Elon Musk’s ‘Dota 2’ Experiment is Disrupting Esports in a Big Way – No Playing Field

Elon Musk’s ‘Dota 2’ Experiment is Disrupting Esports in a Big Way – No Playing Field

They asked this question and they had this knowing smile. They said, do you think you’re going to win? (Music) I assumed that it had to be good. Otherwise, they wouldn’t seem so confident. I was pretty nervous but, I mean, I was excited just because of the implication, right? – On Monday, the bot was beatable by some players. By Wednesday, we played the best player in the world. We won with him, 10-0, 12-0, maybe. – They told me it wasn’t a big deal, because it played thousands of games against itself, so I shouldn’t feel too bad. But I was embarrassed. This thing accomplished in six months what I have set out to do for eight years. (Music) When I first played against AI bot, they didn’t really tell me anything going into it. They said, “you’re going to play a 1v1.” I was pretty nervous but, I mean, I was excited just because of the implication, right? If there’s a bot that can play at a super high level, then it’s great. I mean, it just pushes ‘Dota’ to a new level in general. Eight very high level players played it and all eight of us got stomped in game two. People were upset, they didn’t want to give it the credit that it deserved. Anything that can be programmed in six months to beat eight top players one after another… That thing was… It was unlike anything. (Tournament host)
– You’ve heard about it. Now you’re going to be playing against it. Can you comprehend the idea of a bot being better than you? – We wasn’t sure exactly whether we’ll be able to beat Dendi on stage. (Dendi)
– Okay, this guy is scary. (Host)
– Does it feel like a player, like a person? (Dendi)
– Nope. (Host)
– He’s going in. (Dendi)
– I’m dead. (Host)
– GG! (Applause) – We’d like to find agents that are able to adapt and learn in very new environments and accomplish increasingly more complicated tasks. And it turns out that games are really great for this. ‘Dota 2’ is one of the most difficult games in the world that are available right now. – Nothing that I’ve ever done has quite prepared me to play a game like ‘Dota.’ (Commentator)
– They’re going in once more. Blink, looking for the Crush, the Frostbite, perfect timing from Faith, creating space. – There’s just so much about it, so much you have to learn, so much you have to grind. It takes I’d say a year to just learn the basics, understanding what all the heroes and spells do. It probably takes about, I’d say a minimum of 12,000-ish hours to be a pro. – Even the 1v1 version of ‘Dota 2’ is super complicated The game is real time, which means that in every second, there are 30 decisions to be made. If we could, we would probably hard code all the strategies, but it turns out that is not feasible. We use a concept called self play. So what we do is we start by having an artificial neural network with initially random parameters, and we let it play the game. The only influence that we have on it is we say last-hitting of creeps is good, dying is bad, getting damaged is bad. What happens is that it will walk randomly, trying out lots of things. We can run the games usually much faster than real time, and we can run many copies of them in parallel. Each of the changes of the agent makes it slightly more efficient, slightly better. After about two weeks, they reach the level of professional gamers and then they become much better than any human players. – It did things that we had never seen anybody else do and it had set a type of play style that we pretty much just copy now. (Game audio) When I see the bot make a play, it clicks in my head. I’m like, ‘why aren’t we doing that?’ – It’s actually really fascinating to see the strategies that they learn over time. For example, we noticed that the bot behaves in an extremely silly way. In their first minute, it started taking a lot of damage, and we didn’t know why it was happening. It doesn’t make a lot of sense. The bot was using a baiting strategy. If you see that your enemy is on low health, you go very over aggressive and you think that you’re going to win. And then the bot was defeating its opponents within the next minute or so. When we let it train for a few more hours, it learned to counter that strategy in particular and at that point, no one was really able to beat the bot consistently. – No matter what happened— like, sometimes I would outplay it… It didn’t change anything, it just kept playing the same way. – You cannot tilt the bot in the way that you would humans. It calculates what is the most optimal for the human player to do in the next couple of seconds. So it’s not going to be distracted by random movements of human players or faking attacks. These types of strategies do not work well against the bot. (Game audio)
– First blood! – I think it’s going to be interesting to see how this is going forward in the future. We’ve got many requests from professional players getting access to the bot. It’s possible that in order to be the best in the game, maybe you’ll need to use similar bots for your regular practice. – Because it does certain things perfectly and you have to play perfectly to match those things. In that regard, you can train yourself over and over again to just get these pitch perfect games at a more consistent level. So in terms of high level play, it should make things a lot more mechanically skilled. (Music) – Players do not want to reveal their strategies to their opponents, so when they want test out more crazy ideas, they don’t have too many people to play against. Once we have very good 5v5 bots, you could imagine trying out hundreds of variations of different scenarios and trying to improve your longer term strategies. – I don’t think that it’s ever possible for a group of five bots to ever beat a group of five human players that are any good, just because I don’t think that there’s enough processing power in the world for it to handle that. The interactions between 116 or something heroes, that go in so many different ways that work together in different spots. There’s just too many variables that I don’t think that it can completely gather. – In the 90s, people thought that it’s impossible to write bots that can play chess. And we see the similar theme going forward over the last 10 or 20 years and we are able to solve increasingly more complicated environments and tasks. I think it’s definitely possible. It’s not going to be easy, but if our research is successful, we’ll be more efficient and we’ll be able to tackle more complicated problems in the future and solve more complicated games. It’s always impossible to tell which of the advances turn out to be useful. We’re definitely going to see a lot more applications of these methods in the future, but it’s too early to tell now. (Game audio)
– Outplayed!

