The legend says that Dylan Cope returned after his loyal cultists the Sapling dev, Adapt dev, Simulife Hub(foo52) and the Bibites dev summoned him from the dead.
I'm so friggin mad at the youtube algorithm for not recommending me this for 4 MONTHS! It knows exactly that this is the kind of content that gets me going
This is a fantastic simulation of Evo-Devo, However I wanted to add something. The Salamander/Axolotl test, the reason why this happens is in Mexico, the caves lacked Iodine. Iodine is a very important chemical for Amphibian Development, its what allows for maturation to complete (Axolotls are Salamanders that basically are stuck in puberty because they don't get iodine). I wonder how removing a growth hormone affects evolution, and reintroducing that hormone back after speciation as occured. Like imagine if the Axolotl found a way to enter the salamander stage without Iodine, now give it iodine again. The machinery should still be present, would this cause multiple "growth" stages to occur?
I watched the previous evolution video today. (I actually rewatched it for like 20th time after I got back to my own evolution program starting basically from scratch). This video release caught me by a surprise I have to say, a really nice surprise
The amount of complexity you were able to put into that simulator and still make it stable is incredible. All my attempts, even the simplest ones, on creating evolution, always turned into pure chaos.
Your current system for forming multicelluar organisms seems to be different cells just joining together if they have enough adhesion. But most multicellular organisms in real life come about through a single cell (a zygote) multiplying and staying connected.
@@Groggle7141 Indeed. The purpose of the artificial gene regulatory is to lay down the foundations for reproduction via single-cell bottlenecks to emerge. When talking in evolutionary terms, there is a blurry boundary between colonial organisms that grow and split, and "true" multicellular organisms that rely on zygotes.
I think the real issue with simulating evolution, is that it's based on biochemistry. What are the cell walls made of? (How does it build more to reproduce), how does it produce chemicals that can bridge the hydrophobic space in the cell membrane. It's VERY hard to simulate how that happens with proteins. Not impossible mind you, but very very difficult. And it's the basis for almost all of this idea.
The process of evolution is definitely NOT based on biology. It's based on replicating systems with differential reproductive success whatever the substrate or medium. The simulations we generate (whether it's in computers, or on paper) loosely arrive at those rules first through our grasp of the was it works in biological systems. Genetic change is heritable & essentially irreversible. To prove evolution wrong, you must show what can reach into the DNA of every member of the descendants which inherited some beneficial gene & remove that gene from each of those descendants or send info to each recipient as to which gene to remove. There is no other option. It is only plausibly reversible if it occurs to the original organism in which the change occurred. Once the gene has been passed to multiple offspring, the odds of reversing that mutation in all offspring goes from being off the chart to utterly impossible.
I was laying in bed for the last few hours, unable to sleep, thinking about creating eerily similar simulation. I finally got out of bed, opened youtube, and this is the first video I've seen, lol. I love it!
This network on them is amazing! It’s definitely the most thoughtful evolution simulation I have seen on UA-cam, I thought it is “Bibites” earlier. Good luck with the project!
I stumbled across the previous video on this yesterday, liked it, and tried to check out the other stuff this channel has, only to realize it hasn’t had a new video for a year. AND THIS POPS UP IN MY RECOMMENDED THE NEXT DAY? I’m a blessed human today let’s go
I really hope to see more work done with gene regulatory networks and similar mechanisms in evolutionary computation. It always feels like evolution in these systems is less flexible than you'd hope, and anything providing improvements to the reuse and repurposing of functional components seems like a promising direction.
I'm happy to see that you're back! I think you're the only artificial life UA-cam channel that uses GRNs, which combined with the excellent narration and animations, is enough to put you on top of all other channels in this category. Incredible work!
Very engaging video Dylan! Excited to see this after your presentation at SLCU a few months ago. You do an excelent job getting people excited about evo-devo!
I am writing such a simulation in python. I use RPyC to communicate between nodes in the cluster, which is 256 individual computers (raspberry pi CM4s) using a hypertorus type of routing scheme. Of course I have a single computer that rules them all, presents the main interface like yours. I just found your channel today, I will look into your work, but this was just my first thought. Don't just use one computer, make it where you can use a whole lot of them. Edit: Also amazing work! Subscribed!
