302 neurons doesn’t sound impressive to people who may be used to working with 7B+ parameter neural networks. But those neural networks have about as much in common with a biological neuron as a bicycle had with a horse. They can both travel pretty fast but one evolved naturally through over a billion years of harsh natural selection, and the other is a precisely tuned metal machine with a single purpose.<p>Neurons are similar, they are incredibly sophisticated biological machines, with billions of DNA base pairs controlling their behavior. The emergent behavior of neurons in both biological and AI systems are pretty fascinating
I was lucky enough to do some programming work, very many years ago, in the 1990s, in the laboratory of Ralph Siegel (<a href="https://en.wikipedia.org/wiki/Ralph_Siegel_(scientist)" rel="nofollow noreferrer">https://en.wikipedia.org/wiki/Ralph_Siegel_(scientist)</a>), who among other things worked on this type of worm connectome models. He used the Hodgkin-Huxley equations to simulate neuron responses on the connectome. The Hodkin-Huxley model, as someone explained to me, is kind of like modeling a human leg as three rigid blocks connected by hinges - it's enough to be useful in many models, but of course it's not a full description. Also, it may not the right model for worm neurons, because worm neurons are non-spiking, and the HH equations describe neurons that produce trains of spikes; they exist in more complicated nervous systems. The HH equations are used in simulations because it's the mathematical model we have, and it seems that they're still used by the OpenWorm project. (I am not very sure about properties of worm neurons, I heard about this a long time ago and the information may be out of date).<p>I think it's great that this work is still going on, it may produce insights about functioning of nervous systems. But the difficulties are fierce, and we're making very slow and difficult progress in an immense unknown area.
I like these bottom up approaches, as they demonstrate very well how much we _don't_ know yet about life. Important to mention here Craig Venters minimal cell project syn3.0, where the team synthetically created a livable cell comprising 473 genes. It was done to a large part with trial and error, the function of many of those genes is still not known. A recent review from the same team is to be found at <a href="https://doi.org/10.1016/j.cell.2022.06.046" rel="nofollow noreferrer">https://doi.org/10.1016/j.cell.2022.06.046</a> .
Noam Chomsky says there are two kinds of AI.<p>1. The first kind of AI research is more like engineering for example creating self driving cars, language translation & object recognition.<p>2. The second kind of AI research is trying to replicate the intelligence of living organisms (humans, worms) with models that are consistent with what cognitive scientist have.<p>An example for such a system is one that would pick up any human language with very little supervision. Like children for example.<p>Open worm seems like no. 2. Any one have any interesting resources for no. 2 type AI? I would love to explore it some more.<p>[0]: <a href="https://youtu.be/TAP0xk-c4mk" rel="nofollow noreferrer">https://youtu.be/TAP0xk-c4mk</a>
I looked at this a few years back. I've always kept some hope that the Kurzwellian eschatology had some validity. You know, first we simulate a C.elegans and once that's done, it's only a matter of scale up before we can simulate a human and then boom singularity sky. And, really, how difficult is it to simulate a silly worm? after all the connectome is there, we know all the neurons and their connections, should be easy.<p>Well, when I looked at it I was shocked: it doesn't work! sure it could replicate some basic movements but many things that the stupid worm actually does where still a mystery. The docs didn't seem like people were close to figure it out either. And sure enough, a few years later seems like they gave up. And afaik that hyped European brain emulation project also folded in the midst of corruption allegations no less.<p>I think we don't understand any of this and we seem very far from it too. I think it's back to science fiction novels for a while.
Reminds me of the first episode of Devs ( <a href="https://www.imdb.com/title/tt8134186/" rel="nofollow noreferrer">https://www.imdb.com/title/tt8134186/</a> ) where an artificial intelligence engineer does a demo about syncing, and then predicting the future movements of a nematode worm.
Great show by the way!
Connectome model of fruit fly brain was just recently published[1].<p>[1] <a href="https://hub.jhu.edu/2023/03/09/scientists-complete-first-map-of-an-insect-brain/" rel="nofollow noreferrer">https://hub.jhu.edu/2023/03/09/scientists-complete-first-map...</a>
I went to an OpenWorm mini-conference of some kind in London a few years back - a couple of days talking about worm locomotion. Knowing nothing about worms or animal locomotion in general, I was a complete layman with an amateur interest in life simulation and understood next to nothing. But it was still fascinating, and was so cool to meet people passionate about this.