100 thoughts on “Elon Musk’s ‘Dota 2’ Experiment is Disrupting Esports in a Big Way – No Playing Field

  1. Gamer says he thinks there is not enough computing power to beat a team of humans in 5v5…oh boy how you will be dissapointed my friend

  2. Honestly I'm more interested in bots that can play like human beings. Bots that remember who you are, who learn your patterns. Who can actually build relationships in lane with you through a match. Who don't just want to win, but don't want to lose to you. So that we as people can become better persons. Rather than we as people become better bots.

  3. I'll be brutally honest, I'd much rather watch teams of programmers compete by pitching their own uniquely programmed AIs against each other within the framework of a tournament, than platoons of nerdy teenagers exuding a collective miasma of social awkwardness potent enough to match the stench of the french people.

  4. They failing is understandable.

    They have been so used to playing against humans. AI is TOTALLY different.

    I bet the DotA pros can defeat this AI given they are allocated time to learn this AI. AI's are to some extent predictable, even more than humans

  5. Coming from league some of these kills look so stupid. Like the first one with the bot vs the player, he just walked up and attacked him and the player died. Its like, wtf, the lowest players in league know how to counter that

  6. You tell it that taking hits is bad. Suddenly it intentionally takes hits to bait the opponent in….. oh god it's become self aware!

  7. Alright I’ve been following this since the first video on YouTube was up. My question is: is OpenAI playing with human characteristics?

  8. Nope nope nope nope nope nope nope nope nope nope. I'm legit terrified right now. Kept seeing Blitz as every military intelligence officer in the world on the day we all die. 2 weeks to surpass 6 years? 8 years was it? Can we build it? Yes. Should we? Hell no.

  9. That pro 'I don't think it'll ever be possible for 5 AI to beat 5 top pros, theres too many variables…' — Yeah right, they'll be beating you within a handful of years buddy.

  10. One day we'll have countries vs countries with AI controlled robots. One "genius" programmer would insert program that devise the learning logic on how to win a war as fast and efficient as possible into the system, of which the AI decide the most logical things is to eradicate the ones who start wars before wars happens.

  11. This video makes it seem like the bot was somehow more than a gimmick. The bot match had severe restrictions like having to face the same hero in one lane the entire time… the same hero can’t even be chosen in real matches and nobody stays in one lane

  12. 1. Is it programmed within the game itself? 2. some program that reads the display? 3. an actual robot that reads the display?

    I'm way less impressed if its number 1. Be more like cheating at that point.

  13. I always suspected these games tournaments are just bot gameplay with payed actors hamming up in a auditorium of people who know they are there for a job of sorts. Its like the game ai manifested a bio synth form to beat you IRL but they are also your society, the game is consumerism and social media distractionism, you are already trapped if you think its winnable.

  14. Why we dont have an AI-teams competition? it's an amazing way to force the development of the tecnology.

    And of couse, like any tecnology based sport (like F1), it's a realy good show, imagine many diferent aproachs to beat another team trained AI in an game i would love to se that

  15. Pro level gamers ceratainly have much expertise and skill level when it comes
    to the usual tourneys.

    You need geniuses to play(since they can come with minute decisions a lot faster)and get to a pro tier they could probably have a far more even match against the bots.(Remember No Game No Life?)

  16. There is nothing embarrassing about losing to a real AI. They work differently than human. We will never be as advanced as AI because we can not use the full potential of the brain.

  17. Figures an asian is the best player in the world, and AI starts with A so technically the AI is Asian, should be AAI artificial Asian Intelligence

  18. Shouldn’t feel to bad I mean if this was done in a shooter like csgo then you might as well as just sit there and take in the bullets

  19. In learning ai should not involve in fighting or violent sports it could have that ways of neural patterns

  20. What bemusses me is that Elon Musk has all this negative shit he talks about AI and yet he started and controls one of the 1st competitive AI companies in the world! Is Elon going to actually be the one that destroys the world with his AI?

  21. Old town roads is full of weird comments…

  22. I would honestly love fighting more AI in all games. I dislike the idea of PvP because something about being outplayed or absolutely destroyed by another human just makes me upset. But if it's an AI you're like "ok it's programmed to do that".

  23. ok so whats this dota 2 thingy, call me crazy but long time ago they were doing pretty tough game that you can't beat like Tetris lol

  24. And in 2019 a team of 5 bots beat 5 best dota2 players. The dots player had no idea what hes talking about saying there isn't enough processing power

Leave a Reply

Your email address will not be published. Required fields are marked *