AI Learns to play the Worlds Hardest Game

  • Published on Jul 12, 2018
  • If it wins does that make it the worlds best AI?
    Run the AI in your browser

    Check out my tutorial on genetic algorithm

    Follow me on twitter

    Become a patreon to support my future content

    Check out my Discord server

Comments • 7 018

  • Code Bullet
    Code Bullet  3 years ago +3176

    Hey Guys
    Its not more levels but it is the next best thing

  • Septimus ii
    Septimus ii Year ago +329

    I feel like they haven't really learnt the rules of the game, they've just learnt to follow that exact path

    • Nova
      Nova 9 days ago +1

      @HarryD neurons are kind of like wires… it depends on how you think about it.

    • G G
      G G 6 months ago +2

      it literally a single man made AI, now imagine what Elon Musk crew can do

    • HarryD
      HarryD 7 months ago +4

      actually, in my opinion, ai dont actually “think”, if u give it a quest, it will try until it get it done, it don’t understand the quest or how its done the quest cuz well its kinda simple to how cpu work, its just lights up lights, its need the correct combination. if it die or didnt. win it cant get the right combination

    • Greeb McFleeb
      Greeb McFleeb 7 months ago +6

      ​@SuomuSintti i feel like you could give the ai the ability to "see" the dots and know their positions. the only problem might be that if you give it that information, a program doesn't have the delayed reaction time and imperfect judgement of a human so it might be too easy once it learns the rules. Idk how ai works really but I feel like this is super simplified

    • SuomuSintti
      SuomuSintti 7 months ago +2

      @okay Yeah and more advanced AI could avoid the dots and move away when they are going to hit

  • Rusty Nail
    Rusty Nail 2 years ago +4332

    You should create AI to play Geometry dash

    • CowboyVoicem523
      CowboyVoicem523 6 days ago

      There is auto. Auto

    • Abhijit Prajapati
      Abhijit Prajapati 23 days ago

      @Léo i agree its basically flappy bird (which he did) but with slightly different neural networks

    • Matrodoli 10
      Matrodoli 10 3 months ago

      @Léo Yeah, but it would be hard for the AI.

    • E E
      E E 5 months ago


    • Spark Jex
      Spark Jex 5 months ago


  • Wraein
    Wraein 2 years ago +692

    “Most players have figured out how to get out of the starting area”
    *Me, being fully represented by the cubes still ramming into the walls:*

  • Squishmellow 3
    Squishmellow 3 2 years ago +942


    • Jan_Papież_Polak
      Jan_Papież_Polak 5 months ago +1

      Yes he fukin didit

    • Blobs2
      Blobs2 7 months ago

      BRUH Play the og

    • lmao
      lmao 11 months ago +2


    • Acrobro
      Acrobro Year ago

      cool math games bruh

    • CyberCider
      CyberCider Year ago +3

      Wait, it’s all 12 year olds?

      *Always has been*

  • Code Bullet
    Code Bullet  3 years ago +6299

    Holy shit, settle down guys I get it. I’ll do more levels.

  • Spider King
    Spider King Year ago +264

    Code bullet: can't get past the first level

    Everyone who grinded the game for hours: pathetic

    • Beetal
      Beetal 7 months ago

      Me and the boys in class grinding during math class

    • Kooora World
      Kooora World 7 months ago +1

      @Jesse Moss 15 minutes i swear on god i finished it on one minute or two the maximum

    • proyoutubr 1
      proyoutubr 1 7 months ago +1

      @Jesse Moss true

    • Jesse Moss
      Jesse Moss Year ago +1

      @duceysanem it's not tho
      Look for the pattern's of the levels

    • duceysanem
      duceysanem Year ago +3

      @Jesse Moss cap

  • Kira for the rescue
    Kira for the rescue 2 years ago +69

    I've played this game and I learnt how to beat the level easily from the 1st try by copying what your AI did and following exactly the same moves WOOOOOOOOOOOO.

    • Moon Rell
      Moon Rell 2 years ago +1

      @69ho69 world's hardest game

    • 69ho69
      69ho69 2 years ago

      Whats this game called mate?

