Artificial Life

[00:00:00][music] >> And a river went out of Eden to water the garden.

[00:00:06]And there it divided and became four rivers.

[00:00:10]The natural world is bursting with biological complexity. [music] From birds to bees to beetles to bacteria and back.

[00:00:21]This overwhelming diversity of complexity all emerges from [music] a remarkably simple mechanism. A mechanism so simple that it can be replicated in a computer [music] program.

[00:00:33]With it, we can make typically rigid logical machines behave all soft and squishy.

[00:00:40]Today, I want to talk about these kinds of soft programs. Gloopy, gloppy, globby programs that look surprisingly biological. They emerge from a strange field of study [music] that is known as artificial life.

[00:00:57]>> [music] >> Artificial life or A-life is the study of human-made living or life-like systems. Not life as it is, but life as it could be.

[00:01:25]A-life is a big field that overlaps with many other fields like robotics and genetic editing. But here, we're going to focus on virtual life, simulated life, the kind that emerges entirely within a computer program.

[00:01:39]This video is a sequel to my previous video on emergent complexity where I emerged all over the place.

[00:01:47]Today, I want to talk about the next kind of emergent complexity, the emergence of life itself.

[00:01:54]Welcome to Emergent Garden.

[00:01:57]Wow, is that a link to my Patreon where you can support the channel and get exclusive content like Claude Fable building stuff in Minecraft? Or you could just subscribe and share this video with your friends? That's incredible. I'm going to do that right now.

[00:02:11]In this video, we'll be diving deep into the open ocean of artificial life and glimpse some of the strange creatures that live here.

[00:02:20]To do this, we'll be playing with a couple different kinds of simulations.

[00:02:24]Some will be cellular automata like Stephen Wolfram's rules, so to speak, or John Conway's Game of Life, or Bert Chan's Lenia.

[00:02:34]These are all essentially grids of pixels that are updated using local [music] rules. Pixels are affected by nearby pixels.

[00:02:43]We'll also be playing with entity-based simulations, which means you're simulating a bunch of entities like particles or shapes or creatures [music] that move around and interact independent of any grid.

[00:02:57]Many of these programs are available for you to play with right now. I made a few websites you can explore, and I've linked a lot of other people's work down below as well.

[00:03:05]I won't be going into much detail about how individual programs work, but rather I'll focus on the more general emergent behavior and its implications.

[00:03:14]I especially want to shine a light on what is considered the holy grail of artificial life, >> [music] >> which is open-ended evolution, a forever algorithm that you could run for a day, a month, a year, a million years, and it would endlessly produce surprising, clever, subtle complexity, just like real biological evolution.

[00:03:37]Now, typically on this channel, I aim to piss off mathematicians, [music] but the day has finally come to also piss off biologists. I'll have to play a little fast and loose with biological concepts, and suffice to say I am talking about simulated biology, not real biology. I'm not a biologist, I'm a computer nerd who wants to simulate the Cambrian explosion.

[00:03:59]We'll have to dance around a lot of questions that are open to scientific debate, like the very definition of life itself.

[00:04:06]And this is a YouTube video, so I'm not going to be giving you a rigorous definition of life, but we can at least talk about what life looks like, and see if real biology has anything in common with these weird, squishy programs.

[00:04:21]So, what does life look like? We'll have to answer this question gradually, building up from simpler forms.

[00:04:28]>> [music] >> In my last video, I talked about the many different kinds of complexity that emerge from simple systems, [music] both in computer programs and in the real world. You first need some basic building blocks that come together and interact in interesting ways. They could be bits or pixels, >> [music] >> LEGOs or logs, ants or people, atoms or chemicals. All of these things can be combined to form a variety of weird structures and patterns. Some of these emergent patterns are flat [music] and boring.

[00:05:06]Others are repetitive and predictable.

[00:05:11]Others are chaotic and random.

[00:05:15]And some, [music] the most interesting ones, are complex yet organized, a mix of order and randomness. There is some messiness, some repetition, and some surprising structure.