I think you're somewhat right about evolution of traits in your introduction. I was kind of looking forward to seeing if you implemented something a little more complicated than static traits that could be expressed and tweaked. I don't know how it could be done but being able to have emergent evolution of traits would be really promising, right now you have 6 traits and they can all be expressed to varying degrees and at different times, but that's it there's no way for new types of structures to come into being, no ways for different types of adaptations humans didn't design to appear or evolve spontaneously or by parts. Don't get me wrong, designing that is a HARD task, I mean it comes close to needing to fully simulate cellular biology at a molecular level or even something analogous. The interactions of tiny molecular changes on macro structures and their functions is not in anyway simple to simulate or even model. But given your Introduction I was somewhat hopeful you'd made some sort of strides towards this with your simulation. I am subscribing though looking forward to learning more
Thank you for your comment and your subscription! :) I agree that the project so far has only made small strides towards the ideas that I outlined at the top of the video. However, I do think that in the current version of the simulation, new/undesigned structures could potentially emerge as multicellular systems that compose the hard-coded six traits in novel ways. So indeed individual cells may only have limited functions, but together they may form something more interesting! If you check out my paper in the description, I go through a case-study from the simulation with a multicell structure that was able to self-replicate quite reliably! Also, I agree that only having six possible traits is limiting - and I would like to add more - I don't think adding any number of fixed traits would really change that picture! Anyways, that's what I was thinking of as my initial strides towards the problem - maybe you disagree? I would be curious to hear what direction you think I should head in next :)
@@dylancope Thanks for the quick and detailed response. I totally agree with you about the emergent properties of multicellular life, there are certainly possibilities there. I don't know how I'd go about emergent traits, there are a few models I can think of but ultimately I'm starting to understand (thanks to what you wrote) they're what you have already on a different scale, picking a selection of traits that can interact at some scale to facilitate the simulation of evolution at a larger scale. I don't think it'd actually be any more useful as a model of real evolution to do that without going to something like an engine that can fully emulate some kind of molecular interactions with the possibility of emergent complexity which isn't guaranteed. That said I might consider the following changes to your designs to reduce the number of traits rather than increase them. Pressure -> if you can have directed osmosis with the environment you could have emergent muscle analogues, a tensed muscle being a one with a higher internal pressure (more round) and a relaxed muscle having lower pressure and so smaller size (more stretchy/deformable by macro structure), this could actually be done by 'simplifying' the existing phagocytes, have phagocyte be the trait and then have a size of chemical be the target, Water is very small, sugar is small, proteins are medium, chunks of dead creature are large to very large... you could even have these be reversible so emitting molecules rather than absorbing them, transferring to neighbouring cells etc. In real cell chemistry I think that each of these ports is dedicated to a given molecule and directional but I don't think your simulation is that granular yet, but if you did go that far then the evolution of pheromones and synapse are theoretically possible so rather than relying on direct data transmission between connected nodes the communication protocol between different cells could be negotiated and evolved, co-operatively. An I want sugar pheromone responded to with a sugar exchange for example, these pheromones might then be hackable by other organisms which isn't the case in the networked connections your nodes create. Stickiness -> a cell that can be selectively sticky to the environment rather than to co-operative cells would allow for something similar to grip, move, release, grip move release, maybe Mucocytes, but these might also be a possible adaptation of the 'simplified' phagocytes. The ability to release chemicals would necessitate being able to modify the cell membrane to be resistant to certain chemicals as well as similar for internal cell structures to contain toxic chemicals for release Simplification of flagella and the spikes, these could be the same structure, the spikes are stiff and the flagella are not, the spikes get longer with exertion the flagella flex with exertion. This could allow for other expressions with perhaps other functions to be emergent. Though I'm not seeing obvious uses for these yet, spikes that change direction might rip holes into cells? added length providing range but also making the spike less strong. While flagella that change in length might be able to interact as sensory appendages? especially if they're sensitive to chemicals in a different way than the cell membrane I think the photo receptors could be chemical in nature instead of nodal, if there's the possibility for the growth of chloroplasts then measuring the sugar levels in the cell is analogous to a none directional light sensor rather than an explicit trait. So internal sensors for temperature, pressure, salinity, glucose levels etc, gives the ability to develop more emergent macro properties. To have sensory directionality the lifeform must be multicellular to block a sense from one or more directions. Rather than sensor components being built into a cell's trait list, especially for light as no individual cell is really opaque to light. Even without chloroplasts any measurable reaction that is facilitated by photons could serve as a way to sense light without giving direct access to sight to cells. Single Cells can follow gradients towards or away from a stimulus but complex features like sight shouldn't be baked into the design, in order to follow those gradients having a working memory if only of the last few time steps is critical, I go in this direction the gradient is followed, go faster, I go in that direction the gradient is reversed, go slower/change direction. This only works if it's possible for the cell to compare the situation now to the situation a couple of timesteps before, the bigger the gradient the bigger the response.
