How I created an evolving neural network ecosystem 

You could use shape to show the speed. A faster Bibit could be more slender.

I would suggest adding three things: 1. Sexual reproduction, this would allow for much higher genetic diversity along with competition for mates 2. Objects other than food that the bibites can manipulate, allowing for the construction of structures 3. Vision? I don't know how easily this could be done, but if you could accomplish this, a whole range of new possibilities for the creatures would be opened up

Since it's a computer program shouldn't they be called Bibytes?

This is pretty cool, I am working on a similar (Very similar, in fact, down to using NEAT) project myself. One thing I have noticed in your video at around 3:35 is that your bibites can collide and grab each other, I suggest that you look at energy sharing, i.e. if two creatures are colliding and each are willing enough, they share their energy in proportion of how much energy the other has. I think it just might evolve synergy organisms or even multi-cellular groups of bibites, which would be very cool.

Concerning the colors: You can have different viewing modes where colors represent different states like the current action (mating, laying egg, searching food) or other states (energy level, diet composition, ...)

Add a pellet to individuals so they canibalize. Make it run on an android with screen-sized ecosystems and provide communication by bluetooth to other devices so they can migrate to other ecosystems. And let them make audio noise by intentionally bumping an on-screen red button, and hear noise by listening to the device microphone.

Did I search for this? No. Did I watch the whole thing? YES

While nutrients on earth are a closed loop system, Energy is not, plants absorb incoming energy from the sun, capturing it in chemical bonds. Maybe you could make a system where the food pellets are also evolving to cover maximum area to capture light.

I love this, its so interesting, i could see it for hours, if more variables were added id see it for even longer.

use graphs to represent what traits are most used, this would show what traits are the most favored for survival.

1:36 I love how there's just that one that just runs around in circles trying to get a pellet

Will you share the source code on git in description ? I'd be curious to see how it works. Maybe people can contribute

I just found this and I haven't watched your other videos yet, but it's exactly the kind of thing I've been learning coding for! I love the idea of ecosystem simulators as sort of digital fishtank where you can experiment with different organisms and how they react to environments and each other, with no real animals getting involved. I'm excited to see where this goes!

This is cool. Have you considered using CPU cycles as your "energy"? I don't know how to implement this but it would be aa more "real" limitation than an arbitrary amount of pellets.

I like the one that's running in circles but still out surviving most of the others.

Seeing this old prototype for the simulation is oddly surreal.

When I saw this in my feed I was like cool a simulated evolving ecosystem I haven't seen yet! Then I saw the channel name and I was like ah bin tabarnak ! Very well done, this looks promising

Brilliant. Thank you for this, its mesmerising. Great use of neural networks and I'm looking forward to future developments. Thanks again

This video got me going crosseyed in a good way. Good job!

Bro, RU-vid suggest me your video and I'm glad it did it ! I will devore all your content, it seems so interresting OMG

Merci beaucoup d'avoir entièrement traduis la vidéo en français. En plus, tu as très bien simulé l'évolution naturelle. Super vidéo

you gave me some ideas and inspired me to do some project

What a great video! It mixes so many disciplines!

Beautiful project my friend! :) Bibites are so cute! Thank you for sharing!

As far as having something to show their 'stats' in a visual fashion the first thing you need to do is limit the number of stats you want to show. Some are already visible. For instance speed. They already move on screen at their speed. The same thing with size as you can visually check that. If you can get the number down low enough then depending on how you have it set up you might be able to just have a few dots on their back of appropriate color. That way you can maintain their unique coloration. Also you have probably already read or heard this but adding a predator will not make your population problems go away. Your closed system idea will however likely do it for you (if the ecosystem doesn't collapse somehow).

you could have the antenna or mouths change color along the spectrum to indicate whether they're an omnivore or what. i'm really interested to see if they'll develop as predators or scavengers when the food dropping feature is added. Great work!

Great Work! Love your stuff

Excellent and inspiring. Thanks for taking the time to make a video

i think the colors should stay random. if it really works properly then the colors will end up being representative of different traits like speed anyway. My suggestion is to give the bibits a way to determine the fitness of other bibits, and prefer mating with ones who are more fit. another idea i had after watching many similar videos to this is maybe add more than one type of food. Like maybe 1 type is more numerous but provides less energy and another type is the opposite. Eventually I guess the idea would be that some bibits would eventually specialize in one food type (or maybe not, who knows). i really like the idea of a closed energy loop where dead bibits and/or waste turn into new pellets eventually. I'm really interested in this project and like how you want to work with a community to develop it more.. I subscribed.

