'From a machine learning point of view, StarCraft provides an ideal environment to study the control of multiple agents at large scale, and also an opportunity to define tasks of increasing difficulty, from micromanagement, which concerns the short-term, low-level control of fighting units during battles, to long-term strategic and hierarchical planning under uncertainty. While building a controller for the full game based on machine learning is out-of-reach for current methods, we propose, as a first step, to study reinforcement learning algorithms in micromanagement scenarios in StarCraft. '
Yeah, for some reason I always think Soumith Chintala works at Deepmind ( admins could you change the title ?). It's a carbon copy of methods those guys used anyway. I need to sleep.
They are looking at BW specifically. The article mentions dragoons and wraiths on page 4.
Changed the title. And indeed, they're using BW (and BWAPI) which is pretty exciting.
Here is a short (non-technical) summary that might be more interesting to people here: - In all tasks, we control all units from one side, and the opponent (built-in AI in the experiments) is attacking us - Tasks are: 5 marines vs 5 marines, 15 marines vs 16, 3 zealots + 2 dragoons on both sides, 15 wraiths vs 17, other scenarios with X marines vs Y marines and X wraiths vs Y wraiths - In 15 marines vs 16 scenario, the advantage is on our player’s side - 81% win rate - whereas in 15 wraiths vs 17 is hard - 10% win rate - 5 marines vs 5 marines scanario - 85% win rate - 3 zealots + 2 dragoons vs 3 zealots + 2 dragoons - 48% win rate - The subsequent experiments will include self-play (training and evaluation), multi-map training (training more generic models), and more complex scenarios which include several types of advanced units with actions other than move and attack - Finally, the goal of playing full games of StarCraft should not get lost, so future scenarios would also include the actions of “recruiting” units (deciding which types of unit to use), as well as make use of them
So looks like they're just getting started! Hopefully they come far enough to start participating in the BW AI tournaments.
Very interesting, and strange... seems like a weird publicity move to choose broodwar given how little attention it gets from the mainstream nowadays, but still super cool. maybe it'll bring more people to bw when they realize that facebook is tinkering with it
On September 13 2016 16:58 Endymion wrote: Very interesting, and strange... seems like a weird publicity move to choose broodwar given how little attention it gets from the mainstream nowadays, but still super cool. maybe it'll bring more people to bw when they realize that facebook is tinkering with it
Technically simpler (game engine), with more previous work to piggyback on. It's a logical choice given that BW has been theoretically interesting to the AI community for a while now.
i know there's always being a lot of work done on broodwar AI in the community but i didn't realize that it was large enough to actually get recognized by a company like facebook for investment, everyone in the botting community must be pretty excited about the level of these new eyes on them
It's Djem5 vs. some computer opponents. Basically the summary of all games is, "computer does cookie cutter strategy, human counters it without much effort."
2Pacalypse, thanks for this. I'll post a detailed summary of the article when I get a sec, just wanted to give everyone a heads-up that this is happening.
On September 13 2016 16:58 Endymion wrote: Very interesting, and strange... seems like a weird publicity move to choose broodwar given how little attention it gets from the mainstream nowadays, but still super cool. maybe it'll bring more people to bw when they realize that facebook is tinkering with it
So I have a better answer for you as to why, if you read that U Alberta link. Basically the reason is BWAPI and how Blizzard does not allow any similar software to modify SC2, per the EULA. So it's a legal issue.
On September 13 2016 16:58 Endymion wrote: Very interesting, and strange... seems like a weird publicity move to choose broodwar given how little attention it gets from the mainstream nowadays, but still super cool. maybe it'll bring more people to bw when they realize that facebook is tinkering with it
So I have a better answer for you as to why, if you read that U Alberta link. Basically the reason is BWAPI and how Blizzard does not allow any similar software to modify SC2, per the EULA. So it's a legal issue.
