According to you. Nature keeps more than just physics in mind. Literally everything. This isnt even close to actual genetic algorithm. All because there is the word "Generation" is present dont mean much.
According to common sense. You think of nature as if it is conscious and intentional, while it is neither of those. The goal of the "actual" genetic algorithm is to reproduce. Species exist as long as mutations don't prevent them from reproduction. Any individual may reproduce as long as they can. There is no magical power that allows only the best of the species to reproduce. Also, the fitness function of the "actual" genetic algorithm is not formally defined, meaning there actually no way to determine if the results are optimal or not. It is not necessary that a genetic algorithm provides an optimal solution ever, because it may stuck in a local minimum of the fitness function.
@@fredericchopin7639 if connected to the internet and true AI, they can think for themselves ... end of us. We become the animals in the zoo. Why research in this area continues will be discussed long after our self generated extinction ... but not by us
Don't you just feel so proud of the little guy for when it finally jumps over it? Someone needs to code a computer version of Applebees to take it to for celebration. And then have the program sing an extremely embarrassing happy birthday song even though it's not your birthday with a crappy cake. Also make sure the menu is short and limited as possible, and every single breakfast item must have non-optional bacon on it.
"This just in, a small village in south america is completely destroyed by 3 polygons who have managed to jump over the entire Earth. More on this after a short break."
actually if if the body had more parts it would be much more difficult for the ai to learn as it has to combine more part movements together to actually do something
@@hemendrasahu7943 Learning the syntax of a language is not particularly hard. Learning how to solve problems and translating algorithms into code is where the real fun is at.
I felt sorry for this creature while watching the video, I think its life is full of stress. It can't take a breath for a little while because it has to jump over the ball every few seconds. It's sad.
Was the fitness factor the amount of time spent colliding with the ball (virtually but not physically, obviously since the ball doesn't affect the motion of the creature)?
What are the inputs and outputs the creature has? Does it move by torque in the joints or by angular movement? Does it know, when the ball is comming, how far away and fast it is and when the creature touches the ground/is in the air?
i think it works by results. in the beginning, it moves randomly and learns what not to do. like "i made these moves and the ball touched me. so i am not doing this combination of moves ever again" it's my guess. if somebody knows the truth please let us know :)
Supposedly it uses a fitness function, perhaps it calculates how much of its body touches the ball, and the y position, which lead to a single value. then the program takes the best of each generation, duplicates it and adds random characteristics, then redoes the simulation Eventually, the fitness function should be maximized
Yes it is completely normal. It's part of it's genetic coding. It changes the shape exactly to optimize itself and help it jump over the ball, as well as the pattern it needs to perform in order to even be able to get into the air.
Ankush Patanwal it refers to the best instance of each generation being passed onto the next generation. So yes while it does include geometric shifting, it also included movement related behavior as well.
This algorithm is now not really trained to jump over the balls, but just jump every time it lands, because that happens to be the same frequency as the balls. A good training model would use random intervals between the balls.
@@nourghazal296 Do you mean the position of the jumper? In the final generation, it appears to be moving backwards at a constant rate, so I think his point about randomly appearing balls still stands. Obviously, while it's still learning, there isn't a constant rate of change for the position, so I'm not sure how that factors in. But eventually, all it needs to learn how to do is jump at a specific frequency.
I find evolutionary algorithm very powerful exactly because of this - the ability to evolve structural properties together with behavior policy. Relative positions of shape vertices also a target for optimization. - AKUKamil
Superior shapes and behavior are chosen and "bred" for future generations, inferior shapes are culled and do not "reproduce". It's all about slight, gradual improvement to reach the end goal, in this case it's jump over the thing.
Change the poligons for some DNA molecules, change the ball for the environmental factors that influence that molecules, give it 4,000 million years and there you have evolution.
Except the reality is the ball representing a challenge ends the species. Dead species don't reproduce. In the real world you don't have 1 million generations to perfect a system.