As a founding member of OpenWorm, I emulated the connectome and applied it to robotics almost 10 years ago (<a href="https://youtu.be/YWQnzylhgHc" rel="nofollow noreferrer">https://youtu.be/YWQnzylhgHc</a>). In this experiment, each neuron is represented as a single program and uses UDP messaging to communicate between the individual neurons (programs). There was a number of issues I ran into and probably #1 is the fact that connectomes are highly recurrent which in my experiment lead to what I called UDP stacking; i.e. the activations (UDP messages) come in faster than the program can process them.<p>I have downloaded the complete (so far) Drosphilia connectome and working to create an emulation. I am exploring Hypervectors to emulate neurons or neuropils but still working on the concept. I have tried many other ideas including Adjacency Matrices and Function programming where each Function is some part of the animal nervous system (e.g. with C elegans think Sensory, Interneuron and Motor as 3 distinct functions and programs).<p>The C elegans emulation was also done in a single Python app, translated to several other programming languages, that also showed clear emulation in many different robots. Ablation tests demonstrated how the emulated nervous system is congruent to the biological nervous system through observable behaviors of both.<p>There are two extremes on the evolutionary continuum of nervous systems with the worm's brain on one end and the human brain on the other. However, even with the Worm's nervous system, we can clearly see general intelligence and a gateway to AGI.
It's always fun for me to see C. elegans meet software. This is amazing.<p>My undergrad degree capstone project was a flow-based visual C. elegans strain builder[1]. The team worked with two researchers who taught us a lot about genetics and basic C. elegans biology. They are a fascinating model organism, and it was a super fun project to work on. Even though it's got a very small potential userbase, it did <i>have</i> a potential userbase (which was more than you could say about most capstone projects). We used some interesting technology to build it (Tauri[2]: Rust + Web Frontend), learned some biology along the way, and ended up with a great prototype.<p>Since none of the software team had any background in genetics, modeling the data was pretty difficult. We'd meet with researchers, they'd teach us new genetics concept, we'd build our models, then the next week they'd say "OH we forgot to tell you about this caveat", then we'd go back to the drawing board, update the schema (thank heavens for migrations), rinse and repeat. It was a lot of fun though :) I couldn't have asked for much more out of a capstone project.<p>[1] <a href="https://worm-world.github.io/" rel="nofollow noreferrer">https://worm-world.github.io/</a>
[2] <a href="https://tauri.app/" rel="nofollow noreferrer">https://tauri.app/</a>
As an amateur scientist with an interest in neurodegenerative diseases, I am interested in neurons simulations. As a model with neurons and muscles, Openworm looks very interesting for application in ALS (Lou Gherig's disease).
Neat. I'm wondering how much of inner cell biology can be abstracted away.<p>How accurate can neuron simulation be without underlying chemistry and physics?
Fascinating to see this again it's been at least a decade since I first found it. Projects like this always make make me glad to see because it's almost "useless" work that I imagine will be a foot note to some seriously impressive future technology. Very cool work in the mean time. As an aside - this brings back a weird feeling of nosatlgia, but this repository also brought me back to a time where I was stuck in an over-engineered .NET shop and i'm now debating if we start the day off with whiskey after having a flashback to being asked to implement "Ninject" into a massive spaghetti banquet of a code base (collective hope that dep injection fixes bad code).<p>In case anyone is worried I write software in Rust and Golang now and my life has improved significantly since the origins of this worm and people taking dep injection frame works seriously. :D
I love it when anything Caenorhabditis elegans (C. Elegans) related pops up because this little biological organism sits at this beautiful intersection between technology and biology and philosophy. The successful emulation of C. Elegans would represent a concrete step towards whole brain emulation and all the transhuman and ethical and moral quandaries that would bring. The general idea is that the human brain has billions of neurons, Elegans has hundreds (and we've had them mapped since 1986). If one can successfully "upload" Elegans, then humans are just a matter of scale.<p>However, it should be noted that the field, and specifically this line of research, hasn't produced much in the way of results in 10+ years. University of Oregon planned (though I can't tell if they ever developed) NemaSys[0] ~1997. OpenWorm has been exploring this since 2011. Project Nemaload explored it a bit from 2011-2013.