The piece I'm not quite understanding is. To what extent is this really simulating the brain accurately?
It appears to behave accurately but it seems from reading the background which is largely above my head that those behaviours are in effect implementations not directly from data.
I.e. a pattern in the neural simulation is interpreted as groom. And then a groom behaviour is in effect played back
Walk pattern recognition is translated into walking (effectively this is a animation).
Does it matter? I'm not sure but I think it's the difference between playing a running animation or running in QWOP
They've basically taken three separate models, one for fly vision, one for the fly brain, and one for the fly body, and bolted them together Frankenstein style.
They've taken the connectome, which is a map of how neurons in the brain are connected to each other, and then created a fly brain using artificial spiking neurons connected together using that same connectome.
So the neurons are not remotely accurate. The interesting point they're making is that even with these simplified neurons they still see plausible behaviours (i.e. simulate the presence of sugar by stimulating a gustatory neuron -> neurons associated with lowering the proboscis for feeding are triggered). So they make the case that a lot of information is encoded simply in the connectome.
The body isn't connected up to the brain in the way we'd expect. "Input" comes from a completely separate neural network which they've trained to simulate appropriate CNS neurons, and output is looking at "descending" (efferent I guess) neurons in a very basic way. It's not completely playing an animation, but the level of connectivity is very low dimensional. It's not clear how much control they have, but for example I imagine they have a spiking threshold for the proboscis below which it's lowered and above which it's raised, which is sort of like you being able to stick your tongue completely out or pull it completely in but nothing else.
So it's not especially bioplausible. The most interesting part is that leaky-integrate-and-fire connectome-based brain model which they're using, even though it is also very limited (for example, it doesn't learn).
Demo looks very cool though. Much credit for being so explicit about what's going on in it in that post. And I was immediately filled with ideas about what they could do next to improve it, which to me is a signal of good research.
There is sooo much hype around this, when the Eon team did almost nothing new compared to the published research. (For context I did my PhD in a fly motor control lab and still follow this field. My colleagues are authors on the research papers Eon used.)
I want to highlight some limitations of the current state of research as well, based on questions that many neuroscientists in this field are struggling with.
- The movement of the legs is not modeled at a fine level, just whether fly is moving forward or turning. This is because we don't have great data on fly leg movements on all these situations, and ventral nerve connectome is still in progress
- By the way the brain connectome still has a lot of errors and needs more proofreading. Also the identity of many neurotransmitters and synaptic strength of connections is unknown. Many are identified through our knowledge of fly genetics which won't translate to humans. In current research, scientists add some finetuning to match some behavior to account for those unknowns.
- There's definitely fly behavior data at the level of making decisions, but not much at level of limb kinematics. Even where data is available, it's unclear how to evaluate the simulated fly against the data. How do you know you got it right?
When I saw the Eon announcement, I was curious how they tackled these challenges. Seems like they didn't. It looks like they forked a few research repositories and vibe coded something to combine them.
I'll give them props for the videos and marketing though, it's crazy to see so many people interested in this research field!!
> The fly body is not currently driven by the full downstream motor hierarchy of the biological fly. Instead, we use a small number of descending outputs as a practical interface between the connectome model and the biomechanics
> [...] Steering in our model is driven through the neurons DNa01 and DNa02 (Yang et al., 2024), which are implicated in turning. Forward velocity is modeled by activation of oDN1
It seems that currently only WASD control is working. But even that is impressive! This is essentially an NPC driven by a real connectome.
I had exactly the same question. In the linked tweet the CEO claims it's not an animation. But in the article they imply certain animations are played when a specific signal is detected.
I think a full simulation without these precreated animations would be more convincing that they actually fully simulated a fly's behavior. It's very easy to make it look like real behavior using animations, videogames do exactly that.
I saw a short clip on this project the other day and was thinking along the same lines! They animated the flies’ mouth part (proboscis?) lowering and I was wondering if they were truly reading the motor commands for that from the brain.
I think the fact that it's chunks of exist and work plugged together is what makes it so impressive to me in the sense that it means that we're actually building somewhat reusable modular architecture for doing this that people can actually repeat and build on separately and so on. And thanks for the future, I think. Also, I have to say, excellent article.