@@johnydlamazing ideas. Something I was pondering. Modifying physical properties is really interesting. Alien project is simpler in many way but does have muscle connections which combine with multiple cells for swimming and other behaviour like potentially trapping food in a net. I think a lot of that may be programmed rather than evolved. You kind of want a ‘Turing complete’ physical world. But simulating such a thing at speed becomes hard. Systems like Core Wars and Avida try and simulate a more open ended world by using more general programming languages. It’s certainly a good goal to be open ended. So combining Avida in a more realistic physics engine looks to be a good direction. Hard stuff though
@@oystercatcher943 I don't claim all the credit, I've been thinking about this since reading permutation city a few years ago it had an "autoverse" with a whole new simplified chemistry (no fission/fusion) and evolution from a designed single celled life form. The thing was even in the story so complicated that running experiments with more than a handful of microbes was a super computer endeavour so HARD is just not sufficient, but Greg Egan is a genius. His route describes cellular automata as the underlying physics and the emergent atoms be on a scale sufficient that pixelation of reality has a negligible effect on chemistry. But we're talking a running quadrilions of cells for just a single atom, and probably that cubed for a single life form xD not at all feasible for 2024 computation even with supercomputer access.
Speaking as a person who's been low-key obsessed with genetic algorithms and genotype->phenotype mappings since college, this is fascinating. One day, if I ever have the time, I hope take Douglas Hofstadter's "typogenetics" and invent "artificial molecular biology" out of it, which I _think_ will allow for very powerful and subtle mutation operations. Does your system allow for horizontal gene transfer, i.e., the sharing of DNA between cells, i.e., the only way bacteria can have more than one parent? My own [outdated!] experience with genetic algorithms tells me that crossover between parents can really supercharge evolution.
Nice video 👍. You sort of opened my eyes a bit for the "dynamic-world -》 gene regulation." part of EvoDevo. I often think about its role in modularity and symmetry/reuse. The program looks cool. Cant belive you chose Java for it. Still, amazing work 👍
Super inspiring stuff. I will be reading your paper. I’ve got a long way with my own GPU based simulation. I think your thoughts on GRN and the environment is super interesting. I’m interested in the similarity of GRN and neural networks with a GRN potentially having longer memory more like a RNN. The modular idea is brilliant and something I hope you don’t mind me copying and playing with. I’m also interested in how far evolution can learn to use the full complexity of your simulation world
Once again, very cool project. I do wonder if evolution can take advantage of all the solution space that you created. These videos need a Natural Geographic section 😅
Great observation! As with most ALIFE simulations, evolution tends to keep the population with within a somewhat narrow part of the solution space. I'm definitely going to continue exploring the parameters of the simulation and tinkering with features to see how a more diverse ecosystem can be maintained :)
13:13 I notice the GRN nodes move. How is this controlled I wonder. Graph layout like graphviz is complex because of avoiding overlaps etc. is there some spring based relaxation happening? Very interested to know thanks
Good question. To be honest, I kind of hacked together a solution to that problem. It's not a proper spring-based model. Each node in the graph repels neighbours that are too close and attracts ones that are far away. If I recall correctly, there are some edge cases to handle the nodes moving with the cells and attaching them to anchors on the surface nodes.
Lol Darwin didn’t actually study the finches until after he returned and asked for samples from other people there to draw his conclusions. But it makes for a nice story even if it didn’t happen that way
The "good" or "bad" ness of a gene is completely & totally dependent on the context (read "environment"). Within the context of one singular environment, gene alternatives have -no- VERY LITTLE space for being differentiated from the optimal species for that one environment (whether dynamic or not). The hardest part of simulating *the variety* produced by evolution is simulating a plethora of different environmental gradients encompassing any of several extremes. This is true for single celled (SCS) as well as multicellular species (MCS). SCS have few physical barriers other than the gravity well we live in (although evidence shows they could very well survive hitching a ride to other planets in the solar system). MCSs on the other hand are more or less limited to reproduction with local members within the confines of their local environment.
