One of the greatest problems I had with "evolution simulators" is that they are almost always limited to a few stats, perhaps a neural network, and left there. Only one other evolution simulator actually evolved body parts in a meaningful matter(The Life Engine), and even then not to the extent which you have. Well done.
Cheers! Your comment hits on one of the reasons I'm motivated to work on this project; to find ways for evolution to really surprise me. As you say, many simulations are essentially just optimising over some fixed stats, which really leaves very little room for surprise.
@@dylancope You are right, but I think this is why this type of project is scary to undertake. Because if you want to simulate life in a somewhat accurate way, you'll need to simulate genes, molecules, and to get to complex life you'll need billions and billions of these which would require super computers. Also you are using neural networks do handle cells, but those cells are supposed to become neurons afterwards lol, but it would be very tedious to evolve cells without neurons.
What is that “evolution simulator” called about evolving body parts in a meaningful matter? I’d like to look for any evolution simulators that can evolve not just their brain, but also body parts, stats, etc. Basically very similar to the real evolution.
The complex molecules thing is brilliant! As other sims have shown, predation is hard - but in our world the number of vitamins we would really prefer to get from other organisms is way more than is usually simulated (as in, not at all). A brilliant addition.
this is brilliant. Closest thing to first-principles evolutionary sim I've seen thus far. I could even see a point in your sim where it looked like your individual cells almost glued together into a stomach of sorts. They didn't hold out in that shape, but it was quite close.
@@revimfadli4666 Oh Life Engine is also brilliant in its utter simplicity, but I think this one being grid-free is a really good thing that can't quite be compared with that one
This video was top tier. Got really excited when you started mentioning allopatric speciation, I like this topic and seeing it in simulation is very nice) Keep up the good content
Loved it. Thanks! I do biotechnology and this is my favorite evolution simulator a have encountered (even at this very initial state) also I would love (I NEED) to see the evolution playout as you accompany the events. In this video you mostly explained how it works. Now I want to see it in action. Keep this amazing work!! The epigenetics will be a challenge
You're definitely gonna be a "big" youtuber, or at least much bigger than what you are now. Like come on you have 400 subs and this video quality is just incredible. Great job!!!
I've always wanted to create a life simulation, but I never knew how to start. I mean, I know how to code and program, but there are just too many factors in a simulation, and I wanted to create one entirely on my own. Watching this helped me get a general idea of what things to focus on when starting my project. I will certainly try to work on it now.
Anything you make with any rules has a good chance of having some sort of pattern of varying levels of interest, just try applying any old set of rules and you'll likely end up with something even if it's not what you wanted.
Projects and videos like these are what make me always come back to programming and learning about the relationship between chemistry and biology, thank you for making this :) The music is so fitting too, I just stared at the screen like a hypnotized chicken the whole time
Found this vid a week ago and can not stop thinking about it. I don't often rewatch the same video but this one has me hooked on the presentation and concept of this cell evolution simulation.
This video has reached a rather large audience for your channel, Congratulations! I found it very pleasing to watch and adding to the great network of RU-vid evolution simulator videos and channels. One thing though that I think you might benefit from is something Michael from Vsauce said recently, and I'm going to paraphrase here, " you shouldn't underestimate your audiences comprehension of a subject matter, but instead their knowledge of the vocabulary". If you take this into account I think you will be able to reach a wider audience for your upcoming videos on this subject by spending a touch more time explaining each step of more complicated ideas, especially in future chapters similar to this video's "Understanding evolution", and "Looking at data from the simulation". Your use of 3B1B's Manim is excellent in a quest to teach the masses and your graphs are very helpful, though, perhaps easing viewers into the graphs by explaining the concepts they are showing before the graph instead of during. That will give them less of a distraction and allow from them to grasp the concepts better before going into the actual data. Lastly, one section I think you should be carful of is during your explanation of Pooled Standard Deviation (10:33 -ish), taking a bit more time to explain what that exactly is, with simpler diction, for general audience members who are not as math savvy. Obviously that should apply to any future math concepts covered by this series going forward. I hope these pointers aren't taken as an attack on your work and instead merely as constructive criticisms that I noticed during my viewing experience. I also hope that this series will continue and that your channel will grow to an even wider audience. I look forward to seeing that become a reality.
