Slashdot Mirror

← Back to Stories (view on slashdot.org)

OpenAI's Bots Defeated Former Pro E-Sports Players At Dota 2 (theverge.com)

Posted by BeauHD on from the another-one-bites-the-dust dept.

On August 5th, OpenAI's bots defeated four former professional Dota 2 players in two of the three matches played. "There were a few conditions to make the game manageable for the AI, such as a narrower pool of 18 Dota heroes to choose from (instead of the full 100+) and item delivery couriers that are invincible," reports The Verge. "But those simplifications did little to detract from just how impressive an achievement [the] win was." From the report: The OpenAI Five triumphed in convincing fashion in the first game, not allowing the human players to even destroy one of their defensive towers. The humans recovered a little in game two, conquering one tower, but they still got demolished. And finally, in a game three played purely for pride, the humans managed to squeeze out a win. What stands out when you watch the matches is the apparent intelligence of the AI's decisions and the inhuman absence of any indecision. The typical Dota 2 game, even on the professional tier, involves quite a bit of equivocation about whether to engage in a fight, try and shift it to a more favorable battleground, or run away from it completely. The OpenAI team just doesn't need the processing time that humans require, which made its play appear unnatural -- but only in the speed and crispness of the decision-making, not in the content of those decisions.

30 of 49 comments (clear)

  1. Had me checking for April Fools by uCallHimDrJ0NES · · Score: 2, Interesting

    Teaching bots to miss convincingly was the first problem we had to solve back when we were constructing quakeworld bots. It's hard for me to believe that it's some kind of news when bots defeat humans.

    --
    Cloudiot: A person who does not see offsite storage as a way to lose control over access to his or her own data.
    1. Re:Had me checking for April Fools by Kjella · · Score: 1

      Teaching bots to miss convincingly was the first problem we had to solve back when we were constructing quakeworld bots. It's hard for me to believe that it's some kind of news when bots defeat humans.

      That's like saying I built a bot to play tic-tac-toe, it's hard to believe Deep Blue and AlphaGo made the news. Making an aimbot is trivial, making a system to balance a number of complex objectives, form strategies and adapt to rock-paper-scissors choices is hard.

      --
      Live today, because you never know what tomorrow brings
    2. Re: Had me checking for April Fools by CaffeinatedBacon · · Score: 1

      Not if the AI adapts to that previous weakness.
      Or if it was just trolling you with a pretend weakness just for the lols.

    3. Re:Had me checking for April Fools by uCallHimDrJ0NES · · Score: 1

      Lol, "making an aimbot is trivial". So, old quakeworld bots are tic tac toe players, and modern bots are systems that balance a number of complex objectives? Is that what you're getting at?

      --
      Cloudiot: A person who does not see offsite storage as a way to lose control over access to his or her own data.
  2. Sample size too small by SuperKendall · · Score: 5, Interesting

    So wait a second, the pros beat the bots on round three...

    Doesn't that seem to mean that the pros simply needed time to learn how the bots thought, and reacted to different situations? And that they did so by the third game, where after two sound thrashings they pulled out a win? Even if close that is quite a performance leap they managed.

    I would have been a lot more interested in the results of a ten-game series where the pros had adequate time to understand the inhuman patterns of play the bots had.

    --
    "There is more worth loving than we have strength to love." - Brian Jay Stanley
    1. Re:Sample size too small by Lab+Rat+Jason · · Score: 2

      You missed the part where the audience selected the "keystone cops" avatars for the computer to use during round 3.

      --
      Which has more power: the hammer, or the anvil?
    2. Re: Sample size too small by bmimatt · · Score: 1

      Bullshit. This is /. - we don't RTFA.

    3. Re:Sample size too small by SuperKendall · · Score: 1

      Ok, that makes lots more sense but would have been nice to note in the summary which implied something very different.

      --
      "There is more worth loving than we have strength to love." - Brian Jay Stanley
    4. Re:Sample size too small by SirMasterboy · · Score: 5, Informative

      For games 1 and 2 the AI got to draft it's own choice of 5 heroes out of the allowed 18 in an alternating picking sequence with the human opponents as is traditional. When each game started, the AI announced that it had a 95% confidence that it would win based solely on the heroes it had choosen against the human's choices.

      For game 3 the audience picked 5 "random" heroes that had little to no synergy as a DotA team lineup and not good options for each of a DotA team's "roles" (pushing, ganking, tanking, supporting, etc). At the start of game 3 the AI announced that it calculated only a 2.9% chance that it would win against the human team with the heroes it was given.

      Even with the severely gimped hero choices, it did an admirable job at trying to make it work by doing some unusual strategies. The professional casters even commented on how the AI was seemingly doing all the right things that it could have done given its poor position and that it actually did better in the match than anyone had expected it to do given the heroes that it was.

    5. Re:Sample size too small by michelcolman · · Score: 2

      Indeed, at the very start of the game the computer gave itself a 2% chance of winning with that selection. In the preceding two games it started out with more than 90% confidence of winning. It still played very well, but that was just an unwinnable game.