  • Turtle Wurtle
    Turtle Wurtle Year ago +132

    “The worlds hardest game” is actually on scratch. You can see by the flag and stop button

  • Luis González
    Luis González 2 years ago +19

    I got through the first level in 19 tries then got thought to the fourth level of that game & got tired of it. The first one was the hardest one though

  • Spencer Stream
    Spencer Stream 3 years ago +5065

    Adding the incremental learning was a neat trick to get around the players stuck in the start. Overall very neat but I agree with others, I'd like to see a video where the AI learns the rules of the game, not the path of a particular level. This would be more fun to watch with more levels going forward.

    • SpaceGateX
      SpaceGateX 3 years ago

      @John Hunter No, never give the AI any information. Use artificial neural networks so it learns the game's rules without any algorithm / script to begin with. It should not even follow your playing. Same thing while talking about stock trading or people. People do not learn to walk so that someone tells it to them.

    • Stephen
      Stephen 3 years ago

      @Rawkfist1501 Good joke. Interesting philosophy.

    • Julio C.
      Julio C. 3 years ago

      But that is the way you as a human beat the game, learning the path of a particular level...

    • Pere Ginebra
      Pere Ginebra 3 years ago

      @Campasaurus Rex the coder defined what was good or what was bad to the AI to make it faster and easier to code just for demonstration purposes, but he could actually just tell it that loosing is bad and it would end up associating getting hit by balls as a bad thing. If you look it up on the internet, you will easily find and AI that beat a team of professional DOTA 2 players by just being told that dieing was bad and letting it learn.
      Also you don't gain credibility by telling people with different oppinions to yours on the internet to go hang themselves and using insults such as looser. These things just make me think you are a very close minded 10 year old. Have you by any chance been sheltered and homeschooled by a pair of flat earthers?

    • Pere Ginebra
      Pere Ginebra 3 years ago

      @Tayler Robinson it's bold claim to assume he's older than 10...

  • Flexible Games
    Flexible Games 3 years ago +1

    I love watching all the actors evolve over time, keep the best a different color, but would love to see the evolution of the entire set.

  • CyberNZ
    CyberNZ Year ago +4

    Me and my friends played this game in school for hours. I think we could get though the first 5 levels no deads. Then we learned that there were sequels we played those for along time we never got as good as we did on the first one tho

  • Jocose Sonata
    Jocose Sonata 2 years ago +6

    Imagine a game with a self-learning AI controlled swarm. Or in fact just a game with self-learning AI. Like the game ECHO but more.

  • Utkarsh Puniani
    Utkarsh Puniani Year ago +1

    I see AI play games and perform the task given to them such as reaching the goal in least time frame possible or simply survive. What if we give AI a role just to gain knowledge and enhance itself just by it's knowledge it gained and giving it the platform "web" to work on.

  • Goel
    Goel 3 years ago +445

    I would really like to see more levels, but I also want to see a neuroevolution approach to this. You would train it on the first 5 levels lets say, then start running it and see how well it can figure out levels it hasn't seen yet!

    • Goel
      Goel 3 years ago +1

      Thanks! People do seem to have trouble understanding that when something isn't generalized, it still is learning. There are many different types of machine learning, a genetic algorithms being one of them. Genetic algorithms (when not paired with a nerual network) are also commonly not general. I made a simulation almost exactly like this, and if you changed the course then it wouldn't be solved, since it was not a general solution, but it is still learning to complete the course.

    • TheMaskedNinja
      TheMaskedNinja 3 years ago +2

      Finally someone with this idea that comments without saying it’s not learning

    • Goel
      Goel 3 years ago

      That does make sense, all the distractions an colors and everything are really just for the human player.

    • Spencer Osgood
      Spencer Osgood 3 years ago +1

      idk I've played a lot of super hexagon and theres only 3 outputs the ai needs to calculate from 6 inputs. Move left, don't move, and move right based on where the next opening/s is/are. The patterns aren't really important and are more just to keep us humans entertained. Would be fun to watch all the same but a super trivial AI is going to have no problem with it

    • Goel
      Goel 3 years ago

      In that case, you would need a Neural Network and it would be really difficult. I really would like to see Code Bullet tackle Super Hexagon now!

  • Phoebe Ong
    Phoebe Ong 2 years ago +89

    How did Code Bullet even come across scratch???

    Also why not code the AI by remixing that game?