Someday the stars will be extinguished, and in my opinion, if something is worth competing for, it is for who will be the ancestor of the last life. I'd like to do research on entropy and replication, but idk where to start
The paper probably answers these, but for the algorithm: Are the GRNs always shallow? If they can also process signals like a neural network, then multicellularity can serve as a way to gain network depth (with some latency) Can the adhesion nodes change their length? Can they reposition in realtime (not between generations)?
The GRNs can have arbitrary depth - they evolve using the NEAT algorithm. But yes you're right that multicellularity is a way to gain depth. Also, depth within a cell has latency. Secondly, the current adhesion system cannot be repositioned. Although connections can be broken and remade.
@@darlingortiz2956 the only AI generated content was the first 10 secs where I used a stable diffusion model as a part of a custom animation script. The rest was stock footage from various sources and custom animations
i love these projects but frankly it feels that they aren't really simulations of evolution per se as there isn't space for novel functions to develop, just chosen functions to be modified. If anything it's more akin to adaptation. Although it's way easier to point at problems than solutions and having building blocks with defined physical properties that can be assembled from code like amino acids forming proteins would, i suspect, be a computational nightmare beyond actually coding it so you wouldn't get many generations/organisms even if you got it working in a way where they had the capacity to generate novel functions instead of being stacks of silly string dangling off of cells. more constructively (although not particularly important): diffusion is the word for the passive movement of solutes across the cell membrane based on concentration gradient. Osmosis only refers to the diffusion of water.
It’s a very hard thing to grapple with. However this simulation is totally able to generate evolutionary novel behaviours if not structures. But even then it could make interesting multicellular structures. I guess you have decide the level at which you wish to run the simulation. Evolution is going on at multiple levels, within cells and at the organism level. CoreWars and Avida may be more open-ended because they use more general programming languages but in a much less biologically inspired environment. Doing both would be great but very hard to do I think. Also Battle-of-the-clans on YT is interesting in this respect but still on a simpler grid based world. I think every style has its place. IMHO No one simulation can do everything without as you say being hopelessly complex and slow ❤
To be honest, I don't think that any of those languages would make a huge difference. Java is pretty performant if you don't abuse the GC too much, and it's easier to do multiprocessing that C (not sure about Rust). And that's the key issue for this program. Collision engines are hard to parallelise. The only way it would be significantly more performant is if I managed to find a better backend physics library, as I moved away from my own implementation. Ultimately as this is a side project, I don't have time to write everything from scratch so relying on physics engines and UI libraries has been very helpful.
The legend says that Dylan Cope returned after his loyal cultists the Sapling dev, Adapt dev, Simulife Hub(foo52) and the Bibites dev summoned him from the dead.
foo52 is cool
I love this "community"
@@davidzaydullin yea
the bibites dev
@@josephvanhaaften7710 edited
I'm so friggin mad at the youtube algorithm for not recommending me this for 4 MONTHS! It knows exactly that this is the kind of content that gets me going
This is a fantastic simulation of Evo-Devo, However I wanted to add something. The Salamander/Axolotl test, the reason why this happens is in Mexico, the caves lacked Iodine. Iodine is a very important chemical for Amphibian Development, its what allows for maturation to complete (Axolotls are Salamanders that basically are stuck in puberty because they don't get iodine).
I wonder how removing a growth hormone affects evolution, and reintroducing that hormone back after speciation as occured. Like imagine if the Axolotl found a way to enter the salamander stage without Iodine, now give it iodine again. The machinery should still be present, would this cause multiple "growth" stages to occur?
Thank you for providing this interesting context!
What if we just did ontogeny on a chimp face? The opposite of what Humans had occur.
Amazing work! Glad to see a new video!
I literally said "Wow!" when you showed off the inter-cell gene regulatory network. Amazing!
I watched the previous evolution video today. (I actually rewatched it for like 20th time after I got back to my own evolution program starting basically from scratch). This video release caught me by a surprise I have to say, a really nice surprise
Thank you for the support! Good luck with your project 😊
I watched the previous one yesterday and was pleasantly surprised to find out this video was uploaded. Also good luck!
Omg, i just posted some days ago to pls continue this, and after over a year, you returned! ❤
Time to watch the yearly upload!
I hope you can make more videos on this project! Even if you don't have many updates, I'd just like to see what can evolve in this!