  3. Team of casters by Tough+Love · · Score: 4, Informative

    Those are hardly professional players, that is a team of casters who never practice together. It is doubtful that any of them could get onto even a second tier team today (well, moon, but then again maybe not). Still an impressive benchmark for AI and top tier pros are sure to get beaten sooner or later too, but what is the point of misrepresenting what actually happened?

    --
    When all you have is a hammer, every problem starts to look like a thumb.
    1. Re:Team of casters by SirMasterboy · · Score: 2

      Most of them have played and practiced together quite a bit actually.

      https://liquipedia.net/dota2/V... [liquipedia.net]

      Blitz, Merlini, and Captitalist most recently, but Fogged used to play with them as well and Fogged has played with Merlini for a very long time.

      Blitz and Moon are certainly "pros". Merlini used to be pro, but sure, he is not as good as he used to be and Fogged and Capitalist have played on teams at least semi-professionally.

      I mean, they have to start somewhere, this is just more testing to get the AI ready to play top current pros at the exhibition matches later this month obviously.

    2. Re:Team of casters by Tough+Love · · Score: 1

      Right, still impressive as I said, and to be fair the headline was "former pros" and the article goes out of its way to note the progression from casual players up to the current ex-pros. Reaction time is a huge part of Dota and a huge advantage for a bot, also noted in the article.

      --
      When all you have is a hammer, every problem starts to look like a thumb.
    3. Re:Team of casters by Tough+Love · · Score: 1

      inhuman absence of any indecision

      That, too, is reaction time.

      --
      When all you have is a hammer, every problem starts to look like a thumb.
    4. Re:Team of casters by lgw · · Score: 1

      No, that's initiative. Reaction time implied something specific to react to, and a trained response to that signal. Reactions don't involve conscious decision-making. Initiative is picking a strategy from many that might work given an opportunity.

      --
      Socialism: a lie told by totalitarians and believed by fools.
  4. Game three was just for fun by martyros · · Score: 5, Interesting

    One thing the summary didn't point out was that game three they didn't let the AI choose their own heroes. The humans basically conceded defeat; so for the third game they were just experimenting: The let the "audience" choose the heroes, and they chose heroes specifically which would be poor at playing the style which the computers had played so far, just to see if it could change its playing style to adapt to the new heroes. And the human team chose exactly the heroes that the AI chose for the first two games. After that draft, the AI's rating of its own chance of winning was 2.3%, based only on the draft. The AI adapted somewhat, but not much; and near the end of the game, the AI seemed to be doing a bunch of fairly sub-optimal things; like, it knew it couldn't really win, so it didn't know what to do except random micro.

    So, it was an interesting data point -- particularly the importance of choosing the right set of heroes. But it was certainly not a victory for humans. The AI soundly trounced them except when it was purposely crippled.

    Matches, post-game commentary, and other information available on the OpenAI Blog about the match.

    --

    TCP: Why the Internet is full of SYN.

    1. Re:Game three was just for fun by zlives · · Score: 2

      so... its not AI, just another database.

    2. Re:Game three was just for fun by liquid_schwartz · · Score: 2

      One thing the summary didn't point out was that game three they didn't let the AI choose their own heroes. The humans basically conceded defeat; so for the third game they were just experimenting: The let the "audience" choose the heroes, and they chose heroes specifically which would be poor at playing the style which the computers had played so far, just to see if it could change its playing style to adapt to the new heroes. And the human team chose exactly the heroes that the AI chose for the first two games. After that draft, the AI's rating of its own chance of winning was 2.3%, based only on the draft. The AI adapted somewhat, but not much; and near the end of the game, the AI seemed to be doing a bunch of fairly sub-optimal things; like, it knew it couldn't really win, so it didn't know what to do except random micro.

      So, it was an interesting data point -- particularly the importance of choosing the right set of heroes. But it was certainly not a victory for humans. The AI soundly trounced them except when it was purposely crippled.

      Matches, post-game commentary, and other information available on the OpenAI Blog about the match.

      Sounds like a classic case of not all heroes being equal (ie some are OP / some are too nerfed) if the odds were that obvious after only character selection.

    3. Re:Game three was just for fun by lgw · · Score: 1

      Sorry that's what "AI" means. Just like "hacker" means someone who attacks computer systems. Some linguistic battles aren't winnable.

      --
      Socialism: a lie told by totalitarians and believed by fools.
    4. Re:Game three was just for fun by zlives · · Score: 1

      i am not giving up just yet,

  5. Too easy with raw access? by Anonymous Coward · · Score: 1

    I'll be impressed when they create an AI that can only use mouse and keyboard to play.

  6. Macros by that+this+is+not+und · · Score: 3, Insightful

    The only way to make a competition like this fair is to let the human players use as many powerful macros and software aids as they liked.

    The other alternative would be an airgap between the AI and the gaming computer in the form of electromechanical servos to drive the controllers and a camera with optical recognition software to see the screen.

    Without one of these two things, it's an uneven match and really just a joke.