    • Gaming with Liam
      Gaming with Liam Year ago

      @Pnod Gaming you can also play it on coolmathgames

    • Slimeball
      Slimeball Year ago

      By searching Google.

    • kalbrooooo
      kalbrooooo Year ago

      YH khoo You can! And I have

    • YHkhoo
      YHkhoo Year ago

      you can't code ai in scratch

    • Pnod Gaming
      Pnod Gaming 2 years ago

      @kalbrooooo Okay. Each to their own.

  • leavemeandmybreadloafalone

    I’ve completed that game before, it only took 30 mins... *I FEEL SPECIAL NOW*

    • F0X
      F0X 2 years ago

      Youre not

    • F0X
      F0X 2 years ago

      Yourr not

    • rando prior
      rando prior 2 years ago +2

      Bullshit it took you 30 mins. It has like 40 levels.

  • Drew Darrington
    Drew Darrington Year ago +71

    1:04 “The design is pretty straight forward it’s just a bunch of squares.”


  • Xen
    Xen 3 years ago +2159

    if *red_player touch(blue)*
    then *try again thx*

    • Curiosity Tube
      Curiosity Tube Year ago +1

      Actuallly, in C#, it's
      public void OnCollisionEnter2D(collider collisionInfo)
      if(collisionInfo.collider.tag == "Enemy")

    • Piano DNA - James
      Piano DNA - James Year ago

      I would love to code it tho I can’t type it all here cause it’s too hard and crazy

    • BlueFire Animations
      BlueFire Animations Year ago

      @Eap ???

    • Eap
      Eap Year ago

      @BlueFire Animations bibliiitece

    • BlueFire Animations
      BlueFire Animations Year ago

      @Eap I see you made a mistake, it's std::cout

  • johnny onthespot
    johnny onthespot 3 years ago

    What software are you using to make the game environment? Love your videos!

  • Mitchell Smart
    Mitchell Smart 2 years ago

    more levels please bullet, i love this stuff

  • MaxTheMaximumGamer
    MaxTheMaximumGamer Year ago +4

    Your AI went to the Prometheus School of Running Away From Things and had to unlearn everything it learned there to beat this level.

  • johnny onthespot
    johnny onthespot 3 years ago

    Love to see more levels!

  • cookie123
    cookie123 3 years ago +1991

    please make more levels! turn this into a series already!

    edit 8 hours later: holy shit this got 1k likes? you guys are crazy.

    • Oby-Sver
      Oby-Sver Year ago

      Yeah GD my harder Game. But i GG

    • Jonah LT06
      Jonah LT06 Year ago

      Congrats on 1k likes! I’m happy you’re not one of the people who edit every 50 likes they get

    • George K
      George K Year ago


    • white
      white Year ago

      1 year in a row.

    • Mr Sweed
      Mr Sweed Year ago

      Wtf is your pfp

  • Noor Nasri
    Noor Nasri 2 years ago

    I'd like to see the game have a bit more randomness to it and the positions of the enemies being used for the decision, it would be really cool seeing all the ninjas becoming better and better at dodging stuff instead of just finding the right path

  • Balder
    Balder 2 years ago

    I finally understand machine learning. Thank you!

  • Anayz
    Anayz 2 years ago

    I rarely comment, but I genuinely want to see the next levels. It’s oddly satisfying watching the thing learn

  • GenieStream
    GenieStream 2 years ago

    I actually liked it when you were showing us the entire population as well. Shows the collective improvement they were doing

  • Venkatesh G
    Venkatesh G 3 years ago +1305

    AI learns to code and develop more AI.

  • Grace17
    Grace17 2 years ago

    what an intense episode o-o
    thanks a lot for putting this video up :)

  • Anna Nimmous
    Anna Nimmous 3 years ago

    pretty good, but I think the incremental learning stepped up too fast. I'd say add 1 move per 10 generations, starting at 10.