The amount of complexity you were able to put into that simulator and still make it stable is incredible. All my attempts, even the simplest ones, on creating evolution, always turned into pure chaos.
Your current system for forming multicelluar organisms seems to be different cells just joining together if they have enough adhesion. But most multicellular organisms in real life come about through a single cell (a zygote) multiplying and staying connected.
@@Groggle7141 Indeed. The purpose of the artificial gene regulatory is to lay down the foundations for reproduction via single-cell bottlenecks to emerge.
When talking in evolutionary terms, there is a blurry boundary between colonial organisms that grow and split, and "true" multicellular organisms that rely on zygotes.
The cells in the colonies would need a way to share genetic code among each other.
I think the real issue with simulating evolution, is that it's based on biochemistry.
What are the cell walls made of? (How does it build more to reproduce), how does it produce chemicals that can bridge the hydrophobic space in the cell membrane.
It's VERY hard to simulate how that happens with proteins. Not impossible mind you, but very very difficult. And it's the basis for almost all of this idea.
The process of evolution is definitely NOT based on biology. It's based on replicating systems with differential reproductive success whatever the substrate or medium. The simulations we generate (whether it's in computers, or on paper) loosely arrive at those rules first through our grasp of the was it works in biological systems.
Genetic change is heritable & essentially irreversible. To prove evolution wrong, you must show what can reach into the DNA of every member of the descendants which inherited some beneficial gene & remove that gene from each of those descendants or send info to each recipient as to which gene to remove. There is no other option.
It is only plausibly reversible if it occurs to the original organism in which the change occurred. Once the gene has been passed to multiple offspring, the odds of reversing that mutation in all offspring goes from being off the chart to utterly impossible.
I was laying in bed for the last few hours, unable to sleep, thinking about creating eerily similar simulation. I finally got out of bed, opened youtube, and this is the first video I've seen, lol. I love it!
This network on them is amazing! It’s definitely the most thoughtful evolution simulation I have seen on UA-cam, I thought it is “Bibites” earlier.
Good luck with the project!
I stumbled across the previous video on this yesterday, liked it, and tried to check out the other stuff this channel has, only to realize it hasn’t had a new video for a year. AND THIS POPS UP IN MY RECOMMENDED THE NEXT DAY?
I’m a blessed human today let’s go
Haha now there are two videos I'm hoping that people realise that the project isn't dead, I'm just slow to edit videos 😅
I really hope to see more work done with gene regulatory networks and similar mechanisms in evolutionary computation. It always feels like evolution in these systems is less flexible than you'd hope, and anything providing improvements to the reuse and repurposing of functional components seems like a promising direction.
New videos please, even if it is, just the protozoa moving from one side to the other with you commenting on top, PLEASE, i'm miss you 😢
He's frigging back! I had nearly lost hope, but the great Dylan Cope has graced us with his amazing content once again!
I have massive respect for anyone who can do anything close to this, your actually a legend bro
Im so happy for another installment! Not enough evolutionary simulator content on YT
Welcome back
Nice. God, evolution, technology, and art, all in one package.
We will watch your career with great interest.
he's back boys
Great work
I'm happy to see that you're back! I think you're the only artificial life UA-cam channel that uses GRNs, which combined with the excellent narration and animations, is enough to put you on top of all other channels in this category. Incredible work!
MY GUY IS BACK OH MY GAHHHH I LOVE THIS
He’s back! Can’t wait for more
Wow, literally only found the first vid yesterday, how lucky is that
fantastic as always
happy to see you back its bean a while😁
I just rewatched your previous video out just to remember what it was about and saw this one. Nice.
Thank you!
Really interesting. I'm glad these kinds of simulations are being created... Definitely will be following the project!
Very engaging video Dylan! Excited to see this after your presentation at SLCU a few months ago. You do an excelent job getting people excited about evo-devo!
Thank you Alexandre! I really enjoyed my visit to the lab - hope everyone is doing well :)
I am writing such a simulation in python. I use RPyC to communicate between nodes in the cluster, which is 256 individual computers (raspberry pi CM4s) using a hypertorus type of routing scheme. Of course I have a single computer that rules them all, presents the main interface like yours.
I just found your channel today, I will look into your work, but this was just my first thought. Don't just use one computer, make it where you can use a whole lot of them.
Edit: Also amazing work! Subscribed!
Incredibly interesting !
THE GOAT RETURNS
Awesome!
Fascinating! Thank you
Oh you’re still active! Amazing!
