In these simulations the places where the children were “born” were completely randomized but in reality places of birth are also “inherited” from parents! If your parents lived at the Equator you wouldn’t suddenly appear at the North Pole! I think that is a very significant detail that should be added!
I think that both perspectives must be taken into consideration. The simulations one exclude that fact and by doing this it isolate somehow the brain from the nature so in that way we can study much faster the brain development related to itself. If we also study the brain by adding the missing factor we emulate better the development in real life (where we are passing information not only by genes but also by environment) which is also very important. We can even learn more by studying both sides and compare them.
@@olarmariusalex I agree. It all depends on the end goal: if you want to study one particular feature isolating factors might help narrowing down the amount of information that you’ll have to process but it will make the model less accurate. Adding factors will make processing harder but the model will be closer to reality. And there’s one catch - in real life all factors could be connected and it might be that in order to get some kind of result you would need to simulate all of them because with just one of those missing we might not have the right conditions!
Note, that it's a simple simulation. But you're right. Another issue: generations don't born at the same day, don't live same duration and don't die together. Generations overlap. Also, a gene should control the lenght of their lifes, another should define the age when they make children. Another issue: they should use and consume energy (by eating each other). Who runs out of energy, he or she should die. Can't stop: they should produce, waste (shit). Others should consume it, it would fill the energy, but not as much as if he or she eating other. A gene should control whether the particle can kill other ones (predator), or just eat, well, other's waste (vegan).
This model is simple about how early single cell organisms could interact and swim away from hot dangerous places. Food could passively diffuse into them can be assumed. Overlapping generations would maybe only relevant when the age sense is enabled and when age variable has some useful functionality.
Unfortunately, that wouldn't work here. The simulation is set up to see if the population can navigate to a certain area. If they generated close to their parents you would introduce a bias. For example, in the first simulation, all generations that began in the east would just keep reproducing regardless of their genome. You could change the selection criteria, but I think the way it is now lets you see what's happening clearly.
He lost me at being the first person to ever claim breeders are evolved an focusing on breeding is evolution. I don't think anything outside of this software he programmed to act this way supports this. Even breeders hate being called breeders yet according to this software it would be like being Kobe Bryant.
Really? I found it sort of sad. Poor guy is doing a simulation that proves the value of death, yet it is lost on him. What he should do is give the creatures compassion and hatred and then hardwire them so that hatred and, ultimately kill signals, are generated proportional to the product of the creatures compassion, the genetic similarity of any death in front of it and are directed at the assailant. Then he might learn something.
@@benearhart1224 i too am disappointed in the conclusion of the video. Author's own simulation tells him that violence works as good as peace, if not better, but he still associates it with feral instincts and regards it only as a tool of injustice and disharmony.
Its just sooooooo impressive that just a 4 genome computer programmed organisation could evolve that much for just a hyper random selection. Great job!
Still only 12600 views. This guy still has less than 500 subscribers. YT's algorithm is seriously broken. It recommends tons of mindless junk and doesn't prioritize anything of educational value. I guess it makes sense though, considering YT is more concerned with disabling useful features than they are with enabling people... *Cough* *Cough* "dislike buttons" *Cough*
Wow. The production quality, quality of content and narration - I'm speechless. This content is amazing and I wish you nothing but success. Very inspiring
i knew a lot of this bc ive watched a LOOOOOT of evolution vids. but im soo happy he assumes nothing. i would prefer that. this is a grade A video for people who know nothing about this thing! And he keeps it engaging too
The reason why "kill" trait was bi-stable is because there's no selection pressure associated with it. Would be cool to add a "kill the killer" trait and see whether it leads to a larger group surviving. Or will it start wars? :)
you could either get a mob mentality that swarm the killers, (without consequences, ie 'legal kill') OR a couple of 'sherifs' evolve that seek out the killers........
just what I was thinking, you should try to set it up to see if you can promote cooperation. One thought: if you have the kill gene, other individuals can see that and kill you ahead of time, you can call that the self defense gene. Another idea: Maybe then add Size gene to make individuals more likely to win a fight, However, being bigger will require you to collect more resources in order to survive. You can show it graphically by altering the actual size of the dots!
I did the same but even if i did not want to because i was so brainwashed by the tv i still qould have ended up this way because of how much truth can be proven and how everyone has more then 7 brainrotting seconds to open your mind to new worlds
Wait, so this is basically the only video he published? That's insane. Such an interesting topic and expertly explained, yet all done in a very accessible way. Thank you so much for sharing your work!
I am an old Irishman and can safely say this is a highly evolved and sophisticated individual who ain't got time for no youtube shenanigans! LoL Imagine them as like the hottest woman in the room. They ain't got no time for my silly shenanigans! It would take something special to draw their attention, or you must be lucky and witness their magnificence as it unfolds, as it did here with this video. Just discovering this myself, I imagine he works doing interesting things and, much like in this video, I imagine him doing these types of fun things in a world that doesn't involve youtube overly much. My own world doesn't include those things either. I am aware of this youtube thing right here, obviously, and love exploring the knowledge contained herein, but it isn't a priority. I am often busy just living life, in a world far far away, powered by an infinitely small, very subtle and basically nonexistent, singularity, within the singularity. It's something I fondly call the duality, the duality actually is Within the singularity. My world is like a black hole trapped inside a black hole, event horizon within event horizon. I couldn't make this up if I tried. How deep can we go? Perhaps not objectively real like the singularity, nor as magnificent, but my world is real enough for this old Irishman, and still magnificent because it was derived from such a perfect hottie of a singularity! Wow, what a babe, that source of my world. Can you imagine the magnificence of a hottie that is powerful enough to be the source of universes? Now THAT is a total babe right THAR! Right Thar! LoL haha sorry, shenanigans!
nothing gets you more excited than seeing some random dots trying to adapt to survive some random scenario! and seeing the way their brains were wired up to do so was just icing on the cake! perfect video!
I think this is the best video I've ever found on UA-cam. Not only for the content, although I love both programming and biology, but the way it's presented is genius. You have these structured parts, the simulation, but in between the simulations you have smaller parts explaining everything, really smooth. Thanks for making this video.
This was amazing. At no point in the video was I confused, bored, or even slightly distracted. I didn't even realize an hour had passed till you mentioned the the programs you used to make the simulator. Amazing talent and great idea, I loved every second of this. I can't wait to see the next video!
Sin leads to hell, keep focused, the devil is on earth to destroy your soul. But God wants to give you everlasting Joy. But our sin is keeping this from happening. You must stop sinning and turn to Jesus Christ he is your only hope. He can save you from eternal suffering under the Earth, where hell is hot.. Not everyone who calls me their Lord will get into the kingdom of heaven. Only the ones who obey my Father in heaven will get in. Matthew 7:21.....
Just when I think there is no hope left for the UA-cam algorithm, it throws this absolute gem my way. Thank you so much for such a good video! I have done many programing thought experiments in my head trying to do something just like this but there were a few things I didn't quite understand or know how to achieve and this cleared them up for me. I'm very excited to get the code and try some things out for myself!
I ran 50 generations and changed a couple of the config parameters. Excited to play around with it more when I get home. Edit: Ignore my previous edit. I did compile it in Windows, and will help anyone that wants to, but I found a repo that built this with SFML (usually used to make video games) and it's much better suited to a linux development environment. I really want to toy around with this code. There is a lot to learn from this guy.
I never ever saw a hour long video on UA-cam. But I was glued to my screen. Excellently edited, fun, inspirational and entertaining. Please make more videos. This is what I’m here for. 🙏
13:10 "The vast majority of them, the moment they're born, they have the inborn instinct to head east and just keep going until they can't go any further." - As a German, I know that feeling.
As one of the "programmer's friends", I very much appreciate the details providing just enough info to give me the urge to just do it myself as well... very nice video. Thanks to Posthumanist for "reviving" it and hence bringing it to my attention.
Interestingly, I think this also explains why some creatures have smaller brains than others… in your four corners experiment you would get to the point where having a massive brain has no advantage, if all they need to do is go is get to the corners reasonably competently. Our brains consume a large amount of energy, so in some situations having a larger brain to solve certain problems starts to have a disadvantage. It would be interesting to see a simulation that allows neurons to be added, but at a cost to find more ‘food’ then ramp up the challenges.
yeah the next step of those might be a cool scenario where mutation can change the number of connections and having "food" factor, maybe making creatures that found more food, then they consume survive and reproduce. then having something like 1 food is worth like 10 connections. so a creature that have like 300 connections need 30 food to survive.
A big brain is definitely evolved and maintained in a population only when life is a challenge that requires a brain. When survival is simple and intelligence is not required, a brain is an expensive luxury. There are primitive chordates called ascidians that have a brain as free-swimming larvae and lose most of it when they become stationary adults.
One way to sort of simulate the effect of a larger brain could be to add a delay of one simulation step to each internal neuron. It would enable more complex behaviors with loops but cause more complex networks to potentially move slower. It would require modifying the simulation code quite a bit though as the state of each network would have to be carried over to the next simulation step.
This is genuinely one of my favorite videos on UA-cam. If I could point to one piece of media that genuinely changed my life forever, it would be this video, because it launched my interest in programming neural networks, which eventually led to me going from a freelance artist background to a data engineer.
What's pretty interesting to me is that this exact method of programming neural net AIs is currently one of the most successful and promising machine learning models for many applications. It's a pretty perfect (and cool) example of the NEAT algorithm and a great illustration of how it works as a general method to make neural net AIs that are effective at solving all kinds of problems..
That’s amazing, what’s even more interesting is he hasn’t uploaded in 3 years. Bro probably has no clue he just changed the trajectory of your entire life
Just an update about Dave, for anyone wondering why he's not uploading: after making that video, he ascended in to Godhood and is now running his own universe.
He just made east side survivalist(beware west), Btw why he just halved their world? He could have drawn graph lines from top to bottom and from left to right up until it still reaches 50% of the world land and then the mutations wont have a simple survival instinct to run east, they will be forced to have a more complex survival instinct as the whole world will be looking like square boxes of inhabitable lands .
I've seen hundreds of simulation videos like this in UA-cam. this is by far the greatest one I've ever seen. You are a talented scientist, programmer and educator.
This was amazing to watch! As a biologist its sometimes hard to encapsulate the enormity of the genome but with small neural networks like this it can show evolution so succinctly, thank you!
800 generations at max and then drops nearly 1000x in 100 generations? and you are not doubting at all? :D im sorry to pull you back into reality but Jesus Christ is on the throne and the world is about to get judged.
@@lt3742 wtf are you on about? If God is real, he's a psychopath. I'll step to the left if his bastard son shows up, no way in hell would I spend an eternity with an emotional maniac like that.
Wow, I cannot believe I haven’t come across this video sooner. My interest in artificial intelligence began with the idea of natural selection simulation. You show your processes in such a way where there is inspiration to all viewers. I’ve taken away the logic and mathematic concepts in order to perform my own simulations in a more efficient and well thought out manner. I hope to see more AI from you in the future.