[1] But each project ran into three problems:<p>- Knowing the connections isn't enough. We also need to know the weights and thresholds. We don't know how to read them from a living worm.[2]<p>- C. elegans is able to learn by changing the weights. We don't know how weights and thresholds are changed in a living worm.[2]<p>- Funding [3]<p>The best we can do is modeling a generic worm - pretraining and running the neural network with fixed weights. Thus, no worm is "uploaded" because we can't read the weights, and these simulations are far from realistic because they are not capable of learning. Hence, it's merely a boring artificial neural network, not a brain emulation. Relevant neural recording technologies are needed to collect data from living worms, but they remain undeveloped (but in progress?[4][5][6]), and the funding simply isn't there.<p>OpenWorm got the idea to plug their connectome into a Lego robot[7] and got it to exhibit the tap-withdrawal behavior of the nematode, but it had technical limitations preventing easy modification of the connectome or introduction of new models of neural dynamics. JHU Applied Physics Lab extended the work by using a basic integrate and fire model to simulate the neurons and assigned weights by determining the proportion to the total number of synapses the two neurons on either side of the synapses shared and in the end got the simulated worm to reverse direction when bumping into walls.[8] At this point, humanity seems to have abandoned emulated worm driven mechanisms which is honestly kind of a loss.<p>There's no real ending to this comment. Love this project, loves what it stands for, looking forward to seeing progress in this field. And a lot of this information was pulled from this blog post[9] which was also mentioned in the comments somewhere.<p>[0] <a href="https://web.archive.org/web/20030115124331/http://www.csi.uoregon.edu/projects/celegans/" rel="nofollow noreferrer">https://web.archive.org/web/20030115124331/http://www.csi.uo...</a><p>[1] <a href="https://github.com/nemaload">https://github.com/nemaload</a><p>[2] <a href="https://www.jefftk.com/p/we-havent-uploaded-worms" rel="nofollow noreferrer">https://www.jefftk.com/p/we-havent-uploaded-worms</a><p>[3] <a href="https://www.quora.com/Is-Larry-Page-funding-any-neuroscience-mind-uploading-projects-besides-the-NEMALOAD-project" rel="nofollow noreferrer">https://www.quora.com/Is-Larry-Page-funding-any-neuroscience...</a><p>[4] <a href="https://arxiv.org/pdf/2109.10474.pdf" rel="nofollow noreferrer">https://arxiv.org/pdf/2109.10474.pdf</a><p>[5] <a href="https://onlinelibrary.wiley.com/doi/10.1002/cyto.a.24483" rel="nofollow noreferrer">https://onlinelibrary.wiley.com/doi/10.1002/cyto.a.24483</a><p>[6] <a href="https://www.sciencedirect.com/science/article/pii/S0959438820301689" rel="nofollow noreferrer">https://www.sciencedirect.com/science/article/pii/S095943882...</a><p>[7] <a href="https://www.cnn.com/2015/01/21/tech/mci-lego-worm/" rel="nofollow noreferrer">https://www.cnn.com/2015/01/21/tech/mci-lego-worm/</a><p>[8] <a href="https://ccneuro.org/2018/proceedings/1149.pdf" rel="nofollow noreferrer">https://ccneuro.org/2018/proceedings/1149.pdf</a><p>[9] <a href="https://www.lesswrong.com/posts/mHqQxwKuzZS69CXX5/whole-brain-emulation-no-progress-on-c-elgans-after-10-years" rel="nofollow noreferrer">https://www.lesswrong.com/posts/mHqQxwKuzZS69CXX5/whole-brai...</a>
Related:<p><i>OpenWorm</i> - <a href="https://news.ycombinator.com/item?id=29045198">https://news.ycombinator.com/item?id=29045198</a> - Oct 2021 (110 comments)<p><i>OpenWorm – Create a virtual C. elegans nematode</i> - <a href="https://news.ycombinator.com/item?id=8949408">https://news.ycombinator.com/item?id=8949408</a> - Jan 2015 (12 comments)<p><i>OpenWorm: A Digital Organism In Your Browser</i> - <a href="https://news.ycombinator.com/item?id=7613732">https://news.ycombinator.com/item?id=7613732</a> - April 2014 (47 comments)<p><i>Openworm: c.elegans worm simulation</i> - <a href="https://news.ycombinator.com/item?id=4208454">https://news.ycombinator.com/item?id=4208454</a> - July 2012 (14 comments)
I would love to see a webassembly version of this engine so we can see these worms on a webbrowser. Wonder how hard it would be to do so.<p>---<p>edit: Possibly baked already at <a href="http://wormsim.org/" rel="nofollow noreferrer">http://wormsim.org/</a>
That we can run a simulation of an organism that <i>looks and acts like the real thing</i> evokes a rare sense of wonder I thought I could never again experience from technology.
many years ago this was interesting. Nowadays, with the progress in deep models i m not sure if there is much to learn from c.elegans. What is this useful for? For high level cognition, it's much more fruitful to study how deep models do it. For low level brain diseases, c.elegans is too simplistic to tell us anything we dont already know