Yaaaaay!! More computer evolution simulators :D
3b1b music is seriously just the best
that is some deep symulations
The Legend has returned!
w generational upload
Oh this is much better than my version of this... I love this, gonna have to take a look at the code.. mine was still very Darwin-bots like
I think you're somewhat right about evolution of traits in your introduction. I was kind of looking forward to seeing if you implemented something a little more complicated than static traits that could be expressed and tweaked. I don't know how it could be done but being able to have emergent evolution of traits would be really promising, right now you have 6 traits and they can all be expressed to varying degrees and at different times, but that's it there's no way for new types of structures to come into being, no ways for different types of adaptations humans didn't design to appear or evolve spontaneously or by parts.
Don't get me wrong, designing that is a HARD task, I mean it comes close to needing to fully simulate cellular biology at a molecular level or even something analogous. The interactions of tiny molecular changes on macro structures and their functions is not in anyway simple to simulate or even model. But given your Introduction I was somewhat hopeful you'd made some sort of strides towards this with your simulation.
I am subscribing though looking forward to learning more
Thank you for your comment and your subscription! :)
I agree that the project so far has only made small strides towards the ideas that I outlined at the top of the video.
However, I do think that in the current version of the simulation, new/undesigned structures could potentially emerge as multicellular systems that compose the hard-coded six traits in novel ways. So indeed individual cells may only have limited functions, but together they may form something more interesting! If you check out my paper in the description, I go through a case-study from the simulation with a multicell structure that was able to self-replicate quite reliably!
Also, I agree that only having six possible traits is limiting - and I would like to add more - I don't think adding any number of fixed traits would really change that picture!
Anyways, that's what I was thinking of as my initial strides towards the problem - maybe you disagree? I would be curious to hear what direction you think I should head in next :)
@@dylancope Thanks for the quick and detailed response.
I totally agree with you about the emergent properties of multicellular life, there are certainly possibilities there.
I don't know how I'd go about emergent traits, there are a few models I can think of but ultimately I'm starting to understand (thanks to what you wrote) they're what you have already on a different scale, picking a selection of traits that can interact at some scale to facilitate the simulation of evolution at a larger scale. I don't think it'd actually be any more useful as a model of real evolution to do that without going to something like an engine that can fully emulate some kind of molecular interactions with the possibility of emergent complexity which isn't guaranteed.
That said I might consider the following changes to your designs to reduce the number of traits rather than increase them.
Pressure -> if you can have directed osmosis with the environment you could have emergent muscle analogues, a tensed muscle being a one with a higher internal pressure (more round) and a relaxed muscle having lower pressure and so smaller size (more stretchy/deformable by macro structure), this could actually be done by 'simplifying' the existing phagocytes, have phagocyte be the trait and then have a size of chemical be the target, Water is very small, sugar is small, proteins are medium, chunks of dead creature are large to very large... you could even have these be reversible so emitting molecules rather than absorbing them, transferring to neighbouring cells etc. In real cell chemistry I think that each of these ports is dedicated to a given molecule and directional but I don't think your simulation is that granular yet, but if you did go that far then the evolution of pheromones and synapse are theoretically possible so rather than relying on direct data transmission between connected nodes the communication protocol between different cells could be negotiated and evolved, co-operatively. An I want sugar pheromone responded to with a sugar exchange for example, these pheromones might then be hackable by other organisms which isn't the case in the networked connections your nodes create.
Stickiness -> a cell that can be selectively sticky to the environment rather than to co-operative cells would allow for something similar to grip, move, release, grip move release, maybe Mucocytes, but these might also be a possible adaptation of the 'simplified' phagocytes. The ability to release chemicals would necessitate being able to modify the cell membrane to be resistant to certain chemicals as well as similar for internal cell structures to contain toxic chemicals for release
Simplification of flagella and the spikes, these could be the same structure, the spikes are stiff and the flagella are not, the spikes get longer with exertion the flagella flex with exertion. This could allow for other expressions with perhaps other functions to be emergent. Though I'm not seeing obvious uses for these yet, spikes that change direction might rip holes into cells? added length providing range but also making the spike less strong. While flagella that change in length might be able to interact as sensory appendages? especially if they're sensitive to chemicals in a different way than the cell membrane
I think the photo receptors could be chemical in nature instead of nodal, if there's the possibility for the growth of chloroplasts then measuring the sugar levels in the cell is analogous to a none directional light sensor rather than an explicit trait. So internal sensors for temperature, pressure, salinity, glucose levels etc, gives the ability to develop more emergent macro properties. To have sensory directionality the lifeform must be multicellular to block a sense from one or more directions. Rather than sensor components being built into a cell's trait list, especially for light as no individual cell is really opaque to light. Even without chloroplasts any measurable reaction that is facilitated by photons could serve as a way to sense light without giving direct access to sight to cells.