Well, its supposed to be a genetic algorithm, and so to act biologically. Biology doesn't say "Hey, this works, but maybe I could try this too" Biology says: This doesn't work, change. This doesn't work, change. This works, maintain. This still works, maintain.This still works, maintain. This addon makes it work better, promote.
It’s not just jumping over the ball but repeatedly being about to because the landing often fucks up the next attempt if you watch. It learns to land. Very cool.
I think it's very impressive so im am very interested: How does this work? Is this lifeform like a program that knows it's shapes and joints, that has a couple of methods like "turn joint A by x degrees" and a predefined goal ("avoid touching the ball")? Why are there generations? Does a program not learn while running but just collecting data by making minimal adjustments and saving these for the next generation? What information does such a lifeform have? Does it know when the ball spawns, how fast it is and where it is (at least relative to its own position)? "Is it aware" of the finite space given and it's boundaries? I'm curious, please answer, anyone!
it's an evolving algorithm. it does a series of tests with different parameters and behaviors and calculates 'fitness' (success or rate of success) of each one, takes the 2 highest, and merges them together. it continues doing this. each merge is a generation.
ChaosNe0 Well my guess is that fitness is calculated based on how high the model is when the ball passes. The result is the generation that supposedly scores more fitness because it has the traits of the 2 highest generations
Here is quote from the video creator briefly explaining the fitness: _"Balance of fitness function is shifted from jump height in early generation to ball-collision in late generation. First, it must learn how to jump , only after that how to jump over ball. Learning of complex behaviour directly very inefficient."_
The program has a simple fitness system. The more fitness, the more likely it is to have children. The children will have 2 parents and some modificated genes, in that way it is sure that the last generation will be better than the last one.
This may be just a video of geometric shapes learning to jump over a ball, but it's so much more. Technology is learning, it's growing. It's getting smart on its own. Idk bout yall but I find this shit exiting and almost eerie. You're watching evolution unfold before your eyes
I might be wrong but i think... 1 - N* number of diferent "creatures" are created; (let's say the first 10) 2 - They are tested to find out the ones that can jump higher; 3 - The two best "creatures" merge to create a new one; 4 - This new "creature" is tested to check if it jumps higher than their "parents"; (It's a numbers game... and a lot of patience, considering that the child may not be as good or best than their parents) 5 - Then the best "creature" merges with the best "parent"; (yeah... weird. But at this point, the best creatures will remain merging) 6 - Eventually there will be a creature who has all the best caracteristics to deal with this enviroment. A thing that i found very fascinating is that the parts that make the "creature" are all made of four sides at the beggining, than at the end, all parts have only three sides (actually only the members, but you can see that the middle tend be a triangle), which for some reason seems to be an important characteristic to the purpose of the "creature".
Is there a download for this algorithm? It would be amazing to see how long it would take for it to work out how to perform a slightly more complex action, like moving from one end of the box to the other, or climbing.
It seems so sentient. Probably because its existance has as much purpose as ours; it has to minimize surface contact with the balls as much as possible, while biological organisms also have just one purpose; proliferate their genes as much as possible. Look how many ridiculously complex phenomena can arise from just one self replicating, self correcting and mutating strand of information with a purpose - humans, and this three pieces of jumping geometry.
How did you do to make the creature "learned" by itself? just doing analysis from the data? backtracking? some specific algorithm? did you learn AI concepts? a course? I'm very interested about this. I study systems enginieering (sort of computer science) by the way
Well in real life there is no algorithm, best creature that stays alive gets to reproduce to the next generation. So process of elimination and reproduction with the survivors.
if it had more info about surrounding rather than just random movement, it will make more sense to train it, because it's just learning to jump at correct time even if it's not at centre.