Watch carefully what happens around 40 minute 🙂 increased capabilities doesn’t really affect organisms performance already, not nearly as much as going from 2 to 8 neurons. You can have whole computation power in the world, but it wont really unfold any discoveries. Regular computers are already very capable of doing very complex simulations, even what’s shown is amazing. Having more power is cool, but I guess what I want to say is that you can do a lot even with mediocre hardware, don’t think you need a supercomputer to carry those experiments.
38:30 How it works is when N0 activates, N1 suppresses N0, and suppresses itself as well. This little solution here has this result: N0 triggers, triggering N1, which suppresses N0 and itself, effectively resetting the neural network.
Don't be discouraged by your level of understanding of coding. I've been a developer for 10 years and 95% of my work is very rudamentary. It will also get better as you build new things. Take python and your best ideas and run with it! I took lots of biology classes in college and there's plenty of work to do in your field!
This man is so wholesome in how he explains all of this stuff, it's like a 2nd Bob Ross. I wholeheartedly loved it. I love that I went down this rabbithole of weird minecraft glitches to Trackmania Tool Assisted Speedruns and then this gem of a video. You know, I hate to admit it, but sometimes the Algorhythm does not fail.
I wish I discovered this video while studying neural networks and genetic algorithms in school! This was very informative and very well done. I'd love to see more of your work in the future!
This is one of the best videos I've ever seen on youtube. And that's a lot. CONGRATULATIONS, amazingly well explained, programmed, everything. You can be sure that I'll share it as much as possible. Keep going!
I wish for an Evolution game like this where even non-programers can make their own input & output actions and many other things like resource, environment, speed, size... I seriously can't wait to see what I and many other people could come up with such a game. Love this video.
Evolution and AI seems to be barred from games as an intelligence inhibitor.. which really sucks.. a game such as worldbox would be a great starting point but it, an many like it are run by.. evil. I'm not sure on the best strategy to fight this yet. Suggestions welcome.
@@richardward6747 The ol' "if you want something done right, you gotta do it yourself" seems like the simplest way. Find some people that aren't greedy and evil and who share the same passion and do it. I'd help but... Today i was wrecking my brains trying to understand how to create classes and to count how many times a number occured in a list... In python... So yeah, good luck. I'd love a video game with hyper realistic characters and consequences. Also a great crafting, harvesting/gathering and skills system. But then again, if you were to acomplish what i imagine... You'd create a whole new, real world with real people, although seen as npcs by many of us (and i think we all know how gamers treat npcs)
@@migolan6606 thanks man.. I would love to program a good game, while I ain't a great programmer I probably could.. but more important things require my attention at this time.. maybe eventually.
@@richardward6747 do give me notice when you start, if you still remember. If i survive the wars, the ¥|RU$ and whatever else this decade throws at us and we still live in a 'peaceful' world, i'd love to make such a game too
4:00 DNA works not in single letters, but in "Codons" - that is the actual "letter" read out is comprised of a triplet of the 4 molecules. That opens up quite a lot more values per position in the datachain. That brings us to 16 values per molecule, times 4, makes 64 Codons/Values per actual read-out position. That is way denser information than hexadecimal (16 values per position). Two of these Codons (I'd have to look up the exact ones) act as "START" and "STOP" indicators for the Rhibosomes (molecule factories). Add to that that the Rhibosomes read forwards and backwards on both strands of DNA at the same time. This also allows for a lot of white noise in the DNA-"Code" to happen, which gives room for mutation (positive and negative) as well as padding-insulation against damage. Many "production-instructions" on DNA are also present multiple times for the same reason of contingency against error and damage, as well as for production quantity.
There is just one start codon: AUG, and there are actually three end/stop codons: UGA, UAA, and UAG. Also, it's spelled "ribosome"(unless you are talking about the band) Good info though 👍
Does that not also exponentially increase the complexity and need for precise compatibility to survive and be a compatible mate with another genome (because the video states one requirement is to find a mate to reproduce)?
I was hoping in the radioactive example that the mutation rate went up the more radiation a critter was exposed to. :) You totally inspired me. I am going to be coding instead of sleeping now. :)
Holy moly. After grinding coding challenges for three years I got a bit tired of the whole ordeal, but this one sure relights the fire. Can't wait to create a simplified version of this.
This was great to watch. I'm really glad the algorithm worked in your favour. I wish there was an easily approachable way to try out your software, as a microbiologist I am not that versed in getting stuff to run on linux. In any case, this is a great video, your narration and explanations are awesome :) Thank you for all the effort you put into this
Don't you mean this was "great great great great great great great great great great great great great great great great great great great great great great great great great great" to watch?
@@Ty-mf3vz the term worked in someone's favour doesn't imply that it isn't subjective or unbiased. It means exactly what you want it to mean. It basically means that despite the inherent randomness he got lucky that it worked out the way it did.
@@Ty-mf3vz Unbiased? What makes you think the UA-cam algorithm is unbiased? In fact, they make a point of telling us of at least some of the biases they employ.
I think this video just finally helped me understand what i was struggling most to understand about what it takes to develop my own ai models from scratch. Thank you very much! Now to put these thoughts into practice
David , you have had two and one half million views in less than 2 years. That should tell you that you are very good at mtaking a complicated subject and explaining it in a way that people will listen to. It is a shame that you called it quits.I for one wish you had not made that decision. Best to you and thank you for all the effort and work you put in.
38:32 oh my God. I think that's a primitive form of memory. I was thinking about the loop in the first run how the internal neuron had a negative loop to itself. I intuited that it was basically a flipper that flipped the neuron back and forth so that it wojld move randomly every other step. This seems to basically remember what it did for the last 2-3 steps. It's got the same negative self flipper but it also inverts itself into the neuron linked to thst sensor. N1 basically inverts itself every step while also adding influence back to N0. Which means certin locations of x over time will do different things, depending on the cycle. Basically it alternates between moving west and moving random, but not every cycle. Usually it will do this 50/50 but the higher Lx gets, the stronger the strength of each subsequent flip will be. But N1 also inhibits N0 so it only really does this every other step. OH! Because it only wants to strengthen it when it's already strong, otherwise it would become neutral on the off cycle. Brilliant! I see that Rnd (random?) is linked to move random. So basically the stronger N1's flips are it will eventually overpower Rnd and move west more often instead of random. Or if Lx is shrinking, it will move random more often than west.
I'm trying to make my own (much simpler) project following the same ideas, but I'm stuck at the internal neurons. I don't understand what I want from them, and it seems like you might. I tried poking around in the linked code, but haven't found it yet. If you could point me in the right direction, or share a resource that explains the concept, I'd appreciate it :)
@@Dnallohes They're just there to be generic intermediate nodes that are able to connect to each other and pass signals around. Just having them available for this increases the number of different possible ways that the input signals can flow through the brain and be linked to each other, and therefore the possible complexity of the AIs behavior. If you're familiar with electrical engineering, you'll see quickly that one of the things neurons can do, just by being connected to each other in certain ways with signals going through them, is form various kinds of circuits, including things like logic gates (AND/OR/XOR/NOT, etc), timers, and memory circuits. The signals from the inputs going through these circuits allow the neural net brains to hold and act on more possible informational states, process the inputs in more complex ways, and even learn to correlate them (since the internal neurons can connect to each other as well). The more neurons & connections available in a neural network, the more complex its circuitry, and therefore its processing, can ultimately get. It can develop memory, learn to recognize patterns, adapt to changing conditions, get and use feedback on the sensory results of its previous actions, etc, ALL just depending on precisely how the neural circuits in its brain are connected. The neat thing is you don't have to understand what it's doing or program the neurons to do any of this, just let them connect however they want and mutate randomly. Over generations of selection, the successful surviving neural networks will become smarter and smarter (up to their capacity), evolving the right wiring to succeed (and/or call down SkyNet... but that's what you get for activating the 'kill' neuron...)
Evolution lol. Show me creating life from non life. You dont get that as a gimme. Evolution trying to explain what happens AFTER life is already there is fun to think about....but its not science without showing life can just happen out of nowhere.
@@KWifler Explanations are not science. You have to do it physically or it isnt in reality, just in your brain. Which is the entire point of my comment lol.
Hoooly! I've been thinking about trying genetic programming for a while now, and this is probably my favorite video on YT now! Thank you so much for this! This is incredible.
Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont
40:45 One interesting thing is that in reality, neurons take energy to use so they increase the needs of the organism. And as you shown in the exemple there is diminishing return when it comes to the result of increasing brain size. The smallest brain size was awful, but just a few neurons were enough to make it almost as good as the biggest brain. This probably is one part of the explanation why most organisms in nature are extremely simple. Only the most successful creatures with bigger brain benefit from it, while smaller brains require less energy and have less pressure to succeed. It's obviously more complex than that but there is definitely a tension between the cost and benefit of brain size. While in your simulation if brain size could evolve alongside brain structure (a gene causing extra genes or neurons to appear in the next generation) there would probably be a complexity creep that would never stop because there is no limit to growth, at least until it becomes unsustainable and crashes your computer and oh my god this became a global warming analogy! Except that in this case they would rip the fabric of spacetime without even having any way to realize they have an effect on it. Creepy...
An interesting related idea is the observation that brain size changes tend to lag behind body size changes, that is to say, if your body is growing in size over evolutionary history, your brain size doesn't increase at a proportional rate. This had led to the hypothesis that reduction or gain in body size primarily leads to disparities between brain and body sizes. I seem to recall reading about size reductions in our primate evolutionary history for instance. Non-migratory birds tend to have more encephalization when compared to migratory birds - likely because nervous tissue is too energetically costly when competing with similarly expensive migration patterns where birds might have to fly thousands of kilometers. Our large brained non migratory birds dont have to worry with competing costs, they just need to support their larger brains. It's hypothesized that larger brained birds are more behaviorally flexible and can capitalize on niches that aren't available to smaller brained individuals. This would stratify where certain individuals of a species can survive (based on individual brain differences), acting as a soft reproductive isolation. Which may have contributed to the diversity of avians fauna today
So this guy came out here and dropped this absolutely banger of a video, got 3M views and 40K subscribers and said “I’m out!” I was hoping I could binge a ton of his videos. This was so captivating, entertaining and educational!
As a Computer Science student and also a person who is interested in evolution and biology, this video is absolutely fantastic. I usually get bored and just stop watching if the video just gets boring but not this one.
I started watching the vidéo, and it was already the end, that was really good, I didn't see the hour pass at all! Thank you for this great guide and for even sharing how you coded it, it really is great materiel to learn from!
@@stanleydavidson6543 though the credibility of your belief lacks legibility or factual reference, I’d dedicate time to follow up any sources you have if you can provide links as a counterpoint to this extremely detailed and widely respected area of study. Much more worthwhile to share links and peer reviewed scientific data than argue our relative understanding of this phenomenon
"and it also has to be sufficiently socially adept to find a mate" ok man, there's no need to attack me like that. memes aside, this is a super fascinating video, i have skipped a total of 0 seconds.