This for sure, as well as in general, when training neural networks on pixels, you want as little of those as possible ( as it makes the statistical estimation process easier ), both in terms of resolution and FPS. So if the same game was available to play on an old school Nintendo versus a Playstation 4 HD, you'd want to use the former for AI training purposes. BW makes more sense by that logic too.
bw's eula allows 3rd party input but sc2's doesn't?? i would have expected both of them to not allow modification.. i wonder if hdbw's eula will be the same as sc2s
On September 14 2016 17:10 Endymion wrote: bw's eula allows 3rd party input but sc2's doesn't?? i would have expected both of them to not allow modification.. i wonder if hdbw's eula will be the same as sc2s
Well, back when BW eula was written, DLL injection wasn't widespread knowledge. So since DLL injection doesn't alter the .exe itself it is technically allowed in that sense.
But even if Blizzard tried to stop it they can't. Since BW has private servers, and you can play via DirectIP or LAN with Hamachi. Whereas in SC2 you have to use Battle.net to play multiplayer.
So I got a good chance to read through the paper. They're basically applying more modern machine learning techniques to very simple micro scenarios as described by 2Pac. They talk about how Starcraft represents a rather interesting challenge for AI because of its large, almost world-like state space with uncertain information, but that they're basically going to abstract all of that away by focusing on the tiny problem of reliably fighting the built-in AI under simplified scenarios (map is visible etc).
On the one hand, it's nice that they're not arrogant enough to believe that they're going to solve a game that is well beyond the ability of modern AI capability to play well (computers have no sense of strategic thinking and you can't just search a tree like in chess/Go to find an optimal/high-quality move because Starcraft just isn't so statically defined). And their learning algorithm approach is certainly interesting. I'd like to see what they come up with, on video. On the other hand, even by AI standards beating a built-in AI (i.e. no micro whatsoever) under simplified circumstances is nothing special. I look forward to seeing what they come up with, but I think it's more relevant to their own research work than to Starcraft AI-making. I might be wrong but I'm going to take the "skeptic by default" position here until they prove otherwise; hype usually leads nowhere unless there's evidence that it's not just fluff.
I'm one of those kooks who think singularity, qualitative thinking and so on being irreducible capacities of the intellect, are forever impossible to machine systems. We've been trying for so long, that I doubt anything really new will come out of this, just incremental progress in systems handling. But, that's important! The capital accumulation of code that can successfully 'abstract' will continue developing.
On September 15 2016 03:39 Jaedrik wrote: I'm one of those kooks who think singularity, qualitative thinking and so on being irreducible capacities of the intellect, are forever impossible to machine systems. We've been trying for so long, that I doubt anything really new will come out of this, just incremental progress in systems handling. But, that's important! The capital accumulation of code that can successfully 'abstract' will continue developing.
We've been trying, but our hardware was primitive. Not only is it getting faster, but quantum computers are on the horizon too.
We probably do need some kind of breakthrough still. But I don't see why it shouldn't be possible. The brain is a proof of concept afterall. On top of that, there's no reason to believe the human brain is the pinnacle of intelligence. We might surpass its capabilities quite easily, eventually.
I sometimes wonder whether AI will end up similar to people trying to fly.
All early attempts tried to imitate nature (birds) with some limited success. We are presently at a similar stag in AI. Trying with some limited success to imitate what we think the brain does. However the first actual flying machine didn't fly at all like how it is done by nature.
On September 15 2016 03:39 Jaedrik wrote: I'm one of those kooks who think singularity, qualitative thinking and so on being irreducible capacities of the intellect, are forever impossible to machine systems. We've been trying for so long, that I doubt anything really new will come out of this, just incremental progress in systems handling. But, that's important! The capital accumulation of code that can successfully 'abstract' will continue developing.
We've been trying, but our hardware was primitive. Not only is it getting faster, but quantum computers are on the horizon too.
We probably do need some kind of breakthrough still. But I don't see why it shouldn't be possible. The brain is a proof of concept afterall. On top of that, there's no reason to believe the human brain is the pinnacle of intelligence. We might surpass its capabilities quite easily, eventually.