  • fanie isaac
    fanie isaac 2 years ago

    I laughed too hard at "the players aren't learning, some of them are just lucky"

  • random boi
    random boi 2 years ago +1

    i beat the first level and i was so happy but found out there was a second level

  • Ridwan
    Ridwan 3 years ago

    I actually like watching the horde of mindless squares sprawling across the level, so if you do more levels, please keep em

  • Sezen Tutum
    Sezen Tutum 2 years ago +2

    Do you ever watch while your AI tries to learn stuff and then scream like "IT'S ALIVE!!!!" bc it felt like it was alive

  • Komment King
    Komment King 7 months ago

    Excellent. I'm sorry to say I'm a has-been programmer. This is the type of machine learning coding I should have learned in the 80s. Back then I remember it was a big deal when somebody wrote a program where a small frog character would learn to not hit the edges of the monitor display. Well anyway this guy did a great job, and off the cuff - without having to really think about it.

  • exxo1598
    exxo1598 3 years ago

    What I would like to see is AI that learns the correlation between variables. This type of learning (natural selection) is great because you basically brute force, and select the best outcomes, but it only works in an environment where you control all the relevant factors (such as the speed of the dots, and where/when they start relative to the actions of the hero).
    If you had the dots continually moving despite the time of death I'm pretty positive this program wouldn't get very far, because the dots are only able to learn relative to a constant path, not a changing one.
    I'm not quite sure what the name is of that type of variable-based learning is but I think that would be totally helpful to learn.
    Although, I must say, I'm still very impressed, because I know for me to whip up this type of algorithmic program I would have to completely outline the process, and do a ton of research to understand what exactly is being implemented. Simple to understand but harder to implement.

  • kyle hickmen
    kyle hickmen Year ago +19

    Me after 5 muinets: oh yea i forgot hes a programmer lol

  • joseph jackson
    joseph jackson 3 years ago

    More levels? I want to see the whole bloody game!!

  • Alec Taylor
    Alec Taylor 3 years ago


  • bouncy house
    bouncy house 3 years ago +1

    Random learning is fun to watch, even after it gets past the final obstacle it just fucks around in the corner until it dies

  • Fair Maiden Games
    Fair Maiden Games 3 years ago +1135

    Hey Code Bullet, I'm the developer that made this game. Thanks a lot for the awesome video, really enjoyed it!! :)

    • 1233344
      1233344 3 years ago +1

      No I made this game :)

    • 2Lame4Name
      2Lame4Name 3 years ago

      Me and a buddy teamed up once, Many Many years ago 10+ in the first "hardest game in the world" took us 16 hour's, but tagteamed, we managed to get to the last level, to only give up because of exploding headeache, and not wanting to kill ourself yet.

    • Dimitri Gómez
      Dimitri Gómez 3 years ago +1

      The World's Hardest Game this is... official?

    • john Ali
      john Ali 3 years ago

      The World's Hardest Game dude you’re awesome

    • Ivan Rush B
      Ivan Rush B 3 years ago

      I made this :)

  • Rodrigo Rodriguez
    Rodrigo Rodriguez 3 years ago

    I used to play this a lot in middle school with my cousin😂😂 Nostalgia for dayss.. We were surprisingly good at it.

  • Aarón Pérez
    Aarón Pérez 2 years ago

    This was awesome i want more levels please

  • Johnathan Hulse
    Johnathan Hulse 2 years ago

    But hang on a sec, what if the one rectangle was like the size of the moon. Would it still be faster to draw this one or 700. You should program an AI to make some equation for this

  • GregTom2
    GregTom2 2 years ago

    Give it more input nodes so it knows its own position, the position of the dots, and the position of the walls relative to itself.

  • Asdrubale Girolamo
    Asdrubale Girolamo 3 years ago +170

    When you can't beat a level so you recreate the whole level and add AI to beat it for you. Nailed it.

  • Rathuel
    Rathuel 2 years ago

    I think it's fascinating watching ai learn

  • Diamond Epic
    Diamond Epic 2 years ago +12

    at this rate they are never going
    beat the level

    Can you spot the mistake?

  • pp poo poo e e oo oo

    It's funny to think that back then people thought that when we had technology we would make useful things for humanity but here we are teaching a computer to play a video game :)

  • Starlight
    Starlight 2 years ago

    I so happy that I am living in the era of AI 😍🤗

  • Bozocow
    Bozocow 3 years ago +1706

    The fact that he remakes the game in like 4 seconds tells us a lot about the original.

    • _voidPTR
      _voidPTR 9 months ago

      @Yeah KeeN if both tic tac toe players are even SLIGHTLY COMPETENT, it will result in a tie 100% of the time. It's a fundementally flawed game

    • Jenn Mason
      Jenn Mason 2 years ago

      @TTime685 ahright calm down

    • Awkward Pizza Dude
      Awkward Pizza Dude 2 years ago +3

      Copying something is a lot easier than being original.