David, I watched your video this morning after learning about The game of life. You stepped it up with the little brains and neural connections, genes and mutations. You are excellent in didactics. Please go on. Your channel is awesome.
I would love to see a continuous simulation instead of a generational random redistribution. Instead, I think, adding a birth rate along with a slowly changing environmental challenge would be more analogous to what we see here on earth. This would replace simulation cycles with time, more like a real world circumstance.
You have the genuine skills of a documentary maker - I could happy sit through another hour of similar stuff. Also I'm super appreciative of the explanations, it's literally the first time my knowledge of neural networks has surpassed 0% (though I still don't get the "internal neuron" purpose).
Me too! I've been combing through the code trying to figure out what the internal neurons do, and I can't find them. What I've been able to find in the code comments is what's explained in the video. I'd love to make something similar on a simpler scale, but I just don't know what's going on there.
check out some videos about neural networks (keyword: perceptrons) which mention the XOR problem… sounds complicated when you don’t know and/or/not/xor logic, but it turns out to be really simple when explained well. The internal neurons allow XOR which is not possible with only two layers of neurons.
@@Dnallohes For that you need to learn bit about neural networks.... I would suggest watch video 'Perceptron' by Carnegie Mellon University Deep Learning
@@j.j.maverick9252 Yes. Perceptrons have only one layer, so they can't determine parity differences. The more inside layers of interconnected neurons, the more intelligence. Humans have many layers in their visual processing, and that's before the visual information even makes it to the actual brain.
As a programmer myself, with a strong interest in biology, I want to thank you for producing the best video I have ever seen on UA-cam. Fantastic work. I learned a lot and have tons new of ideas and inspiration
This is a very nice and interesting insight into evolution. I don't find it surprising though, that the kill neuron soon predominates the respective simulation. David accidentally incentivized killing contemporaries. The reason is, that his populations have the same overall size from generation to generation, but only survivors do reproduce. This means, that if fewer entities survived, the survivors will create more offspring (per survivor). So by killing others, his entities have more offspring, while reducing the offspring of the competition. It was unavoidable that his little creatures became murderous :) - on the plus side: Different environmental conditions (like in the real world) would incentivize different behavior, and thus not have this drastic an effect.
I was sort of wondering about the reproduction factor too! Like what if the population is allowed to grow/shrink based on the number of survivors at the end of each generation, rather than entirely repopulating the beginning of each generation? It seems to me that this would have a significant impact on the applicability of the simulations' outcomes to nature. But then, I also see the advantage beginning each simulation with a standard population size. It just makes things so much cleaner and easier to compare, not to mention keeping CPU usage manageable. And I guess you can think of them as samples of a larger, unsimulated population? I feel like there's something wrong with that view but I can't put my finger on it. Super interesting video though! Makes we want to try all sorts of different configurations
The real world almost certainly favours other traits than killing, until congestion happens. Unfortunately, direct cooperation is many tiers higher up in orders of required logic, and can therefore be preceded by the occurrance of congestion.
Poor guy is doing a simulation that proves the value of death, yet it is lost on him. What he should do is give the creatures compassion and hatred and then hardwire them so that hatred and, ultimately kill signals, are generated proportional to the product of the creatures compassion, the genetic similarity of any death in front of it and are directed at the assailant. Then he might learn something. We might all learn something of highly contemporary value. Because, nothing spills more blood than a bleeding heart.
I think it's counterintuitive that the kill neuron did not actually "kill" the other cells. That is, each next generation starts with the same number of cells. Also, killing usually has some kind of benefit associated with it. Most killing happens for sustenance. If a kill was rewarded with a higher reproduction rate, things would get more interesting.
Those that are killed are replaced with the children of those who survived. If few survive they will have many children to get back up to full population size.
the pheromones and genetic similarity sensory inputs could be used to preserve total population while minimizing deaths with more complex reproduction conditions that better utilize the extant pieces, you could easily get more interesting results. ironically, the built in evolution of the sensors was the least effective of all, as it was designed for an environment that the cells did not exist in. arbitrarily setting a zone to reach, then spreading the offspring at birth is foreign to say the least. the radiation example is better designed, as it allows the age or oscillation sensor to better direct the cells. tying any conclusion from this simulation to the real world, while the selection pressures are so vastly different between the two felt pretty silly as a cap to the video, though i am grateful that there is room for me to mess around with the code and answer my own questions
I would love to see sims with some kind of consumption requirement introduced. For example maybe, every 2 cycles inside a generation each individual must consume at least 1 bit. At the beginning of each cycle 95 bits to 100 individuals are introduced to the environment. Maybe a random mutation could allow for space to stockpile or another to only need 1 but in 3 cycles.
That would be really cool but every additional behavior will make it take exponentialay harder to simulate. We need to get him more computing power so he can keep it going.
@@wastedtalent1625 yes I can understand that. Just adding cycles inside the generations almost makes a single generation relative to an entire current sim. Still, I know he could do it. Wish I could somehow contribute even if it was just time proofreading code but I don't know how and I'm not working... There are now other sim games somewhat similar but I really like the user friendly results of these sims. Also, I feel the others are often biased towards increasing sales to non-scientific interest.
@@wastedtalent1625 wonder if this could be done on a "folding at home" type thing. I forget the name, but there is a tool that can be used to convert a regular program into a "folding at home" type thing
"Those who don't [reproduce] were just unlucky with mutations, and they don't have the brain wiring to know how to make it to the spawning areas" And i took that personally
This video changed my perception of life and the universe. Really great stuff, I re-watch it occasionally and this time I took notes. You've inspired me to finally try to create my own simulation. Thank you for a truly great video!
This one went straight to my favorites. I'll watch it and re watch it so many times I'll know the words from the top of my head. You, sir, are an awesome square on our grid.
As a game designer whos very enthusiastic about procedural processes and evolution this was very informative and entertaining. I really liked the structure you gave the video and the way of narration. Although I am very inexperienced with neural networks I feel like I could reproduce this after this video and who knows maybe ill give it a try. Good Job!
The reason the killing creatures thrived is because there's no repercussion and almost no cost for killing, it only takes up one neural connection slot. There's also the fact that the more they kill, the more they get to reproduce. It would be interesting to see what happens if there was more mechanisms related to killing, for example: - make creatures unable to move after killing for some turns - give creatures the ability to choose to avoid reproducing with killers - give creatures the ability to defend themselves
My man, you are no less equal to Primer and other talented simulation programmers out there. How comes you have been so underrated. I hope the algorithm aids you as it has guided me to this masterpiece.
Not exactly a video game and I'm not sure how much/where it is still available but there used to be a pretty neat 3D evo sim called 3DVCE that evolved 3D creatures composed of evolved physical bodies that moved with simple, evolved neural networks. The selection options were pretty limited but you could get some really really interesting results if you ran the sim long enough.
This is great! I might have to dig into this because there's several things I want to play with (individually and in conjunction): * setting mutation rate as an output gene (what's optimal?) * setting neural complexity (# of both connections and internals) as output genes * making it so that reproduction happens specifically only between individuals ending next to each other (curious about how that would effect murder rates in particular, and would they evolve to use genetic similarity input to avoid killing potential mates) * food chain simulation; neural complexity as output gene as above, but higher complexity can starve and must "eat" (kill) to survive (curious how many will prioritize simpler non-eaters versus complex eaters) * like the radioactive challenge, but mutation rate depends on amount of exposure (will they evolve behaviour specifically to find best mutation rate, and will it match first item above)
Wow. Perfect presentation. The best I have seen. You sir deserve an award of something. Super interesting, well explained, it keeps one's attention all the way to the end. A master piece, well done. This was an inspiration.
Wow. The video came up in my recommended last night and I added it to my watch later list. I'm glad I didn't try to watch this last night because I would have gone to bed an hour later than I wanted. The presentation in this video was amazing. You're explanations are great. And thank you for sharing the code.
I'm so glad this video is finally getting views. This was an amazing watch. Honestly kinda sad more complex survival criteria weren't used. I would've loved to see solutions that make use of pheromone as a result.
@Aurora Peace Yes. I was one of the first few to watch it. It sat at less than a thousand views for a really long time. It's only recently that the algorithm picked it up.
I had to go look up what a hyperbolic tangent function is, which I needed to look up what Eulers equation was, which I then needed to look up what eulers number meant, all because I wanted to understand what was happening in the video. This is one of the first videos I’ve watched that made me so interested in something.
This has made me rethink what "learning" means. You can see here that learning could be considered a complex form of filtering. It also makes intelligence seem a lot less special, because intelligence is a side effect of survival selection based on progressively larger environmental challenges. The more environmental complexity there is, the more "intelligence" must develop by filtering to navigate it. Fascinating.
Yep, and one machine learning technique used to design AIs involves doing exactly this. Genetic algorithms, which run many generations of concurrently running and competing AIs that likewise can be neural networks, but don't have to be, as long as they have a way to process inputs and map them to actions, a sufficiently complex brain, and a random factor (like mutation), which as we've seen here is all that is needed to allow you to allow the AIs to evolve adaptively in order to get better and better at solving a problem.
Im not sure I would use the term "learning" - Learning as I understand it implies some conscient action with the goal of expanding ones understanding. This is adaptability which in itself doesnt require any form of goal or conscient understandings. It is stored as innate abilities that the gene carries with them regardles of the mental capacity and/or intentions the carrier brings to the table. Its the same if its a earthworm or a human. And I dare claim we cant learn an earthworm anything. But they do have the reflexes/autonomous behavior that prevents them from drowning when it rain, or fry if the get exposed to high temperatures. They have adapted..
Yes I was thinking on the similar lines that teaching and learning seem to be added functions of evolution . To transfer knowledge of survival, knowledge on top of your genomes knowledge/information on how to survive ( or the ones that could not be transferee through the genome )
@@martinwinther6013 Traditional neural network implementations 'learn' more in the way you are thinking of... the system is 'trained' with input and programmed to get varying levels of reward from the results of its chosen actions which it uses to reinforce or weaken connections in its brain. This lets it learn what works to solve the problems it needs to... This is also more like how a real brain learns during an organism's lifetime. There are some problems with implementing this though, it gets complicated figuring out how many layers deep you should make the connections, how malleable the connections are, etc... The genetic algorithm technique here builds the connections in a different way, using selection, randomness, and evolution to build brains that already have wiring that can solve the problem... This is more like how real organism's brains evolve their base wiring, the things like instincts and effectively controlling the body. This has shown a lot of promise in developing AIs and is pretty much the new way AI developers are going. The term machine learning here is still valid though, since the end goal is getting an AI that is properly trained to solve a problem. This is still a process of giving the AI all kinds of input, observing their output, and rewarding successful results in some way (in this case, with 'survival' of its better qualities to pass on to further 'generations'), still these generations were really just repeated training iterations for the working AI that is the end result... When you look at this effect in nature, you see how it also almost has to result in intelligent (or rather, at least sufficient for survival in their environment) behavior in any species which can evolve. A species 'learns' its basic behaviors, those required for it to survive, as a population (or rather, the surviving portion of it...), in much the same way these things do.