2:59 Neuro Evolution of Augmenting Topologies 3:16 Stanley K. O. 2002 3:19 RT-NEAT

this was a very entertaining video, i hope you keep producing such video.

Very nice thoughts on how every feature brings with it an efficiency penalty! I'm very excited to see this unfold!

For some reason most of the comments are people with experience with similar projects or at least have experienced programming such things. Wow

I think that "baby's color" should be something chosen by the neural network, so color could act as a job assigner, like in ants

Came here from reddit, this is a really interesting project! Thanks for sharing.

its amazing to see what how much this have evolved from the beginning!

what about cannibal bibites who eats another bibites and choose target by color

Add classical physics Like basic Newton laws so they interact with environment Just spawn few 2D objects that just interact and pass organism's momentum to it. But floor should have slight friction as a fundamental property of basic area where it exists. So while organism takes in food, the part of food which is taken is converted to waste as it moves more and more cause it's getting energy wasted from friction, but here's a catch, were will the energy go after it goes to friction? Well that is take by the background (area where everything simulated exist) and that energy is collected, once enough of it is there background takes that energy and spawns an food particle, with random place. This is a great convectional cycle that allows your organism to interact with objects to create things on 2D plane! But things get tricker if u add something more....but what? Adding a condition! If and organism of specific type pushes few objects that could enclose an area but a little bit is still opened through which they can come in and out, then the background triggers to spawn food mostly inside that structure, this let's our organisms think and start reasoning for survival ! There is no limitations of what to expect as we really don't know how that kind of life would react to this type of change, *they are the once who change things not something already programmed to change without their will!* In a nutshell *you give them freedom of choice, a path that divides into two, a will* You say to them that, we grant you this power only if you fulfill this task, and so on. So this is it. I Hope this will be helpful THANKS FOR READING SO FAR GREAT PERSON! 😊😊😊😊😊

photosynthesis bibits that making eggs and green things

What a cool idea. I'm fascinated by this story of the bibits.

And we need to add sunlight to keep the " human simulation going We recently updated the global warming patch to see how the humans react

Feromones - add diffusion and hlaflife.

Super boulot, hate de voir la suite !

Simply amazing! Keep up the good work! :)

This looks very interesting. Have you considered using the GPU for computation? Algorithm would have to be changed a lot, but it would incredibly improve the performance. Or just try to utilize all CPU cores. Also you could add patterns to bibits to make them look more pretty, like dots, checkered, crosses and many others. Also you could add more needs than just food, like there could be a day-night cycle, weather, seasons, terrain with water sources and stuff. It would make the system so more complex but it would be incredible. You could make eggs need certain environments and make egg cost dependent on how quickly it hatches, how big of a range of temperatures it can handle and stuff. You could add more ways that bibits can communicate, for example sound, which transmits information instantly in certain range. You could also split world in chunks for optimization and allowing much bigger worlds. I know, all of the stuff I said is hard to implement and would increase project scope a lot, but I just think that it would be fun to experiment with those features. This project is just really awesome, I want to see it update and change over time. I wish I could run something like this on my pc.

They needs a purpose to collaborate, look for the exploration exploitation dilemma, zero sum games. And thanks for the French subtitles :)

Remember to post updates frequently! even if it's a small feature addition, since development isn't so fast i'm sure we'd love to see any small changes

Hey man, this project is really interesting!!. I love the idea. My only suggestions is add predation like you describe at the end of the video. Mix food/predation would be interesting. Another thing would add obstacles in their path. Another interesting is adding a seed which conduct to creation of pellets/food.

How is there only 1 comment this is CRAZY COOL!!!

I think the obvious Additions would be flocking or non flocking, and housing, like a bee hive or anthill

I really like this project. Keep us updated please :)

Such amazing project

i would be very interested to see some source code. I dont know if you are ever planning on making money with this or not, but if you publish it on github, a lot of people could help out and build up from your already very impressive base.

hi

2 years ago, I'm so happy that I found this channel

This looks great man.

they are already fighting each other for food. but they are reproducing asexually. my suggestion: try adding sex and sexual selection... at one point they'll start farming ;)

yay colour now we can have racism! AI really is becoming more advanced these days

For the colours: you can have the body colour being independent, and have another colour for the tail representing speed stats, mouth colour for strength, etc.

This is amazing!

A while ago I played a game “Sim life“ it made me want to program something like your simulation. I can think of a whole bunch of ways to make their behaviors more interesting, basically it’s make the world more complex and give lots of options for successful life strategies.