[00:05:29]And in that structure, you sometimes find little local structures that seem to do their own thing, like a glider in Conway's Game of Life.

[00:05:38]These structures are distinct from their surrounding environment. They have boundaries. [music] They are stable. They maintain themselves. They don't just fall apart, at least for a while.

[00:05:49]But, they're also not static. They're dynamic. They change over time. They move around, interact with their environment, and they even grow or spread.

[00:05:59]These distinct, stable, dynamic structures I will call life-like [music] patterns. They're a kind of proto-life, and they're surprisingly common in lots of simple computer programs. You can often find things that look a lot like ants or worms or mold or bacteria.

[00:06:17]But, to go from merely life-like patterns to more genuinely living patterns, [music] we need another ingredient: adaptation.

[00:06:26]Adaptive complexity is the kind of complexity that is good for something and gets better at it. It adapts to some environment or some goal or task or problem or constraint. [music] It optimizes and explores. It acts and reacts. It learns and remembers and it protects itself and repairs itself.

[00:06:46][music] Over time, adaptations can stack on top of adaptations, resulting in extraordinarily competent, capable complexity.

[00:06:56]This is how you get eyes that can pinpoint prey, wings that can dive-bomb that prey, and talons that can tear it to shreds, all in the same organism.

[00:07:07]This concentration of adaptive complexity does not emerge overnight. It can only be acquired gradually through a process of evolution.

[00:07:17]Evolution itself is an emergent adaptive process. [music] It emerges from a remarkably simple mechanism that I alluded to at the beginning of this video. And that mechanism is the replicator.

[00:07:31]A Darwinian replicator is a life-like structure that does three things.

[00:07:36]First, it replicates. Duh. The structure is copied, and those copies make their own copies.

[00:07:42]Second, there is variation. Some of those copies might be imperfect copies, [music] just a little bit different from their parents or their siblings.

[00:07:51]And third, [music] there is natural selection. Only some of those copies go on to make their own copies, while others die out. If any variant happens to be better at replicating than its rivals, then it will [music] outcompete those rivals.

[00:08:05]In short, a replicator is copied, varied, and selected. As a consequence, the world automatically becomes full of successful replicators. Over millions of generations, adaptations [music] can accumulate and compound. That is evolution in a nutshell.

[00:08:25]I should say that this is a very selfish gene view of evolution, meaning it emphasizes replication and adaptation.

[00:08:32]I'm very Richard Dawkins-brained in this regard, but know that there are complications to this perspective, some of which we'll see in this video.

[00:08:40]Regardless, we're just going to move forward with the replicator as the central mechanism for the evolution of life. And that's partially for a practical reason. The replicator, as I just defined it, is an algorithm. It's a set of logical statements that can be cleanly translated into computer code.

[00:08:58]It's easy to simulate natural selection.

[00:09:01]Darwin among the machines.

[00:09:04]So, let's simulate evolution. We're going to try to design life that designs itself, so to speak.

[00:09:12]So, it's useful here to think about the difference between explicit design versus implicit design.

[00:09:18]In the context of computer programming, explicit design is anything that is explicitly written out in the source code itself.

[00:09:25]Like the rules of the game of life are explicitly designed. They're in the code.

[00:09:31]Implicit design is anything that is not explicit. It's any behavior of the program that is not found in the code itself. A glider is implicit. This particular pattern is not in the code.

[00:09:44]There is no glider function or glider class or glider variable.

[00:09:49]Implicit design merely emerges as a consequence of explicit design.

[00:09:54]Generally, the goal of artificial life is to maximize implicit emergent behavior. We want the program to figure out the details by itself.

[00:10:03]We're going to take a look at three different classes of evolutionary algorithms that are increasingly implicit and emergent. They all live on a spectrum of emergent behavior and there is some gray area between these algorithms that I have to simplify.

[00:10:19]This is my own framing that I find useful for thinking about simulating evolution.

[00:10:23]As we dive deeper into the ocean of artificial life, we will discover more surprising implicit design.