2:00 Evolution doesn't have to explain where the first self-replicating molecule came from. Chemistry does that just fine. Evolution takes place in the latter part of this inevitable chain of events in certain optimal conditions: 1: Atoms form molecules. Happens quite often. 2: Molecules are sometimes formed that oscillate and move. 3: Molecules that oscillate and move sometimes happen to move in ways that construct other molecules. 4: Sometimes molecules that construct other molecules happen to construct copies of themselves. 6: Molecules that self replicate become super abundant. (evolution) 7: Random events lead to errors in the copies (evolution, mutation) 8: Molecules that happen to mutate to replicate faster or better become a the new fittest replicator. 9: As atoms and molecules to build from become scarce, replicators that steal from other replicators keep replicating and become fittest. 10: Replicators that happen to build little shields for themselves survive better and become fittest. It's all inevitable, and doesn't really require proof other than "chemistry" as a whole.
Thanks for the explanation! Now the only deep questions remaining are located before the 1st step, one of which being "how are attoms formed?" But I guess the more we learn about this world, the more there is to question huh?
@@firegator6853 Very true. Why does a universe arise, why does a universe behave as it does, can it behave differently, and so on. There’s no end to that.
So why doesn't it happen now if it was that simple? The hypothesis that early earth conditions somehow facilitated something thwt cannot happen now is just that, a hypothesis.
Some things to note for future experiments: 1. Add some sort of "food" system and run some tests to see how the creatures evolve with limited food. 2. Add an input neuron that can detect when a nearby creature is killed. Perhaps that would lead to creatures "running away" from killers? 2. Add an input neuron (or neurons) that can detect when a nearby creature is not moving where it should be, and a corresponding output neuron (or neurons) to allow for one creature to sacrifice all other actions it would take on that simulation step, in order to move another creature in whatever direction. Basically a "see a creature in need, help that creature" system. Could also lead to some interesting interactions with the proposed food system and killer detection.
This simulation was such an excellent explainer for how evolution works to generate advantageous adaptations in a population, so amazing job! I am a little embarrassed that I started to get emotionally invested in some of the dots though lol I know they aren't real but it was so heartbreaking to see a little dot just hopping around on the south border, never to make it to the spawning point
Hi David, this is such a great video, I not only learned a lot about mechanisms of evolution, but also about didactics. You considered your viewers' backgrounds, presented the whole before the details and made it a wonderful experience. Truly amazing!
You sir are truly a man of science. Your knowledge and attention to detail show profound wisdom. For example, while talking about your postulates for evolution you mention self replication. Over the screen you have a floating hydrocarbon! The molecule theorized to have given rise to cells and isolated organic environments. I have worked in a molecular biology lab for a few years now and I salute you.
As a 3D artist building my way through videogames industry, I think you just changed my future line. I know I may say some dreamy stuff right now but, I see a future where video games will be.. real. Just like in Tron film. Npcs will feel like really understanding their environment. I’m so dam curious to just create some basic 3d mesh with an AI that deforms it based on mutation factors. The AI could even change the rigging structure of the creature based on movement and animation necessities. There is so much free and non explored field in this... just imagine a new SPORE game released with those features
This is my offering to the UA-cam algorithm.
Let's hope we resurrect this channel so David can make more amazing videos.
Seems to be working, 140k views. It's an extremely well put together video
UA-cam algo took me here from a video about WW1 lol
They really want people to see it again
The algorithm has called us.
Count me in for the algorithm summons
boost
In these simulations the places where the children were “born” were completely randomized but in reality places of birth are also “inherited” from parents! If your parents lived at the Equator you wouldn’t suddenly appear at the North Pole! I think that is a very significant detail that should be added!
I think that both perspectives must be taken into consideration. The simulations one exclude that fact and by doing this it isolate somehow the brain from the nature so in that way we can study much faster the brain development related to itself. If we also study the brain by adding the missing factor we emulate better the development in real life (where we are passing information not only by genes but also by environment) which is also very important. We can even learn more by studying both sides and compare them.
@@olarmariusalex
I agree. It all depends on the end goal: if you want to study one particular feature isolating factors might help narrowing down the amount of information that you’ll have to process but it will make the model less accurate. Adding factors will make processing harder but the model will be closer to reality.
And there’s one catch - in real life all factors could be connected and it might be that in order to get some kind of result you would need to simulate all of them because with just one of those missing we might not have the right conditions!
Note, that it's a simple simulation. But you're right.
Another issue: generations don't born at the same day, don't live same duration and don't die together. Generations overlap. Also, a gene should control the lenght of their lifes, another should define the age when they make children.
Another issue: they should use and consume energy (by eating each other). Who runs out of energy, he or she should die.
Can't stop: they should produce, waste (shit). Others should consume it, it would fill the energy, but not as much as if he or she eating other. A gene should control whether the particle can kill other ones (predator), or just eat, well, other's waste (vegan).
This model is simple about how early single cell organisms could interact and swim away from hot dangerous places. Food could passively diffuse into them can be assumed. Overlapping generations would maybe only relevant when the age sense is enabled and when age variable has some useful functionality.
Unfortunately, that wouldn't work here. The simulation is set up to see if the population can navigate to a certain area. If they generated close to their parents you would introduce a bias. For example, in the first simulation, all generations that began in the east would just keep reproducing regardless of their genome.
You could change the selection criteria, but I think the way it is now lets you see what's happening clearly.
It took just under a year for this video to start getting the traction and recognition it deserves
viewers habits evolved to share it?
TRUE STORY:
Our owners like to steer us for a reason.
I don't know where the journey ends,
but the ride is very enjoyable.
Long live Our Owners!
He lost me at being the first person to ever claim breeders are evolved an focusing on breeding is evolution. I don't think anything outside of this software he programmed to act this way supports this. Even breeders hate being called breeders yet according to this software it would be like being Kobe Bryant.
@@willyreeves319 youtube algorithm evolved to feed it to the right crowd
That's pretty quick for UA-cam tbh
Man was consistently like "You can skip to the next part if you don't care about this topic" and I'd say most of us absolutely did not skip ahead.
inventive
did not skip
Attention span passed ✅
Shut up bro
i did
This man created evolution and decided his UA-cam career was complete
Maybe the evolution stopped him 🤯
and Shakespeare
@@lit22006you dont need to have 100% info in a subject in order to talk about it. none of us is perfect, its the progress that counts.
What are you blathering about@@lit22006
He is active on github
I like how David feels sad about these computed creatures dying or getting murdered.
Really? I found it sort of sad. Poor guy is doing a simulation that proves the value of death, yet it is lost on him.
What he should do is give the creatures compassion and hatred and then hardwire them so that hatred and, ultimately kill signals, are generated proportional to the product of the creatures compassion, the genetic similarity of any death in front of it and are directed at the assailant.
Then he might learn something.
@@benearhart1224 thank god your not smart enough to this
@@Taboomix oh man that just made my day. Insults about intelligent in screwed up English - the best.
@@benearhart1224 oh nooo i forgot a word and damn life must be sad then
@@benearhart1224 i too am disappointed in the conclusion of the video. Author's own simulation tells him that violence works as good as peace, if not better, but he still associates it with feral instincts and regards it only as a tool of injustice and disharmony.
This AI, Computer Science, Biology…gold. These type of content should be the one getting viral.
If only he added tiktok sped up music
Facts
@@Lemjanmusicbruh 🌚
@@Lemjanmusichell no
@@soulslip 😂💀
Its just sooooooo impressive that just a 4 genome computer programmed organisation could evolve that much for just a hyper random selection. Great job!
95 views.... it deserves millions!
At least it now has a bit more but still not enough
Soon...soon. It just showed up in my suggestion feed _two days in a row_ ...The algorithm is speaking.
Still only 12600 views. This guy still has less than 500 subscribers. YT's algorithm is seriously broken. It recommends tons of mindless junk and doesn't prioritize anything of educational value. I guess it makes sense though, considering YT is more concerned with disabling useful features than they are with enabling people... *Cough* *Cough* "dislike buttons" *Cough*
16k now
@@flyingsquirrelfpv4866 17K
Wow. The production quality, quality of content and narration - I'm speechless. This content is amazing and I wish you nothing but success. Very inspiring
i knew a lot of this bc ive watched a LOOOOOT of evolution vids. but im soo happy he assumes nothing. i would prefer that. this is a grade A video for people who know nothing about this thing! And he keeps it engaging too
The reason why "kill" trait was bi-stable is because there's no selection pressure associated with it. Would be cool to add a "kill the killer" trait and see whether it leads to a larger group surviving. Or will it start wars? :)
Can we please see this program simulated?
Great insight
you could either get a mob mentality that swarm the killers, (without consequences, ie 'legal kill') OR a couple of 'sherifs' evolve that seek out the killers........
just what I was thinking, you should try to set it up to see if you can promote cooperation.
One thought:
if you have the kill gene, other individuals can see that and kill you ahead of time, you can call that the self defense gene.
Another idea:
Maybe then add Size gene to make individuals more likely to win a fight, However, being bigger will require you to collect more resources in order to survive.
You can show it graphically by altering the actual size of the dots!
@@mattsowerbutts4163 That's a great idea! mob swarm
love it
I have replaced TV with UA-cam 15 years ago. This is one of the best videos I have watched on this platform.
I did the same but even if i did not want to because i was so brainwashed by the tv i still qould have ended up this way because of how much truth can be proven and how everyone has more then 7 brainrotting seconds to open your mind to new worlds
I’m now convinced, we are all in a simulation
Clearly never watched charlie bit my finger 😂
Wait, so this is basically the only video he published?
That's insane. Such an interesting topic and expertly explained, yet all done in a very accessible way.
Thank you so much for sharing your work!
He has a few more on Vimeo...
I am an old Irishman and can safely say this is a highly evolved and sophisticated individual who ain't got time for no youtube shenanigans! LoL Imagine them as like the hottest woman in the room. They ain't got no time for my silly shenanigans! It would take something special to draw their attention, or you must be lucky and witness their magnificence as it unfolds, as it did here with this video. Just discovering this myself, I imagine he works doing interesting things and, much like in this video, I imagine him doing these types of fun things in a world that doesn't involve youtube overly much. My own world doesn't include those things either. I am aware of this youtube thing right here, obviously, and love exploring the knowledge contained herein, but it isn't a priority. I am often busy just living life, in a world far far away, powered by an infinitely small, very subtle and basically nonexistent, singularity, within the singularity. It's something I fondly call the duality, the duality actually is Within the singularity. My world is like a black hole trapped inside a black hole, event horizon within event horizon. I couldn't make this up if I tried. How deep can we go? Perhaps not objectively real like the singularity, nor as magnificent, but my world is real enough for this old Irishman, and still magnificent because it was derived from such a perfect hottie of a singularity! Wow, what a babe, that source of my world. Can you imagine the magnificence of a hottie that is powerful enough to be the source of universes? Now THAT is a total babe right THAR! Right Thar! LoL haha sorry, shenanigans!