Absolutely fascinating work! I can't wait to see this develop further! You can have different amounts of pellets spawning on a timer to simulate seasons. Pellet Spring--> Pellets are numerous, but small. Bibites might select for speed and being able to sense pellets at a greater range. Pellet Summer--> Pellets are numerous and big. Bibites might select for making lots of eggs. Pellet Fall--> Pellets spawn in smaller amounts, but they are big. Bibites might hoard food or even begin farming. Pellet Winter--> Very few pellets spawn at all. Bibites might start hibernating if they can not sense food near them. Or, Pellets might develop seasonal diets, where they are vegetarian in the summer and carnivores in the winter.

I am creating the similar project in 3d in godot engine and youtube suddenly suggested me this video. It highly motivated me and gave me a direction on what i should do next. Thanks

I think it is a great idea to make evolution process to food as well

I think the idea of keeping color non-adaptive so that it can track genetic drift and speciation is really cool! Definitely worth keeping as a feature

Good work!

Neural Evolution of Augmenting Topologies sounds NEAT

I love this. You have a surprisingly complex ecosystem going. Well done. Suggestion: Maybe allow them to pair up and travel together? Maybe even groups of 3 or 4. This would make mating more efficient, and be safer if you implement carnivores. Downside the cost of covering less area when looking for food. It would be very interesting if you had diverse diets. Looking forward to more.

@6:45 color for traits would still result in identifiable species since "families" should be similar in color. you could use the three primary colors to represent 3 different important traits and the mix of how far those traits have drifted will determine color. You could also introduce simple patterns for this like stripe or XX pattern or something.

6:45 - One thing I did was to have two mechanisms available in the UI. Basically, when I click on a creature, I've "tagged it" and a translucent box with internal stats appears that follows it everywhere. I can click on that box to increase/decrease the amount of information present. The other thing I did was have a set of options on which critical things I wanted alerts for. I never finished this, but the ones that I played with was literally writing "Pred" temporarily right over the creature to show an offspring had been produced that evolved from herbivore to predator.

A good way to conserve energy is keeping the corpses around as food for scavengers, as well as allowing eggs to be eaten as they require a certain amount of energy to be produced.

I could imagine them instead of a trail or visual, you just have them "twitch" at each other, kinda like bees, or ant communication. Where they go up to each other, sort of dance, say something, the bees, or bibits, may understand where the food is now. On the map.

This is like thrive but even more of a simulation

The pheromones are a good idea, perhaps a bibite would be willing to give a pellet to a bibite that it likes based off its signal

First of all, this is amazing. I love it. I really enjoyed programming little population simulations back when I started computer science, but that was long before I knew anything about machine learning. As for your question about color-coding, I'm more inclined to think you should leave it. My reason why is that I think there are too many traits/feature-weights to map them to colors and still have those colorful be meaningful. I think displaying weights on input features may be interesting, if you can assign the size of the weight to the size of a shape, and place those shapes as identifiable parts of the Bibites. There's a Steam game called Eufloria, which gives an example of what I'm talking about - though with only 3 traits, not 20- input weights (more with hidden layers). It's not necessarily functionally useful information - because weights often don't tell you much information on their own - but it might allow you to create an even more unique visual character for them. On second thought... maybe you could use a combination of shapes and colors to represent the weights, and thus give a unique visual character to each Bibite.

OOH OOH OOH ! make a really cool custom forest background!!

I created much the same, only added hunters and made the entities talk to each other and experimented with all kinds of neural network hidden layers and cost function. Seeing emerging behaviour after a total random start is awesome :)

Nice simulation! I had similar project 10 years ago but less inputs. I used acceleration on output insted speed and creatures (mine was fishes) looked much more real.

I've liked this video for a while, but the from-scratch way you explained it has convinced me to implement something similar. I'm making a little "fish tank" to be displayed on a screen and run in the background. I plan to increase the maximum number of nodes in their neural nets and add "adding a node" as a mutation so they start off simple and get more complex if they stop improving with the same number of nodes

One of the things I noticed right off is the uniformity of the environment your creatures live in. Excepting the other creatures and food pellets there is nothing available for them to react to. Adding environmental obstacles which can either assist or repress the bibits' ability to survive and/or reproduce would be a great way to add interest to the project. Couple that with characteristics in the creatures that could potentially allow them to more effectively take advantage of the assisting type obstacles along with others that potentially allow them to more effectively resist the repressing type obstacles and I think the project would be far more interesting and diverse. Additionally, it would permit you to see what happens in different environments by altering what obstacles are present and in what ratios etc.. For instance, have two type of food pellets, one which is poisonous. Then implement a characteristic which determines how effective they are at discerning between the two. Also you have pheromone transmitters and receivers, but no antennas to determine the signal strength received or sent. This would be another good way to introduce more diversity into the population. For example, antennas can be either directional so they only detect pheromones in a forward cone or they can be omni-directional but weaker (less detection range). Food appears out of thin air. Instead, have an object that spawns the food somehow when they interact with it somehow. Maybe make it movable and make it so when two of them are positioned adjacent, they lock together and produce much more food, or much better food, than they do when separated. See if the creatures can figure out to push them all together.

cool ces bibites, on attend les coucouilles avec impatience !

