With this and the news of 5 AI players getting Top 100 in LoL, I think either eSport tournaments are going to ban AI players and we'll see only AI tournaments or we'll see best teams competing...
With this and the news of 5 AI players getting Top 100 in LoL, I think either eSport tournaments are going to ban AI players and we'll see only AI tournaments or we'll see best teams competing against AI.
Either scenario is pretty exciting but I would really like to see, a honorary match at least, the winner of the global tournament competing against one of these AI teams
I get exhibition matches between a human team and an AI team, but I am not interested in a mixed tourney of AI and humans. Separate leagues were AI compete with one another and humans with...
I get exhibition matches between a human team and an AI team, but I am not interested in a mixed tourney of AI and humans. Separate leagues were AI compete with one another and humans with themselves works for me; personally, beyond the novelty of an exhibition match, I am never invested enough to care about an AI vs Human game. So banning AI for normal competition is appealing to me. I just feel like it is only a matter of time before an AI will become faster reacting and better than a human player, making mixed competition noncompetitive.
They have an amazing advantage: They can share each of their vision with almost no delay. Humans have to tell what they saw through a mic or with signals, and the receiving end has to react to it.
They have an amazing advantage: They can share each of their vision with almost no delay. Humans have to tell what they saw through a mic or with signals, and the receiving end has to react to it.
Be sure to check out the recent posts about OpenAI Dota 2 bots going from 1v1 mirror-match mode to playing a 5v5 mirror-match at or above human level. IIRC they intend to host a show match against...
Be sure to check out the recent posts about OpenAI Dota 2 bots going from 1v1 mirror-match mode to playing a 5v5 mirror-match at or above human level. IIRC they intend to host a show match against professional players at TI next month.
Last year they did the same for the 1v1 mode (beating some of the best human players), and it took weeks before the human players were able to find weaknesses and start being able to defeat the bot.
I don't know anything about the LoL bots, but OpenAI at least haven't quite reached the point where they can play Dota in its entirety (the mirror-match mode still has other limits like no invisibility, no courier management, no wards or ban/draft stage, etc), but it's already very impressive as it is.
Actually that's the Top 100 bots I was talking about, seems I misremembered Dota being LoL, thanks for the info! Even with those limits, I bet these bots could beat one of the Top 50 teams in Dota...
Actually that's the Top 100 bots I was talking about, seems I misremembered Dota being LoL, thanks for the info!
Even with those limits, I bet these bots could beat one of the Top 50 teams in Dota because their advantages are too big, the biggest one being instant sharing of 5 players vision. Humans have to substitute that with wards and voice comminucation, which is nowhere near what the bots have.
It's actually really interesting reading up on how the bots are implemented, though OpenAI haven't published the full details yet. Apparently the current set of Dota bots don't really communicate...
It's actually really interesting reading up on how the bots are implemented, though OpenAI haven't published the full details yet. Apparently the current set of Dota bots don't really communicate directly between each other at all, the only direct interaction is through a sliding weight governing how much they care about their own individual expected reward function vs the average of the reward functions for the whole team.
The bot API, however, does just feed them all the visible information directly, so they don't need to scroll around the map or communicate to share information, they just all know the same things by definition.
It's really interesting to see how the particular strengths of a bot player can lead to totally different playstyles being viable; I recall a Starcraft Brood War bot using a unique strategy based around Mutalisks (a fairly light fast flying Zerg unit, most often used either very sparingly for scouting or in massed hordes). The bot was able to use dozens of small swarming groups of them to pursue objectives across the entire map simultaneously in a way that no human player could.
I'll have to see if I can dig up that Starcraft article, it was a really interesting read, albeit with fairly conventional programming techniques instead of deep learning.
I do expect the bots to be able to take at least one or two games off the humans, especially since the mirror mode is so different from the game that the professionals spend all their time playing, the selected heroes for the mode play well to the strengths of the bots and the normal counter picks aren't available.
I'd love to play around with something like this, even if I have a measly GTX 1060 that would take weeks to train. DeepMind certainly isn't going to publish a working tensorflow adaptation,...
I'd love to play around with something like this, even if I have a measly GTX 1060 that would take weeks to train.
DeepMind certainly isn't going to publish a working tensorflow adaptation, hopefully OpenAI publishes their current one when they develop a better one.
In the article they mention AI are more collaborative than people, this could make for some interesting gameplay. Exited to see this tech applied to single player games as well.
In the article they mention AI are more collaborative than people, this could make for some interesting gameplay. Exited to see this tech applied to single player games as well.
Interesting development, but less impressive than the OpenAI dota news announces recently. Quake isn't nearly as complex a game. It's also noteworthy that the article didn't talk about in what...
Interesting development, but less impressive than the OpenAI dota news announces recently. Quake isn't nearly as complex a game.
