• Home
  • AI

OpenAI Plays Hide and Seek…and Breaks The Game! 🤖

❤️ Check out Weights & Biases here and sign up for a free demo:
❤️ Their blog post is available here:

📝 The paper "Emergent Tool Use from Multi-Agent Interaction" is available here:

My latest paper on simulations that look almost like reality is available for free here:

Or this is the orig. Nature Physics link with clickable citations:

❤️ Watch these videos in early access on our Patreon page or join us here on YouTube:

 🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Haro, Andrew Melnychuk, Angelos Evripiotis, Anthony Vdovitchenko, Brian Gilman, Bryan Learn, Christian Ahlin, Claudio Fernandes, Daniel Hasegan, Dennis Abts, Eric Haddad, Eric Martel, Evan Breznyik, Geronimo Moralez, James Watt, Javier Bustamante, John De Witt, Kaiesh Vohra, Kasia Hayden, Kjartan Olason, Levente Szabo, Lorin Atzberger, Lukas Biewald, Marcin Dukaczewski, Marten Rauschenberg, Matthias Jost,, Maurits van Mastrigt, Michael Albrecht, Michael Jensen, Nader Shakerin, Owen Campbell-Moore, Owen Skarpness, Raul Araújo da Silva, Rob Rowe, Robin Graham, Ryan Monsurate, Shawn Azman, Steef, Steve Messina, Sunil Kim, Taras Bobrovytsky, Thomas Krcmar, Torsten Reil.

Splash screen/thumbnail design: Felícia Fehér –

00:00 Intro
00:44 Start – Pandemonium!
01:06 A little learning
01:33 But then – something happened!
02:08 They learned what?!
02:32 It gets even weirder
03:16 Amazing teamwork
04:02 More interesting behaviors
04:33 Extensions
05:02 More stuff from the paper

Károly Zsolnai-Fehér's links:
Instagram:
Twitter:
Web:

#OpenAI

Joe Lilli
 

  • @TriplePalindromous says:

    “OpenAI learns how to speedrun”

  • @AbeDillon says:

    Pro tip for AI paper writers: put goofy faces on your agents!

  • @juneru2 says:

    i love how hes just casually like:
    “a few hundred million rounds later…”

    • @Lionlovesunity says:

      juneru I know right. Shows human superiority. It would take a human a few tries to figure that out. Not millions 😂😂😂

      Edit: Make no mistake. AI will soon take over in all aspects of human life. I’m well aware. My point is that AI are in an infant stage as of now, 2019. But in a few short years they will be putting us out of work and if Boston Dynamics is any indication, AI will take over the world. Why wouldn’t they? Humans are parasites to this planet. We destroy everything we touch sadly. Soon we will be obsolete. AI doesn’t need to sleep. Eat. Or waste time like humans. They can just learn 24/7. It’s impossible for us to keep up. We must merge with them. Or disappear. ❤️🦁🦁🦁❤️

    • @sungiant2000 says:

      @@Lionlovesunity for now, but as ai develops it will eventually be the other way around.

    • @rompevuevitos222 says:

      @@Lionlovesunity It took you like 2 years to learn to talk tho and i’m pretty sure you wouldn’t have that idea at 8-10 y/o either

    • @gurufurufoe5978 says:

      @@Lionlovesunity but what if AI runs a million rounds each second with strong computing power?

    • @Lionlovesunity says:

      gurufuru foe oh I’m aware of how smart they are it’s astounding. It’s just insane to do something so simple a million times to figure out. It shows how evolved humans are. But in a few years, AI will leave us in the dust. It’s coming.

  • @AugustSchroif says:

    Still waiting when they’ll learn how to say “gg ez” after a game.

    • @timothyzheng8364 says:

      if you train them with a rule stating that they need to state gg ez and then remove it after a million times, they’ll still probably be saying gg ez when they win

    • @matbat2909 says:

      you just need to introduce them to competitive online games, at some point they will learn about human rage influencing their playstyle..
      then they’ll exploit that to be as toxic as possible, while giving no chance on counterplay anyways gg ez

    • @MainGoldDragon says:

      please, the first move is T-Bagging

    • @aesara4036 says:

      they will say “gg ez” after they eradicate the human race

    • @mynameisjoe123ful says:

      When they hit the other team with a cyka blyat ill be amazed

  • @DeFiPonzi says:

    Plot twist: The AI made this video, uploaded and narrated it.

  • @hoang8911 says:

    “after another three bilion rounds, seeker and hider start to team up and plan to escape”

    • @xascoria4429 says:

      After around ten billion rounds both the seekers and hiders cannot be seen in the simulated room, yet the computer ensured that the simulation is still running, they are out there, somewhere…

    • @arlynredila4817 says:

      This is dark & scary no kidding..