[00:10:31]Let's start with genetic algorithms.

[00:10:38]>> [music] >> Genetic algorithms are optimization algorithms. You have some clearly defined problem and you want to optimize a solution to that problem.

[00:10:50]So, first you have to define a fitness function to evaluate the performance of a given solution or the fitness of a given genome.

[00:10:58]For instance, I want to create a blocky creature that moves as far away from the starting point as possible. So, I define the fitness function as the final distance of the creature from the starting point after a certain amount of time. We start by just generating a bunch of random solutions and evaluate fitness for each one. You pick the best one, copy it and make small random changes to create a population of mutated offspring.

[00:11:22]You then evaluate performance again and pick the best offspring for the next generation.

[00:11:27]Repeat this over and over and they will optimize and adapt. We have replication, variation, selection, and adaptation. I think that makes them a genuine form of artificial life.

[00:11:39]Now, what I just described is really a stupidly simple genetic algorithm called a local search. A real genetic algorithm is more sophisticated and specifically tries to maintain genetic diversity.

[00:11:52]A genetic DEI program, so to speak.

[00:11:56]For instance, rather than only picking the best solution, you pick a diverse population of pretty good solutions. You can perform crossovers between these solutions and mix and match different fit parents to hopefully produce even fitter offspring.

[00:12:10]You can split populations into different species or gene pools that are genetically isolated from each other and then maybe reintroduce them later.

[00:12:20]The theory is that maintaining diversity lets them discover and combine a variety of useful strategies, and it helps keep the algorithm from getting stuck in a rut, a local optimum.

[00:12:33]So, here's a hot take that's going to cause some drama down in the comments.

[00:12:36]I don't really like genetic algorithms.

[00:12:40]They're just kind of boring.

[00:12:42]Now, I am simplifying the concept for the sake of this video. You can use them for much more interesting open-ended stuff beyond straight optimization.

[00:12:50]But just as optimization algorithms, they're usually not very good. They work, but they're slow and inefficient and kind of a pain to deal with. The sooner you can figure out how to use something else, like reinforcement learning, the better.

[00:13:04]They're also not very emergent. The only things that are implicitly designed in genetic algorithms are the final optimized genomes and I suppose the individual mutations along the way.

[00:13:16]Everything else must be explicitly defined by the programmer. The fitness function, the genetic structure, the mutation and crossover algorithms, and any system for speciation or diversity.

[00:13:28]And that last one is the real kicker for me. New species or niches don't just emerge by themselves. They have to be programmed in with some diversity metric that is imposed from the outside.

[00:13:41]To be fair, genetic algorithms absolutely can have lots of weird, interesting emergent behavior.

[00:13:47]But overall, they are necessarily constrained [music] by predefined goals, explicitly handcrafted systems, and relentless optimization.

[00:13:57]I guess part of what annoys me about genetic algorithms is that they're the most well-known kind of evolutionary algorithm, so they limit people's imaginations.

[00:14:06]But there is a different kind of Evo Algo that is more emergent and implicit and surprisingly popular, even right here on YouTube. And that is the evolution simulation.

[00:14:21]This is my own definition of an evolution simulation. It's not a formal program, but honestly, I think it could be.

[00:14:28]We'll use my old evolution simulator, the life engine, and also the Bibites simulator, which has its own YouTube channel that inspired me to make this YouTube channel. The Bibites is an actual game that you can get on Steam right now, and you should.

[00:14:41]An evolution simulation is a large simulated environment that is shared by many organisms and their genes.

[00:14:48]They interact with each other and with any resources or entities in their world. You still have to explicitly define the organisms and their genetic structure, but unlike a genetic algorithm, there are no distinct phases for reproduction, mutation, and selection.

[00:15:05]Instead, each individual organism lives, dies, and reproduces [music] on its own time, independent of any fitness function.

[00:15:14]And this is [music] the biggest difference from a genetic algorithm.

[00:15:17]There is no fitness function, no single way [music] to calculate and rank the fitness of a given organism. Instead, you have a reproduction function.