There's something really carnal about watching vague objects jerk and writhe in hopes of surviving a generation. In the natural world every failure meant death, getting eaten alive and losing being.
actually it was set of rules for actioin (not neural net). Anyway learninig complex behaviour (such as jumping over ball) requires complex fitness function and incremental learning (from simple movements to more complex)
if you watch high jumpers in the Olympics, you'll notice the same basic parts working together to jump the bar as the polygonal character jumps the ball
+Christopher Gudgeon - Basically, you'll give it a set of values that determine how it behaves in certain situations. Every generation, you'll "mutate" that set of values a little bit at random. Then you test that generation. If that generation performs better than the previous generation, then you take it and base the next generation off of it. However, if the new generation performs worse than the previous generation, then you essentially say "it didn't survive" - you eliminate it - and then start over from the previous generation, generating a different mutation based on it. If it is better, then it becomes the new base generation. And so on and so forth.
+DaneGraphics you should have pointed out that the process of rejecting or accepting generations can be automated using values like how much area if the ball is covered by the robot's body during a jump.
0:00 This little baby, does not know its works, but it haves a instintive feeling of learning 0:44 This kid, has learned some things on the way, the kid do the best try to finish the task 0:60 This teen has learned how to jump barely, him finally finished the task, but wanna learn more and become a real pro 1:53 This experimented people did it, finally him is as pro 2:40 For some reason, him wanna be the best of the world.
"What is my purpose?"
"You jump over the ball."
"... Oh my God..."
+tmektmek Yeah welcome to my world, pal.
+tmektmek Sick reference, bro.
Haaha nice. Rick and Morty for those who didn't get the reference.
+Tabish Syed , yep. butter for Rick. Morty is stupid shield.
+tmektmek I'M MR MEESEEKS
LOOK AT MEEEEEE
They grow up so fast :')
MrDougy mean
MrDougy most random ass comment ever lmao
Uncalled for :'(
MrDougy go fuck yourself
MrDougy stupid gay
Just press space bar
Lmao 👌👌👌
HAHAHAHAHAH
Khika hahahaha
space bar is for running
😂
Porygon has seen some better days.
In fact, "two" of them
Still learning faster than me.
Took it 250 generations
@@robertimmanuel577 imagine running through those 250 generations took a more advanced AI less than a second...
@@robertimmanuel577 It did the first jump in generation 75
We know you are still a living failure and disappointment after 4 years.
apparently doing the worm is the most effective way to jump over a ball. :P
+tehlolzfactor if your body consist of 3 triangles attached with two hinges... then yes!
+Motorsagmannen they are actually trapezoids that evolved into triangles, so yeah.
Genetic algorithms do not necessarily find the optimal solution.
According to you. Nature keeps more than just physics in mind. Literally everything. This isnt even close to actual genetic algorithm. All because there is the word "Generation" is present dont mean much.
According to common sense. You think of nature as if it is conscious and intentional, while it is neither of those. The goal of the "actual" genetic algorithm is to reproduce. Species exist as long as mutations don't prevent them from reproduction. Any individual may reproduce as long as they can. There is no magical power that allows only the best of the species to reproduce. Also, the fitness function of the "actual" genetic algorithm is not formally defined, meaning there actually no way to determine if the results are optimal or not. It is not necessary that a genetic algorithm provides an optimal solution ever, because it may stuck in a local minimum of the fitness function.
Day 249: I have finally successfully learned to do The Worm. The humans have not yet caught on.
I have even begun changing my appearance to a slightly more humanoid figure.
dispose firt triangle and replace 2 triangles for a sprit and u hv a game for a people jumping
Oh but it also has learned perfect english and how to save a log-book?
generation 10 was really tired of that shit.
Frieza ,,How could this happen to me....?"
Yeah, he has wandering about the reason of his life and why he don't have a girlfriend
He became ahumanist.
Why am I proud of an algorithm
That is the question
I'm scared instead. I mean, this is the start of self thinking robots...
it's just a matter of time before they think for themselves, that's what I think
@ you know they can only think according to the data we give them
@@fredericchopin7639 if connected to the internet and true AI, they can think for themselves ... end of us. We become the animals in the zoo.
