How the Eon Team Produced a Virtual Embodied Fly

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.

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.

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!!

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.