a living being…online"), further intensifying the fervor. Subsequently, content farms proliferated, repackaging the material as "news" celebrating the "first-ever brain upload" and provocatively questioning, "Are humans next?" (These narratives also invoked "The Matrix," though, spoiler alert, humans are not next).

"This is, in our view, a real uploaded animal."

The internet was abuzz with speculation. However, the supporting evidence remained confined to two brief videos on X. When a claim of such monumental scientific significance—potentially one of human history's most profound milestones—is made, it necessitates substantial and verifiable proof.

Andregg attempted to provide clarification via an X thread, which proved to be a heterogeneous mix of caveats, vaguely described scientific terminology, and seemingly concrete figures like "91% behavior accuracy." Despite my own background in animal behavior, having dedicated a significant portion of my master's degree to its study, the precise meaning of this metric remains elusive. Nevertheless, he consistently maintained that "this is, in our view, a real uploaded animal."

I contacted Andregg on LinkedIn seeking further details. His response was a link to a recently published Eon blog post titled "How the Eon Team Produced a Virtual Embodied Fly." While not a peer-reviewed scientific paper, it at least offered some form of explanation.

For the experts interviewed by The Verge, the blog post was far from sufficient, though it adopted a considerably more cautious tone than the initial X posts, notably refraining from declaring it a "real fly." Shahab Bakhtiari, a professor leading the systems neuroscience and AI lab at the University of Montreal, observed that while the initial posts "obscured critical" details, the new blog offered more context. "But it arrived a bit late, and remains insufficient to fully validate the claims," he stated. He would have anticipated a detailed technical report encompassing specifics on software, code, and simulation environments, enabling other scientists to reproduce and evaluate the work.

Alexander Bates, a research fellow in neurobiology at Harvard Medical School specializing in fly brains, echoed Bakhtiari's sentiments. He characterized the group's efforts as "under-delivered." While the blog shed more light on their methodology—integrating existing large-scale projects such as a detailed map of a fly's brain, a physical simulation of a fly's body, and models to simulate their virtual interactions—Bates emphasized, "for a claim of this magnitude, I would expect something that should spell out the whole approach in specifics."

Bates further asserted that the virtual fly's behavior should be rigorously evaluated against real-world data using "clearly defined metrics," noting that the 91 percent figure remained unexplained in the blog post. He added pointedly, "Also, the fly does not fly."

Bates informed The Verge that while he understands "strong framing and hype can matter for fundraising," he stressed that Eon’s assertion of a "real uploaded animal" lacks credibility. Aran Nayebi, a professor of machine learning at Carnegie Mellon University, contended that the group was "not even close" to comprehensively capturing the fly's entire brain. He noted that while connections between cells were shown, crucial details like neurotransmitters or the strength of these connections were absent. Nayebi also stated that the motor system was not a "true upload," concluding, "We are not even faithfully simulating its brain in silico."

Even if one were to concede that Eon had perfectly copied the entirety of a fly's brain, every single component, the question remains: do we then possess a digital fly? The answer is elusive—yes, no, maybe. Neither I, nor likely you, nor even Eon itself, can definitively say. The blog conveniently sidestepped the profound definitional questions central to the "upload" claim: What precisely constitutes a "fly"? Our intuitive understanding of a fly extends beyond a collection of behaviors or neural connections; we envision the creature itself. Is the reproduction of a few fly-like behaviors in a simulation sufficient? Does a fully mapped brain in a virtual environment truly qualify? Or does "fly" inherently imply the complete, intricate biological entity—a body, cells, metabolism, and the entirety of its memory and learned experiences over its lifespan?

And that represents the simpler facet of the problem. The entity depicted on screen is not overtly a fly at all. It is a composite of neural wiring, programming, and other information meticulously assembled from multiple distinct animals. While this approach is valuable for modeling, it raises fundamental questions about which specific organism can genuinely be claimed to have been uploaded. Technically, it's also a copy, not an upload, which carries obvious and profound implications conveniently omitted by the prevailing hype: one could hypothetically create two, ten, or ten thousand identical "flies." What then are the implications for identity and individuality?

Ordinarily, one would not expect a startup to resolve a fundamental metaphysical dilemma—a question philosophers have debated for centuries—but Eon is the entity asserting the creation of a "real uploaded animal."

The experts I consulted were unconvinced that the term itself holds meaning. Bakhtiari suggested it remains an "open question" whether a "real uploaded animal" is even a possibility. Jonathan Birch, a philosopher at the London School of Economics, was more direct: "I don’t think we should ever say ‘uploaded animal,’" he declared. He clarified that Eon's objective, in his view, is "whole-brain emulation," which inherently leaves the rest of the animal behind.

"…this fly is conscious in a limited sense, it can smell, see, taste, etc."

Biological factors are crucial to behavior, as noted by Tom McClelland, a philosopher at the University of Cambridge. He concluded that "at best they’ve uploaded some of the fly’s mind and thereby uploaded some of the fly."

After a period following his viral post, I revisited Andregg to inquire if he still stood by his initial claim. "Yes," he affirmed. Moreover, he expanded on his assertion: "We [the research group and its academic collaborators] think this fly is conscious in a limited sense, it can smell, see, taste, etc." (I will refrain from delving into the broader topic of consciousness here.) He characterized the system as a form of "MVP," or minimum viable product, of an uploaded animal, acknowledging it possessed "lots of limitations." The concept of a "minimum viable fly" is perplexing; a fly is an organism, not a software application. "MVP" is the parlance of tech startups, not scientific discourse.

Upon a second engagement with Andregg—this time after consulting with experts and conveying their criticisms—he reiterated his original claim, albeit with even more caveats. He conceded that the work "isn’t a perfect replica of a fly" and maintained that Eon had never claimed it was. "I don’t think of uploading as a binary concept," he explained, describing "different levels" of upload and admitting that the precise amount of biology required to capture essential information remains unknown. "There is a lot more work to be done to achieve the level of upload that we may want for ourselves someday."