Final week, a number of posts a few so-called digital “embodied fly” tore by X, boosted by AI hype accounts and excited commenters who didn’t appear to know what it was they had been enthusiastic about.
The movies got here from San Francisco-based Eon Techniques, which says it’s working towards “digital human intelligence” and claims it desires to construct a full digital emulation of a mouse mind throughout the subsequent two years — a timeline that’s, to place it generously, formidable. Cofounder Alexander Wissner-Gross shared the unique clip publicly, calling it the “world’s first embodiment of a whole-brain emulation that produces a number of behaviors” and hinting at an impending technological singularity. CEO Michael Andregg posted a unique lower, describing it as a “actual uploaded animal.”
And that was it for proof: no detailed strategies, no scientific paper, no unbiased verification, simply movies of what appears to be like like a digital fly strolling round, consuming, and rubbing its legs collectively.
We have uploaded a fruit fly. We took the @FlyWireNews connectome of the fruit fly mind, utilized a easy neuron mannequin (@Philip_Shiu Nature 2024) and used it to regulate a MuJoCo physics-simulated physique, closing the loop from neural activation to motion.
A number of issues I wish to… pic.twitter.com/Qnlu3INs33
— Michael Andregg (@michaelandregg) March 8, 2026
AI-related accounts on X and Reddit amplified the clips and repeated the captions as reality. Predictable drive-by endorsements from the likes of Elon Musk (“wow”), Bryan Johnson (“that is wonderful”), and Peter Diamandis (“it is a residing being…on-line”) added gasoline to the fireplace. Then the content material farms moved in, repackaged the entire thing as “information” celebrating the first-ever mind add and asking “Are people subsequent?” (sure, additionally they referenced The Matrix, and spoiler: We’re not subsequent).
“That is, in our view, an actual uploaded animal.”
The web was abuzz. The proof was nonetheless two quick movies on X. Should you’re going to inform the world you’ve simply hit what could be one of the vital important scientific milestones in human historical past, you higher include receipts.
Andregg tried to supply some readability on X in a thread that was half seize bag of caveats, half vaguely-described scientific phrases, and half concrete-sounding numbers like “91% conduct accuracy.” I’ve sat with that metric for some time, and I nonetheless don’t actually know what it’s presupposed to imply, and I spent an excellent chunk of my grasp’s diploma finding out animal conduct. Nonetheless, he insisted that “that is, in our view, an actual uploaded animal.”
I despatched Andregg a message on LinkedIn asking for extra particulars. He replied with a hyperlink to a weblog submit Eon had simply printed titled “How the Eon Staff Produced a Digital Embodied Fly.” It wasn’t a scientific paper, but it surely was one thing, I assume.
For the specialists The Verge spoke to, that weblog wasn’t practically sufficient, but it surely trod a way more cautious line than the posts on X, in a lot because it doesn’t say “it is a actual fly.” Shahab Bakhtiari, a professor main the techniques neuroscience and AI lab on the College of Montreal, mentioned that whereas preliminary posts “obscured vital” particulars concerning the work, the brand new weblog supplied extra context. “However it arrived a bit late, and stays inadequate to totally validate the claims,” he mentioned. He would have anticipated an in depth technical report that features particulars on issues like software program, code, and simulation environments that will let different scientists reproduce and consider the work.
Alexander Bates, a analysis fellow in neurobiology at Harvard Medical College finding out fly brains, echoed Bakhtiari. He mentioned the group “under-delivered” and that whereas the weblog supplied extra particulars on what the workforce did — they stitched collectively present large-scale tasks like an in depth map of a fly’s mind, a bodily simulation of a fly’s physique, and fashions simulating how these work together in a digital setting — “for a declare of this magnitude, I might count on one thing that ought to spell out the entire strategy in specifics.”
Bates additionally mentioned the digital fly’s conduct must be evaluated in opposition to actual knowledge and “clearly outlined metrics,” including that the 91 p.c determine was nonetheless not defined within the weblog submit. “Additionally, the fly doesn’t fly.”