@@QuantumElectroDynamic why are you so thirsty for this man lol
@@QuantumElectroDynamic one of the best binary drinks I had
@@QuantumElectroDynamic lmao 😭
"Your neurons are valuable!. Take care of them."
0:00 --Introduction
7:57 --Simulation #1 (how it works)
20:07 Brain anatomy details (neural networks)
27:16 Simulation #2 (mutation and adaptation)
32:12 No mutation (mutation rate to 0)
34:17 Different brain sizes
35:52 Simulation #3 (brain sizes)
41:13 Genome encoding
42:42 Simulation #4 (the "kill" neuron) (kinda sus)
49:57 Software used
52:14 Simulation #4.2 (radioactive challenge)
55:35 The end
hardest quote 2021
Your evolved!
@@User-Seven-Teen
But what the hell this is all about? 😂
I don't understand anything
Sussy
@@lorenzo689 Sussy Begula
nothing gets you more excited than seeing some random dots trying to adapt to survive some random scenario! and seeing the way their brains were wired up to do so was just icing on the cake! perfect video!
yes. riveting.
@@VickMcbread I know right!
Just like us
I think this is the best video I've ever found on UA-cam. Not only for the content, although I love both programming and biology, but the way it's presented is genius. You have these structured parts, the simulation, but in between the simulations you have smaller parts explaining everything, really smooth. Thanks for making this video.
This was amazing. At no point in the video was I confused, bored, or even slightly distracted. I didn't even realize an hour had passed till you mentioned the the programs you used to make the simulator. Amazing talent and great idea, I loved every second of this. I can't wait to see the next video!
Sin leads to hell, keep focused, the devil is on earth to destroy your soul. But God wants to give you everlasting Joy. But our sin is keeping this from happening. You must stop sinning and turn to Jesus Christ he is your only hope.
He can save you from eternal suffering under the Earth, where hell is hot..
Not everyone who calls me their Lord will get into the kingdom of heaven. Only the ones who obey my Father in heaven will get in.
Matthew 7:21.....
@@SINLEADSTOHELL Bro that's way off topic
anyhow i loved when he just went to onto gen 50 and said "those are the great great- oh never mind"
@@SINLEADSTOHELL uh what?
what does this have to do with the topic of the vid?
@@Mark-Wilson your average extreme zealot I guess
1:45 The conditions for evolution
8:00 Simulation #1 - How it works
20:28 Brain anatomy
27:18 Simulation #2 - Mutation and adaptation
34:20 Brain sizes
35:50 Simulation #3 - Brain sizes
41:15 Genome encoding
42:44 Simulation #4 - The "KILL" neuron
50:00 Software used
52:15 Simulation #5 - Radioactive challenge
Thanks
Thank you! This should be top comment
Just when I think there is no hope left for the UA-cam algorithm, it throws this absolute gem my way. Thank you so much for such a good video! I have done many programing thought experiments in my head trying to do something just like this but there were a few things I didn't quite understand or know how to achieve and this cleared them up for me. I'm very excited to get the code and try some things out for myself!
If you want more, check out carykh evolution on UA-cam
Facts
My guy wrote a whole paragraph
Zimmerman St. Charles is dead
@@KitchenSinkGaming73 Holy shit a few words are an entire paragraph to you? This says a lot about you...
I ran 50 generations and changed a couple of the config parameters. Excited to play around with it more when I get home. Edit: Ignore my previous edit. I did compile it in Windows, and will help anyone that wants to, but I found a repo that built this with SFML (usually used to make video games) and it's much better suited to a linux development environment. I really want to toy around with this code. There is a lot to learn from this guy.
how do use the file?
@@Zigiely you have to compile it or run it in docker. Docker is probably easier
@@Zigiely if you still need help, just ask. Learning about how a compiler works is frustrating for sure
I feel honored whenever he says "Some of my fellow programmers...."
Are you a progammer
@@cookoreo6890 no, he es a football player
@@MartinPirizDrums thought he was an astronaut in the ocean 😒
@@MartinPirizDrums Really? I got more of an artistic vibe from him
a engineering student
Dot: *Disappears*
David: I can't bear to watch this violence
304 thumbs up, and here I am to leave the very first comment. People are getting silent these days, tired of violence.
Someone, think about the children!
What GOD must feel as well.
He’s the God we need lol
Timestamps
1:45 The conditions for evolution
8:00 Simulation #1 - How it works
20:30 Brain anatomy
27:15 Simulation #2 - Mutation and adaption
34:20 Brain sizes
35:50 Simulation #3 - Brain sizes
41:15 Genome encoding
42:45 The "KILL" neuron
50:00 Software used
52:15 Simulation #4 Radioactive challenge
Thank you
There wasn’t a minute i wanted to skip.
The conditions for evolution. 1 = create an intelligently designed software program. haha. the irony.
I was just stamping for myself originally. thanks for the likes though 😀
I never ever saw a hour long video on UA-cam. But I was glued to my screen. Excellently edited, fun, inspirational and entertaining. Please make more videos. This is what I’m here for. 🙏
13:10 "The vast majority of them, the moment they're born, they have the inborn instinct to head east and just keep going until they can't go any further." - As a German, I know that feeling.
But isn't Argentina south west of Germany?
@@ianmeade7441 goated reply
@@ianmeade7441 it's north-east if you go far enough
Fake you must be Russian
Auf jeden Fall
As one of the "programmer's friends", I very much appreciate the details providing just enough info to give me the urge to just do it myself as well... very nice video. Thanks to Posthumanist for "reviving" it and hence bringing it to my attention.
Interestingly, I think this also explains why some creatures have smaller brains than others… in your four corners experiment you would get to the point where having a massive brain has no advantage, if all they need to do is go is get to the corners reasonably competently. Our brains consume a large amount of energy, so in some situations having a larger brain to solve certain problems starts to have a disadvantage. It would be interesting to see a simulation that allows neurons to be added, but at a cost to find more ‘food’ then ramp up the challenges.
yeah the next step of those might be a cool scenario where mutation can change the number of connections
and having "food" factor, maybe making creatures that found more food, then they consume survive and reproduce.
then having something like 1 food is worth like 10 connections.
so a creature that have like 300 connections need 30 food to survive.
A big brain is definitely evolved and maintained in a population only when life is a challenge that requires a brain. When survival is simple and intelligence is not required, a brain is an expensive luxury. There are primitive chordates called ascidians that have a brain as free-swimming larvae and lose most of it when they become stationary adults.
One way to sort of simulate the effect of a larger brain could be to add a delay of one simulation step to each internal neuron. It would enable more complex behaviors with loops but cause more complex networks to potentially move slower. It would require modifying the simulation code quite a bit though as the state of each network would have to be carried over to the next simulation step.
Add a consume action, so they can potentially cannibalize each other
@@reyariass I think that would result in the population splitting into a "herbivore" majority and a "carnivore" minority.
David come back please! We need more videos like this one!
This is genuinely one of my favorite videos on UA-cam. If I could point to one piece of media that genuinely changed my life forever, it would be this video, because it launched my interest in programming neural networks, which eventually led to me going from a freelance artist background to a data engineer.
That's amazing. I'm happy for you, friend
What's pretty interesting to me is that this exact method of programming neural net AIs is currently one of the most successful and promising machine learning models for many applications. It's a pretty perfect (and cool) example of the NEAT algorithm and a great illustration of how it works as a general method to make neural net AIs that are effective at solving all kinds of problems..
😂
That’s amazing, what’s even more interesting is he hasn’t uploaded in 3 years. Bro probably has no clue he just changed the trajectory of your entire life
You should be making videogames bro! At least one indie game from you would probably rock
David: we won't be simulating weather.
Also David: Throws a comet at half the planet.
Jim: Billy the weather channel said we are gonna be hit by a meteor within a few day
Billy: God Dammit not again
@@BoomBoomMushroom Just another day in Florida
This is by far the most elegant description of genetic algorithms I have ever seen. Thank you for sharing.
Hopefully he programming the software made him able to explain how it works
Couldn't agree more!
Just an update about Dave, for anyone wondering why he's not uploading: after making that video, he ascended in to Godhood and is now running his own universe.
😂😂😂😂
understandable
The guy became the god of creation
He just made east side survivalist(beware west),
Btw why he just halved their world? He could have drawn graph lines from top to bottom and from left to right up until it still reaches 50% of the world land and then the mutations wont have a simple survival instinct to run east, they will be forced to have a more complex survival instinct as the whole world will be looking like square boxes of inhabitable lands .
@@WhitePerson-he didnt think about it?
I've seen hundreds of simulation videos like this in UA-cam. this is by far the greatest one I've ever seen. You are a talented scientist, programmer and educator.
This was amazing to watch! As a biologist its sometimes hard to encapsulate the enormity of the genome but with small neural networks like this it can show evolution so succinctly, thank you!
I see you want to create cat girls
@@lollidomoni9523 Don't we all?
@@fatitankeris6327 Lmao. Underrated comment.
01:45 The Conditions for Evolution
08:00 Simulation #1 - How it Works
20:30 Brain Anatomy
27:15 Simulation #2 - Mutation and Adaptation
34:20 Brain Sizes
35:50 Simulation #3 - Brain Sizes
41:15 Genome Encoding
50:00 Software Used
52:15 Simulation #4 - Radioactive Challenge
adoption?
@@TheElectricalls Thank you for pointing that out.
@@brandonmidkiff8200 o>
You, Sir, are a gentleman and a scholar
@@ChaineYTXF You are certainly most welcome.
Well it took 4 years for mr. Algorithm to understand that this content is valuable. Better late than never
Generally happens to small channels where the creator died.
YT is dirty.
Don't expect any new videos.
That was without doubt, one of the most fascinating videos I have seen on UA-cam for some time. Thank you for creating and posting it, David.
800 generations at max and then drops nearly 1000x in 100 generations? and you are not doubting at all? :D im sorry to pull you back into reality but Jesus Christ is on the throne and the world is about to get judged.
@@lt3742 uh?
@@matteosposato9448 seek for Jesus Christ of Nazareth while He may be found
@@lt3742 Errrm, what the heck has this to do with my reply. Nothing is what.
@@lt3742 wtf are you on about? If God is real, he's a psychopath. I'll step to the left if his bastard son shows up, no way in hell would I spend an eternity with an emotional maniac like that.
Wow, I cannot believe I haven’t come across this video sooner. My interest in artificial intelligence began with the idea of natural selection simulation. You show your processes in such a way where there is inspiration to all viewers. I’ve taken away the logic and mathematic concepts in order to perform my own simulations in a more efficient and well thought out manner. I hope to see more AI from you in the future.
Same
Shut
According to Vid-IQ UA-cam wasn't really promoting this year-old video until two months ago and then suddenly - Boom! Tons of views.
You should explore complex systems
This is really good. You need a Patreon!
