Multi-Agent Hide and Seek

We’ve observed agents discovering progressively more complex tool use while playing a simple game of hide-and-seek. Through training in our new simulated hide-and-seek environment, agents build a series of six distinct strategies and counterstrategies, some of which we did not know our environment supported. The self-supervised emergent complexity in this simple environment further suggests that multi-agent co-adaptation may one day produce extremely complex and intelligent behavior.
Learn more:




  • But can they cure cancer?

    deep minddeep mind3 saatler önce
  • I love the cute little faces of joy every time the seeker finds the hider.

    Jaebez BleahJaebez Bleah4 saatler önce
  • Wow, would like to play this game in multiplayer

    M4rkozM4rkoz6 saatler önce
  • How do we speed this up?

    Jason LimaJason Lima8 saatler önce
  • Simply fascinating...

    João PachecoJoão Pacheco9 saatler önce
  • Box surfing seems a bit glitchy.

    Nicko G.Nicko G.9 saatler önce
  • In a couple of years, AIs will hack their reward code to give themselves infinite reward

    Nathan HuismanNathan Huisman10 saatler önce
  • Let's play this on a huge server! With AI agents!!!

    김재욱김재욱11 saatler önce
  • What if humans were playing with ai as well

    Miguel Carlo GallanoMiguel Carlo Gallano12 saatler önce
  • Ai is dumb as fk... why don't the hiders trap the seekers in instead?

    TheEpicSandwich gocTheEpicSandwich goc21 saatler önce
  • Dead by daylight really downgraded

    Caleb CruzCaleb Cruz23 saatler önce
  • Dang I wish normal people could run this simulation

    Im On Da WebIm On Da WebGün önce
  • Fake news

    AgitatedAgitatedGün önce
  • Yo this game looks sick is it on steam?

    MartyMacaroniMartyMacaroniGün önce
  • Ai really be learning how to glitch

    Don’t touch my phone GamerDon’t touch my phone GamerGün önce
  • I’m very disappointed too say I’m here from TikTok 😔

    Jordan BigbyJordan BigbyGün önce
  • That's freaking cool

    zbobz12zbobz12Gün önce
  • This is not in-depth enouuugh

    dxxPacmanxxbdxxPacmanxxbGün önce
  • -until the hider are smart enough to lock the seeker inside the cage

    xX_Kjcomputer_XxxX_Kjcomputer_XxGün önce
  • *This will be implemented in future robots and then they will learn that we are destructive to yourselves. And then decide that they are the ones best suited to protect us from us. And thus we begin our journey into robotic slavery.*

    The Other SideThe Other SideGün önce
  • He attacc He protecc but most importantly He surfs in buccs

    Pseudo XPseudo XGün önce
  • So basically it’s slavery with extra steps

    Lucas RLucas RGün önce
  • Okay seriously wtf, all TRvision comments are just quoting the videos now. This is seriously weird.

    DolankDolank2 gün önce
    • What?

      mattox huttomattox hutto4 saatler önce
    • IKR

      Lucas RLucas RGün önce
  • Better . Far far better

  • Tech them to speak

    i love love songi love love song2 gün önce
  • Which 3D simulation program did they use? Pretty cool stuff though!

    Shivam DhootShivam Dhoot2 gün önce
  • You should make this a Videogame somehow

    SpaceDave1337SpaceDave13372 gün önce
  • Me: Just surround the seekers with walls AI: *Circuits Blown*

    Mr. MindReaderMr. MindReader2 gün önce
    • cool stratagy

      김재욱김재욱11 saatler önce
  • Im surprised they didnt lock in the seekers

    gangster gandalfgangster gandalf2 gün önce
  • Everyone else: AI is learning to hunt us down. Me: AI learned speed run exploits.

    fl00fydragonfl00fydragon2 gün önce
  • terminator age is coming. And it's looking so cute.

    The PotatoThe Potato3 gün önce
  • Remember when humans use to play hide and seek?

    HackTorHackTor3 gün önce
  • I wonder if AI will learn how to ABH...

    HarryHarry3 gün önce
  • How do I learn to do this?

    vijay vittalvijay vittal3 gün önce
  • these little creatures, reminds me of little big planet Sackboy :,D

    DuoBV ChannelDuoBV Channel3 gün önce
  • Hiders can box the seekers ,problem solved for seekers that use other object to jump over and totally in lockdown

    mb kmb k4 gün önce
  • didn't expect people to be meme-ing down here not complaining tho •ᴗ•

    Ee Cheng LEEEe Cheng LEE4 gün önce
  • now, this is a open world game i would like to play

    LoopLoop5 gün önce
    • @John DC ofc they can, whole AI system is actually based on reward and penalty system

      LoopLoop4 gün önce
    • @Loop even better if the NPCs can somehow learn to give players apporopriate quests and rewards based on what they want. Everything would basically be procedural and you would actually be shaping your own world alongside the NPCs.

      John DCJohn DC4 gün önce
    • ​ John DC Exactly, and as a developer, instead of building boring and liner quests, you would only implement game dynamics and let NPC's decide for them selves what they want to do.