@@dylancope Once again amazing video and im excited to see more on the multicellular concept as a lot of the evo-sim channels are doing mostly single cell stuff.
@@dylancope yeah, firstly thank you: superb content, excellent simulator. Sadly, as English is non-native for me, even though I mostly watch more dense contents at .75 speed and even pause and ponder over new concepts, there were a few words/phrases I couldn't make out even after repeated view of the section (Not even with the help of automatic subtitles could I make sense of those). Minor stuff, but I can imagine that it may cause some discomfort for those not fluent in English....
@@dylancope Okey, I've re-evaluated my response: I've figured out what was distractive for me: the way how you pronounce some words with "r" inside them....("curve" sounds like "code") and also the endings of some words were "cut off". Once in every second, third sentence: nothing to worry about, at least I practiced some more real life English. Cheers
I've been wanting to find something like this for so long! and just now found it! I know you have thousands of potential fans waiting to see this, please keep making more videos! Your audience and subs will grow
Halfway through the video, I thought about suggesting a way for cells of the same species to differentiate and specialize, but I can see you already thought of that! Nice simulation.
Absolutely brilliant, can't wait for the continuation of this series! If you keep this quality up and start posting relatively frequently, you are sure to grow to become a large youtube channel in no time!
Definitely one of the better simulations ive seen. I appreciate youve largely resisted the urge to gamify this simulation, which most youtube simulations do. I feel the beauty of these simulations is laying out the rules, and watching the behavior naturally occur from them. Not constructing the rules to force the behavior.
Thanks! Yeah I'm definitely not making a game. The UI exists to watch and analyse the simulation. The only interaction that has been implemented was some stuff I added for debugging - you can move things around and kill cells. The only plans I have for the UI involve a screen for setting up a new custom simulation, loading an existing one, and hopefully a dashboard to see the statistics over time.
This project is incredible and even reminds me of a simulator called "cell lab" but perfected with proper cell adhesion and if i could i would suggest making certain cells be able to sacrifice self viability aka specialisation for a specific role such as fast movement , defence or even increased central control almost like a brain the potential is incredible and i would love to see something like that in the future and the potential for evolutionary surprises would be incredible! i also applaud you for the project and thank you for all your effort!
Cell lab simulate the morphology evolution and very simple embryogenesis very well, but a bit lacking and limited in the genetic encoding and the limited "landscape" a dish can provide. And I've done experiments using android simulator to run for hundreds of thousands of hours purely on random mutation. The result is less impressive without "human design" creatures. Although they did evolve multicellular linking with plant like "blankets", but few mobile cell lineages (although some evolved very surprising ways of movements by "farting" small child to push them toward different directions, it seems to be a better way of control its hunting range, than Flagellocyte which could waste energy and hard to "stop"). And the PC version is still being working on I believe.
@@countingtls from my experimentation some interesting forms did emerge like ones you mentioned such as farting predators however im surprised you didnt mention swimmer evolution as those seem to be pretty common , tho i will admit your right , no matter what method ive used true complex creatures never emerge due to how low the chance is and that even the transitional forms wouldn't survive long enough to give form to a future form. That is one of the many problems with evolving in cell lab tho evolving human made creatures do allow for more interesting results
@@obiomajronyekwere4469 The emergence of the simple swimmer rely heavily on the starting condition, where nutrient particles are provided with sufficient quantity and distribution, but not too much where the splitting force is enough to survive. In my experiments, I want to see if hierarchical species can evolve purely from light to photosynthesis and then "herbivores" and different levels of predators. Once there are enough plants, the chance of evolving swimmers will be reduced quite a lot and the farting movement creatures seem to dominate the mobility methods. Maybe there are certain density and other factors combined can make their coexistence more likely, but in general for swimmers to evolve, the conditions of the dishes have to change depends on the current evolution stage to make them emerge, which will be interventions at the time I didn't wish to impose.