There is a level where the criticisms of the neuroscientists are both entirely legitimate and at the same time probably not really valid.
Again, to go to an SF reference, the Australian hard-SF writer and mathematician Greg Egan has gone into this at some depth. I can't call which story to mind now, but he imagined a scenario in one of his where the tech is available to do a full whole-body in-silico emulation of a human: every dendrite of every brain cell, every action potential propagating along it, every neurotransmitter diffusing across every synapse.
And what the people running the simulations AND THE PEOPLE BEING SIMULATED discover is that you don't need it most of the time.
In the story, doing the neocortex of the brain and a coarser sim of the underlying structures is enough to support full consciousness. For the peripheral nervous system, an even lower-level sim is enough: your limbs still feel right. For the sympathetic nervous system, you don't bother at all -- just simulate overall excitation levels.
Don't bother doing whole muscles, as most people aren't consciously aware of them anyway, even when running or doing sport -- any more than we're aware of breathing.
You downgrade the sim of everything except the important bits to a coarse low-res approximation and most of the time you wouldn't be able to tell -- but it's much faster and takes much less CPU time, so the same compute substrate can simulate more people faster for a given amount of resources.
No, this is not a full simulation of all the nerve cells in a fly brain, but it seems like it does more or less what a fly would do anyway. It seems quite possible that for a fly, a coarse low-level generalised simulation might be enough to produce something that walks like a fly, feeds like a fly, and maybe flies like a fly and breeds like a fly.
I don't think a fly "knows" which leg to move when walking. I suspect a horse doesn't. I barely do, and I only have 2 of them.
Crude abstracted low-level sims with the right structure might be enough to get the desired behaviour and have a model that's good enough that its behaviour is indistinguishable from the original.
In human terms... if we ever get to the point that we can simulate a brain in a jar, if the sim "lies" to the brain and tells it a body is there, and is doing the normal body stuff and walking and swinging its arms and whatever, that might be enough to "feel real" to the mind in the simulated body.
Do it for an athlete or sports player or a gymnast or a bodybuilder and they'd notice. They'd know. But most of us never would.
You can't feel your individual toes unless you stub one. So don't simulate them.
A low-res fly brain model connected to a lower-res overall ventral nerve cord connected to a trivial fly muscle body which doesn't even simulate individual muscles might be enough that the fly does everything a fly can do.
I think there's another layer to this too, which is that — if you've already scanned and uploaded a person's mind, the original person is presumably dead, or at least a completely separate person from the one in the simulation. Moreover, really, no matter how much versimilitude you pour into your simulation, it's never really going to be exactly the same. Hence, if we're being clear-eyed about this, presumably the point of doing a brain upload like that isn't to get immortality for a particular person or something, its just to get a much more advanced and human-like form of artificial intelligence.
At which point, if we don't need a perfectly accurate neurological and physical simulation to get something that mostly "walks," talks and acts like a human, or something very similar, why would we bother doing it at all?
And yeah, although you could obviously assume that perfect simulation of all of the physical and chemical and neurological processes is necessary to get something to even slightly functions, considering the fact that people can function with 90% of their brain entirely gone, I'm not sure how accurate that is.
And yet I'm not exactly the same person as the one who hadn't yet read your comment, nor the same as the one an hour ago, not the one that woke up this morning.
Maybe exactly the same isn't the right metric after all.
This is true. But I'm talking about fundamental substrate and the rules by which it operates, not just the flow of time — that's probably both a qualitative and quantitative difference. Additionally, even if it's not about being the exact same, although you could view brain uploading as a continuation of the species, in some sense, then — you still couldn't view it as a continuation of the same individual... because that guy/gal is right there!
It might well be, yes -- it's a long time now, I have read all his published novels except maybe in the last couple of years, and I am not sure which one is which.
> and while I am not a neuroscientist, I tried to explain this is just a combination of chunks of existing work
Which most complex technology is these days, even the most impactful. Docker is a fancy wrapper around Linux cgroups, Kubernetes combines that with etcd, and stuff like AWS EKS combines that with hypervisors (which I think were based on Linux KVM for a long time).