    • Naud van Dalen
      Naud van Dalen 2 years ago +2

      I think it's mostly level design that costs time instead of programming or graphics because Code Bullet doesn't have to come up with the levels. The game is just simple movements with simple shapes, so that takes very little time.

    • Kriss Wolf
      Kriss Wolf 2 years ago +2

      I have no idea how to do any of this but i got a degree in CS halp

  • q0zM0s
    q0zM0s 2 years ago +3

    AI can solve 1x1x1
    AI can solve 17x17x17

    But most important:

    *Can't solve world's "hardest game"*

  • GrumpleslamAlabaster
    GrumpleslamAlabaster 3 years ago

    Could you have the AI try to better its time until it goes as fast as possible?

  • DrawesomeJulia
    DrawesomeJulia 7 months ago

    This needs a sequel for the next level!

  • Alex Blas
    Alex Blas Year ago

    Can you transfer AI learning from one player to the rest so that from one mistake/solution all players come to be aware of it

  • Albert Renshaw
    Albert Renshaw 3 years ago +3612

    It didn't really learn to play the game though, it just learned an exact path to beat that exact one level.

    • mateo turcios
      mateo turcios Year ago

      If the path is changed speed or the dots are moved it will move the same way,wich means this is a ai,ai's cant learn so it will go the same path but since its AI its a bot so it knows everything and try to avoid everything that isnt supposed to be done.

    • mateo turcios
      mateo turcios Year ago

      @Ichijo Festival yo no need to go hard

    • mateo turcios
      mateo turcios Year ago

      @Krzysztof Piasek It is learning but not really since it already knows the path,but the main thing obout it is to complete it,this is one level

    • mateo turcios
      mateo turcios Year ago

      It learned the path wich means it still learned the game bro

    • Oreo17
      Oreo17 Year ago

      @blackham7 I would argue that it is. It doesn't automatically know the rules or how to win so it just tries stuff and uses what works. Thats how you learn.

  • K1naku5ana3R1ka
    K1naku5ana3R1ka 2 years ago

    Now try continuing the evolution but gradually clamp down on the number of moves they are allowed, and see how efficient you can get them.

  • OH
    OH Year ago

    I’ve forgotten the general term, but it looks like this algorithm will never improve its generic skill at handling new levels, unlike human players. Is such an algorithm possible?

  • Peter J-S
    Peter J-S 9 months ago

    If you were to watch all the attempts per generation, once that first one won, would all future attempts on all future generations win too? I.E. would the "blob" of attempts all instantly become streamlined?

  • Unknown Account
    Unknown Account 3 years ago

    Even without feelings I still believe this AI raged

  • Kevin Zhang
    Kevin Zhang 3 years ago +38

    Problem is I don't think ai even knows what an enemy is and that they should avoid it. Using that same ai on a different level would be catastrophic lmao.

    • X Ray
      X Ray Year ago

      Although his method isn't as advanced as NEAT, the AI would eventually figure out the second level, and any level after that, unless the dots would be randomized every attempt

    • Microhacker
      Microhacker 3 years ago +1

      Tycho Wozniaki, NEAT is an acronym for Near Earth Asteroid Tracking or more based on AI NEAT can mean NeuroEvolution of Augmenting Topologies. NEAT is a genetic algorithm for the generation of evolving artificial neural networks. (From Wiki)

    • Tycho Wozniaki
      Tycho Wozniaki 3 years ago

      What is NEAT?

    • Bucket_head
      Bucket_head 3 years ago

      Kevin Zhang 2

    • Kevin Zhang
      Kevin Zhang 3 years ago


  • Crewsiunia
    Crewsiunia Year ago

    no joke I just tried the first lvl and beat it without breaking a sweat

  • RGB Animations
    RGB Animations 7 months ago

    I was actually able to complete the first couple levels by myself

  • Mai Phan
    Mai Phan 3 years ago +9

    I’m just going to say....

  • Max MusterSpace
    Max MusterSpace 3 years ago

    But! So they learned how to avoid just this pattern of blue dot movement. Wouldn't it be even cooler to randomly change the movement of the enemy and figure out how to beat that?

