Yes I don't really know it. But I understand the processing powe of it is what can give it an edge. But in a really complex game u can only teach a bot so much. And you can only "teach" or write something you know. So if a bot doesn't know how to respond to a certain move it will fail. It means the programmer would have to have anticipated it and accounted for it in his algorithm. The other alternative is machine learning. But I just have my doubts. There's only so much u can achieve with AI
Thats the thing, programmer doesnt need to know everything, he just needs to implement the algorithm and give the bot some training set. After that the bot can teach on its own. Every game, win or loss makes him a better player. That being said, i think you could surprise him with smoke ganks etc at first but after several attempts he will know - ok, there are supports missing on the map, better play safer.
AIs nowadays can be really complex, very good at one specific task (in this case, playing dota mid lane) and should not be underestimated
but then it turns into a numbers game. he will hide sometimes, he will show sometimes, you just have to know what you're showing him and what hes concluding and fucking outplay him. It's the same old dota but with a retarded bot that can't adapt
That retarded bot that cant adapt just outplayed several top dota players because he actually can adapt.. and yes, even dota can be turned down to mathematics and statistics (which is what is artificial intelligence about)
dota from algorithmical pov is a multidimentional chess sort of game and it sure can be turned down to pure mathematics because it IS maths in the first place
LMAO benao
The whole fucking point of machine learning is that the bot will adapt to it's previous mistakes
Given enough time against even the cheese strats it'll figure out a way to beat you.
Every single decision in dota and even life can be boiled down to statistics and numbers.
Every
Single
One
It is made that way, please get some knowledge on how neural networks works in general, you can start with bayesian theorem, it is relatively easy to understand.
The bot ONLY KNOWS WHAT it SEES. WE GOT FUCKING FOG IN DOTA
USE YOUR FUCKING BRAIN
edit: the fog is just 1 variable btw. There's so much more!! I might just end up explaining with so much fucking bullshit you guys write about a bot being unbeatable. Hint: The fact it relies in values is the main reason it WOULD lose and it's main weakness
idk i dont think many of u guys understand how this works
the bot doesnt improve itself if you cheese it, the bot played 1v1 against itself up to yesterday, where the final version got put on an usb stick and got used in the 1v1s
Ok, I'll teach you for free. How do you think the bot would make it's decision? Based on what? Probabilities right? That in itself means it doesn't win everytime, but ok let's go with it. There's an 80% chance you are ganking it based on the minimap and movements prior to that ok? It decides based on its subhuman calculations it's better to hide right? What if you knew that? Then you would definitely NOT gank and gain an advantage from farming woods while it doesn't do what it should be doing if it knew you'd just go out of vision just to confuse it. And so forth with who knows how many patterns. What about positioning in a fight? Fight constantly in a delta split formation might give a 70% of winning the fight in any case scenarion, but what if the other team goes 5 man yolo on a single target? That same formation might just have a 20% chance of winning the fight and so on. So yeah, go to fucking hell with your fucking delusional thought and conclusions. The fact it's based in numbers is its greatest fucking weakness. It's a fucking BOT
what if the players' knowledge of the approach to probability concept AI has is also a knowledge available to the AI, though?
at least thats how chess AIs work afaik
@triple then it just turns into a real game where the bot has better mechanics which again, doesn't secure a win.
edit: it also turns into a game where the bot side ends up relying on the mechanics advantage which might in turn change the moves to unilinear. Turning it into a more stupid bot than smarter in terms of dota.
Benao
If u jungle while it Lanes it'll come out ahead and take ur tower?
Obviously the way u beat it is by doing something unorthodox it hasn't encountered yet. But if the devs were to run it in training against bots who wud do the same cheese as the players who beat it 50 times at Seattle, then the bot wud learn how to beat it.
It kinda is. Can you with certainty talk about purpose? What makes people different? Though I recommend not dwelling on that. You'll just get all fucked up. Just do your own shit :P
Ppd said that pajkatt defeated the bot, and the guy on Russian stream said pajkatt tricked the bot. He kept the mango on the ground so when the bot checked pajkatt's inventory and courier he didn't see the mango.