      LoopLoop4 gün önce
    • Dude imagine if you just had an open world game that also included learning NPCs that have neural nets. You'd have a whole world that changes artificially from the players and naturally from other AIs. Probably gonna be a PC killer though lol

      John DCJohn DC4 gün önce

    Igor GabrielanIgor Gabrielan6 gün önce
  • And this my gamers is the *recommended page*

    HarryHarry6 gün önce
  • Uncomfortable

    Late night talk show with the BronsonLate night talk show with the Bronson6 gün önce
  • Is that a game!?!?!?!?!

    mr. grootexmr. grootex6 gün önce
  • Very nice, i wold like to see more strategy games...

    João Ramon Gomes Da SilvaJoão Ramon Gomes Da Silva6 gün önce
  • competition? it was cooperation to survive that led us to where we are u dingus.

    4ammofo4ammofo7 gün önce
  • That's intelligent, yet scary. applying such algorithms on machines. you know the rest.

    Bratteries and SnignalsBratteries and Snignals7 gün önce
  • Elon’s brain nightmares are coming back to haunt him.

    ChuckNorris100000ChuckNorris1000007 gün önce
  • スゲェ…

    kiko synthkiko synth8 gün önce
  • That's insane...u can drop this last AI generation in Mars & let them build simple buildings & wiring throw the walls...insane

    Next Gen GamesNext Gen Games8 gün önce
  • SkyNet liked this video

    Ganymede, Jupiter IIIGanymede, Jupiter III9 gün önce
  • beautiful

    Kick LeeKick Lee9 gün önce
  • Instead of hiding from the red ones they should locked the red ones by the blocks .

  • If you know the rule of the game, it's not hard to figure out the hiders ultimate strategy: lock all blocks and wall themselves. On the contrary, these RL agents learn these simple strategies by playing millions or perhaps billions of games. This is NOT how humans or other animals perform problem-solving. We do not solve puzzles by attempting them several million times. We simply cannot afford to do so. Instead, we solve problems by abstracting them and reason about them. That is called intelligence. RL is NOT the golden path to intelligence, it is a path to problem-solving with NO intelligence, contrary of what the vision of general artificial intelligence is aiming for.

    Xuezhou ZhangXuezhou Zhang10 gün önce
  • Imagine someday OpenAI will work with Boston to make Sky net.

    Jay SukumalchanJay Sukumalchan11 gün önce
  • bomba kimi

    Azeri LyricsAzeri Lyrics11 gün önce
  • The fact that it learned all of that by itself is insane and a huge step towards self aware ai.

    YoseiHitoYoseiHito11 gün önce
  • alternate title: making AI use Half-Life 2 speedrun strategies

    Gustav Isak AbrahamssonGustav Isak Abrahamsson11 gün önce

  • Expecting spontaneous combustion with the agents as saying auto-intelligence will emerge with more simulation. The maximum of what they can is bound by the physic rules of the environment perceived by these agents. Their call is confined to one layer of the environment that makes them interact the way they do.

    WulfCryWulfCry12 gün önce
  • This is witchcraft! WOW!

    Jack NapierJack Napier12 gün önce
  • Idk how this cane up on recommended but it's actually pretty cool

    Football addictsFootball addicts12 gün önce
  • Hiders atlast ran out of tht stage....?? Is tht so

    Bhuvanesh s.kBhuvanesh s.k12 gün önce
  • PPL 50 years ago:- science can never explain feelings and thoughts like love, logic etc etc.... Currently... Reinforcement Learning an mathematical model...!!! Can mimic tht process imagine the power we are literally speeding up the evolution of millions of years to few weeks with these simulators and fast TPUs or GPUs... This is crazyyy

    Bhuvanesh s.kBhuvanesh s.k12 gün önce
  • Welcome to the Aperture Science computer-aided enrichment center.

    Abe AlexanderAbe Alexander12 gün önce
  • Seeing them yoink the ramp from the seekers is so funny for some reason lol

    Leeroy JenkinsLeeroy Jenkins12 gün önce
  • oh yeah, this is big brain thime

    David BaumannDavid Baumann12 gün önce
  • Can someone make this a game

    Bloodcrow 100Bloodcrow 10012 gün önce
  • 1:52 They're starting to think like Gurdan Freemon

    OcraelOcrael12 gün önce
  • "One day, truly complex and intelligent agents will emerge." I hope not. Skynet will not be a picnic.

    Ephraim CullenEphraim Cullen12 gün önce
  • Im surprised they didn't trap them

    Anson ChanAnson Chan12 gün önce
  • I don't think we'll reach 'truly intelligent' .. I can't foresee designing an environment that mimics "real life"

    Jamil MadanatJamil Madanat12 gün önce
    • @Jamil Madanat I see what you're saying but I've heard many times that the data required for self awareness is achievable, it's just way too much information for today's technology, the ai you see right now is aware of its environments that's why it's capable of reacting to it without programming so at some point in life, it's gonna be capable of comprehending life, I don't think it's impossible.

      YoseiHitoYoseiHito11 gün önce
    • @YoseiHito self awarness is precisely what i find impossible to achieve.. We dont understand consciousness nor where it comes from. How can we assume that self-learning will be followed by self-awarness?

      Jamil MadanatJamil Madanat11 gün önce
    • If the ai "self learn" techniques keep evolving, it can get to the point where they become self aware of themselves, humans, emotions etc and that probably would make them able to mimic humans and other beings.

      YoseiHitoYoseiHito11 gün önce
Multi-Agent Hide and Seek