[00:15:28]A reproduction function decides how an organism in its current state should reproduce [music] and it's triggered by satisfying certain criteria. Has it eaten enough food or lived for long enough or traveled far enough or defeated enough enemies? How many children should it have? How should they be positioned? How should they be mutated? The reproduction function should answer all of these [music] questions. And the goal of each organism is to successfully reproduce as much as possible.

[00:15:56]But to keep the population from endlessly growing, you also need to define a death function that is triggered under its own criteria. How and when should an organism die? Is it starving? Has it lived too long? Has it taken too much damage?

[00:16:12]Combined, the reproduction and death functions allow for the emergence of natural selection and evolution. The simulated world will inevitably become full of organisms that are able to avoid death [music] and successfully reproduce.

[00:16:25]Now, while we have replaced the fitness function, we have not replaced fitness.

[00:16:30]We still have survival of the fittest, but that fitness changes as the environment changes [music] and as the gene pool changes. Fitness is relative and temporary and it's much harder to measure.

[00:16:43]You can approximate it, but that would be descriptive fitness, not prescriptive. In other words, fitness is now implicit, not explicit. It is not defined anywhere in the code. It only emerges as a consequence of reproduction and death.

[00:17:00]And I think this captures something important from real biological evolution.

[00:17:05]In the real world, there is no explicit fitness function that's written down in the laws of physics or in the genetic code or in the Bible. Fitness is an emergent property of replicators living in a complex, dangerous environment.

[00:17:20]Evolution simulators capture this emergence better than genetic algorithms.

[00:17:25]They're also better at encouraging speciation and diversification.

[00:17:29]In a large world with a large population, it's natural for replicators to branch out and evolve in different directions. They can flourish and coexist with radically different branches of life.

[00:17:41]They do this by finding different niches. They can colonize different parts of the world or consume different resources or fill different roles in the ecosystem. You can have predators and producers and scavengers all sharing the same space. And I mean, they're not exactly [music] friendly, but they coexist and they might even depend on one another.

[00:18:01]This allows for interesting emergent relationships between organisms, like symbiosis or camouflage or parasitism.

[00:18:10]Even populations of exact clones, which would be very wasteful and redundant in a genetic algorithm, can do surprising things in a shared environment. In fact, that's how real multicellular life evolved as colonies of cooperating clones.

[00:18:27]But if I'm being honest, I've never seen multicellular life emerge in an evolution simulation.

[00:18:33]Obviously, there are limits here. We are still constrained by lots of explicit design choices, like the genetic structure, the mutation algorithm, the world itself, and of course, the reproduction and death functions.

[00:18:46]These impose all kinds of silent restrictions on what can actually evolve.

[00:18:51]Like, you will never see the evolution of, say, sexual reproduction unless I program it in.

[00:18:57]Earlier, I made it sound like diversification is guaranteed, but it's really not. The entire ecosystem can still be dominated by a single species.

[00:19:06]You have to tune the rules and resources to encourage diversity.

[00:19:10]And even when it does diversify, it can still flatten out into simple, stable, boring states.

[00:19:17]And that's just what evolution does sometimes. Life does not need to be complicated in order to survive. We'll talk more about this later.

[00:19:26]Regardless, evolution simulators can be very emergent, and I think they're a step in the right direction towards open-ended evolution. You can even combine elements from genetic algorithms and evolution simulations to get more interesting results, like using a fitness function in a shared environment.

[00:19:44]But to really push the envelope, we have to start stripping away as much explicit design as possible. We have to go deeper in our search for artificial life.

[00:20:03]I will call this final class of algorithms emergent A-life simulations.

[00:20:08]I know that might be a little confusing since like, aren't all of these A-life simulations? But the keyword here is emergent. In these simulators, there are precisely zero explicit biological systems. There are no genes, no organisms, no species, no mutations, no fitness functions, no reproduction functions, nothing.

[00:20:29]So, where does the life come from?

[00:20:31]It emerges from the lower-level rules of the simulation. Really, a good A-life simulation is a good physics simulation or chemistry simulation.