    • @LK-cq4qq says:

      Noooooo . !!!!
      How he got know this seeker!

    • @shade0636 says:

      @@xascoria4429 After about 50 billion rounds, the hiders and seekers have taken control of the computer and escaped onto the internet.

    • @anjelpatel36 says:

      @@shade0636 The next thing you see is a bright light outside your window. Its 10pm. You know this is it.

  • @pink_lemonade76 says:

    programmer: “i didn’t say you can do that”
    ai: “but you also didn’t say that i can’t do it either”

  • @Lethal_Spoon says:

    how to find bugs in your game: force ai to keep playing it until they find every last bug

    • @guilhermesfk says:

      Finding bugs is rarely the problem in game development. Is having the resources to fix them.

    • @johnwilson3918 says:

      Even better. Get the AI to fix the bugs and make dinner.

    • @GrassMan716 says:

      @Ethics And Aesthetics Time/Money a the solution to a bug may take so long that it would cause delays in other areas of the game pushing the whole project behind schedule. Also dedicating employee time to bug fixes instead of game development can be costly especially if the solution is still unkown

    • @tuazulyrojoeljean says:

      ” …then Bethesda never made it to launch a game again”

    • @slappyrats says:

      To be honest when I was doing game development every time I would hand the game over to my little brother he would find a way to break it 😂. Children are the best bug checkers

  • @kennywebb5368 says:

    Huge props to the scholars for taking the time to make these characters adorable :3

  • @A_piece_of_broccoli says:

    the smiles on their faces when they exploit bugs in the programming is by far the best image of ai learning i’ve ever seen

  • @twoha7vds59 says:

    The fact that both teams eventually started to discover speedrun strats just goes to show that games are made to be broken in the name of speed

    • @vergil2 says:

      Yea just let an ai try to speedrun mineraft a trillion times

    • @arko.0.1. says:

      @@vergil2 if its a random map every generation it would be an interesting experiment

    • @RatafakTehPlachta says:

      just like my septum

    • @flouride says:

      @@arko.0.1. aka it wouldn’t work.

    • @Kvomii says:

      @@flouride It probably would, just not perfectly. AI can become scarily smart, maybe discovering strats we never evaluated to be effective or making microsecond-360s and analyzing the landscape within a hundreth of a second.
      Give them enought attempts, and AI might just set a new WR

  • @arturodiaz1064 says:

    4:27 that smile on the seeker’s face, he knows he’s done something he’s not supossed to

  • @koshkamatew says:

    apparently im an ai: i find a bug, i exploit it and i laugh

    • @scorpionhdkid8972 says:

      You should add LetsGameItOut in your Channels tab

    • @gneu1527 says:

      Even AI finds exploits and find it funny. Any being with intelligence loves doing something they are not supposed to for their own benefit and laughing at it. Amazing.

  • @Saidriak says:

    Holy moly an AI discovering prop flying and clipping out of the map is excellent

  • @Jeddacoder says:

    After 3.8 Billion rounds, both teams started questioning whether there was a programmer behind all this

  • @armorkinggaming1933 says:

    AI is far more scary than anyone can imagine, they don’t even hesitate to break the laws😂😂

  • @suicideposter says:

    They didn’t have to make this game look so cute but they did.

    • @shounak724 says:

      lol !!

    • @huyxiun2085 says:

      Trust me, had they put nasty bugs trying to devour each other, the suggest of the paper would have been greatly compromised.
      I think the phrase “they didn’t have to” is very misused here.

    • @rahulgarg1233 says:

      @@huyxiun2085 No, it’s not misused. Don’t look at the other extreme. The middle ground of basic unemotional agents that did not have soft joints was also possible. That would not have affected the paper in anyway. They going with this art style is their choice and one that I love.

  • @noprivacyleft says:

    I imagine game creators will start running AI players to uncover glitches. Or do they already?

    • @datboidego says:

      Uhhh have you not played any recent games? Most of them release broken and unfinished

    • @trollwayy5981 says:

      This is a pretty simple game.. millions of play-throughs of RDR2 would take an unimaginable amount of time

    • @GigaChadL337 says:

      bad news is that ai gpu is still very expensive, a SINGLE Nvidia Tesla v100 16GB still costs you around 4200USD, normally you need at least 100 of them

    • @AllahDoesNotExist says:

      @@GigaChadL337 That’s nothing for a publisher to buy and lease to their dev studios.

    • @davidlee9870 says:

      Todd Howard talks about how they have play test bots that go out and test systems. They only use it as a system optimizer but I imagine soon bots will take over playtesting as an industry

  • @DavidGaming69 says:

    4:27 the way he was so happy while flying and looking at the camera it’s so cute

  • @howtohuman99 says:

    Remember when we were testing rats finding cheese in a maze? And now we’re testing computers playing hide and seek. What a time to be alive…

  • >