  • future deals
    future deals Year ago

    Oghh that brings back memories from when I was I kid. I completed all lvl a couple of times, I think my best time ever was about 40 deaths when I completed the last lvl 💪🧠

  • Malachi Clifton
    Malachi Clifton 3 years ago

    I would like to see where it learns how to play the game, not just survive that level

  • T Y E
    T Y E 3 years ago

    You should have an ai learn to play solitaire, that would be pretty cool

  • Ana Weber
    Ana Weber Year ago

    Imagine being smart enough to program this...cant relate and never will, good job bro

  • XsaivierWolfe
    XsaivierWolfe 3 years ago +568

    Do the players actually "See" anything, or are they just following the pattern that the best one got?
    This AI probably couldn't go on to the second level and attempt to go through it with what it's learned on this level, could it?

    • Joshua Storlie
      Joshua Storlie 3 years ago

      It couldn't apply what it's done, but what it could do is solve any other level that follows the same basic rules, and it could do it in a blink of an eye.

    • vitfirringur
      vitfirringur 3 years ago +5

      This is an evolutionary algorithm. Each player starts out with a predetermined set of moves, goes through the level and whoever gets the furthest gets used to breed the new generation.Then there is some randomness involved to create new players and so on. There are a bunch of different variants of evolutional algorithms like this. But in any case, it didn't really learn anything. This is essentially a probabilistic algorithm to generate a player that can pass the level. You wouldn't even need the level data as input to the algorithm.

      That doesn't mean this isn't cool, though. Evolutionary algorithms are really cool and can solve a lot of problems neatly.

    • KrebseBubrek
      KrebseBubrek 3 years ago

      Ai is gay matey...focken learn it man...or yar never gonna be succesfull in focken nit focken jooouuuken ya mate...nuee...fock off

    • A Gaming Channel
      A Gaming Channel 3 years ago +3

      Technically in this example, overfitting is desired as the solution is the overfitted goal.

    • Joel
      Joel 3 years ago +4

      You're not wrong, but you have to think in terms of small goals. Think of this as a new chemical pathway for a bacteria to be able to digest a new food source in its environment. As it gets closer and closer to the objective, correlate that to an enzyme getting more and more adapted to tapping that source of energy until it gives the bacteria a big leg up on all its competitors. It's thousands of small changes across a population that lead to macro changes. Changing the level is equivalent to radically changing the environment all at once. Of course the population would die out all at once. That's just how extinctions be. If you slowly changed the first level into the second, the program would adapt with it and the strong (and lucky) would survive.

  • KitsuneHawk
    KitsuneHawk 3 years ago +3

    Once it got to a certain point near the end, I kept yelling at my screen for it to go down instead of right for a moment lol.

  • Bảo Nguyễn
    Bảo Nguyễn Year ago

    After I watch this video i can finally beat that level

  • Pedro Vaz
    Pedro Vaz 3 years ago

    This was a lot of fun to watch!

  • noam monmarche
    noam monmarche 3 years ago

    Hey im not very smart and all but i was just asking myself a question.. wouldnt they be faster to learn if each square of every generation just got the pattern of the best of the previous generation?

  • cion
    cion 3 years ago +4124

    All you have to code is

    If “redRectangle” touch “blueCircle”
    Then {don’t}

    • X Ray
      X Ray Year ago

      All these people who don't get the joke

    • KirbyCreep
      KirbyCreep 2 years ago

      Fake, that doesn't work.

      if (bluecircle = coming for you)

    • hue
      hue 2 years ago


    • Chappie
      Chappie 2 years ago

      @as ter it wouldn't do anything cause that's not how code works.

    • SeireiART
      SeireiART 2 years ago

      Harvard : *Hey bro, Want a scholarship ?*

  • Ya Boiz
    Ya Boiz 2 years ago

    I can’t even begin to try to figure out how this works 😂

  • blackham7
    blackham7 3 years ago

    This is only good if the game rules stay fixed when it changes like in the subsequent levels where it's more than just dodging squares moving back and forth it'll struggle again hence why I'm presuming you only showed the first one. You'd have to rewrite its code again and again this AI is only good if the constants don't change but real life isn't like that. I don't think it understands what's happening.