@@countingtls That is very interesting and you are right on the swimmers somewhat it does rely heavily on starting conditions making single cells viable but slowly allowing random chance to produce swimmers. However i doubt the ecosystem you are attempting to make is viable , with my experimentations such ecosystems almost always crash due to devouros aka predation and a light centered ecosystem always seems to be unstable
This is one of the most fantastic videos I have seen on the subject. Most projects don't have any space for emergent behavior. For example in most projects, predation is hard coded and two types of species exist (apart from plants) but here the idea of manufacturing complex molecules is a very brilliant idea. Best of luck for future projects.
I have toyed around with removing the distinction between plants and protozoa entirely. They both inherit from the same base class and I could just make "chlorophyll" a complex molecule that forces you to be green and produces mass and energy idly. However, one of the problems is the computational cost of the simulation. Currently, in order to run it I have to cap the types of cells individually - e.g. let there only be 5000 plants and 2000 protozoa. The plants tend to reach this cap pretty fast and stay there. The problem is that when a population is up against this cap evolution essentially stops working very efficiently as attempts to split often end up with the children not being produced. They fail to be added to the environment because to do so would exceed the cap. I have a couple ideas to get around this - I just implemented the idea of "local caps" that limits capacities to each region of the world. I could also debuff plants, but that might lead to more extinction events. Anyways there's lots to follow-up on!
@@dylancope local caps sound more realistic... I'm not a programmer, but couldn't it be coded that after the local cap is reached , newly spawned plants would be compared with other plants in the region according to some evolutionary trait and simply replace the least fitted, which would disappear (let's say this would imitate plants overgrown by others not getting enough sunlight). Such trait could be the calorie/energy content which would be thus selected both positively and negatively by being a more preferred food of the protozoans... Just thinking 🤔
@@dylancope I've been daydreaming about this kind of project for years, but never got around to coding one. So, take this with a grain of salt: First: Is there a way to have a time varying, per-individual "metabolic rate" and/or probability of cell division? Could this be a function of something in the environment? Second: You already have a system for pheromones -- what if you expanded that to include release + absorption of oxygen and/or CO2? If you find a sufficiently-clever way to combine those two concepts, you could set up a system where overpopulation releases too much CO2 into the water, which slows down metabolic rate and with it reproductive rate. Just a thought. LOVED the video!
@@dylancope It would be nice if it was a non-rigid cap where it's possible to go over the cap, but doing so results in worsened environmental conditions. For instance, divide a total amount of sun to all the plants meaning when there are more plants there is less sun per plant.
Thank you! I'm glad you enjoyed it. I started the project in April 2016 and periodically came back to it occasionally over the years. But it wasn't until May 2022 that I started seriously working on it again, and even then I was working in bursts so it took a while to get the video made (which is the current situation again with the V2 I've been working on 😅)
This is one of the best evolutionary simulation videos that I have ever seen and I am always anxiously waiting for the next one, I hope I can see it soon (:
The animations are really good. I can't even think how much time did you take to do it haha. I really like evolution simulators and it would be great if the subjects could generate different traits from zero based on their environment. And when I said "from zero", I mean without any prescribed code behing
@@realbrickbread Perhaps a pool of features could be created that appear randomly in the following generations and that these features can mutate in size or shape also randomly in the following generations if natural selection allows them to persist in the environment. For example, a vestige of a limb appears that will later grow or change its shape.
Between this and the Bibites, the cellular automata genre has gotten incredibly interesting. Thank you so much for taking a detailed mathematical approach; its amazing!
This is cool, thanks for sharing the code. It's interesting that you use Java Swing to do the graphics - back 10 or so years ago I made a simulator for electric charges that used Java Swing to do real time graphics and I would have expected that to be slow but it worked pretty well at a decent FPS. Have you looked into using GPU acceleration for the simulation? I would guess if that could be done it would speed things up a huge amount.
Hey - since releasing this video I've been working on a 2.0 of the project and have ported it over to libgdx and box2d. Performance is better so far with most of the intense features re-added. I've offloaded more of the rendering to pre-rendered sprites and shaders. I've also been learning some JCuda programming and that has accelerated the background chemical field massively!