This man needs to rent a super computer to really do some amazing things. Lets get him there.
let him cook
agreed need a super computer for sure
Watch carefully what happens around 40 minute 🙂 increased capabilities doesn’t really affect organisms performance already, not nearly as much as going from 2 to 8 neurons. You can have whole computation power in the world, but it wont really unfold any discoveries. Regular computers are already very capable of doing very complex simulations, even what’s shown is amazing. Having more power is cool, but I guess what I want to say is that you can do a lot even with mediocre hardware, don’t think you need a supercomputer to carry those experiments.
UA-cam algorithm's selection has finally brought this video to my recommendations!
As a Biologist myself and with only a rudimentary knowledge of coding this is very interesting and a joy to watch!
Cheers :)
38:30 How it works is when N0 activates, N1 suppresses N0, and suppresses itself as well. This little solution here has this result: N0 triggers, triggering N1, which suppresses N0 and itself, effectively resetting the neural network.
Don't be discouraged by your level of understanding of coding. I've been a developer for 10 years and 95% of my work is very rudamentary. It will also get better as you build new things. Take python and your best ideas and run with it! I took lots of biology classes in college and there's plenty of work to do in your field!
I hope you didn't waste your time on the learn to code meme since chatgpt can do all the stuff for you now lmao
@@Noqtisyou are why idiocracy is a documentary not a comedy smh
This man is so wholesome in how he explains all of this stuff, it's like a 2nd Bob Ross. I wholeheartedly loved it.
I love that I went down this rabbithole of weird minecraft glitches to Trackmania Tool Assisted Speedruns and then this gem of a video.
You know, I hate to admit it, but sometimes the Algorhythm does not fail.
I love how he calles his community "friends"
I kid you not that is the same exact route I took to get here, Minecraft, TrackMania and then this beauty
im actually confused, i watched someone build a computer in minecraft, watched a video about cheating in trackmania speedruns and then ended up here
1pqqq
@@larafields5169 I agree! Couldn't have said it better myself.
This is one of the most well-explained videos I've seen in UA-cam lately. You have a talent that should be shared with the world!
ie, He needs to reproduce wih some scientist or mathematician
Bro He shared his creation in UA-cam so it will reach most of viewers around the world.
These kinda videos make us "the programmer friend" curious, inspired and HAPPY!! Thank you sir.
I wish I discovered this video while studying neural networks and genetic algorithms in school! This was very informative and very well done. I'd love to see more of your work in the future!
Indeed so do i, too bad we were too scared to fail and copy pasted everything instead of being inspired to learn more!
This is one of the best videos I've ever seen on youtube. And that's a lot. CONGRATULATIONS, amazingly well explained, programmed, everything. You can be sure that I'll share it as much as possible. Keep going!
Yes! Amen to the video, and amen to you for noticing and appreciating the greatness of the video! Kudos all around!! 😁👍
These mfing bots in the comment section having coversation 😭😭
I wish for an Evolution game like this where even non-programers can make their own input & output actions and many other things like resource, environment, speed, size... I seriously can't wait to see what I and many other people could come up with such a game. Love this video.
Evolution and AI seems to be barred from games as an intelligence inhibitor.. which really sucks.. a game such as worldbox would be a great starting point but it, an many like it are run by.. evil.
I'm not sure on the best strategy to fight this yet. Suggestions welcome.
@@richardward6747 The ol' "if you want something done right, you gotta do it yourself" seems like the simplest way. Find some people that aren't greedy and evil and who share the same passion and do it. I'd help but... Today i was wrecking my brains trying to understand how to create classes and to count how many times a number occured in a list... In python... So yeah, good luck. I'd love a video game with hyper realistic characters and consequences. Also a great crafting, harvesting/gathering and skills system. But then again, if you were to acomplish what i imagine... You'd create a whole new, real world with real people, although seen as npcs by many of us (and i think we all know how gamers treat npcs)
@@migolan6606 thanks man.. I would love to program a good game, while I ain't a great programmer I probably could.. but more important things require my attention at this time.. maybe eventually.
@@richardward6747 do give me notice when you start, if you still remember. If i survive the wars, the ¥|RU$ and whatever else this decade throws at us and we still live in a 'peaceful' world, i'd love to make such a game too
@@richardward6747 also good luck with whatever requires your attention
4:00 DNA works not in single letters, but in "Codons" - that is the actual "letter" read out is comprised of a triplet of the 4 molecules. That opens up quite a lot more values per position in the datachain. That brings us to 16 values per molecule, times 4, makes 64 Codons/Values per actual read-out position. That is way denser information than hexadecimal (16 values per position). Two of these Codons (I'd have to look up the exact ones) act as "START" and "STOP" indicators for the Rhibosomes (molecule factories). Add to that that the Rhibosomes read forwards and backwards on both strands of DNA at the same time. This also allows for a lot of white noise in the DNA-"Code" to happen, which gives room for mutation (positive and negative) as well as padding-insulation against damage. Many "production-instructions" on DNA are also present multiple times for the same reason of contingency against error and damage, as well as for production quantity.
There is just one start codon: AUG, and there are actually three end/stop codons: UGA, UAA, and UAG. Also, it's spelled "ribosome"(unless you are talking about the band)
Good info though 👍
Does that not also exponentially increase the complexity and need for precise compatibility to survive and be a compatible mate with another genome (because the video states one requirement is to find a mate to reproduce)?
Ooh
Sure
I was hoping in the radioactive example that the mutation rate went up the more radiation a critter was exposed to. :)
You totally inspired me. I am going to be coding instead of sleeping now. :)
He's using radioactive Cytosine already. @3:42 The Nitrogen has 5 bonds rather than the usual 3 NH3= :-)
Holy moly. After grinding coding challenges for three years I got a bit tired of the whole ordeal, but this one sure relights the fire.
Can't wait to create a simplified version of this.
When you do can you link it? im genuinely wanting to run this for myself. Ill pay
@@tylermanning4321I'm thinking about it too 😂
how far did u get
Since this isn't happening I'll just make it. Give me like a week tops.
@@hellvet3 u got thisss
This was great to watch. I'm really glad the algorithm worked in your favour. I wish there was an easily approachable way to try out your software, as a microbiologist I am not that versed in getting stuff to run on linux. In any case, this is a great video, your narration and explanations are awesome :) Thank you for all the effort you put into this
Don't you mean this was "great great great great great great great great great great great great great great great great great great great great great great great great great great" to watch?
worked in his "favor"? Shouldn't it be subjective, unbiased??
I’m sure there will be iPhone apps for this in a few years
@@Ty-mf3vz the term worked in someone's favour doesn't imply that it isn't subjective or unbiased. It means exactly what you want it to mean. It basically means that despite the inherent randomness he got lucky that it worked out the way it did.
@@Ty-mf3vz Unbiased? What makes you think the UA-cam algorithm is unbiased? In fact, they make a point of telling us of at least some of the biases they employ.
I think this video just finally helped me understand what i was struggling most to understand about what it takes to develop my own ai models from scratch. Thank you very much! Now to put these thoughts into practice
David , you have had two and one half million views in less than 2 years. That should tell you that you are very good
at mtaking a complicated subject and explaining it in a way that people will listen to. It is a shame that you called it
quits.I for one wish you had not made that decision. Best to you and thank you for all the effort and work you put in.
I was so disappointed when I went to his channel and couldn't binge years worth of content...damn it
he quit? whered you find that info? did he say why he quit?
Men in black took him in as a researcher. Now doing top secret work.
"two and one half" never understood this americanism. wonder if you ever say "two and two halves" or "two and three halves"? plus it just sounds wrong
@@artanaillazentujin3449 he only has 2 videos on his channel and hasnt posted in 2 years, pretty safe to say he either quit or died or something
38:32 oh my God. I think that's a primitive form of memory.
I was thinking about the loop in the first run how the internal neuron had a negative loop to itself. I intuited that it was basically a flipper that flipped the neuron back and forth so that it wojld move randomly every other step.
This seems to basically remember what it did for the last 2-3 steps. It's got the same negative self flipper but it also inverts itself into the neuron linked to thst sensor.
N1 basically inverts itself every step while also adding influence back to N0. Which means certin locations of x over time will do different things, depending on the cycle.
Basically it alternates between moving west and moving random, but not every cycle. Usually it will do this 50/50 but the higher Lx gets, the stronger the strength of each subsequent flip will be. But N1 also inhibits N0 so it only really does this every other step. OH! Because it only wants to strengthen it when it's already strong, otherwise it would become neutral on the off cycle. Brilliant!
I see that Rnd (random?) is linked to move random. So basically the stronger N1's flips are it will eventually overpower Rnd and move west more often instead of random. Or if Lx is shrinking, it will move random more often than west.
Good catch. It works similar to a computer memory latch
40:18 lol you should check this one out, it looks extremely impressive however if we compare that to our brains its not even like 0.001%
I'm trying to make my own (much simpler) project following the same ideas, but I'm stuck at the internal neurons. I don't understand what I want from them, and it seems like you might. I tried poking around in the linked code, but haven't found it yet. If you could point me in the right direction, or share a resource that explains the concept, I'd appreciate it :)
@@Dnallohes They're just there to be generic intermediate nodes that are able to connect to each other and pass signals around. Just having them available for this increases the number of different possible ways that the input signals can flow through the brain and be linked to each other, and therefore the possible complexity of the AIs behavior. If you're familiar with electrical engineering, you'll see quickly that one of the things neurons can do, just by being connected to each other in certain ways with signals going through them, is form various kinds of circuits, including things like logic gates (AND/OR/XOR/NOT, etc), timers, and memory circuits. The signals from the inputs going through these circuits allow the neural net brains to hold and act on more possible informational states, process the inputs in more complex ways, and even learn to correlate them (since the internal neurons can connect to each other as well).
The more neurons & connections available in a neural network, the more complex its circuitry, and therefore its processing, can ultimately get. It can develop memory, learn to recognize patterns, adapt to changing conditions, get and use feedback on the sensory results of its previous actions, etc, ALL just depending on precisely how the neural circuits in its brain are connected. The neat thing is you don't have to understand what it's doing or program the neurons to do any of this, just let them connect however they want and mutate randomly. Over generations of selection, the successful surviving neural networks will become smarter and smarter (up to their capacity), evolving the right wiring to succeed (and/or call down SkyNet... but that's what you get for activating the 'kill' neuron...)
@@blissful4992 it’s insane, the idea that their brains are that complicated for them to do something so simple makes me appreciate us as people
"I'm an innocent and inoffensive dot..."
David: "Shut up, let's play a game..."
And I guess for people who believe in God, God is just doing a David move with all of us
@@AratjaUjotOurstories and the kill gene was always on!
Evolution lol. Show me creating life from non life. You dont get that as a gimme. Evolution trying to explain what happens AFTER life is already there is fun to think about....but its not science without showing life can just happen out of nowhere.
@@sunnydlite-t8b he lied. There are lots of explanations for making life from non-life.
@@KWifler Explanations are not science. You have to do it physically or it isnt in reality, just in your brain. Which is the entire point of my comment lol.
Hoooly! I've been thinking about trying genetic programming for a while now, and this is probably my favorite video on YT now! Thank you so much for this! This is incredible.