  7. Great by bmimatt · · Score: 1

    Let machines play machines, while we watch. Make that a sport.

    1. Re: Great by that+this+is+not+und · · Score: 1

      And for vintage matches let's all just watch IBM Card Sorters compete for time.

  8. Bots Cheating by Luthair · · Score: 5, Informative

    It should be noted that the bots are cheating and have a different level of access to the information in the game that players do not have. e.g. they have no need to select players to know inventories, precise health, mana etc. In addition to perfect possible information they also have perfect control input.

    They also make a big point about how the draft is considered to be one of the more difficult parts of dota and how impressive it is their bots can draft... but restrict the pool from the normal 115 to 18 and eliminate bans rendering it entirely unlike dota drafting. They also restricted 2 items, no summons/illusions, 5 invulnerable couriers (instead of a team having 1, which can be killed) and no scanning (a time-limited radar like ability).

    Don't get suckered into the news hype cycle until the bots use a screen with USB input and play the actual game instead of an arbitrary one.

    1. Re:Bots Cheating by SirMasterboy · · Score: 3, Interesting

      People need to get more informed. There are limitations yes, but there previously were a LOT more limitations and they are continuing to lift them.

      4 months ago the bots could only play 5 specific heroes against the same 5 specific heroes, they couldn't ward and they couldn't use or play against invisibility or fight roshan.

      Those limitations have been lifted as well as lifting the 5 static heroes to a choice for both teams of 18 heroes. They will obviously keep lifting these restrictions until the AI can play against illusions/summons and with the full hero pool.

      Also, the AI has a 200ms delay on every action, so I'm not sure I would hardly call it "perfect".

      But DotA is so complicated that just playing mechanically perfect is not enough to win. You need to play the proper strategies to win and whats amazing is that the AI is playing a strategy that it formed organically on it's own simply by playing itself for millions of games with very simple rules about what is good and what is bad.

  9. Next Year's Headline by mentil · · Score: 3, Funny

    AI defeats top Generals at military wargame, put in charge of country's military strategy.

    --
    Corruption is convincing someone that the selfless ideal is the same as their selfish ideal.
  10. They should have played more games by misnohmer · · Score: 1

    First 2 games AI wins, but the description seems like the humans were learning quick and getting better every game, winning by 3rd game. The question would be, would humans continue to win in subsequent games?

  11. DOTA? Not very impressive. by HeckRuler · · Score: 1

    "But those simplifications did little to detract from just how impressive an achievement [the] win was."

    Which is... not very impressive.

    Ok, so I haven't actually played Dota2. I'm assuming it's a twitchy game where reaction speed and knowing when to attack/run is vital. That's how Zergling Blood and the first Dota were. When this whole genre was a mod of starcraft and warcraft, it wasn't very deep. And journalists are just legendarily bad at hyping AI and not undestanding games. Way too many damn people think that "games" are just "kid's stuff". And on the flip side, way too many kids are too easily impressed.

    I don't even know why they limited it to 18 heroes. Wide search spaces is something that self-learning is supposed to be able to tackle with ease. ...Unless the entire field of "big data" is an over-hyped myth.

    1. Re:DOTA? Not very impressive. by HeckRuler · · Score: 1

      I don't really understand your reasoning. What game is deep for comparison?

      Chess is EXPTIME-complete.

      Check out Game Complexity or Complexity in general if you want to dive down the rabbit hole.

      I read up a little on DOTA2, it's given some of the units powers which make the game complex. Poker sort of stuff where bluffs are possible. Psychology is a bitch and a half to model, so these researchers specifically banned those powers.... making this even less impressive.

      Dota 2 is a much slower game than shooters like CS.

      You're not quite thinking about that the right way. You're talking about the speed rate like the human scale matters. CS isn't even a game from a AI perspective. Maybe there's some shenanigians with grenades, but otherwise: Spin();If(LOS to head){click}else(GoToBomb()). That's it. That's a perfect game. It's solved. Consider a version of DOTA where you had an infinite amount of time to plan and calculate things between every tick or frame of the game. That's essentially how computers operate.

      Beating a human in a game that depends on reaction time and wide search-spaces just isn't impressive.

      And the choice of heroes (that is also asymmetrical) changes the game a lot. Not to mention that the choice of items and the order to get them can easily change a game from a win to a loss.

      hmm, the versions I remember, this wasn't a real deep decision space. I always had a preferred build. Crunch it out and there's.... likely a optimal set. Maybe an optimal set per every configuration of the opponent's selection. And then an optimal set depending on what you expect the opponent to expect your set to be. ONLY on that last step does it even become a game. Everything before that is just math. As boring as "is 2.7 dps more than 4.7dps?". With an obvious answer that involves zero AI. Statistics. But this game might be deeper, I honestly don't know.

      Picking the wrong item can lead to a loss... The same way that picking the wrong square in tic-tac-toe can lead to a loss. But that doesn't make it a deep game.

      It's orders of magnitude more complex tactically than i.e. chess

      Uh huh.