Single Cells can follow gradients towards or away from a stimulus but complex features like sight shouldn't be baked into the design, in order to follow those gradients having a working memory if only of the last few time steps is critical, I go in this direction the gradient is followed, go faster, I go in that direction the gradient is reversed, go slower/change direction. This only works if it's possible for the cell to compare the situation now to the situation a couple of timesteps before, the bigger the gradient the bigger the response.
@@johnydlamazing ideas. Something I was pondering. Modifying physical properties is really interesting. Alien project is simpler in many way but does have muscle connections which combine with multiple cells for swimming and other behaviour like potentially trapping food in a net. I think a lot of that may be programmed rather than evolved. You kind of want a ‘Turing complete’ physical world. But simulating such a thing at speed becomes hard. Systems like Core Wars and Avida try and simulate a more open ended world by using more general programming languages. It’s certainly a good goal to be open ended. So combining Avida in a more realistic physics engine looks to be a good direction. Hard stuff though
@@oystercatcher943 I don't claim all the credit, I've been thinking about this since reading permutation city a few years ago it had an "autoverse" with a whole new simplified chemistry (no fission/fusion) and evolution from a designed single celled life form. The thing was even in the story so complicated that running experiments with more than a handful of microbes was a super computer endeavour so HARD is just not sufficient, but Greg Egan is a genius.
His route describes cellular automata as the underlying physics and the emergent atoms be on a scale sufficient that pixelation of reality has a negligible effect on chemistry. But we're talking a running quadrilions of cells for just a single atom, and probably that cubed for a single life form xD not at all feasible for 2024 computation even with supercomputer access.
It's back!!!
Speaking as a person who's been low-key obsessed with genetic algorithms and genotype->phenotype mappings since college, this is fascinating. One day, if I ever have the time, I hope take Douglas Hofstadter's "typogenetics" and invent "artificial molecular biology" out of it, which I _think_ will allow for very powerful and subtle mutation operations.
Does your system allow for horizontal gene transfer, i.e., the sharing of DNA between cells, i.e., the only way bacteria can have more than one parent? My own [outdated!] experience with genetic algorithms tells me that crossover between parents can really supercharge evolution.
HE HAS RETURNED!!!
Amazing project ! Thanks also for the biology explainations, they are extremely clear and engaging.
Nice video 👍. You sort of opened my eyes a bit for the "dynamic-world -》 gene regulation." part of EvoDevo. I often think about its role in modularity and symmetry/reuse.
The program looks cool. Cant belive you chose Java for it. Still, amazing work 👍
love it
Super inspiring stuff. I will be reading your paper. I’ve got a long way with my own GPU based simulation. I think your thoughts on GRN and the environment is super interesting. I’m interested in the similarity of GRN and neural networks with a GRN potentially having longer memory more like a RNN. The modular idea is brilliant and something I hope you don’t mind me copying and playing with. I’m also interested in how far evolution can learn to use the full complexity of your simulation world
Once again, very cool project. I do wonder if evolution can take advantage of all the solution space that you created. These videos need a Natural Geographic section 😅
Great observation! As with most ALIFE simulations, evolution tends to keep the population with within a somewhat narrow part of the solution space.
I'm definitely going to continue exploring the parameters of the simulation and tinkering with features to see how a more diverse ecosystem can be maintained :)
YOOOOO HE'S BACK!!!
YOO ive been waiting forever for this video
Hey i tried to download it but it wont even open even after i run it as administrator. Is it because im still on windows 10??? Sad :(
HES BACK
JOHN HOLLAND MENTIONED!
Got to show some love for the OG evo sim guy
13:13 I notice the GRN nodes move. How is this controlled I wonder. Graph layout like graphviz is complex because of avoiding overlaps etc. is there some spring based relaxation happening? Very interested to know thanks
Good question. To be honest, I kind of hacked together a solution to that problem. It's not a proper spring-based model. Each node in the graph repels neighbours that are too close and attracts ones that are far away. If I recall correctly, there are some edge cases to handle the nodes moving with the cells and attaching them to anchors on the surface nodes.