Edit: This is wrong according to blitz who was near pajkatt.
Blitz: Pajkatt won by buying a faerie fire and wand and tricking the bot into trading and hitting wand and faerie at last second lol I was sitting right behind him during this time
I think we should put stuff into perspective.
1 hour of playing against itself, it already can beat the built in bots of the game.
2 weeks after, it can go head-to-head(albeit with limited rules) against the best mid players in the world.
They have now a whole year for the bot to train by itself. Sure, it's 5v5... but its probably plentyof time to sift through a number of strats. Who knows, the bot might pull out something it would make by itself that no one else has ever done before.
As someone who has studied AI, I can tell you its unbeatable by a human being, no matter how skilled you are. Take Paparazzi, the 1v1 Tournament winner who defeated the likes of Miracle, Sumail and Sccc. Or even better, you take all the professional mid players, and even players like Arteezy if you like, and put them in a Valve Major focused solely on 1v1 Mid matches and produce a victor, the "best" mid player in the world, and he won't be able to beat the bot.
Two weeks is an insane amount of time for an AI to learn Dota. Especially because they only trained the AI for 1v1 solo mid mode and probably only on SF. Its not even nearly close to a human being's training in the same period of time. The bot actually decided to AFK in base at one point of time during the training, thereby removing any possibility of dying, but learned that you will lose the advantage due to that. You could go to a different lane entirely and just constantly auto-attack creeps to push the tower and win, it won't work.
Like someone above just mentioned, Dota is like a chess game, but the variables are far too vast to calculate since there aren't just 64 squares and 32 pieces. Its a huge map with 113 heroes. That increases the complexity to an insane degree, but that doesn't mean its completely unpredictable or incalculable. I'm pretty sure if they complete the project of a 5 man BOT team, they will be unbeatable, even by any TI champions of the past. Its just that simple.
P.S. Look up the computer "Deep Blue". The song named after it is also very good. ;)
A curious question tho - while everyone was busy beating the bot to get those 50 arcanas, was the bot still in 'learning mode' or a finalized version? Cos it sure would have gotten a hell lot smarter from all those games imo.
there's too much uncertainty in dota. its not like chess where everything can be calculated. that's why in a pure 1 v 1 u can actually design a good bot who can take advantage of all the calculations with limited uncertainty.
its a probability game, and not a pure mathematics game. if bots have no information how are they going to play? they have to decide on what's optimal and calculate probability. and that's something thats really hard to program into bots
Hi. I am Benao and everyone must reflect my absolutely correct opinion. If you have opinions slightly different from mine I will anally f0ck you across cyberspace and make you regret ever exercising your individuality.
Hi. I am Benao and everyone must reflect my absolutely correct opinion. If you have opinions slightly different from mine I will anally f0ck you across cyberspace and make you regret ever exercising your individuality.
its a probability game, and not a pure mathematics game
There are RNG mechanics, yes. But even those can be predicted by really good players who count their hits. And if human beings can guess enemy movements, calculate numbers to determine how many and which spells would kill a hero, so can bots.
It's not mainly because of RNG that it's a probabilities game you sad fucking shit. That's what you still don't get. Diox even pointed it out to you. There is limited information, the game is ever changing. It's all about adapting and at that, a bot would always be at least 1 move behind, in everything!!
Doesn't mean it can't get close but for that you need a personality, or 5 of them. But at that point, do you really have a bot? (Not that I see it feasible, from what I can see, it will always be a static copy of something outdated). edit (actually i solved it, but the only way to make it original and again optimal would make it program it's own bot the same way new generations of men are created, teach it and so on. Still would need a few generations to have a better version and then again, can you really say that's a bot? Because at the point of conceiving, you would've already trascended that which you call a bot.