@@dylancope highly recommend against java for this sort of thing, its meant for business-applications, its quite bad at succinctly expressing mathematical models
A bit of correction/addition on development and pluripotency of initial cells: there are, in fact, a minimum of three approaches in Nature: one of them is similiar to what you've described - starting cells are totipotent and can potentially differentiate into any other cell, heavily depending on external (to exact cell, but maybe not to organism as a whole) factors. After final differentiation there is no redifferentiation possible (although some near-end stem cells would stay in such near-end stem condition for tissue to be able to reproduce), but if you kill/delete some cells from early embryo it will still develop as usual (some animals can even assemble scattered cells into new embryo if it is an very early stage - some sea urchins would be an example). Another example would be us, vertebrates. The second approach is way more rigid, hovewer - even zygote has anisotropic distribution of specific chemicals in it and even after the first division first two daughter cells are bound to their fates - one will inevitably grow/develop into anterior part of an organism and the other will inevitably grow/develop into posterior. All the other divisions only make this fate more precise for new daughter cells. If you remove cells from developing embryo of such type your end-point organism would lack some bodyparts/organs/tissues, for example, removal of any of the first two daughter cells would result in development of only a half of an organism - anterior or posterior depending on which one you remove. A good example of such taxa is Spiralia (mostly known for molluscs and annelids). Another good example is nematodes - some of them even delete some genes from some cells as they differentiate - so they not just stop expressing this genes, they physically lack them. Both of this systems have division into germ and somatic cell lines in early development. Germ cells would become functional part of sexual organs (but not necessarily their structural parts, i.e. genitalia - these are often made from somatic ones) and will produce gametes and ultimately zygotes which are totipotent again. Somatic cells will become the whole other parts of an organism, something like a space suit or space ship for germ cells if you want an analogy. The third system lacks such dramatic cell line division and/or end-point differentiation and "somatic" cells can become germline or just totipotent under some conditions. A good example would be hydra polyp that form gametes (that will eventually form hibernating zygotes) in autumn and buds that will develop into small polyps (the main mode of hydra reproduction) in summer both from just normal somatic cells - they lack any specific germ cells. Another good example would be plants - that's why you can reproduce many of the plant species by dividing and replanting adult organism (ofcourse there might be some constraints in technical details of this process). This three-type classification is also a bit simplifying, but it catches the main ideas of developmental biology of multicellular organisms and especially animals. Hope you'll find it useful. And about your simulation - the incorporation of complex organic molecules as required/signalling components is great and awesome, never saw anything like this in any other evolution simulator.
Thank you for all this great information! I have been doing quite a bit of reading on differentiation but have not yet settled on how I'm going to implement differential pluripotency, so your comment is really valuable! My current idea is to implement a basic form of cytoplasmic determinants that are distributed to children upon cytokinesis. Cells will be able to produce determinants or destroy determinants already present. Gene regulatory networks will then be able to take determinants as inputs, thus allowing parents to attempt to condition the expression of genes in the next generation.
Keep it up! I am amazed at the group of people this video has collected. I’ve dreamt of doing a very similar project, regulatory networks and all! Kudos to you for getting started and making it look beautiful, with great narration to boot!
I have always wanted to see a decent evolution simulator. I'm old enough to remember what a huge let-down Spore was at release. Always seemed like there was potential for a great simulator, but it never got mined -- until now! This is an incredible bit of engineering -- really well done! I will download and play with it when I get a chance
It’s kind of annoying no game like spore has come out recently, only games that explore one aspect have (single celled, civilization, resources, and space colonization), no game has truly explored the level of player controlled evolution and survival since then, that’s why I still play it sometimes. It’s disappointing, but the only option for its experience.
I really like the video - a lot of thought and consideration has gone into an interesting design for this evolution sim :) I would however like to point out that the full-commitment interpretation of Waddington's landscape shown in the video is no longer the putative model of cell differentiation. This one-way street concept was disproven by the work of Nobel Laureate Shinya Yamanaka on the creation of induced pluripotent stem cells - showing that differentiation may be reversed in some cases!