Have a look at this comment from Ken Hayworth, a highly respected scientist in the field, which I am copying here from a twitter post by him (https://x.com/KennethHayworth/status/2032604687212392562). I also just came back from the Cosyne 2026 conference, and the work was unfortunately not met with great enthusiasm, despite the media attention: My statement regarding the misleading EON Systems “fly upload” video:
The hundreds of researchers who make up the Drosophila neuroscience community are making good progress toward eventually understanding how the intelligent behaviors of a fruit fly are produced by computations in its neural circuits. Obtaining the structural connectome of the fly brain and ventral nerve cord was a significant milestone in that quest, as was obtaining an estimate of neurotransmitter types for each cell type. What is currently most lacking is a catalog of the precise electrophysiological and molecular dynamics of each neuron and synapse type. Dozens of on-going electrophysiological, genetic, and behavioral experiments are beginning to fill in those details. But completing that task will likely take many years, possibly decades, of more research. At the end of that long road, I have no doubt, there will be a detailed paper, published in a high-quality journal with full details and carefully peer-reviewed, which will at long last make the true statement “we’ve uploaded a fruit fly”. And that future paper will have a supplementary video much like the EON Systems one, showing a fly navigating a virtual environment. But, unlike the misleading EON Systems video, that future video will be real… all 100,000+ neurons displaying dynamics that reflect those that would occur in the real fly engaged in the same sensory-motor behaviors. That paper will represent the crowning achievement of a successful Drosophila neuroscience field.
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, all in the hopes of raising some cash from naive investors who think they might get to human uploads soon, and all while riding a tide of hype they generated in the gullible public. The result has been a wave of secondary reporting that grossly mischaracterizes the current state of neuroscience progress, implying that it is much further along than it currently is.
As a member of the Drosophila research community, and as a long-term advocate of brain preservation for eventual mind uploading, I feel it is my responsibility to call out this reprehensible behavior. Neuroscience technology is progressing fast enough that we are now able to obtain structural connectomes of small organisms like the fruit fly. But neuroscience understanding is progressing much more slowly. True uploading, even for a fruit fly, is likely years to decades away. Even obtaining a mouse connectome seems likely to be a decade or more away. Human uploading is simply not on any reasonable research or investment timeline, unless such a timeline includes many decades of methodical basic neuroscience research. Of course, we can preserve human brains today using aldehyde fixatives as is done in all of today’s connectomics studies. But we will not be able to upload a human brain for many decades, perhaps centuries to come.
Please do not let today’s real scientific progress in connectomics and brain preservation be drowned out by misleading hype.
This seems petty. Basically he's annoyed that a commercial entity made a video based on published research rather than an academic group.
As far as I can tell the blog post does a good job of citing sources - they go over and above what a commercial entity is required to disclose (I.e. nothing). No good deed goes unpunished.
I thought the whole point of academia was to do research for the benefit of society. Research is published so that society can make use of it. Not to give academics "thunder".
If you don't want anyone to read your papers there's a simple solution: don't publish them.
Look at it from Ken's perspective though. He's one of the few people who have moved this field forward by solving very hard problems over decades.
Now a startup comes in and publishes a cool video that claims to solve a big chunk of what he wants to do / contribute to over the rest of his career, like "hey that was easy!".
He says the video is very misleading, and that's just a fact.
It's hard to see for someone outside of this how insulting that can feel to someone in his position, and that's why it can come across as petty.
Also, your summary of his criticism isn't quite right, read it more carefully. It would apply equally if an academic group had done that. It's more a case of academic standards for communication meeting silicon valley bluster.
Petty? He's accusing them of fraud, and if he's right then yeah we should all be disappointed in Eon's deceptive marketing.
Fully simulating a drosophila has been a high goal for a very long time and great claims require great proof, but Eon has been stingy with the details (and no this blog post does not reveal much beyond chaining together lots of impressive sounding words).
Imagine the skepticism on HN if someone declared they invented AGI. Similar level claim.
The piece I'm not quite understanding is. To what extent is this really simulating the brain accurately?
It appears to behave accurately but it seems from reading the background which is largely above my head that those behaviours are in effect implementations not directly from data.
I.e. a pattern in the neural simulation is interpreted as groom. And then a groom behaviour is in effect played back
Walk pattern recognition is translated into walking (effectively this is a animation).