I believe :p
I'm also one of those kooks who think the predominant models of physics are contradictory, have faulty metaphysical foundations, and are pseudo-scientific. Quantum computing will never come to bear: the universe is not based on probabilities or virtual particles to make our fancy QED equations balance, it is based on cause and effect and a proper method of conducting physical sciences in real experimentation. Susskind says: "Where common sense and intuition failed, we had to create a new form of intuition based upon abstract mathematics. When common sense fails, we must create uncommon sense." I think this is sophistry, even if out of good intention. Where they may continue to irrationally reify all things, all fields, as being mediated by particles, I reject it. Now, what might be on the horizon is vector-based resistors. Anyone can do this experiment... well, anyone with the funds. Get a quartz crystal (can put it on a string to see the effect easier) and a strong magnet (100 Gauss NDIB is enough), and turn it different ways and approach the magnet. There will be more or less resistance depending on what angle it's turned at. I've seen the experiment done in person and cannot deny its veracity.
On September 15 2016 10:36 Equalizer wrote: I sometimes wonder whether AI will end up similar to people trying to fly.
All early attempts tried to imitate nature (birds) with some limited success. We are presently at a similar stag in AI. Trying with some limited success to imitate what we think the brain does. However the first actual flying machine didn't fly at all like how it is done by nature.
Making a computer 'think' may turn out similarly.
That gets into the debate of which function a computer should perform; the modern dichotomies are thinking vs. acting and humanly vs rationally. The consensus for most mainstream AI is to act rationally, which is ultimately a "birds vs. airplane wings" type of conclusion. So the consideration has been made.
On September 15 2016 03:39 Jaedrik wrote: I'm one of those kooks who think singularity, qualitative thinking and so on being irreducible capacities of the intellect, are forever impossible to machine systems. We've been trying for so long, that I doubt anything really new will come out of this, just incremental progress in systems handling. But, that's important! The capital accumulation of code that can successfully 'abstract' will continue developing.
We've been trying, but our hardware was primitive. Not only is it getting faster, but quantum computers are on the horizon too.
We probably do need some kind of breakthrough still. But I don't see why it shouldn't be possible. The brain is a proof of concept afterall. On top of that, there's no reason to believe the human brain is the pinnacle of intelligence. We might surpass its capabilities quite easily, eventually.
I believe :p
I'm also one of those kooks who think the predominant models of physics are contradictory, have faulty metaphysical foundations, and are pseudo-scientific. Quantum computing will never come to bear: the universe is not based on probabilities or virtual particles to make our fancy QED equations balance, it is based on cause and effect and a proper method of conducting physical sciences in real experimentation. Susskind says: "Where common sense and intuition failed, we had to create a new form of intuition based upon abstract mathematics. When common sense fails, we must create uncommon sense." I think this is sophistry, even if out of good intention. Where they may continue to irrationally reify all things, all fields, as being mediated by particles, I reject it. Now, what might be on the horizon is vector-based resistors. Anyone can do this experiment... well, anyone with the funds. Get a quartz crystal (can put it on a string to see the effect easier) and a strong magnet (100 Gauss NDIB is enough), and turn it different ways and approach the magnet. There will be more or less resistance depending on what angle it's turned at. I've seen the experiment done in person and cannot deny its veracity.
Regarding quantum physics: It's true that there's contradictions between quantum theory and relativity theory. Therefore, there has to be an even more general theory about nature. Such a theory would undoubtedly include both quantum and relativity theory as special cases, because there's very precise experimental verifications of both theories.
So, quantum mechanical "probabilities" will always show up somehow. But their interpretation is questionable. You can read them as probabilities, or an average over a multiverse, or whatever...
Regarding your quartz crystal experiment: Just to clarify, you do talk about electrical resistance, right? So you are trying to send a current through the crystal and measure it depending on the crystal's alignment in a magnetic field?
There is a difference between creating a machine that is thinking and a machine that makes ppl believe it is. Modern AI solutions are second variant at best but we are improving. Not only the deep learning becomes possible these days, also we have more powerful hardware. Deep Learning is obviously not a solution but a significant improvement over the previous attempts. Well engineered AI systems are hybrid, a combination of statistical and rule-based approaches.
On September 15 2016 23:02 letian wrote: There is a difference between creating a machine that is thinking and a machine that makes ppl believe it is. Modern AI solutions are second variant at best but we are improving. Not only the deep learning becomes possible these days, also we have more powerful hardware. Deep Learning is obviously not a solution but a significant improvement over the previous attempts. Well engineered AI systems are hybrid, a combination of statistical and rule-based approaches.