Thank you for your comment, and bringing up the nuances regarding Waddington's Landscape! I purposefully left out complications such as Yamanaka factors as I thought the video was already quite heavy, although they are a completely fascinating aspect of modern epigenetics. Maybe at some point in the future I may address it :)
Oooh! Epigenetics and gene-to-gene interaction/recursion have been something I've always wanted to see in one of these (and something I've played with in the plans for my own evolution simulator that I'll probably never get around to coding). Excited to see further updates!
The movement restriction theory really gives me something to think about, it makes a lot of sense seeing as many both simple and complex animals grow inside of rocks and other structures, like for example the ''living rock''.
That was a class video. I'm a science student myself so it was right up my street. I love evolutionary biology and have always found it fascinating. Great job 👌
@@justineezekielbolivar As someone who codes himself, I’m well aware of that. I just thought he’d make more, shorter videos. But I like the quality over quantity approach. On top of that he is not just coding smth. He’s technically researching a relatively new and untouched field of computer simulation.
Love this! I've watched a lot of evolution sims but never seen one that aims to simulate multicellular evolution. It's a shame you didn't allow the sim to evolve retinal cells as specialised cells though. Maybe this is just me misunderstanding semantics(?) but giving cells the ability to evolve internal "retina" seems to be at the wrong level of complexity (retinae being collections of light-sensitive cells). Allowing light sensitivity is certainly good, but I think it would be interesting if this were kept simple so that retinae, and maybe organs like eyes could evolve as specialised cell collections within a multicellular organism. Even better if the shapes of the cells could evolve in ways that influence and were influenced by e.g. the physical properties of simulated rhodopdin molecules (size, shape and photoelectric properties) so the ability to capture light and generate electrical impulses would evolve, as opposed to being predefined (thinking how rod and cone cells evolved to be long thin "organic fibre optics", essentially, but also act as nerves). Perhaps a future version? Anyway, excellent work and very beautiful too! Well done!
awesome! it's been a while since i saw a new life simulation. and you have put in so many cool aspects! gene regularion, cell chemisty, multicellularity, oh my! i am very excited where this will go.
rly looking forward to the rest of this, and playing around w it when i can. i've seen so many other life sims, but none of them are this in depth, and i can't wait to see what u do w it
really cool vid, there's a lot to evolution that i would never have learned in education cause i just didn't study down this field... so getting to see a cool visualisation is reall neat. dipping our toes into many different fields to see what impact they have on your model. easy sub, great vid.
this is stellar work. Most of my interest in evolutionary systems (and biology in general) is how certains system will be favorised, why and how, and what's the result that come out of it. I hope to see this project turn into a professional project being worked on full time, paving the way for simulations trying to try to recreating environnements closer to reality instead of basing themselves on variables for what would happen in a real environnement (the more realistics simulations more difficult due to requiring way more precise knowledge). And the creation of a third dimension as well. Would be interesting to see what comes out of it.
Always love me a good evolution simulation. Glad to see a working "into carnivorous" solution via simplifying the collection of nutrients instead of making it themselves. So many of these fall into "It's just better to eat the grass" due to basic energy systems.
This is the first evolution sim I've seen which actually works at a biochemical level, it's amazing! Since you were talking about regulated gene expression, I strongly encourage you to check out Michael Levin's talk on the SEMF youtube channel if you haven't already. He goes into the biochemical processes of cell specialization and the formation of multicellular organisms.
Another project that seems to attempt something similar to what you’re building, though far more broad and game-y (it’s attempting to be a Spore reimagining) is THRIVE. Been in development for years, and from what I remember it just recently reached multi-cellular.
You should check out the game Thrive. It's based in biochemistry of single celled organisms currently as a open project and they are working on implementing mutlicellular life soon. It's like the game Spore but much more realistic and based in chemistry and such.
Are things like increasing size, rotation speed, movement speed, etc handled similarly to construction projects, requiring specific complex molecules? And can the Protozoa tell which creatures are the same “species” or are they all the same species?
Changing those variables doesn't require complex molecules. Increasing size requires mass and energy, speed and rotation just requires more energy. Protozoa can't tell each others "species", but they could potentially evolve to infer that through sensing the colour.