    It's more like when the teacher in class makes you blindly copy what's written on the board the answer to an exam question but you don't understand the logic behind it. In the exam if you got the exact same question you'd know which boxes to tick but if the question was worded differently and had it's values changed you'd be lost.

  • Zac McDaniel
    Zac McDaniel 2 years ago

    I played this game and made it past the first level and couldn’t get any farther than that

  • Dum Neptune Bacon

    I played the game once and got to the second level with 500+ deaths, and got so mad I quit...

  • Sergio Franco
    Sergio Franco 3 years ago +2351


    • Lovinggd750 GD
      Lovinggd750 GD Year ago


    • Bojun Choi
      Bojun Choi 2 years ago

      Kevinofrepublic its objective is to get to the end and it just needs to have alot of trials so you dont need to edit the code

    • Owlblocks David
      Owlblocks David 3 years ago

      Actually, if the AI's had a rendered map as their input, they could learn to dodge...

    • Eugene V
      Eugene V 3 years ago

      Wrong answer. If all levels are designed to move one red square from left to right, then the algorithm should be good for all levels. (And I know it does not)

    • Roninecostar
      Roninecostar 3 years ago

      Kevinfrepublic. In this instance, this algorithm will do exactly as you say, but making a different one with deep neural network will most probably work on all levels.

  • Daniel Roberts
    Daniel Roberts 3 years ago

    Congrats, it took the AI an edited cut to 7 minutes to beat a level that took me like 3 tries. Humans>AI

  • okay
    okay 8 months ago

    The part at 2:40 where there were a 100 AIs to simulate natural selection made me feel weird. Looked like a bunch of ants crawling around on my screen.

  • Aubrey Ahrens
    Aubrey Ahrens 2 years ago

    So the players had no information about other players movements? I noticed when players outside of the starting area moved up the x axis players still inside the starting area also moved up the x axis. Just a timing thing? "Moving up at this point in time gave positive results in previous generations, so this is the point in time to move up."


    4:31 The Swarm of Red Dots with a green core looks like a boss

  • Slothery
    Slothery 3 years ago +22

    This was awesome and I want to see more.
    Also, I kind of like seeing the chaotic mess of AI's scrambling to evolve. Maybe keep the top 3, 5 or 10 to show how they differ?

    • Gamer Man
      Gamer Man 3 years ago

      Meow? Meow. It’s fine, sorry xD

    • CelestialTree
      CelestialTree 3 years ago

      +UltraGaming, just clarifying so nobody misunderstands, no need to get annoyed due to that.

    • Gamer Man
      Gamer Man 3 years ago

      Meow? Meow. Yeah I know we not stupid

    • CelestialTree
      CelestialTree 3 years ago

      Maybe the best, worst, and most average ones?
      EDIT: By that I mean the top one, the bottom one, and the one right in the middle.

  • llucllumador
    llucllumador 2 years ago

    What would happen if you made a level with 2 paths, one path that goes right next to the goal but doesn't reach it and another longer path around the goal that actually reaches it. I think your AI would never complete it because it would choose the shortest path that got closer to the goal and never try the longer correct path.

  • Mxri
    Mxri 2 years ago +159

    "most of the players have figured out how to get out of the STATING area"

    • Budling
      Budling 2 years ago +31

      Spelling errors: CHEK

  • ArtemisTX
    ArtemisTX 3 years ago +11

    3:28 ok so at first I only got a glimpse of the words, so I thought incremental learning said internal screaming

  • Ben Donahue
    Ben Donahue 2 months ago

    Now I wanna see it beat the hardest level

  • Cameron Todd
    Cameron Todd 3 years ago +52

    Me: There’s a second level though.

    Code Bullet: What’s a level?

    • Cameron Todd
      Cameron Todd 3 years ago +1

      Bucket_head I can’t tell if this reply is a joke. :/

    • Bucket_head
      Bucket_head 3 years ago

      Watch til’ end

  • D Sanchack
    D Sanchack Year ago

    Now if only LV 1 out of 30 wasn't so easy that most people get it on like the third try. Still, AI in itself is awesome, so still a cool vid. I gave a like.