Does it matter? I'm not sure but I think it's the difference between playing a running animation or running in QWOP
They've basically taken three separate models, one for fly vision, one for the fly brain, and one for the fly body, and bolted them together Frankenstein style.
They've taken the connectome, which is a map of how neurons in the brain are connected to each other, and then created a fly brain using artificial spiking neurons connected together using that same connectome.
So the neurons are not remotely accurate. The interesting point they're making is that even with these simplified neurons they still see plausible behaviours (i.e. simulate the presence of sugar by stimulating a gustatory neuron -> neurons associated with lowering the proboscis for feeding are triggered). So they make the case that a lot of information is encoded simply in the connectome.
The body isn't connected up to the brain in the way we'd expect. "Input" comes from a completely separate neural network which they've trained to simulate appropriate CNS neurons, and output is looking at "descending" (efferent I guess) neurons in a very basic way. It's not completely playing an animation, but the level of connectivity is very low dimensional. It's not clear how much control they have, but for example I imagine they have a spiking threshold for the proboscis below which it's lowered and above which it's raised, which is sort of like you being able to stick your tongue completely out or pull it completely in but nothing else.
So it's not especially bioplausible. The most interesting part is that leaky-integrate-and-fire connectome-based brain model which they're using, even though it is also very limited (for example, it doesn't learn).
Demo looks very cool though. Much credit for being so explicit about what's going on in it in that post. And I was immediately filled with ideas about what they could do next to improve it, which to me is a signal of good research.
There is sooo much hype around this, when the Eon team did almost nothing new compared to the published research. (For context I did my PhD in a fly motor control lab and still follow this field. My colleagues are authors on the research papers Eon used.)
I want to highlight some limitations of the current state of research as well, based on questions that many neuroscientists in this field are struggling with.
- The movement of the legs is not modeled at a fine level, just whether fly is moving forward or turning. This is because we don't have great data on fly leg movements on all these situations, and ventral nerve connectome is still in progress
- By the way the brain connectome still has a lot of errors and needs more proofreading. Also the identity of many neurotransmitters and synaptic strength of connections is unknown. Many are identified through our knowledge of fly genetics which won't translate to humans. In current research, scientists add some finetuning to match some behavior to account for those unknowns.
- There's definitely fly behavior data at the level of making decisions, but not much at level of limb kinematics. Even where data is available, it's unclear how to evaluate the simulated fly against the data. How do you know you got it right?
When I saw the Eon announcement, I was curious how they tackled these challenges. Seems like they didn't. It looks like they forked a few research repositories and vibe coded something to combine them.
I'll give them props for the videos and marketing though, it's crazy to see so many people interested in this research field!!
> The fly body is not currently driven by the full downstream motor hierarchy of the biological fly. Instead, we use a small number of descending outputs as a practical interface between the connectome model and the biomechanics
> [...] Steering in our model is driven through the neurons DNa01 and DNa02 (Yang et al., 2024), which are implicated in turning. Forward velocity is modeled by activation of oDN1
It seems that currently only WASD control is working. But even that is impressive! This is essentially an NPC driven by a real connectome.
I had exactly the same question. In the linked tweet the CEO claims it's not an animation. But in the article they imply certain animations are played when a specific signal is detected.
I think a full simulation without these precreated animations would be more convincing that they actually fully simulated a fly's behavior. It's very easy to make it look like real behavior using animations, videogames do exactly that.
I saw a short clip on this project the other day and was thinking along the same lines! They animated the flies’ mouth part (proboscis?) lowering and I was wondering if they were truly reading the motor commands for that from the brain.
I wrote this up for the Register, and while I am not a neuroscientist, I tried to explain this is just a combination of chunks of existing work...
https://www.theregister.com/2026/03/16/digital_fruit_fly_bra...
I think the fact that it's chunks of exist and work plugged together is what makes it so impressive to me in the sense that it means that we're actually building somewhat reusable modular architecture for doing this that people can actually repeat and build on separately and so on. And thanks for the future, I think. Also, I have to say, excellent article.
Thanks very much indeed!
There is a level where the criticisms of the neuroscientists are both entirely legitimate and at the same time probably not really valid.