It's also noteworthy that the article didn't talk about in what ways the ai was limited. Usually in instances like these, ai has absurd advantages because it can read the game state directly and doesn't have the lag that a human has in needing to observe and react to what is happening. This is especially important in a more "skill" based game like a shooter where bots with perfect aim can be easily made to beat any human through conventional programming (aka no ml).
10 Years From Now - "So E-117, your team wins again this year against the Mech Pythons! Tell me when are you going to allow humans into the world games?" "Well, we tried engaging humans in the...
10 Years From Now -
"So E-117, your team wins again this year against the Mech Pythons! Tell me when are you going to allow humans into the world games?"
"Well, we tried engaging humans in the games, in fact it was years ago, but they just wanted to play among themselves. We're still waiting for them to propose it..."
With this and the news of 5 AI players getting Top 100 in LoL, I think either eSport tournaments are going to ban AI players and we'll see only AI tournaments or we'll see best teams competing against AI.
Either scenario is pretty exciting but I would really like to see, a honorary match at least, the winner of the global tournament competing against one of these AI teams
I get exhibition matches between a human team and an AI team, but I am not interested in a mixed tourney of AI and humans. Separate leagues were AI compete with one another and humans with themselves works for me; personally, beyond the novelty of an exhibition match, I am never invested enough to care about an AI vs Human game. So banning AI for normal competition is appealing to me. I just feel like it is only a matter of time before an AI will become faster reacting and better than a human player, making mixed competition noncompetitive.
They have an amazing advantage: They can share each of their vision with almost no delay. Humans have to tell what they saw through a mic or with signals, and the receiving end has to react to it.
Pretty sure botting is banned in every game ever. Maybe they could use this to make the training bots not shit.
Be sure to check out the recent posts about OpenAI Dota 2 bots going from 1v1 mirror-match mode to playing a 5v5 mirror-match at or above human level. IIRC they intend to host a show match against professional players at TI next month.
Last year they did the same for the 1v1 mode (beating some of the best human players), and it took weeks before the human players were able to find weaknesses and start being able to defeat the bot.
I don't know anything about the LoL bots, but OpenAI at least haven't quite reached the point where they can play Dota in its entirety (the mirror-match mode still has other limits like no invisibility, no courier management, no wards or ban/draft stage, etc), but it's already very impressive as it is.
Actually that's the Top 100 bots I was talking about, seems I misremembered Dota being LoL, thanks for the info!
Even with those limits, I bet these bots could beat one of the Top 50 teams in Dota because their advantages are too big, the biggest one being instant sharing of 5 players vision. Humans have to substitute that with wards and voice comminucation, which is nowhere near what the bots have.
It's actually really interesting reading up on how the bots are implemented, though OpenAI haven't published the full details yet. Apparently the current set of Dota bots don't really communicate directly between each other at all, the only direct interaction is through a sliding weight governing how much they care about their own individual expected reward function vs the average of the reward functions for the whole team.
The bot API, however, does just feed them all the visible information directly, so they don't need to scroll around the map or communicate to share information, they just all know the same things by definition.
It's really interesting to see how the particular strengths of a bot player can lead to totally different playstyles being viable; I recall a Starcraft Brood War bot using a unique strategy based around Mutalisks (a fairly light fast flying Zerg unit, most often used either very sparingly for scouting or in massed hordes). The bot was able to use dozens of small swarming groups of them to pursue objectives across the entire map simultaneously in a way that no human player could.
I'll have to see if I can dig up that Starcraft article, it was a really interesting read, albeit with fairly conventional programming techniques instead of deep learning.
I do expect the bots to be able to take at least one or two games off the humans, especially since the mirror mode is so different from the game that the professionals spend all their time playing, the selected heroes for the mode play well to the strengths of the bots and the normal counter picks aren't available.
I'd love to play around with something like this, even if I have a measly GTX 1060 that would take weeks to train.
DeepMind certainly isn't going to publish a working tensorflow adaptation, hopefully OpenAI publishes their current one when they develop a better one.
In the article they mention AI are more collaborative than people, this could make for some interesting gameplay. Exited to see this tech applied to single player games as well.
Interesting development, but less impressive than the OpenAI dota news announces recently. Quake isn't nearly as complex a game.
It's also noteworthy that the article didn't talk about in what ways the ai was limited. Usually in instances like these, ai has absurd advantages because it can read the game state directly and doesn't have the lag that a human has in needing to observe and react to what is happening. This is especially important in a more "skill" based game like a shooter where bots with perfect aim can be easily made to beat any human through conventional programming (aka no ml).
10 Years From Now -
"So E-117, your team wins again this year against the Mech Pythons! Tell me when are you going to allow humans into the world games?"
"Well, we tried engaging humans in the games, in fact it was years ago, but they just wanted to play among themselves. We're still waiting for them to propose it..."