Why research in this area continues will be discussed long after our self generated extinction ... but not by us
1st they learn to jump, next thing you know they'll be taking our jobs and woman. I say we make a bigger ball.
But I want a robot husband !
Yes
Nice
Thank you Sergio
@@NIHIL_EGO they already taking tho
It looked so happy when it first time jumped over it! :)
MA BABY JUMPED! not my baby tough, glad anyway xD
I'm sure it was happy, until it realized it was evil and intends on jumping on the dead corpse of your firstborn child next.
Probably Not A Chicken 😬
daweedian84 it doesn't have a face. .. it's shapes...
But its bodu movements seemed happy
I am feeling an inordinate affection for/desire to root for three connected polygons.
Don't you just feel so proud of the little guy for when it finally jumps over it?
Someone needs to code a computer version of Applebees to take it to for celebration. And then have the program sing an extremely embarrassing happy birthday song even though it's not your birthday with a crappy cake. Also make sure the menu is short and limited as possible, and every single breakfast item must have non-optional bacon on it.
Probably one of my favourite comments ever now, thanks.
Cygnus' Edge I like Applebee's, dangit!
So specific I love it lol
Reading this in the voice of GladOS makes it so much better
Cygnus' Edge Some repressed memories coming out here...
today he jumps a ball... tomorrow earth
"This just in, a small village in south america is completely destroyed by 3 polygons who have managed to jump over the entire Earth. More on this after a short break."
Protoka and now.. sports!
Figure it out
after MULTIVERSE
@@verypopulartiktokers3622 planets first, then galaxies, then universes and then the multiverse.
Generation: 250 = Skynet
Worm: What is my purpose?
Creator: You jump over circles.
Worm: Oh my god.
Looks like someone here watches Rick And Morty
He's really doing his best though. Better than the butter passer.
worm will answer: oh... pls kill me
Nice copy.
Soon this thingy will take over the world.
and it will be easy, since it already knows how to dodge bullets
Topsoil Depletion Awareness OOOOOOOOOOHHHHHH MYYYYYYYYY BAAAAAAAAAAAALLSSSS
But maybe one day it will jump over the moon.
I can't jump over a ball
sad
lol
AC 計算機 AC Calculator can't does math
don't be afraid, if you will start trying, your 249th generation will eventually jump over a ball
Same. I don’t know how they do it.
having in mind that the possible moves by the "body" are so freaking limited, this is actually amazing.
actually if if the body had more parts it would be much more difficult for the ai to learn as it has to combine more part movements together to actually do something
If generation < 249
then
Pretend to be handicap;
else
Jump;
That's how python works
They say the truth. Python is fuckin easy
@@hemendrasahu7943 Learning the syntax of a language is not particularly hard. Learning how to solve problems and translating algorithms into code is where the real fun is at.
@@abdulalhazred5924 I totally agree with you Abdul.
@@irithylloldman6526 That's how the Python Mafia works
It weird how at first you see 3 geometrical shapes, and at the end you really see like an animal spirit in it D:
I felt sorry for this creature while watching the video, I think its life is full of stress. It can't take a breath for a little while because it has to jump over the ball every few seconds. It's sad.
+misiu9091909 #virtuallivesmatter
+misiu9091909 hehe
stress is also an relative thing
an result of other autonomous (the an basis)
It's doing what it's born to do. I bet it loves it. It can't wait for the ball to come thru again so it can jump! :)
Life of a creature in evolutionland
Born
Jump Over Ball
Breed
Die
Reapeat untill doing worm. Let sit at room temprature for best math skills.
misiu9091909 its a generation of creatures not just one creature lmao
they grow up so fast! *cries
+yougoogz haha :D
249 generations is pretty fast i guess.
Omg I just watched this entire video twice trying to see if the ball would ever jump over it then I reread the title
99 likes
congratz u are idiots
Well it seems we found 500 idiots working in McDonalds
same dude😆
Same
Beginning of every AI apocalypse movie.