Finally something with evolved emergent morphology from cells joining together. Can they communicate via the joints? Will they have joint actuators, or would cell thrusters already serve that function?
I'm currently working on the binding communication system and I have some really cool ideas in development! Will have a video about them at some point :)
@@dylancope for the modular communications, did you take inspiration from works featured at Otoro's blog post/paper(especially the collective intelligence survey one)? Those were some mindblowing stuff, especially the mujoco and self-assembly ones
I would have like to see more specialization and interdependence because its is such an important step in multicellularity, would also have liked chemotaxis instead of a retina
Thanks for the feedback! I have been working on cell communication and more complex specialisation mechanisms, so there should be a video soonish on those developments. So I didn't mention it in the video, but a rudimentary form of chemotaxis is possible in the simulation, but never really seemed to evolve. The control networks have inputs that specify the chemical density gradients in the X and Y directions, but these didn't seem to be used by the cells. I think maybe fixing the gradients to the global axis could be part of the problem, so I want to try with an axis oriented to the cells direction. It might be that solving the reorientation problem is too tricky (and either way, fixed global orientations aren't realistic).
Nice! I think maybe the prevalence of network size to depth may just have to do with how the simulation itself isn't very complex, at least relative to a real protozoa. That or maybe it's more of a marginal benefit relative to depth increases, and with enough runtime, the size reaches a Nash equillibria, and network depth takes over. I recently learned about the stuff Michael Levin is doing, and his work seems to suggest you can affect the Waddington's landscape (at least temporarily) using electric fields. Maybe some way of making the pluripotency a function of a parameterised vector field is key for emergent control systems.
Question/Suggestion re: 8:02 Is it possible to let the cell effectively decide whether or not to split _with_ adhesion or _without_ adhesion (fission) to allow the potential for specialized reproductive cells? First, regarding splitting _with_ adhesion, I'd have to read through your code more to get better bearings, but I'm looking at Cell java >> burst method (line 474); I suppose you'd have to call createNewBinding(...) on the children when they're being created here to keep them stuck. I'm thinking: 1) for each child, attach to the previously created child, likely before or after the handlePotentialCollision bit (line 497 - 498). This could probably be done like: child.createNewBinding( _cam_, children.get(children.size - 1)), something like that. 2) for each binding in the parent's cellBindings, move the binding over to the nearest child. As for splitting _without_ adhesion, I was thinking that first, you'd need a property on the cell that determines how likely it is to lose adhesion when splitting (a trait that would be passed down and evolved/specialized). To implement, you'd have to alter the aforementioned "binding inheritance" if you may, to take that value into consideration and fail based on that probability. Just an idea I guess 😅
This is fantastic! Are you familiar with Michael Levin's research into morphogenesis? I think his work can offer some great conceptual starting points for cell specialization.
Great vid, would love to see more. May be wrong, may need to see more but it seems like the epigenetics and genetics systems here are functionally equivalant.
I'm very interested in how you're going about processing the large amount of Neural Networks. You mentioned that more layers result in a slower response time for the cells, are you executing NNs layer by layer across multiple ticks? A more detailed explanation would be greatly appreciated, as based on the github repo it looks like you tick and update each neuron every time the Network receives a tick which I would think would result in instant input=>output.
Each node in the network is essentially recurrent - it has a state, and each tick it updates that state according to the states of its inputs. There is one tick per time step, so in order for information to get from sensor to output it would require one tick for each intermediate neuron. It's important to note that each neuron updates their state synchronously with all the others. So at any given moment it looks at the current state of its inputs to calculate it's next state. Hopefully that clears things up a bit! Otherwise, processing the large numbers of networks isn't too hard. It's one update per neuron per cell (and plant cells don't have these networks). It's even embarrassingly parallel so the whole set of cells can be processed very quickly.
Theres another game called cell lab that is similar in that it tries to simulate basic multicellular evolution, though its more of a building game than an evolution sim. Its still very cool.
I've done a tonne of work on the code, wrote a paper about it, and gave a presentation on that at a conference. I'm currently writing, editing and animating another video on all this stuff, but it takes quite a long time (especially as a final year PhD student I'm super busy 😅).