I really like the symbiosis of the blue and cream colourd species one is mouth one is movement
badass
Lol Darwin didn’t actually study the finches until after he returned and asked for samples from other people there to draw his conclusions. But it makes for a nice story even if it didn’t happen that way
Haha yeah that's true - the road to science is always messier (and longer) than simple stories like that😅
he's back!!!
The "good" or "bad" ness of a gene is completely & totally dependent on the context (read "environment"). Within the context of one singular environment, gene alternatives have -no- VERY LITTLE space for being differentiated from the optimal species for that one environment (whether dynamic or not). The hardest part of simulating *the variety* produced by evolution is simulating a plethora of different environmental gradients encompassing any of several extremes. This is true for single celled (SCS) as well as multicellular species (MCS). SCS have few physical barriers other than the gravity well we live in (although evidence shows they could very well survive hitching a ride to other planets in the solar system). MCSs on the other hand are more or less limited to reproduction with local members within the confines of their local environment.
HE UPLOADED YIPEEE
Someday the stars will be extinguished, and in my opinion, if something is worth competing for, it is for who will be the ancestor of the last life.
I'd like to do research on entropy and replication, but idk where to start
This project reminds me of _Cell Lab_ on Android.
The paper probably answers these, but for the algorithm:
Are the GRNs always shallow? If they can also process signals like a neural network, then multicellularity can serve as a way to gain network depth (with some latency)
Can the adhesion nodes change their length? Can they reposition in realtime (not between generations)?
The GRNs can have arbitrary depth - they evolve using the NEAT algorithm. But yes you're right that multicellularity is a way to gain depth. Also, depth within a cell has latency.
Secondly, the current adhesion system cannot be repositioned. Although connections can be broken and remade.
it dosent work. when i start the program i the window just briefly flashes on the screen and dissapears
how do i download new version
HE'S BAAAAAAAAAAAAAAAAAAACK!!!!!
0:52 ¿como se llama esa imagen?
@@darlingortiz2956 the only AI generated content was the first 10 secs where I used a stable diffusion model as a part of a custom animation script. The rest was stock footage from various sources and custom animations
Make that food can have waste value and cells can get poisoned
he has returned.
My waiting gave fruit
i love these projects but frankly it feels that they aren't really simulations of evolution per se as there isn't space for novel functions to develop, just chosen functions to be modified. If anything it's more akin to adaptation. Although it's way easier to point at problems than solutions and having building blocks with defined physical properties that can be assembled from code like amino acids forming proteins would, i suspect, be a computational nightmare beyond actually coding it so you wouldn't get many generations/organisms even if you got it working in a way where they had the capacity to generate novel functions instead of being stacks of silly string dangling off of cells.
more constructively (although not particularly important): diffusion is the word for the passive movement of solutes across the cell membrane based on concentration gradient. Osmosis only refers to the diffusion of water.
It’s a very hard thing to grapple with. However this simulation is totally able to generate evolutionary novel behaviours if not structures. But even then it could make interesting multicellular structures. I guess you have decide the level at which you wish to run the simulation. Evolution is going on at multiple levels, within cells and at the organism level. CoreWars and Avida may be more open-ended because they use more general programming languages but in a much less biologically inspired environment. Doing both would be great but very hard to do I think. Also Battle-of-the-clans on YT is interesting in this respect but still on a simpler grid based world. I think every style has its place. IMHO No one simulation can do everything without as you say being hopelessly complex and slow ❤
How can evolution starting with a complete set of codes? wasn't it supposed to be start from 0 lines?
@@Jabawokiz810 real evolution didn't start until a lot of physics and chemistry stuff was already going on.
So it runs in Java, no wonder you need a pretty powerful computer :(
if only it was written in C, Rust, or C# DOTS...
To be honest, I don't think that any of those languages would make a huge difference. Java is pretty performant if you don't abuse the GC too much, and it's easier to do multiprocessing that C (not sure about Rust).
And that's the key issue for this program. Collision engines are hard to parallelise. The only way it would be significantly more performant is if I managed to find a better backend physics library, as I moved away from my own implementation.
Ultimately as this is a side project, I don't have time to write everything from scratch so relying on physics engines and UI libraries has been very helpful.
t
You could just use Spore the game for this literally lmao