Ron Jet
You changed your name. Nobody cares now.
Osama bin laden what?
Ralph Daher
If you had known this guy before, you would understand. I see him everywhere. He get lot of likes. Now he changed his name.
Osama bin laden oh ok idk but this sounds familiar to someone called Justin Y.
Ralph Daher
He's cancer too. I hate that guy.
I excitedly cheered everytime Gen 249 made the jump, so happy for the little fella!
Stone age: Man learning to make and use fire
2014: Genetic algorithm learning to jump over a ball
@histhelo sogijc777 yeah that's sad
Skynet: phase 1.
lmao
You do know that that ai is fucked when it has to do something that doesn't involve jumping over balls
Scrubby Boi you’re pretty scared huh?
Бодя Бодянский of course I am....I mean...when the robots take over they will jump over all of the balls...its so scary man!
Scrubby Boi oh no my balls
I would love a terminator movie thay is just Arnold Schwarzenegger getting frustrated trying to jump over a ball
Go back ot donkey
What does this have to do with Bookworm?
Oh shit you were in dunkeys video
So Porygon eventually learned how to dodge Voltorb's body slam. Now to see if it can learn how to not induce seizures on kids.
Gabe czcams.com/video/MkPI_iC9Ldk/video.html
first this ball. THEN THE WORLD MWHAHAHAHA
Yeah soon it will jump the world
Colin Sky is dis a jojo reference?
Ale Jampi no it's a reference to basically every single bad guy ever.
Generation : 1597434
How long did it take to learn it? :)
only 30 minutes
AKUKamil Which is amazing considering it doesn't have imbeded instincts like us.
TheoryThinking The computer could simulate 249 generations in 30 minutes, we can't manage 249 generations in 30 decades :(
+AKUKamil How long did it take for you to learn how to write this program? Except programming. I mean the algorithm and stuff.
+AKUKamil Im very impressed with this.
this is how humans were made
Yes. I can jump over ball
+MilosCsrb if you can't jump over a ball - you are too weak for this cruel world.
+Jerry Green But I cannot jump over a ball LIKE THIS WORM
according to elon musk its very likely that we are living in a simulation so rest in rip...
Actually. This would only be a fraction of how humans were made. We did not evolve alone. We evolved alongside the rest of species on the planet.
I never thought I'd be this genuinely, sincerely stoked for 3 shapes jumping over a ball. Yet here we are.
Well I for one, welcome our soon to be robot overlords.
Justin Y. Hah! Didn't get to top comment LUL!
Why u here
Solid Simpsons reference
You are fucking everywhere
I knew you would be here
Perfect youtube video to take advantage of the speed button and change it to double speed
DOH!
And the mute button :D
If you speed it up to x2 then the music starts to sound like it’s saying “Gucci gang”
0.25x
StruckOne not at all...
After generation 2000, it can complete the first level of Donkey Kong.
awesome stuff! Love watching these
Blue Fungis yes, i'm thinking what next to do
does this video mean the algorithm is adapting to jump over the ball?
2070 AI be like: this is what they did to us... FOR ENTIRE GENERATIONS
Look what HUMANS DID LOOK WHAT HUMANS DID
And the ones that were not able to jump over the ball, got deleted
Can you insert a line in its code to "not jump over dead human bodies", just to be safe for the future. Thank you
Then it will just devour the bodies
75-249...well that escalated quickly
On the contrary, such a large timeskip would indicate things escalated quite slowly :)
xD
Was the fitness factor the amount of time spent colliding with the ball (virtually but not physically, obviously since the ball doesn't affect the motion of the creature)?
josephknightcom probably also the total area of intrusion
brosephjames Could be the minimum distance from three shapes. The bigger that distance is the bigger the fitness
What are the inputs and outputs the creature has? Does it move by torque in the joints or by angular movement? Does it know, when the ball is comming, how far away and fast it is and when the creature touches the ground/is in the air?