[00:20:41]You only explicitly define the ingredients for life, the chemicals or atoms or cells or particles, and then you just run it and hope for the best.

[00:20:50]Literally, that's what you got to do.

[00:20:52]You have to just play around and experiment and tune the rules and building blocks until you get some interesting behavior.

[00:20:59]And hopefully, you start to see those life-like patterns that I described earlier in the video, distinct, stable, dynamic structures that emerge without explicit design.

[00:21:10]These are some of my favorite kinds of programs.

[00:21:13]There are too many to mention, but to give you a taste, convolutional cellular automata can look very gloopy and gloppy.

[00:21:21]Multi-neighborhood cellular automata by slacker mans are incredibly beautiful. I plan to build a simulator for these in the future.

[00:21:29]Particle life is an entity based one. I featured it heavily in my emergent complexity video.

[00:21:35]And maybe the apex of the art form is Lenia. Lenia is a fairly complex cellular automaton invented by Bert Chan that is continuous. It uses smooth valued cells and also smooth parameters for time and scale, which is very useful.

[00:21:51]I did build a simulator for Lenia that you can go play with. You can mess with the parameters and you can browse over 8,000 discovered species. By species, I mean a saved configuration of parameters and a starting state. Again, there are no species explicitly defined in the simulation. It's merely a descriptive term used for implicit behavior.

[00:22:13]And it's a good term because they really can look like species, like worms or mold or cells.

[00:22:20]And crucially, some of these species self-replicate.

[00:22:26]Wherever self-replication emerges, it tends to take over. It will spread and it might even evolve. It can transition from merely life-like to genuinely living. A kind of virtual abiogenesis, life from non-life.

[00:22:42]Natural selection [music] can now begin its work to sculpt adaptive complexity.

[00:22:47]And because the entire mechanism of self-replication is contained inside the simulation, it too can evolve. The genetic structure itself can fundamentally change in ways that it couldn't have if it were explicitly defined outside the simulation.

[00:23:01]Theoretically, you really could see the emergence of something like sexual reproduction. I mean, just look at these guys. It looks like they're getting it on.

[00:23:10]Now, inevitably, we will get emergent evolution and competition and speciation and diversification and complexification and endosymbiosis and multicellularity and civilization.

[00:23:23]Right?

[00:23:24]Well, not so [music] fast. Let's take a look at what's happening here.

[00:23:28]These self-replicating blobs have filled the world and that's it.

[00:23:33]Self-replication is actually really common and often trivial.

[00:23:37]>> [music] >> For instance, here's another kind of self-replicating pattern. It starts in the middle and spreads across the world, filling it with copies of various patterns. But, it's trivial and the world is essentially dead. This kind of replication is much more comparable to the growth of a crystal than it is to a living system. Crystals can grow and even seem to self-replicate, but they eventually reach a stable, static, dead state.

[00:24:02]Such patterns are a dime a dozen in the world [music] of simple programs.

[00:24:07]But, remember, a Darwinian replicator needs three ingredients, not just replication, but variation and selection.

[00:24:15]There needs to be some kind of death to control the population and enable natural selection. Usually, you can get this implicitly with some kind of instability where patterns can occasionally collapse or dissolve. This slows down population growth and performs a kind of natural selection.

[00:24:32]But, only if you also have variation.

[00:24:34]There must be some meaningful difference between copies.

[00:24:38]So here, hopefully, you can see that some of these blobs are brighter and some are dimmer or more of a checkerboard pattern.

[00:24:45]This variation is inherited and it can sometimes arise accidentally and naturally during self-replication.

[00:24:52]But, it doesn't make any difference.

[00:24:53][music] This variation has no effect on fitness.

[00:24:57]So, no adaptation occurs.

[00:25:01]Simulating self-replication was actually the original motivation for the invention of cellular automata by John von Neumann. Von Neumann is who Stephen Wolfram wishes he was. Back in the '40s and '50s, he designed a very complex cellular automaton and a machine inside of it that he called a universal constructor.