You could add a seperate sprite to your bibites for every stat they have and then scale that sprite with the stat. Those fin like things at the side could be for speed and scale with speed for example

Some suggestions: - keep the main body colour of the bibets as you have it, I agree with you on that - Have a visual indicator for herbivore/carnivore ness, maybe the outline colour (what is currently black) - to display the pheromones have a "pheromone cloud" basically a circle around the bibet that is the colour that represents the pheromone and translucency of about 5-10% - have your bibets evolve to use the various pheromone and body colours (both the main colour and outline predatoriness) for determining mating i.e. some bibits are attracted to blue bibits and other prefer red or green bibits then you'll get to see if/how these attributes evolve - have some other appearance genes like wing size/shape variations, tail length, spots stripes, include these in the mating as well, then hopefully you can evolve several bibet species - rather than having a generic predatoriness, have the predators evolve preference for certain traits, some bibits might find longer tails more tasty and others might prefer purple bibets others might find red bibets poisonous and other might prefer to eat dead bibets - use waste (excrement & decaying dead bibets) as fertiliser for different pellets that "grow" (now you don't need to spawn them) maybe over time some bibets prefer pellets that grew from excrement and others prefer pellets that grew from decaying dead bibets Thanks for the interesting video hope you take some of these ideas

This is a really really awesome and very inspiring project! Software engineering is a huge passion for me and AI in general is a super interesting topic :) Thanks a lot for making and sharing this project with us, really really apreciated! I have been playing around with neural networks and genetic algorithms lately, so obviously I take a lot of inspiration from your project and have learned a lot watching your videos already. And I know this video is a couple years old now, but I have a few questions, maybe you'll see them anyway :D 1) At around 0:56 in the video you show an overview with inputs and outputs of the neural network. Input 0 is called "constant", what exactly does that mean? Are you feeding in some constant number or is this constant representing some sort of current state of the bibite or something thereabouts? Finding the right inputs is obviously pretty important and I've been thinking about that topic a lot lately, which is why it's very interesting and helpful to me to see how others do it :) 2) Is there any mechanism to make sure that bibites don't wander too far off the area they're supposed to be in? Or is the availability of food really the only thing to make them stay in that area? If so, I imagine that there's a very real risk for them to just die out early on if all of them just happen to race off into the sunset and never return, if you know what I mean :D Or do they just kinda loop back to the other side if they go off screen? 3) Do you normalize all inputs into for example the 0-1 range or are some of them fed in as is? I hope it's ok for me to ask those questions, sorry if they are too specific :) As you can see, I have a very big interesting in this kind of thing :D I would love to see some videos going into more of the technical detail behind these systems :) Though I can understand that you have got more than enough work on your hands already and maybe you might not want to share all your secrets :D Anyways, once again thanks a lot :) I love The Bibites and am looking forward to seeing how the project evolves in the future!

I think that when the bibets eat or move food to eat later, the food should drop 2 seeds, so that the bibets sort of act like "bees" and the flower releases "pollen". Then it need fertilisation to grow and maybe you could even awldd a water system. Little lakes (watering holes), where all the animals go to eat, some carnevoures maybe even would wait to ambush their prey there (by random evolution). Or maybe the foods only release seeds if they are moved around. This mechanic might force toffee to bibets to learn they have to move it around.

Complexity in behavior: you could add push, grab, release, break, try to eat.To really get things interesting you could make the environment more complicated so there are structures they can hide in or build to store food or protect their eggs, or whatever else they can evolve.

Something that you could add to your function is like a nest or shelter, allowing their to be a certain growthpoint to where's there's a nest Queen and adding like a apex predator to go along with hunting

As a French Canadian I find the name especially funny, as bibitte means critter to us. Great video btw.

This is very nice, for gonna check out newer videos. Ideas so far are: Take some inspiration from Gratuitous Space Battles, where some ships can be set to follow, escort or move in formation with other ships and some sort of aggression-avoidance parameters that make the flee or swarm enemies like piranhes. It should be doable to create swarm behaviour by copying/adjusting movement based on movement of creatures similar to them. Introduce an alpha-beta gene (scale 0-1) where alpha's will be bigger, stronger and possibly aggressive to other alpha's. Beta's will be more docile. Either will still be aggressive or protective against creatures that don't look very similar or have different pheromones. Introduce plant-creatures that use photosynthesis to produce seeds. Introduce some sort of shelter only small creatures can move through. And/or create a toxic environment that one can get resistance too. This can be combined with the pheromones into three different kinds of substances that affect different genomes differently.