Bates informed The Verge he understands that “sturdy framing and hype can matter for fundraising,” however pressured Eon’s declare of a “actual uploaded animal” isn’t credible. Aran Nayebi, a professor of machine studying at Carnegie Mellon College, mentioned that the group was “not even shut” to capturing the complete mind of the fly, displaying connections between cells however not essential particulars like neurotransmitters or how sturdy the connections between totally different nerve cells are. The motor system isn’t a “true add” both, he mentioned. “We’re not even faithfully simulating its mind in silico.”
Tremendous, let’s say Eon truly did it. It copied the fly’s mind completely. The entire thing. Each final half. Do we have now a digital fly now?
Sure. No. Possibly. I don’t know, and also you most likely don’t both. Neither does Eon. The weblog conveniently glossed over the weighty definitional questions on the core of the add declare: What, precisely, counts as a fly? Once we consider flies, we don’t consider a bunch of behaviors or neural connections. We consider, nicely, a fly. Is it sufficient to breed a number of fly-like behaviors in a simulation? Does a completely mapped mind in a digital vat depend? Or does “fly” imply the entire, messy organic bundle — a physique, cells, metabolism, and no matter counts as “reminiscence” or discovered expertise over the course of its life?
And that’s the simple model of the issue. The factor on the display isn’t clearly a fly in any respect. It’s a composite of the neural wiring, programming, and different info stitched collectively from a number of totally different animals. That is helpful when modeling, however exactly what organism can we meaningfully declare has been uploaded on this case? Technically talking, it’s additionally a replica, not an add, which comes with apparent and profound implications the hype conveniently skips over: You might make two, or 10, or 10,000 of the “similar” fly. What then?
Usually, I wouldn’t count on a startup to resolve a significant metaphysical drawback — philosophers have been squabbling over that one for hundreds of years — however they’re those saying they’ve received a “actual uploaded animal.”
The specialists I spoke to weren’t even satisfied the time period is smart. Bakhtiari mentioned it’s nonetheless very a lot an “open query” as as to whether a “actual uploaded animal” is even doable. Jonathan Birch, a thinker on the London College of Economics, was blunter: “I don’t assume we should always ever say ‘uploaded animal,’” he mentioned. What Eon is aiming for, he mentioned, is “whole-brain emulation,” leaving the remainder of the animal behind.
“…this fly is acutely aware in a restricted sense, it might probably scent, see, style, and many others.”
That biology is essential to conduct, mentioned Tom McClelland, a thinker on the College of Cambridge. “So at greatest they’ve uploaded among the fly’s thoughts and thereby uploaded among the fly.”
With a while behind his viral submit, I requested Andregg if he stood by his declare. “Sure,” he informed me. The truth is, he went additional: “We [the research group and its academic collaborators] assume this fly is acutely aware in a restricted sense, it might probably scent, see, style, and many others.” (I’ll completely not be going into the entire consciousness factor.) He described the system as a type of “MVP,” minimal viable product, of an uploaded animal, one with “a number of limitations.” I don’t fairly know image a minimal viable fly. It’s a fly, not an app. MVP is tech startup communicate, not science.
Once I went again to Andregg a second time — this time after talking with specialists and relaying their criticism — he nonetheless stood by the unique declare, however with much more caveats. He conceded the work “isn’t an ideal duplicate of a fly,” and added that Eon had by no means mentioned it was. “I don’t consider importing as a binary idea,” he informed me, describing “totally different ranges” of add and admitting that we don’t but know the way a lot biology is required to seize the data that issues. “There may be much more work to be performed to attain the extent of add that we might want for ourselves sometime.”
Comply with subjects and authors from this story to see extra like this in your customized homepage feed and to obtain electronic mail updates.
- AIShut
AI
Posts from this subject will probably be added to your every day electronic mail digest and your homepage feed.
Comply withComply with
See All AI
- ReportShut
Report
Posts from this subject will probably be added to your every day electronic mail digest and your homepage feed.
Comply withComply with
See All Report
- ScienceShut
Science
Posts from this subject will probably be added to your every day electronic mail digest and your homepage feed.
Comply withComply with
See All Science