If you still don't get it then fuck it. It's a waste of time talking to braindead people that don't want to learn.
1) Learning based AI if it meets a situation it has never encountered will be at disadvantage. But it cannot be discounted that after millions of game cycles playing against copies of itself or other human players, the space for discovering entirely new situations decreases over time. Meaning it will get better as unfamiliar situations decreases.
2) Even information that is not available can become material for deductive decisioning. (E.g. No heroes visible on map, stay safe)
3) AI is best in perfect information environments. But this is generally only true for expert rules based AI, which is what deep blue was. But this AI is not that kind of AI.
4) AI can develop "imagination". This is where humans still have a huge advantage. Imagination in AI is being developed but is still in its infancy today. But make no mistake it is already possible.
5) With time, imaginative ability combined with a large bayesian network, AI can even learn to cheese without ever having encountered it before.
Even if not by the next TI, i believe 5v5 AI that goes up against our best is only a matter of years. Not decades.
As it is, AI already has huge mechanical advantages over humans. And the gaps in creativity will gradually close.
Added after posting: Did you know AI is already used to create new AI?
Learning based AI if it meets a situation it has never encountered will be at disadvantage. But it cannot be discounted that after millions of game cycles playing against copies of itself or other human players, the space for discovering entirely new situations decreases over time. Meaning it will get better as unfamiliar situations decreases.
Hi. I am Benao and everyone must reflect my absolutely correct opinion. If you have opinions slightly different from mine I will anally f0ck you across cyberspace and make you regret ever exercising your individuality.
Yes I don't really know it. But I understand the processing powe of it is what can give it an edge. But in a really complex game u can only teach a bot so much. And you can only "teach" or write something you know. So if a bot doesn't know how to respond to a certain move it will fail. It means the programmer would have to have anticipated it and accounted for it in his algorithm. The other alternative is machine learning. But I just have my doubts. There's only so much u can achieve with AI
Thats the thing, programmer doesnt need to know everything, he just needs to implement the algorithm and give the bot some training set. After that the bot can teach on its own. Every game, win or loss makes him a better player. That being said, i think you could surprise him with smoke ganks etc at first but after several attempts he will know - ok, there are supports missing on the map, better play safer.
AIs nowadays can be really complex, very good at one specific task (in this case, playing dota mid lane) and should not be underestimated
but then it turns into a numbers game. he will hide sometimes, he will show sometimes, you just have to know what you're showing him and what hes concluding and fucking outplay him. It's the same old dota but with a retarded bot that can't adapt
That retarded bot that cant adapt just outplayed several top dota players because he actually can adapt.. and yes, even dota can be turned down to mathematics and statistics (which is what is artificial intelligence about)
That's where you're wrong and proof you're fucking stupid.
Math and statistics, that the main reason the bot would lose if it was made that way. Sigh, why do you stupid people even post. Think a bit ok?
dota from algorithmical pov is a multidimentional chess sort of game and it sure can be turned down to pure mathematics because it IS maths in the first place
I'm sure that the machine learning hasn't gone through every possible weird scenario, so it wudnt know what to do when people pulls this cheese.
LMAO benao
The whole fucking point of machine learning is that the bot will adapt to it's previous mistakes
Given enough time against even the cheese strats it'll figure out a way to beat you.
Every single decision in dota and even life can be boiled down to statistics and numbers.
Every
Single
One
It is made that way, please get some knowledge on how neural networks works in general, you can start with bayesian theorem, it is relatively easy to understand.