This. When Google's programs auto-learn to play Breakout, they perform millions and millions of convolutions and rectifying non-linearities, roughly, but they certainly don't philosophize about it or could 'tell' us what they learned when they did. But pragmatically, they play better than most humans. Is this intelligence if the car self-drives ? Insofar as intelligence is process and result, yes. The high-level conceptual stuff will no doubt come, but later.
On September 15 2016 03:39 Jaedrik wrote: I'm one of those kooks who think singularity, qualitative thinking and so on being irreducible capacities of the intellect, are forever impossible to machine systems. We've been trying for so long, that I doubt anything really new will come out of this, just incremental progress in systems handling. But, that's important! The capital accumulation of code that can successfully 'abstract' will continue developing.
Yes. It's too early to tell if we're gonna go S-curve or exponential when all's said and done. The latest developments in computer vision are pretty startling, but also pretty primitive ( small animals do shape detection and classification too ). On the other hand, if strong AI comes to pass, even peaceful, the whole of civilisation is instantly remade - no need to ever do physical or intellectual work again, and to an extent, no point too, maybe - hence it's such a big deal. Just the economic consequences of automation are going to define our era.
On September 15 2016 03:39 Jaedrik wrote: I'm one of those kooks who think singularity, qualitative thinking and so on being irreducible capacities of the intellect, are forever impossible to machine systems. We've been trying for so long, that I doubt anything really new will come out of this, just incremental progress in systems handling. But, that's important! The capital accumulation of code that can successfully 'abstract' will continue developing.
We've been trying, but our hardware was primitive. Not only is it getting faster, but quantum computers are on the horizon too.
We probably do need some kind of breakthrough still. But I don't see why it shouldn't be possible. The brain is a proof of concept afterall. On top of that, there's no reason to believe the human brain is the pinnacle of intelligence. We might surpass its capabilities quite easily, eventually.
I believe :p
I'm also one of those kooks who think the predominant models of physics are contradictory, have faulty metaphysical foundations, and are pseudo-scientific. Quantum computing will never come to bear: the universe is not based on probabilities or virtual particles to make our fancy QED equations balance, it is based on cause and effect and a proper method of conducting physical sciences in real experimentation. Susskind says: "Where common sense and intuition failed, we had to create a new form of intuition based upon abstract mathematics. When common sense fails, we must create uncommon sense." I think this is sophistry, even if out of good intention. Where they may continue to irrationally reify all things, all fields, as being mediated by particles, I reject it. Now, what might be on the horizon is vector-based resistors. Anyone can do this experiment... well, anyone with the funds. Get a quartz crystal (can put it on a string to see the effect easier) and a strong magnet (100 Gauss NDIB is enough), and turn it different ways and approach the magnet. There will be more or less resistance depending on what angle it's turned at. I've seen the experiment done in person and cannot deny its veracity.
Regarding quantum physics: It's true that there's contradictions between quantum theory and relativity theory. Therefore, there has to be an even more general theory about nature. Such a theory would undoubtedly include both quantum and relativity theory as special cases, because there's very precise experimental verifications of both theories.
So, quantum mechanical "probabilities" will always show up somehow. But their interpretation is questionable. You can read them as probabilities, or an average over a multiverse, or whatever...
Regarding your quartz crystal experiment: Just to clarify, you do talk about electrical resistance, right? So you are trying to send a current through the crystal and measure it depending on the crystal's alignment in a magnetic field?
Most crystals have anisotropic properties for various reasons. On top of that, magnetic fields can significantly influence e.g. a crystal's structure.
There's much weirder stuff going on in solid state physics. Just think about superconduction, superfluidity, etc etc
I doubt it would include either. GR says nothing can go faster than light, and QM says instantaneous action at a distance. But it goes far deeper: GR says the universe is relative, QM says it's probabilistic, both quantitative models, but they're opposed on a metaphysical level where there can be no reconciliation, so ideas which flow from their premise must be mutually exclusive. I'm not convinced about precise experimental verification, but I'd be more than happy to look at a paper or something if you wish. In my experience, there've been a few cases where the explanation for a phenomena might be adequate, but there being an alternative explanation which is just as adequate. As you say, interpretation is questionable. Rather, there are many good descriptions and explanations from the same evidence, but one is ultimately correct. I anticipate an overthrow of the entire establishment at some point in the future, such events have happened all throughout the history of science, and I don't think there's any good reason to think it wont happen again. Though, I don't mean to say that change is a guarantee, I mean to caution against a post hoc error which glorifies the present and belittles times past. I think the likes of Maxwell, Heviside, Faraday, Tesla, and the old fogies who believed in the ether will be generally vindicated.