[00:25:20]It was a structure that could build any other structure inside of the simulation, including a copy of itself.

[00:25:27]By the way, he designed this entire system without a computer that could actually run it. He built it in a cave with a box of scraps. Von Neumann was a beast.

[00:25:37]I wanted to run the universal constructor myself, but turns out that it's extremely complicated. So, let's look at something easier.

[00:25:44]A simpler version of a self-replicating machine in a cellular automaton was designed in the '80s by a fellow named Chris Langton. We've met him before. He invented Langton's ant. He coined the term artificial life, and he also invented Langton's loops.

[00:26:00]These strange loops are small self-replicating patterns that have a kind of implicit genetic code. This inner strip here acts as a sequence of instructions for self-replication.

[00:26:12]It's very cute, but it has all the same problems we've seen before. It fills the world with exact copies of itself and then dies out. It's too perfect. It never makes mistakes, and so it never introduces variation.

[00:26:25]It's also too brittle. If you manually introduce genetic changes, it completely destroys the structure. All mutations destroy fitness.

[00:26:35]These same problems also plague von Neumann's universal constructor. It's too perfect and too brittle, and so it never evolves.

[00:26:44]However, we have one more trick up our sleeve, a variation of Langton's loop called EvoLoop that was introduced in the '90s by Hiroki Sayama. It's the same idea, but the rules have been tuned to make the replicator less perfect and less brittle, [music] and therefore more open to evolution.

[00:27:01]Offspring can be different from their parent and actually survive.

[00:27:05]>> [music] >> Some are smaller, some are larger, some die, some don't, and some replicate even better than their parent. There is genuine replication, variation, and selection. Evoloop really does evolve with no explicitly defined biology. It is one of the closest things we have to truly emergent, open-ended evolution.

[00:27:30]But, it's not quite there.

[00:27:34]First off, it's not totally emergent.

[00:27:36]You have to explicitly encode the initial gene into the starting state.

[00:27:40]It's not in the rules, but it's extremely unlikely that a replicator would just emerge from random soup.

[00:27:46]More importantly, it's not open-ended.

[00:27:49]The loops quickly evolve to become very small so they can replicate faster than their rivals, and that's it. They converge to a very simple, stable, boring state. It's a shame because I can imagine a version of Evoloop that continues to complexify, with loops growing instead of shrinking, [music] maybe wrapping around and strangling each other, or stretching out and slithering like [music] snakes. But, they don't.

[00:28:13]There are other variants of Evoloop like sexy loop, and they all basically do the same thing. They converge [music] and simplify.

[00:28:21]As I said before, that's just what evolution does sometimes. Replicators don't have to complexify. [music] Sometimes smaller and simpler is the better survival strategy.

[00:28:32]But, in our own evolutionary tree, it's a simple fact that life has complexified. There are cells inside cells that [music] form colonies of cells that form colonies of organisms.

[00:28:44]Short of a supernatural explanation, why is it that real biological life has complexified so much while every form of virtual life stalls out? Evidently, that simple recipe for a Darwinian replicator is not quite enough. [music] We are missing something, and no one really knows what that something is.

[00:29:06]There are some more obvious reasons why this could be the case. Real physics and chemistry are just way more complicated than any simulation. They produce much richer emergent complexity even outside of biology. The non-biological world is chaotic and organized and dynamic and complex with many different environments and resources. [music] It's a chemical playground for life.

[00:29:29]It's also huge and old. Keep in mind that it took billions of years for life to become multicellular. It was single-celled for the vast majority of biological history.

[00:29:40]It could be that one of the programs in this video actually would produce open-ended evolution if you ran it for literally a billion years in a massive simulator.

[00:29:50]But, I doubt it. And you'd want something faster.

[00:29:54]There are many threads to pursue here.

[00:29:56]Life uh finds a way.

[00:29:58]Like a version of EvoLoop where the world beyond the replicator is more dynamic and interesting might encourage more open-ended evolution. A 3D EvoLoop also seems promising. [music] But, I bet the future of A-life belongs to smooth simulators like Lenia. It just seems fundamentally more complex, [music] and it can actually be trained with machine learning methods. I haven't even talked about neural cellular automata.