The bot ONLY KNOWS WHAT it SEES. WE GOT FUCKING FOG IN DOTA
USE YOUR FUCKING BRAIN
edit: the fog is just 1 variable btw. There's so much more!! I might just end up explaining with so much fucking bullshit you guys write about a bot being unbeatable. Hint: The fact it relies in values is the main reason it WOULD lose and it's main weakness
xD
what
id say its his main advantage but w/e, i give up on explaining it to you
idk i dont think many of u guys understand how this works
the bot doesnt improve itself if you cheese it, the bot played 1v1 against itself up to yesterday, where the final version got put on an usb stick and got used in the 1v1s
or maybe i misunderstood it, but i doubt that
Ok, I'll teach you for free. How do you think the bot would make it's decision? Based on what? Probabilities right? That in itself means it doesn't win everytime, but ok let's go with it. There's an 80% chance you are ganking it based on the minimap and movements prior to that ok? It decides based on its subhuman calculations it's better to hide right? What if you knew that? Then you would definitely NOT gank and gain an advantage from farming woods while it doesn't do what it should be doing if it knew you'd just go out of vision just to confuse it. And so forth with who knows how many patterns. What about positioning in a fight? Fight constantly in a delta split formation might give a 70% of winning the fight in any case scenarion, but what if the other team goes 5 man yolo on a single target? That same formation might just have a 20% chance of winning the fight and so on. So yeah, go to fucking hell with your fucking delusional thought and conclusions. The fact it's based in numbers is its greatest fucking weakness. It's a fucking BOT
https://www.reddit.com/r/DotA2/comments/6t8qvs/openai_bots_were_defeated_atleast_50_times/
what if the players' knowledge of the approach to probability concept AI has is also a knowledge available to the AI, though?
at least thats how chess AIs work afaik
pajkatt killed him once but bot immediatelly returned even better and didn't die ever again, it learns from it's mistake
@triple then it just turns into a real game where the bot has better mechanics which again, doesn't secure a win.
edit: it also turns into a game where the bot side ends up relying on the mechanics advantage which might in turn change the moves to unilinear. Turning it into a more stupid bot than smarter in terms of dota.
i wonder what the point of this is, though? it seems supe ruseless except for demonstration purposes only
devs can learn from it
btw by definition, playing dota is useless too =)
Benao
If u jungle while it Lanes it'll come out ahead and take ur tower?
Obviously the way u beat it is by doing something unorthodox it hasn't encountered yet. But if the devs were to run it in training against bots who wud do the same cheese as the players who beat it 50 times at Seattle, then the bot wud learn how to beat it.
L e a r n t o r e a d .
btw by definition living is useless too
It kinda is. Can you with certainty talk about purpose? What makes people different? Though I recommend not dwelling on that. You'll just get all fucked up. Just do your own shit :P
its too late
Ppd said that pajkatt defeated the bot, and the guy on Russian stream said pajkatt tricked the bot. He kept the mango on the ground so when the bot checked pajkatt's inventory and courier he didn't see the mango.
Edit: This is wrong according to blitz who was near pajkatt.
Blitz: Pajkatt won by buying a faerie fire and wand and tricking the bot into trading and hitting wand and faerie at last second lol I was sitting right behind him during this time
I think we should put stuff into perspective.
1 hour of playing against itself, it already can beat the built in bots of the game.
2 weeks after, it can go head-to-head(albeit with limited rules) against the best mid players in the world.
They have now a whole year for the bot to train by itself. Sure, it's 5v5... but its probably plentyof time to sift through a number of strats. Who knows, the bot might pull out something it would make by itself that no one else has ever done before.
As someone who has studied AI, I can tell you its unbeatable by a human being, no matter how skilled you are. Take Paparazzi, the 1v1 Tournament winner who defeated the likes of Miracle, Sumail and Sccc. Or even better, you take all the professional mid players, and even players like Arteezy if you like, and put them in a Valve Major focused solely on 1v1 Mid matches and produce a victor, the "best" mid player in the world, and he won't be able to beat the bot.
Two weeks is an insane amount of time for an AI to learn Dota. Especially because they only trained the AI for 1v1 solo mid mode and probably only on SF. Its not even nearly close to a human being's training in the same period of time. The bot actually decided to AFK in base at one point of time during the training, thereby removing any possibility of dying, but learned that you will lose the advantage due to that. You could go to a different lane entirely and just constantly auto-attack creeps to push the tower and win, it won't work.