No, you don't need to run an electric current through the crystal or measure anything like that. Just your hand holding some good quartz crystal approaching the magnet in different ways: you will feel resistance to your approach depending on how you hold the quartz. I watched me bro do it
I'd attribute to all dielectric material (all matter) four qualities: capacitance, resistance, permeability, and permittivity. In the case of 'superconductivity,' what happens when you cool such materials is they gain high(er) capacitance and low(er) magnetic permeability, such that the 'superconductor' becomes electromagnetically impermeable. The most naturally impermeable material is Bismuth.
... Ah, uhm. How about that AI? I'm reminded of Total Annihilation's intro. "What began as a conflict over the transfer of consciousness from flesh to machines..."
Well obviously quantum physics has a big issue as it doesn't explain dark matter. This is a big deal and is why another more advanced theory might come, and hopefully then, mesh with general relativity in a simpler fashion than '10,11 or 12 dimension string-theory where you can't see the extra dimensions because they're probably Calabi-Yau compactified'. Given how smart a guy like Witten is, if he couldn't figure it out, I doubt we'll see that in our lifetime - unless computers take over, maybe.
I doubt it would include either. GR says nothing can go faster than light, and QM says instantaneous action at a distance. But it goes far deeper: GR says the universe is relative, QM says it's probabilistic, both quantitative models, but they're opposed on a metaphysical level where there can be no reconciliation, so ideas which flow from their premise must be mutually exclusive.
SR (I forgive your mistake) predicts that nothing can move faster than light in our 3dimensional space (+1 time dimension). GR already adds another layer of abstraction on top of that, such that matter can actually warp that 4d space. (Giving rise to several theories of warp drives, lol.)
In a sense, QM adds another layer of abstraction, showing that you can have "non-local" correlations.
As far as I know, those thoughts are not the incompatibilities between QM and relativity. I'll admit I never dug into the math of where it goes wrong, so I won't pretend here. You can do a google search for the reasons, if you feel like it.
I doubt it would include either.
It has to! For the most parts at least. Just like both QM and SR include Newtonian Mechanics.
No, you don't need to run an electric current through the crystal or measure anything like that. Just your hand holding some good quartz crystal approaching the magnet in different ways: you will feel resistance to your approach depending on how you hold the quartz. I watched me bro do it
Was it setup like this? You have a powerful magnet, a quartz and hold a magnetic metal in your hand. Metal gets attracted to speaker, but appears to be pushed away by the quartz?
Spoiler: You and your bro have been scammed ^^
I'd attribute to all dielectric material (all matter) four qualities: capacitance, resistance, permeability, and permittivity. In the case of 'superconductivity,' what happens when you cool such materials is they gain high(er) capacitance and low(er) magnetic permeability, such that the 'superconductor' becomes electromagnetically impermeable. The most naturally impermeable material is Bismuth.
Most importantly, what happens is that electrical resistance drops to absolute 0. Not "very small", but absolute 0. Quite remarkable.
On September 16 2016 22:39 MyLovelyLurker wrote: Well obviously quantum physics has a big issue as it doesn't explain dark matter. This is a big deal and is why another more advanced theory might come, and hopefully then, mesh with general relativity in a simpler fashion than '10,11 or 12 dimension string-theory where you can't see the extra dimensions because they're probably Calabi-Yau compactified'. Given how smart a guy like Witten is, if he couldn't figure it out, I doubt we'll see that in our lifetime - unless computers take over, maybe.
Can we now go back to talking about Starcraft AI.
Nothing can explain dark matter ~
We can go back to general AI, if you want. There won't be many exciting news about strong SC AI anytime soon, I suppose ))
On September 17 2016 09:58 beg wrote: Was it setup like this? You have a powerful magnet, a quartz and hold a magnetic metal in your hand. Metal gets attracted to speaker, but appears to be pushed away by the quartz?