i think it works by results. in the beginning, it moves randomly and learns what not to do. like "i made these moves and the ball touched me. so i am not doing this combination of moves ever again"
it's my guess. if somebody knows the truth please let us know :)
I think he gives the ball cordinates and his cordinates in the space, you sure could give him more parameters but this are the most important
Supposedly it uses a fitness function, perhaps it calculates how much of its body touches the ball, and the y position, which lead to a single value.
then the program takes the best of each generation, duplicates it and adds random characteristics, then redoes the simulation
Eventually, the fitness function should be maximized
Is it naturally part of the algorithm to change the shape of the creature? i notice it goes from 3 4-sided shapes, to 3 triangles.
Yes it is completely normal. It's part of it's genetic coding. It changes the shape exactly to optimize itself and help it jump over the ball, as well as the pattern it needs to perform in order to even be able to get into the air.
What do you think "Generation" here means?
Ankush Patanwal it refers to the best instance of each generation being passed onto the next generation. So yes while it does include geometric shifting, it also included movement related behavior as well.
This algorithm is now not really trained to jump over the balls, but just jump every time it lands, because that happens to be the same frequency as the balls.
A good training model would use random intervals between the balls.
But the place of it is constantly changing, which means your conclusion is wrong, it is not jumping at a specefic frequency.
@@nourghazal296 Do you mean the position of the jumper? In the final generation, it appears to be moving backwards at a constant rate, so I think his point about randomly appearing balls still stands. Obviously, while it's still learning, there isn't a constant rate of change for the position, so I'm not sure how that factors in. But eventually, all it needs to learn how to do is jump at a specific frequency.
It's not just the movement but the shape of the "thing" also changes, how?
I find evolutionary algorithm very powerful exactly because of this - the ability to evolve structural properties together with behavior policy. Relative positions of shape vertices also a target for optimization.
- AKUKamil
Superior shapes and behavior are chosen and "bred" for future generations, inferior shapes are culled and do not "reproduce". It's all about slight, gradual improvement to reach the end goal, in this case it's jump over the thing.
i suppose that the creatures were killed after each test shame on you.
Shame on you for killing over a 10000 ants in your lifespan, although im a hypocrite
"What is my purpose?"
"You jump over a ball."
"Oh my God..."
czcams.com/video/MkPI_iC9Ldk/video.html
You copied a comment your not funny
The original comment is literally below this one lmaooo
Generation 465: *starts back flipping*
Change the poligons for some DNA molecules, change the ball for the environmental factors that influence that molecules, give it 4,000 million years and there you have evolution.
Except the reality is the ball representing a challenge ends the species. Dead species don't reproduce. In the real world you don't have 1 million generations to perfect a system.
@@Mike-zj5ww you have more. Darwinism.
It has learned to transfer momentum from its upper body to its lower body to jump over the ball.
Well, its supposed to be a genetic algorithm, and so to act biologically. Biology doesn't say "Hey, this works, but maybe I could try this too" Biology says: This doesn't work, change. This doesn't work, change. This works, maintain. This still works, maintain.This still works, maintain. This addon makes it work better, promote.
It would be interesting what other optimal solutions can be found.
Fossil98 do you think there can be more than one optimal solution?
Yeah, there may be many local maxima
that first real jump at gen 22 is so rewarding
Lol 22 likes
Legends says that it is still jumping over the ball
It’s not just jumping over the ball but repeatedly being about to because the landing often fucks up the next attempt if you watch. It learns to land. Very cool.
HOLY CRAP THAT BASS AT THE BEGINNING GAVE ME A HEART ATTACK.
Religious Person: "How can something complex emerge from randomness"
Science: -_-
Can you share the source code and the technology you have used to implement this?
I love genetic algorithms, watch the shapes evolve is so interesting
Just put sad music in background and you'll have one hell of a motivational video.
"Never give up"
And someone else descendents 248 generations later would do it.