This was incredibly interesting as a biology student. I hope your channel grows and I get to see a lot more content like this. Best of luck to you.
Thanks, and good wishes for your journey in the world of biology.
Marine biology I assume?
Is this a jojo reference!!!
@@howlu9086 maybe...
Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont make a joke about marine biology Dont
40:45 One interesting thing is that in reality, neurons take energy to use so they increase the needs of the organism. And as you shown in the exemple there is diminishing return when it comes to the result of increasing brain size. The smallest brain size was awful, but just a few neurons were enough to make it almost as good as the biggest brain.
This probably is one part of the explanation why most organisms in nature are extremely simple. Only the most successful creatures with bigger brain benefit from it, while smaller brains require less energy and have less pressure to succeed.
It's obviously more complex than that but there is definitely a tension between the cost and benefit of brain size. While in your simulation if brain size could evolve alongside brain structure (a gene causing extra genes or neurons to appear in the next generation) there would probably be a complexity creep that would never stop because there is no limit to growth, at least until it becomes unsustainable and crashes your computer and oh my god this became a global warming analogy! Except that in this case they would rip the fabric of spacetime without even having any way to realize they have an effect on it. Creepy...
This should be a movie
@@jinminetics599 I mean, it's more or less the plot of gurren lagann.
@@UA-camTookMyNickname.WhyNot You do sound drunk indeed xD hopefully you can clarify when you sober up cause I have no clue what you're trying to say
@@Laezar1 😂
An interesting related idea is the observation that brain size changes tend to lag behind body size changes, that is to say, if your body is growing in size over evolutionary history, your brain size doesn't increase at a proportional rate.
This had led to the hypothesis that reduction or gain in body size primarily leads to disparities between brain and body sizes. I seem to recall reading about size reductions in our primate evolutionary history for instance.
Non-migratory birds tend to have more encephalization when compared to migratory birds - likely because nervous tissue is too energetically costly when competing with similarly expensive migration patterns where birds might have to fly thousands of kilometers. Our large brained non migratory birds dont have to worry with competing costs, they just need to support their larger brains. It's hypothesized that larger brained birds are more behaviorally flexible and can capitalize on niches that aren't available to smaller brained individuals.
This would stratify where certain individuals of a species can survive (based on individual brain differences), acting as a soft reproductive isolation. Which may have contributed to the diversity of avians fauna today
So this guy came out here and dropped this absolutely banger of a video, got 3M views and 40K subscribers and said “I’m out!”
I was hoping I could binge a ton of his videos. This was so captivating, entertaining and educational!
fr like why we all getting this suggested now 😭
Fr. Amazing work. The world needs more.
@@VanyaTheSlavic are you okay
@@VanyaTheSlavic He could be dead for all we know.
@@VanyaTheSlavic that.. was a joke.
you need help?
As a Computer Science student and also a person who is interested in evolution and biology, this video is absolutely fantastic. I usually get bored and just stop watching if the video just gets boring but not this one.
I started watching the vidéo, and it was already the end, that was really good, I didn't see the hour pass at all!
Thank you for this great guide and for even sharing how you coded it, it really is great materiel to learn from!
I was so upset seeing that red bar get longer and longer.
Wow, have just watched a minute of this video and know it is going to be exciting
ALL THEORY...Look up outa the tv mandates...g..trainz hew man isum..
@@paulmillard9535 “g trainz hew man isum”
- please articulate if you want anyone to appreciate what you may have to say
its all bull shit data in dna is not undersude i believe in nateral satlicion but not evelution mutions are damiage
@@stanleydavidson6543 though the credibility of your belief lacks legibility or factual reference, I’d dedicate time to follow up any sources you have if you can provide links as a counterpoint to this extremely detailed and widely respected area of study.
Much more worthwhile to share links and peer reviewed scientific data than argue our relative understanding of this phenomenon
For real, is any of the comments on this comment real?
"and it also has to be sufficiently socially adept to find a mate"
ok man, there's no need to attack me like that.
memes aside, this is a super fascinating video, i have skipped a total of 0 seconds.
Being socially adept to find a mate, if included in the simm, it would yelled much different results in the Kill switch simm.
Sad fuck
What a beautiful illustration of natural selection, the importance of mutation, and the value of brain size. Thanks, David!
David, I watched your video this morning after learning about The game of life. You stepped it up with the little brains and neural connections, genes and mutations. You are excellent in didactics. Please go on. Your channel is awesome.
I would love to see a continuous simulation instead of a generational random redistribution. Instead, I think, adding a birth rate along with a slowly changing environmental challenge would be more analogous to what we see here on earth. This would replace simulation cycles with time, more like a real world circumstance.
Step at a time.
Such things always start oversimplified then grow in later iterations :)
Maybe try to search the bibites
Easy. Learn to code and do it!
@@randallwalkerdiaz1002 good idea!
@@Makebuildmodify you’ll basically be up there with Tesla if you do since you’re a great builder😏🤙🏽
You have the genuine skills of a documentary maker - I could happy sit through another hour of similar stuff. Also I'm super appreciative of the explanations, it's literally the first time my knowledge of neural networks has surpassed 0% (though I still don't get the "internal neuron" purpose).
Me too! I've been combing through the code trying to figure out what the internal neurons do, and I can't find them. What I've been able to find in the code comments is what's explained in the video. I'd love to make something similar on a simpler scale, but I just don't know what's going on there.
check out some videos about neural networks (keyword: perceptrons) which mention the XOR problem… sounds complicated when you don’t know and/or/not/xor logic, but it turns out to be really simple when explained well. The internal neurons allow XOR which is not possible with only two layers of neurons.
@@j.j.maverick9252 Ahh thanks a bunch, I thought it'd be way more complex but your explanation makes a lot of sense.
@@Dnallohes For that you need to learn bit about neural networks.... I would suggest watch video 'Perceptron' by Carnegie Mellon University Deep Learning
@@j.j.maverick9252 Yes. Perceptrons have only one layer, so they can't determine parity differences. The more inside layers of interconnected neurons, the more intelligence. Humans have many layers in their visual processing, and that's before the visual information even makes it to the actual brain.
As a programmer myself, with a strong interest in biology, I want to thank you for producing the best video I have ever seen on UA-cam. Fantastic work. I learned a lot and have tons new of ideas and inspiration
This is a very nice and interesting insight into evolution. I don't find it surprising though, that the kill neuron soon predominates the respective simulation. David accidentally incentivized killing contemporaries. The reason is, that his populations have the same overall size from generation to generation, but only survivors do reproduce. This means, that if fewer entities survived, the survivors will create more offspring (per survivor). So by killing others, his entities have more offspring, while reducing the offspring of the competition. It was unavoidable that his little creatures became murderous :) - on the plus side: Different environmental conditions (like in the real world) would incentivize different behavior, and thus not have this drastic an effect.
I was sort of wondering about the reproduction factor too! Like what if the population is allowed to grow/shrink based on the number of survivors at the end of each generation, rather than entirely repopulating the beginning of each generation? It seems to me that this would have a significant impact on the applicability of the simulations' outcomes to nature.
But then, I also see the advantage beginning each simulation with a standard population size. It just makes things so much cleaner and easier to compare, not to mention keeping CPU usage manageable. And I guess you can think of them as samples of a larger, unsimulated population? I feel like there's something wrong with that view but I can't put my finger on it.
Super interesting video though! Makes we want to try all sorts of different configurations
The real world almost certainly favours other traits than killing, until congestion happens.
Unfortunately, direct cooperation is many tiers higher up in orders of required logic, and can therefore be preceded by the occurrance of congestion.
I see this more as a clear simulation of intelligent design. unless this program wrote itself.
Poor guy is doing a simulation that proves the value of death, yet it is lost on him.
What he should do is give the creatures compassion and hatred and then hardwire them so that hatred and, ultimately kill signals, are generated proportional to the product of the creatures compassion, the genetic similarity of any death in front of it and are directed at the assailant.
Then he might learn something. We might all learn something of highly contemporary value. Because, nothing spills more blood than a bleeding heart.
@@benearhart1224 we get it, you think you're smart, no need to copy and paste your own comments
I think it's counterintuitive that the kill neuron did not actually "kill" the other cells. That is, each next generation starts with the same number of cells.
Also, killing usually has some kind of benefit associated with it. Most killing happens for sustenance. If a kill was rewarded with a higher reproduction rate, things would get more interesting.
Those that are killed are replaced with the children of those who survived.
If few survive they will have many children to get back up to full population size.
the pheromones and genetic similarity sensory inputs could be used to preserve total population while minimizing deaths
with more complex reproduction conditions that better utilize the extant pieces, you could easily get more interesting results. ironically, the built in evolution of the sensors was the least effective of all, as it was designed for an environment that the cells did not exist in. arbitrarily setting a zone to reach, then spreading the offspring at birth is foreign to say the least. the radiation example is better designed, as it allows the age or oscillation sensor to better direct the cells.
tying any conclusion from this simulation to the real world, while the selection pressures are so vastly different between the two felt pretty silly as a cap to the video, though i am grateful that there is room for me to mess around with the code and answer my own questions
I haven't seen a video more thoroughly explain this concept, this is honestly a masterpiece.
That little dot at 13:58 going the opposite way of every other dot is super relatable.
I can’t believe something THAT good would have 7k views. I loved It, can’t wait to see more of this.
Biotech student btw
its up to 70k views which is still nowhere near what it deserves
wait, this had just 7k views 5 days ago?!
I would love to see sims with some kind of consumption requirement introduced. For example maybe, every 2 cycles inside a generation each individual must consume at least 1 bit. At the beginning of each cycle 95 bits to 100 individuals are introduced to the environment. Maybe a random mutation could allow for space to stockpile or another to only need 1 but in 3 cycles.
That would be really cool but every additional behavior will make it take exponentialay harder to simulate. We need to get him more computing power so he can keep it going.
@@wastedtalent1625 yes I can understand that. Just adding cycles inside the generations almost makes a single generation relative to an entire current sim. Still, I know he could do it. Wish I could somehow contribute even if it was just time proofreading code but I don't know how and I'm not working... There are now other sim games somewhat similar but I really like the user friendly results of these sims. Also, I feel the others are often biased towards increasing sales to non-scientific interest.
@@wastedtalent1625 wonder if this could be done on a "folding at home" type thing. I forget the name, but there is a tool that can be used to convert a regular program into a "folding at home" type thing
@@_ayohee is that a joke or are you being serious?
@@wastedtalent1625 Well, I oversimplify, but it should be possible to divide the work, no?
"Those who don't [reproduce] were just unlucky with mutations, and they don't have the brain wiring to know how to make it to the spawning areas"
And i took that personally
I like this one
well if it makes you feel better, mutations are just another word for cancer and vice versa.
@@mikkirefur No??😂 Mutations don't necessarily result in cancer.
@@timr3621 no that is right. Usually the hardware & software and self repairing nature of programmed dna can fix itself. What a design !
Or they choose career over family and hit he "wall".