Again, to go to an SF reference, the Australian hard-SF writer and mathematician Greg Egan has gone into this at some depth. I can't call which story to mind now, but he imagined a scenario in one of his where the tech is available to do a full whole-body in-silico emulation of a human: every dendrite of every brain cell, every action potential propagating along it, every neurotransmitter diffusing across every synapse.
And what the people running the simulations AND THE PEOPLE BEING SIMULATED discover is that you don't need it most of the time.
In the story, doing the neocortex of the brain and a coarser sim of the underlying structures is enough to support full consciousness. For the peripheral nervous system, an even lower-level sim is enough: your limbs still feel right. For the sympathetic nervous system, you don't bother at all -- just simulate overall excitation levels.
Don't bother doing whole muscles, as most people aren't consciously aware of them anyway, even when running or doing sport -- any more than we're aware of breathing.
You downgrade the sim of everything except the important bits to a coarse low-res approximation and most of the time you wouldn't be able to tell -- but it's much faster and takes much less CPU time, so the same compute substrate can simulate more people faster for a given amount of resources.
No, this is not a full simulation of all the nerve cells in a fly brain, but it seems like it does more or less what a fly would do anyway. It seems quite possible that for a fly, a coarse low-level generalised simulation might be enough to produce something that walks like a fly, feeds like a fly, and maybe flies like a fly and breeds like a fly.
I don't think a fly "knows" which leg to move when walking. I suspect a horse doesn't. I barely do, and I only have 2 of them.
Crude abstracted low-level sims with the right structure might be enough to get the desired behaviour and have a model that's good enough that its behaviour is indistinguishable from the original.
In human terms... if we ever get to the point that we can simulate a brain in a jar, if the sim "lies" to the brain and tells it a body is there, and is doing the normal body stuff and walking and swinging its arms and whatever, that might be enough to "feel real" to the mind in the simulated body.
Do it for an athlete or sports player or a gymnast or a bodybuilder and they'd notice. They'd know. But most of us never would.
You can't feel your individual toes unless you stub one. So don't simulate them.
A low-res fly brain model connected to a lower-res overall ventral nerve cord connected to a trivial fly muscle body which doesn't even simulate individual muscles might be enough that the fly does everything a fly can do.
Agreed.
I think there's another layer to this too, which is that — if you've already scanned and uploaded a person's mind, the original person is presumably dead, or at least a completely separate person from the one in the simulation. Moreover, really, no matter how much versimilitude you pour into your simulation, it's never really going to be exactly the same. Hence, if we're being clear-eyed about this, presumably the point of doing a brain upload like that isn't to get immortality for a particular person or something, its just to get a much more advanced and human-like form of artificial intelligence.
At which point, if we don't need a perfectly accurate neurological and physical simulation to get something that mostly "walks," talks and acts like a human, or something very similar, why would we bother doing it at all?
And yeah, although you could obviously assume that perfect simulation of all of the physical and chemical and neurological processes is necessary to get something to even slightly functions, considering the fact that people can function with 90% of their brain entirely gone, I'm not sure how accurate that is.
And yet I'm not exactly the same person as the one who hadn't yet read your comment, nor the same as the one an hour ago, not the one that woke up this morning.
Maybe exactly the same isn't the right metric after all.
This is true. But I'm talking about fundamental substrate and the rules by which it operates, not just the flow of time — that's probably both a qualitative and quantitative difference. Additionally, even if it's not about being the exact same, although you could view brain uploading as a continuation of the species, in some sense, then — you still couldn't view it as a continuation of the same individual... because that guy/gal is right there!
I believe you’re referring to Egan’s book Permutation City. It’s pretty decent and holds up well today despite being written in 1994.
https://en.wikipedia.org/wiki/Permutation_City
It might well be, yes -- it's a long time now, I have read all his published novels except maybe in the last couple of years, and I am not sure which one is which.
> and while I am not a neuroscientist, I tried to explain this is just a combination of chunks of existing work
Which most complex technology is these days, even the most impactful. Docker is a fancy wrapper around Linux cgroups, Kubernetes combines that with etcd, and stuff like AWS EKS combines that with hypervisors (which I think were based on Linux KVM for a long time).