  • Coding Hub
    Coding Hub 2 years ago

    Hey code bullet from your previous videos I know that you use JavaScript and can you help me in my project... I want to add collision between two rectangles for my own game development.. please guys tell me if you have any ideas.. any help would be appreciated

  • David Dahl
    David Dahl 2 years ago +4

    Took a computer that long to do something I could do in under a minute. I guess we don't have to worry about skynet yet

  • Dan Kelly
    Dan Kelly 2 years ago

    You should have the AI start at a random location in it's 'home box' otherwise it isn't really learning to play the game.

  • Shamwow -that guy
    Shamwow -that guy 3 years ago +49

    Try to get AI through all 30 levels, it could be a mini series or something

    • Christopher Savignon
      Christopher Savignon 3 years ago

      @Jordan Fischer
      The difference is that the "Ai" starts every level knowing nothing. Any path it worked out in previous levels is useless in the following ones. A human can understand, generalize, and extrapolate the gained knowledge. Just like a human - not really.

      See, this is not an actual "AI" that learns to react to input from the game's world, it's just a heuristic search function.

      It learns nothing. It only memorizes a path.

    • Kaleb Bruwer
      Kaleb Bruwer 3 years ago

      There are 30 of these?!

    • Maxilander
      Maxilander 3 years ago

      Yea like Jordan said, there's a video where Jabtils (an Ai youtuber) creates "Forrest gump" to go through every maze he creates

    • Jordan Fischer
      Jordan Fischer 3 years ago

      Shamwow The bullet wolf it is possible to get one instance of an AI to beat all levels, just like one human brain can beat all levels

  • Mana Potion
    Mana Potion 3 years ago

    What if the aliens come to earth and they see your amazing AI crafting and wish for you to make them an arcade...or the CIA knocks on your door when the world is controlled by AI and they ask you to analyse their skeptical behaviour... Yeah bet this youtube channel won't be fun and games anymore 😂

  • SalocinRevenge
    SalocinRevenge 3 years ago +1

    Hi, Im brasilian and I found your Channel today, and I loved it. But there is a problem, I want to show these videos to my friends, who does not speak very well, so, If you just add some subtitles, I could show them your videos! It's a good ideia isn't It? More View, more subscribers (more money) and even more work :/ . Thanks for these videos ;)

  • Frank Delgrosso
    Frank Delgrosso 2 years ago

    This seems more like pattern recognition than learning. I would like to see more where the Dot knows what it is adverse to and what is favorable. It's impressive for sure but not adaptive to any new environment

    • Pablo A. Diaz
      Pablo A. Diaz 2 years ago

      Do it yourself, smartass

    • TrueMetis
      TrueMetis 2 years ago +1

      Oh, for sure that are better methods to doing this. This is just the method the maker chose.

    • Frank Delgrosso
      Frank Delgrosso 2 years ago

      @TrueMetis Also I think best on selection is a terrible method. It should be best 10 or 20%. Best one selection allows for a poor but sucesfull early choise to permanently prevent the batch from ever learing better ways. We humans are subject to this problem too, If you were to say take the top 20% and make 1/5 the total spawn of each the divergance will lead to several alternative solutions. Those could in turn pit their top 20% in the next batch. This is evident from other runs where we see AI pick up a pattern it can't seem to shake, it is the lack of divergance from generations long ago because all the current gen are 30 decendants later and divergance from the bad habbit is near impossible.

    • Frank Delgrosso
      Frank Delgrosso 2 years ago

      @TrueMetis It is entirely possible to program AI learning with values, likes and dislikes essentially. Which much more models real learning than pattern recognition and route memory. If you assign values to conditions and have the AI seek positive values and avoid negative ones it will behave in time just as living things do. Obviously not by true learning but thousands of trial and error. But this form of learning is adaptive to changing environments. The walls and the balls moving don't have to follow the same pattern for it to adapt. It will "learn" not a pattern route but rather to avoid the ball specifically and the walls and that trying to move forward is good and the goal.

    • TrueMetis
      TrueMetis 2 years ago +1

      Eh sure, but pattern recognition is a good start. People make the mistake of thinking AI means human level intelligence. Really it just means being able to perceive its environment and act to achieve it's goal. And this program did that. People just aren't impressed because we've had 3.5+ billion years of going through the same sort of selective processes to hone our minds into what they are today. This thing isn't a human intelligence trying this out, it's the simplest of single celled organisms.