This video changed my perception of life and the universe. Really great stuff, I re-watch it occasionally and this time I took notes. You've inspired me to finally try to create my own simulation. Thank you for a truly great video!
I'm in biotechnology and this content is exactly what I'm looking for. Make some more of these. You are an amazing creator/teacher
This one went straight to my favorites. I'll watch it and re watch it so many times I'll know the words from the top of my head.
You, sir, are an awesome square on our grid.
As a game designer whos very enthusiastic about procedural processes and evolution this was very informative and entertaining. I really liked the structure you gave the video and the way of narration. Although I am very inexperienced with neural networks I feel like I could reproduce this after this video and who knows maybe ill give it a try. Good Job!
Id love to see your go at it!- :]
@Lukas Eckart: Besides neural networks check out genetic algorithms. This is a combination of both concepts.
Dude uploaded one god tier video and decided that was enough for his UA-cam career
The reason the killing creatures thrived is because there's no repercussion and almost no cost for killing, it only takes up one neural connection slot. There's also the fact that the more they kill, the more they get to reproduce. It would be interesting to see what happens if there was more mechanisms related to killing, for example:
- make creatures unable to move after killing for some turns
- give creatures the ability to choose to avoid reproducing with killers
- give creatures the ability to defend themselves
Starvation, or it's equivalent.
My man, you are no less equal to Primer and other talented simulation programmers out there. How comes you have been so underrated. I hope the algorithm aids you as it has guided me to this masterpiece.
he only has 1 other video.
That was one of the most entertaining things I've seen for quite some time on YT. And the quality of the video is at a really good level
Nicely done!!
This video was really, really well thought out, the flow is fantastic. I have to go to sleep and I'm so annoyed, I don't wanna stop the video hahah
Yeah its absolutely great right?
I love the way his voice becomes so menacing at 30:20. "but NOW, lets upset their peaceful little world". This is some supervillain stuff
Oh my god this is the coolest shit ever. This would make an incredible video game. Instant subscription
Not exactly a video game and I'm not sure how much/where it is still available but there used to be a pretty neat 3D evo sim called 3DVCE that evolved 3D creatures composed of evolved physical bodies that moved with simple, evolved neural networks. The selection options were pretty limited but you could get some really really interesting results if you ran the sim long enough.
@Ayy Lmao cut them some slack
@Ayy Lmao This video but interactive. Who wouldn't want that?
The closest thing that I know of is bibits (check it out on UA-cam)
i would love some kind of proper evolution game, ive been trying each attempt over the years and nothing has come close yet
Just got this recommended to me, i find simulated life extremely fascinating.
Same!
Easily one of the best videos I've ever seen anywhere.
This is great! I might have to dig into this because there's several things I want to play with (individually and in conjunction):
* setting mutation rate as an output gene (what's optimal?)
* setting neural complexity (# of both connections and internals) as output genes
* making it so that reproduction happens specifically only between individuals ending next to each other (curious about how that would effect murder rates in particular, and would they evolve to use genetic similarity input to avoid killing potential mates)
* food chain simulation; neural complexity as output gene as above, but higher complexity can starve and must "eat" (kill) to survive (curious how many will prioritize simpler non-eaters versus complex eaters)
* like the radioactive challenge, but mutation rate depends on amount of exposure (will they evolve behaviour specifically to find best mutation rate, and will it match first item above)
I bet the one about reproduction happening locally would result in a lot of polyamory :-)
@@katiebarber407as tended to occur in human history!
Wow. Perfect presentation. The best I have seen. You sir deserve an award of something. Super interesting, well explained, it keeps one's attention all the way to the end. A master piece, well done. This was an inspiration.
Wow. The video came up in my recommended last night and I added it to my watch later list. I'm glad I didn't try to watch this last night because I would have gone to bed an hour later than I wanted. The presentation in this video was amazing. You're explanations are great. And thank you for sharing the code.
I'm so glad this video is finally getting views. This was an amazing watch. Honestly kinda sad more complex survival criteria weren't used. I would've loved to see solutions that make use of pheromone as a result.
I'm wondering if that radiation simulation makes use of pheromone. So they know how to clump together or if it is all age and border distance based.
@@NeoNthriller remember there are also sensory inputs for population density. They could've used those too.
@@Syz_gy Ah true, I forgot about that one
I guess it made it through the censorship gates and now it's been approved for our viewership by our master Google. God bless Google!
@Aurora Peace Yes. I was one of the first few to watch it. It sat at less than a thousand views for a really long time. It's only recently that the algorithm picked it up.
I had to go look up what a hyperbolic tangent function is, which I needed to look up what Eulers equation was, which I then needed to look up what eulers number meant, all because I wanted to understand what was happening in the video. This is one of the first videos I’ve watched that made me so interested in something.
Any askers?
@@chrisdawson1776 askers?
@@chrisdawson1776 yeh my thoughts exactly I couldn't see anyone asking for your comment
@@chrisdawson1776 you seem like a hateful person
That's good stuff man.
This has made me rethink what "learning" means. You can see here that learning could be considered a complex form of filtering. It also makes intelligence seem a lot less special, because intelligence is a side effect of survival selection based on progressively larger environmental challenges. The more environmental complexity there is, the more "intelligence" must develop by filtering to navigate it. Fascinating.
Yep, and one machine learning technique used to design AIs involves doing exactly this. Genetic algorithms, which run many generations of concurrently running and competing AIs that likewise can be neural networks, but don't have to be, as long as they have a way to process inputs and map them to actions, a sufficiently complex brain, and a random factor (like mutation), which as we've seen here is all that is needed to allow you to allow the AIs to evolve adaptively in order to get better and better at solving a problem.
Best comment I have read today !
Im not sure I would use the term "learning" - Learning as I understand it implies some conscient action with the goal of expanding ones understanding.
This is adaptability which in itself doesnt require any form of goal or conscient understandings. It is stored as innate abilities that the gene carries with them regardles of the mental capacity and/or intentions the carrier brings to the table. Its the same if its a earthworm or a human. And I dare claim we cant learn an earthworm anything. But they do have the reflexes/autonomous behavior that prevents them from drowning when it rain, or fry if the get exposed to high temperatures. They have adapted..
Yes I was thinking on the similar lines that teaching and learning seem to be added functions of evolution . To transfer knowledge of survival, knowledge on top of your genomes knowledge/information on how to survive ( or the ones that could not be transferee through the genome )
@@martinwinther6013 Traditional neural network implementations 'learn' more in the way you are thinking of... the system is 'trained' with input and programmed to get varying levels of reward from the results of its chosen actions which it uses to reinforce or weaken connections in its brain. This lets it learn what works to solve the problems it needs to... This is also more like how a real brain learns during an organism's lifetime. There are some problems with implementing this though, it gets complicated figuring out how many layers deep you should make the connections, how malleable the connections are, etc...
The genetic algorithm technique here builds the connections in a different way, using selection, randomness, and evolution to build brains that already have wiring that can solve the problem... This is more like how real organism's brains evolve their base wiring, the things like instincts and effectively controlling the body. This has shown a lot of promise in developing AIs and is pretty much the new way AI developers are going.
The term machine learning here is still valid though, since the end goal is getting an AI that is properly trained to solve a problem. This is still a process of giving the AI all kinds of input, observing their output, and rewarding successful results in some way (in this case, with 'survival' of its better qualities to pass on to further 'generations'), still these generations were really just repeated training iterations for the working AI that is the end result... When you look at this effect in nature, you see how it also almost has to result in intelligent (or rather, at least sufficient for survival in their environment) behavior in any species which can evolve. A species 'learns' its basic behaviors, those required for it to survive, as a population (or rather, the surviving portion of it...), in much the same way these things do.
2:00 Evolution doesn't have to explain where the first self-replicating molecule came from. Chemistry does that just fine.
Evolution takes place in the latter part of this inevitable chain of events in certain optimal conditions:
1: Atoms form molecules. Happens quite often.
2: Molecules are sometimes formed that oscillate and move.
3: Molecules that oscillate and move sometimes happen to move in ways that construct other molecules.
4: Sometimes molecules that construct other molecules happen to construct copies of themselves.
6: Molecules that self replicate become super abundant. (evolution)
7: Random events lead to errors in the copies (evolution, mutation)
8: Molecules that happen to mutate to replicate faster or better become a the new fittest replicator.
9: As atoms and molecules to build from become scarce, replicators that steal from other replicators keep replicating and become fittest.
10: Replicators that happen to build little shields for themselves survive better and become fittest.
It's all inevitable, and doesn't really require proof other than "chemistry" as a whole.
Thanks for the explanation! Now the only deep questions remaining are located before the 1st step, one of which being "how are attoms formed?"
But I guess the more we learn about this world, the more there is to question huh?
@@firegator6853 Very true. Why does a universe arise, why does a universe behave as it does, can it behave differently, and so on. There’s no end to that.
How were atoms made?
@@ginsan8198 yeah, I don’t know that one either.
So why doesn't it happen now if it was that simple? The hypothesis that early earth conditions somehow facilitated something thwt cannot happen now is just that, a hypothesis.
Some things to note for future experiments:
1. Add some sort of "food" system and run some tests to see how the creatures evolve with limited food.
2. Add an input neuron that can detect when a nearby creature is killed. Perhaps that would lead to creatures "running away" from killers?
2. Add an input neuron (or neurons) that can detect when a nearby creature is not moving where it should be, and a corresponding output neuron (or neurons) to allow for one creature to sacrifice all other actions it would take on that simulation step, in order to move another creature in whatever direction. Basically a "see a creature in need, help that creature" system. Could also lead to some interesting interactions with the proposed food system and killer detection.
I think you just proved inteligente design?
This simulation was such an excellent explainer for how evolution works to generate advantageous adaptations in a population, so amazing job! I am a little embarrassed that I started to get emotionally invested in some of the dots though lol I know they aren't real but it was so heartbreaking to see a little dot just hopping around on the south border, never to make it to the spawning point
Hi David, this is such a great video, I not only learned a lot about mechanisms of evolution, but also about didactics. You considered your viewers' backgrounds, presented the whole before the details and made it a wonderful experience. Truly amazing!
You sir are truly a man of science. Your knowledge and attention to detail show profound wisdom. For example, while talking about your postulates for evolution you mention self replication. Over the screen you have a floating hydrocarbon! The molecule theorized to have given rise to cells and isolated organic environments. I have worked in a molecular biology lab for a few years now and I salute you.
As a 3D artist building my way through videogames industry, I think you just changed my future line. I know I may say some dreamy stuff right now but, I see a future where video games will be.. real. Just like in Tron film. Npcs will feel like really understanding their environment. I’m so dam curious to just create some basic 3d mesh with an AI that deforms it based on mutation factors. The AI could even change the rigging structure of the creature based on movement and animation necessities. There is so much free and non explored field in this... just imagine a new SPORE game released with those features
This is your sign to go and do it!
Free Guy
It’s like the one used in that game that generates planets by mixing the characters up.
My thoughts exactly.
Rain World