Like someone above just mentioned, Dota is like a chess game, but the variables are far too vast to calculate since there aren't just 64 squares and 32 pieces. Its a huge map with 113 heroes. That increases the complexity to an insane degree, but that doesn't mean its completely unpredictable or incalculable. I'm pretty sure if they complete the project of a 5 man BOT team, they will be unbeatable, even by any TI champions of the past. Its just that simple.
P.S. Look up the computer "Deep Blue". The song named after it is also very good. ;)
It has good mechanical skills but bad decision making. Im surely gonna dumpster that bot back to the fucking stoneage.
If you've studied AI and think that maybe you should change your profession to sitting behind a cash register
A curious question tho - while everyone was busy beating the bot to get those 50 arcanas, was the bot still in 'learning mode' or a finalized version? Cos it sure would have gotten a hell lot smarter from all those games imo.
there's too much uncertainty in dota. its not like chess where everything can be calculated. that's why in a pure 1 v 1 u can actually design a good bot who can take advantage of all the calculations with limited uncertainty.
its a probability game, and not a pure mathematics game. if bots have no information how are they going to play? they have to decide on what's optimal and calculate probability. and that's something thats really hard to program into bots
There are RNG mechanics, yes. But even those can be predicted by really good players who count their hits. And if human beings can guess enemy movements, calculate numbers to determine how many and which spells would kill a hero, so can bots.
It's not mainly because of RNG that it's a probabilities game you sad fucking shit. That's what you still don't get. Diox even pointed it out to you. There is limited information, the game is ever changing. It's all about adapting and at that, a bot would always be at least 1 move behind, in everything!!
Doesn't mean it can't get close but for that you need a personality, or 5 of them. But at that point, do you really have a bot? (Not that I see it feasible, from what I can see, it will always be a static copy of something outdated). edit (actually i solved it, but the only way to make it original and again optimal would make it program it's own bot the same way new generations of men are created, teach it and so on. Still would need a few generations to have a better version and then again, can you really say that's a bot? Because at the point of conceiving, you would've already trascended that which you call a bot.
If you still don't get it then fuck it. It's a waste of time talking to braindead people that don't want to learn.
its a joke man... hes blocking lis godlike and its reaction and decision making is .00001 sec reaction time...welp its a bot indeed
1) Learning based AI if it meets a situation it has never encountered will be at disadvantage. But it cannot be discounted that after millions of game cycles playing against copies of itself or other human players, the space for discovering entirely new situations decreases over time. Meaning it will get better as unfamiliar situations decreases.
2) Even information that is not available can become material for deductive decisioning. (E.g. No heroes visible on map, stay safe)
3) AI is best in perfect information environments. But this is generally only true for expert rules based AI, which is what deep blue was. But this AI is not that kind of AI.
4) AI can develop "imagination". This is where humans still have a huge advantage. Imagination in AI is being developed but is still in its infancy today. But make no mistake it is already possible.
5) With time, imaginative ability combined with a large bayesian network, AI can even learn to cheese without ever having encountered it before.
Even if not by the next TI, i believe 5v5 AI that goes up against our best is only a matter of years. Not decades.
As it is, AI already has huge mechanical advantages over humans. And the gaps in creativity will gradually close.
Added after posting: Did you know AI is already used to create new AI?
"copies of itself" is the key word here
also ai is super scary and i'm not sure if i want to see what is it going to do in 10 years
You cannot beat the AI conventionally.
You *must* use unconventional tactics.
Anything conventional it will simply be better than you at, because it's trained it over and over with perfect memory 100,000 times over.
But like others have said, it's braindead when it comes to new territory. You confuse it, and it's as easy as beating a 1k player.
How can ai possibly be capable of creativity ?
Thanks to Benao's genius level "not a cash register guy" posts, this thread has reached the point of no return.
I duno what you're thinking about when u you talk about AI creativity. But the day they have real creativity is the day skynet is born