Have a look at this comment from Ken Hayworth, a highly respected scientist in the field, which I am copying here from a twitter post by him (https://x.com/KennethHayworth/status/2032604687212392562). I also just came back from the Cosyne 2026 conference, and the work was unfortunately not met with great enthusiasm, despite the media attention: My statement regarding the misleading EON Systems “fly upload” video:
The hundreds of researchers who make up the Drosophila neuroscience community are making good progress toward eventually understanding how the intelligent behaviors of a fruit fly are produced by computations in its neural circuits. Obtaining the structural connectome of the fly brain and ventral nerve cord was a significant milestone in that quest, as was obtaining an estimate of neurotransmitter types for each cell type. What is currently most lacking is a catalog of the precise electrophysiological and molecular dynamics of each neuron and synapse type. Dozens of on-going electrophysiological, genetic, and behavioral experiments are beginning to fill in those details. But completing that task will likely take many years, possibly decades, of more research. At the end of that long road, I have no doubt, there will be a detailed paper, published in a high-quality journal with full details and carefully peer-reviewed, which will at long last make the true statement “we’ve uploaded a fruit fly”. And that future paper will have a supplementary video much like the EON Systems one, showing a fly navigating a virtual environment. But, unlike the misleading EON Systems video, that future video will be real… all 100,000+ neurons displaying dynamics that reflect those that would occur in the real fly engaged in the same sensory-motor behaviors. That paper will represent the crowning achievement of a successful Drosophila neuroscience field.
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, all in the hopes of raising some cash from naive investors who think they might get to human uploads soon, and all while riding a tide of hype they generated in the gullible public. The result has been a wave of secondary reporting that grossly mischaracterizes the current state of neuroscience progress, implying that it is much further along than it currently is.
As a member of the Drosophila research community, and as a long-term advocate of brain preservation for eventual mind uploading, I feel it is my responsibility to call out this reprehensible behavior. Neuroscience technology is progressing fast enough that we are now able to obtain structural connectomes of small organisms like the fruit fly. But neuroscience understanding is progressing much more slowly. True uploading, even for a fruit fly, is likely years to decades away. Even obtaining a mouse connectome seems likely to be a decade or more away. Human uploading is simply not on any reasonable research or investment timeline, unless such a timeline includes many decades of methodical basic neuroscience research. Of course, we can preserve human brains today using aldehyde fixatives as is done in all of today’s connectomics studies. But we will not be able to upload a human brain for many decades, perhaps centuries to come.
Please do not let today’s real scientific progress in connectomics and brain preservation be drowned out by misleading hype.
-Kenneth Hayworth
But Eon's tagline is "Solving brain emulation as an engineering sprint, not a decades-long research program"! How could they have ever gone wrong?
This seems petty. Basically he's annoyed that a commercial entity made a video based on published research rather than an academic group.
As far as I can tell the blog post does a good job of citing sources - they go over and above what a commercial entity is required to disclose (I.e. nothing). No good deed goes unpunished.
I thought the whole point of academia was to do research for the benefit of society. Research is published so that society can make use of it. Not to give academics "thunder".
If you don't want anyone to read your papers there's a simple solution: don't publish them.
Look at it from Ken's perspective though. He's one of the few people who have moved this field forward by solving very hard problems over decades.
Now a startup comes in and publishes a cool video that claims to solve a big chunk of what he wants to do / contribute to over the rest of his career, like "hey that was easy!".
He says the video is very misleading, and that's just a fact.
It's hard to see for someone outside of this how insulting that can feel to someone in his position, and that's why it can come across as petty.
Also, your summary of his criticism isn't quite right, read it more carefully. It would apply equally if an academic group had done that. It's more a case of academic standards for communication meeting silicon valley bluster.
Petty? He's accusing them of fraud, and if he's right then yeah we should all be disappointed in Eon's deceptive marketing.
Fully simulating a drosophila has been a high goal for a very long time and great claims require great proof, but Eon has been stingy with the details (and no this blog post does not reveal much beyond chaining together lots of impressive sounding words).
Imagine the skepticism on HN if someone declared they invented AGI. Similar level claim.
[dead]
I think the research Eon is citing will be seen as a much more important step on the path to AGI than language models.
I'm not an expert in the field, but this reads like someone trying to sound impressive by using big words without providing any solid detail.
So how close are we to Cyberpunk-style braindances?