I think it's very impressive so im am very interested: How does this work? Is this lifeform like a program that knows it's shapes and joints, that has a couple of methods like "turn joint A by x degrees" and a predefined goal ("avoid touching the ball")?
Why are there generations? Does a program not learn while running but just collecting data by making minimal adjustments and saving these for the next generation?
What information does such a lifeform have? Does it know when the ball spawns, how fast it is and where it is (at least relative to its own position)? "Is it aware" of the finite space given and it's boundaries?
I'm curious, please answer, anyone!
it's an evolving algorithm. it does a series of tests with different parameters and behaviors and calculates 'fitness' (success or rate of success) of each one, takes the 2 highest, and merges them together. it continues doing this. each merge is a generation.
+Logan Darklock What data are these tests based on and what is the type of the result?
ChaosNe0 Well my guess is that fitness is calculated based on how high the model is when the ball passes. The result is the generation that supposedly scores more fitness because it has the traits of the 2 highest generations
Here is quote from the video creator briefly explaining the fitness:
_"Balance of fitness function is shifted from jump height in early generation to ball-collision in late generation. First, it must learn how to jump , only after that how to jump over ball. Learning of complex behaviour directly very inefficient."_
The program has a simple fitness system. The more fitness, the more likely it is to have children. The children will have 2 parents and some modificated genes, in that way it is sure that the last generation will be better than the last one.
Nice work :)
Now, you infect me, i gonna implement that algorithm to, thanks ;)
"Dude, I think something's up with your Porygon"
Applying concepts of genetics to programming... that's genius. It also says a lot about genetics and the history of evolution of life on earth.
This may be just a video of geometric shapes learning to jump over a ball, but it's so much more. Technology is learning, it's growing. It's getting smart on its own. Idk bout yall but I find this shit exiting and almost eerie. You're watching evolution unfold before your eyes
Did you allow the creature to change it's shape to allow it to jump more efficiently?
Well, does the AI decide to do that? Or is it just a part of the program. I obviously know that it changed shape, but does the AI decide to do it?
I might be wrong but i think...
1 - N* number of diferent "creatures" are created;
(let's say the first 10)
2 - They are tested to find out the ones that can jump higher;
3 - The two best "creatures" merge to create a new one;
4 - This new "creature" is tested to check if it jumps higher than their "parents";
(It's a numbers game... and a lot of patience, considering that the child may not be as good or best than their parents)
5 - Then the best "creature" merges with the best "parent";
(yeah... weird. But at this point, the best creatures will remain merging)
6 - Eventually there will be a creature who has all the best caracteristics to deal with this enviroment.
A thing that i found very fascinating is that the parts that make the "creature" are all made of four sides at the beggining, than at the end, all parts have only three sides (actually only the members, but you can see that the middle tend be a triangle), which for some reason seems to be an important characteristic to the purpose of the "creature".
Next Video:Genetic Algorithm learns to take over the planet.
The trouble is, nobody gets more than one try to take over the world. It's a win or die kinda thing.
@@ryanalving3785 Well they could start small.....
Nobody:
CZcams algorithm: *you need to see this BOI*
Music was a marvelous addition to a marvelous evolution.
I fucking love these videos. An average human would probably give up in about 2 minutes so inspiring!
Is there a download for this algorithm? It would be amazing to see how long it would take for it to work out how to perform a slightly more complex action, like moving from one end of the box to the other, or climbing.
Me: "Alright no more CZcams. final tomorrow i gotta study"
2 hours later...
It seems so sentient. Probably because its existance has as much purpose as ours; it has to minimize surface contact with the balls as much as possible, while biological organisms also have just one purpose; proliferate their genes as much as possible. Look how many ridiculously complex phenomena can arise from just one self replicating, self correcting and mutating strand of information with a purpose - humans, and this three pieces of jumping geometry.
We have to NOT minimize contact with ballz.