[00:30:23]Flow Lenia was an experiment that tried to enable open-ended evolution by letting different Lenia parameters interact and compete. It's a little too complicated to go into here, but it's very cool, and I want to simulate it.

[00:30:37]Hybrid algorithms that combine elements from genetic algorithms, evolution simulations, and emergent A-life are another promising direction. A great example of this is the Alien Project, which is [music] just gorgeous. It's a physics/chemistry simulation that is very emergent all on its own, but it then also injects some more explicit [music] biological systems like evolvable neural networks. Really cool stuff.

[00:31:02]It could also be that simulating energy or metabolism is a necessary building block for open-ended evolution.

[00:31:11]And finally, I've been talking about the kind of life that lives inside a computer program, but I haven't mentioned the kind of life that is a computer program. There's a whole other world of genetic programming, self-copying scripts, computer viruses, and AI agents.

[00:31:27]>> [music] >> But that's a video for another day.

[00:31:31]To wrap this video up, I want to answer a question that you might have rattling around in the back of your brain.

[00:31:36]What the hell is the point of any of this? Who cares if we can simulate evolution? What good does it do? What problem does it solve? What practical application does it have? The whole field of A-life is sometimes criticized for being unscientific. It's just a bunch of dudes and gals sitting around making pretty patterns of pixels.

[00:31:57]And that's kind of fair. It's easy to get a little carried away by their beauty and to draw grandiose conclusions from a few bits on a screen, so to speak.

[00:32:08]But lots of abstract work is weird and useless until it's not. Mathematics is also just a bunch of dudes and gals sitting around drawing pretty patterns on paper. And then a hundred years go by and those pretty patterns change the world.

[00:32:23]I bet that A-life could shed genuine light on real biology, especially the study of abiogenesis, which remains a mystery. The emergence of [music] abstract life might provide clues for the emergence of real life. It can't replace real evidence, but it might inspire the right kinds of questions [music] and experiments.

[00:32:43]So, here's some extremely weird and unscientific advice that just might be useful for someone who's trying to simulate the emergence of life, either virtually or physically.

[00:32:53]Here's the advice. You ready for it?

[00:32:54]You're not ready.

[00:32:55]Be on the lookout for [music] worms.

[00:32:59]Yep, that's it. Be on the lookout for worms.

[00:33:02]Why?

[00:33:03]Well, if you see worm-like structures, you're probably onto something. If you can mess around with your experiment and get the worms to start to break apart and split into sub worms, then you might see something that looks a lot like budding bacteria, self-replication.

[00:33:19]Where there are worms, there just might be replicators hiding somewhere nearby.

[00:33:24]I've seen this happen a few times in different simulators, so take it for what it's worth.

[00:33:29]Even if A-life is scientifically useless, there's nothing wrong with making pretty patterns of pixels. It's beautiful and it's fun. I think one of the most practical applications of A-life that can actually make money is game design. It's a really unique experience to play around in a world that evolves with or without you. I think it's an underserved market. You might even be able to fund A-life research by making your work playable.

[00:33:56]But for me, artificial life doesn't need any more justification. Evolution is a brutal but beautiful process that concentrates and composes complexity.

[00:34:07]>> [music] >> It's an emergent property that explores emergent properties. We are the products of this [music] process. We are the survivors, kin to all life on Earth, the river out of Eden.

[00:34:35]>> All right, thanks for watching everyone.

[00:34:36]I hope you enjoyed hearing me pronounce cellular automata 700 times. There's a lot more to talk about here, so be sure to subscribe if you'd like to see more and share this [music] video with your nerdiest friends. If you want to support me, I have a Patreon and you can use this link from Scrimba to learn how to code and get a 20% discount on a pro subscription.

[00:34:55]Goodbye.

[00:34:59]>> Those are some of the [music] things that molecules do given 4 billion years of evolution.