This is so cooool!!! I've always wondered if we can do simulation of evolution. It would be awesome if we can play this as standby back ground on TV or desktop. It would be like an aquarium - then people would cultivate them like pets, where they have to monitor the parameters/environment to make sure they're alive or purposefully evolve cool looking species with lights or whatever

This is awesome! For the vision, you should use raytracing to draw a visual field for each creature with each ray as RGB input, although that may be too complex to optimise with the NEAT algorithm

Could you share this code you made ? It's dope and i would like to see how this is done.

I have two points : -Pressure is the key word to evolution. As you know, the reciprocal pressure of carnivorous-herbivorous is kind of mandatory. In order to make them have a pressuring behaviour, your should give them the capacity to detect not only close siblings but more importantly their place in your prey-predator gradient. But the point I wanna get to here is irl is more like a 2 pressure points ecosystem, as herbivorous and plants also do share a prey-predator relationship. As i'm maybe too much perfectionnist, I would suggest u to think about the ways dynamicly evolving species of "food" may be implemented, as i'm sure they can have surprising and interesting effects on the ecosystem if well thought. I recommand u to look at the "paper-cosplay" i did for fun long ago. I spent 3days of 18h creating plants ecosystems on "bad old" python 2, and I'm proud enough of my results to share them. It takes 2 minutes to fully understand so it can only help : www.ambryal.fr/ambry-flow-Une%20Petite%20Planete -One of the driving concept of life on earth is individual evolution (egg-baby-adult). It may not seem obvious, but this dynamic has major consequences, as it really is one of the main reasons the "species meta" gets a bit complex : species comes with many different ways to play around their youth weakness ! Implementing this kind of dynamics is really not simple as you would for the most part have to rethink the core. Although, I suggest you to think of some simple ways that can get you to approach this thematic : -Eggs costs ; time before born ; number of childs ; (child quality (hard with your core, but for example "do they become more quickly adults")). -Eggs as a privilegied source of food. Eggs are defenseless, and nutrient, plus the fact u kill ur ennemies before they are born. It sounds complicated, but I really think bibite's getting interest on eating others eggs and preventing others to do so (for example simply by laying far from others (and close to food sources etc...)) can grow the simulation to a next level. As a final word, i'm warning u that the main problem u will face is that "200 is not enough for a non trivial evolution simulation". Even if you only add prey-predator mechanics, i do believe 200 is not enough to get good results. I have unfortunately no tip for you about this issue, the only thing i can do is wishing u success...

I once did a similar thing, but this happened: After all the food was gone, leaving only a few bits left, all of the Bibites would clump together and share their energy while laying their eggs. When the food came back, they all rushed to consume it, and rinse and repeat.

8:50 "Ok lets rework the energy system to be more realistic." *shows a freakin LIGHTING bolt*

This is really cute and I love the idea of it. One thought I had, is that it might be nice if the sprites of the bibites themselves were also evolved, and if some sort of simple aerodynamic sort of analysis could determine things like their mobility. for example, you could check the two sides, seeing which pixels are bibites edges, and which are empty space, and look for any diagonal two pixels, and add those up to determine how forward pointy they are, to get some sort of speed score on that side. You could treat the internal color area as an Energy Storage capacity, to incentivise them to be large, but treat any holes in their edges as a deadly mutation so they need to be strong and well defined too. You could possibly even relate different colors to different traits, which might cause them each to evolve interesting patterns as they trade a pixel here and there for changing their stats

This is amazing! As someone who's just getting into coding, this is absolutely mind-boggling. Might I suggest using the color scale to indicate how carnivorous/herbivorous the bibites are (the more red, the more carnivorous, etc) and having stripes/polka dots as an independent gene that can identify species? All in all, very entertained. Good luck and keep working!

Perhaps consider adding static obstacles, as well as a method where the brighter their colors ae the more likely they are to be spotted and eaten by other organisms or vice versa, would be rather interesting

Very inspiring video! The 2nd NEAT-style video over 24 hours - it's a sign I need to start writing my very own AI aquarium. =D

You could add some personality with characteristic traits, such as tails. You could use a sin-wave tail that varies in length relative to the bibite's speed. In terms of pheromones, they can spawn discrete pellets that broadcast with a geometric falloff in intensity, and fade over time. You could also make it so the A, B, and C pheromones have individual characteristics in intensity and duration.