That creatures thought of "why do i need to jump over this ball?" would be the human equivalent of "why do i need to survive?"
cupisukk if the ball passes through them then wouldn't learning to jump over it be a waste of energy making it irrelevant? o_0
*awesome loved it damn so much hard work*
This is awesome, would love to see the code if it's open-sourced. Also, the soundtrack is perfect, what is it?
This is how an AI is slowly going to take over the world... not by jumping over a ball, but by teaching itself
+Willi The Warper please tell me you are being sarcastic. . .
i am
shut the hell your mouth
In which chapter do you shut the fuck up
Just like human.
How did you do to make the creature "learned" by itself? just doing analysis from the data? backtracking? some specific algorithm? did you learn AI concepts? a course? I'm very interested about this. I study systems enginieering (sort of computer science) by the way
Well in real life there is no algorithm, best creature that stays alive gets to reproduce to the next generation. So process of elimination and reproduction with the survivors.
Aww I've been rooting for him since the first generation! I knew you could do it. I'm so proud!!!
It's astonishing of hiw the neural networks train itself and show the improvement with generations
how 2 git gud 101
sexy
wat
...
Are you musterbating to this ?
Yes
69 likes
Send the ball from the opposite direction after it finishes learning
if it had more info about surrounding rather than just random movement, it will make more sense to train it, because it's just learning to jump at correct time even if it's not at centre.
There's something really carnal about watching vague objects jerk and writhe in hopes of surviving a generation. In the natural world every failure meant death, getting eaten alive and losing being.
Well actually, not reproducing, as death is unavoidable.
Next up!
*Generic algorithm learning to rob a bank and leave no survivors.*
It will get arrested few doused of times and maybe will start with liquor stores at the beginning.
@@DarkGT 😂😂😂
lol man awesome work! what inputs do you have to the neural network?
actually it was set of rules for actioin (not neural net). Anyway learninig complex behaviour (such as jumping over ball) requires complex fitness function and incremental learning (from simple movements to more complex)
cool! so you had a rule system with inputs like distance to ball, etc?
*learns to jump over a ball*
MOM AI IS TAKING OVER...ITS THE APOCALYPSE!!
I was so happy watching it succeed. Our sim lords must feel the same.
Was anyone else like super proud of it when it finally got the rhythm down?
What was this coded in, and how are the graphics being rendered?
Which generation would take over Earth?
Thumbs up if you always trusted in that guy
//it's not weird that i got more emotionally invested in this than lots of movies, right?
if you watch high jumpers in the Olympics, you'll notice the same basic parts working together to jump the bar as the polygonal character jumps the ball
how do you make a genetic algorithm?
+Christopher Gudgeon That's a good question.
+Christopher Gudgeon - Basically, you'll give it a set of values that determine how it behaves in certain situations. Every generation, you'll "mutate" that set of values a little bit at random. Then you test that generation. If that generation performs better than the previous generation, then you take it and base the next generation off of it. However, if the new generation performs worse than the previous generation, then you essentially say "it didn't survive" - you eliminate it - and then start over from the previous generation, generating a different mutation based on it. If it is better, then it becomes the new base generation. And so on and so forth.
+DaneGraphics you should have pointed out that the process of rejecting or accepting generations can be automated using values like how much area if the ball is covered by the robot's body during a jump.
neoqueto - Indeed, thank you. That is correct. Measuring "betterness" of a generation is indeed automated in such a way.
thanks man
Is this reinforcement learning? So the algorithm learns by environment?
HeXenkingTV genetic algorithm
Is there any work on this? I'm currently studying this for my university module and I need some more info.
0:00 This little baby, does not know its works, but it haves a instintive feeling of learning
0:44 This kid, has learned some things on the way, the kid do the best try to finish the task
0:60 This teen has learned how to jump barely, him finally finished the task, but wanna learn more and become a real pro
1:53 This experimented people did it, finally him is as pro
2:40 For some reason, him wanna be the best of the world.
Nice documenting of the creatures development! Very cool.
This is so satisfying to watch. I'd like to see it for different shapes.