No, not at all. Again, to clarify, a large, clear quartz crystal, held in the hand, I think he used a 50 gauss NDIB magnet, just by itself with no attachment. Depending on which way my brother held it in his hand, he felt resistance (not in the electrical sense, but in the mechanical sense, as a 'force' pushing away) on his hand on approach. We weren't 'scammed,' what we did was an experiment. And, nothing in the physical sciences can explain dark matter because it's unverifiable by definition, hence the darkness--it's a post hoc rationalization created for blind faith that gravity be the dominant force in the universe.
On September 13 2016 16:58 Endymion wrote: Very interesting, and strange... seems like a weird publicity move to choose broodwar given how little attention it gets from the mainstream nowadays, but still super cool. maybe it'll bring more people to bw when they realize that facebook is tinkering with it
So I have a better answer for you as to why, if you read that U Alberta link. Basically the reason is BWAPI and how Blizzard does not allow any similar software to modify SC2, per the EULA. So it's a legal issue.
This for sure, as well as in general, when training neural networks on pixels, you want as little of those as possible ( as it makes the statistical estimation process easier ), both in terms of resolution and FPS. So if the same game was available to play on an old school Nintendo versus a Playstation 4 HD, you'd want to use the former for AI training purposes. BW makes more sense by that logic too.
Actually, they're not doing this by image recognition. They hook up directly to the game and get the units information and input that way.
On September 17 2016 09:58 beg wrote: Was it setup like this? You have a powerful magnet, a quartz and hold a magnetic metal in your hand. Metal gets attracted to speaker, but appears to be pushed away by the quartz?
No, not at all. Again, to clarify, a large, clear quartz crystal, held in the hand, I think he used a 50 gauss NDIB magnet, just by itself with no attachment. Depending on which way my brother held it in his hand, he felt resistance (not in the electrical sense, but in the mechanical sense, as a 'force' pushing away) on his hand on approach. We weren't 'scammed,' what we did was an experiment. And, nothing in the physical sciences can explain dark matter because it's unverifiable by definition, hence the darkness--it's a post hoc rationalization created for blind faith that gravity be the dominant force in the universe.
Regarding quartz crystal: Create a setup where you can actually measure the effect. "Your brother felt it" is not very convincing. I don't think quartz is a magnetic material. And if it was, why would you think this contradicts science?
Dark matter: You say "it's unverifiable by definition, hence the darkness". No. First of all, it's called "matter", because you can fix problems in cosmology by adding more mass (=matter) to a certain equation. But then, you'd also expect to find that matter when you actually look into the sky. But we don't see any, hence it's called "dark", as in "invisible".
We might need a breakthrough in physics to find out what's really going on. But it's not unverifiable by definition.
This thread is now about physics ~ I apologize ~~ Will return to AI discussion shortly :D
im pretty convinced that an AI with enough logic written into it could perfectly emulate humans. it just takes a brilliant architect for optimizing the massive amounts of processes that would be needed and a thorough unbderlying understanding of the subject material. emulating humans as whole is probably pretty far off, but emulating a perfect starcraft player could be done much sooner in my opinion. when it's written correctly i dont' think a human player will ever be able to beat an ai..
now whether it should be done is a whole other question.. personally, i think the fun and beauty of starcraft comes from the players' abilities to overcome stress and multiple points of attention on the map to form a cohesive gameplan, using their game knowledge as well as intuition. when you break it down into pure logic it becomes less interesting, imo, which is why i find less intense rts a lot less interesting.
On September 18 2016 23:25 Endymion wrote: im pretty convinced that an AI with enough logic written into it could perfectly emulate humans. it just takes a brilliant architect for optimizing the massive amounts of processes that would be needed and a thorough unbderlying understanding of the subject material. emulating humans as whole is probably pretty far off, but emulating a perfect starcraft player could be done much sooner in my opinion. when it's written correctly i dont' think a human player will ever be able to beat an ai..
now whether it should be done is a whole other question.. personally, i think the fun and beauty of starcraft comes from the players' abilities to overcome stress and multiple points of attention on the map to form a cohesive gameplan, using their game knowledge as well as intuition. when you break it down into pure logic it becomes less interesting, imo, which is why i find less intense rts a lot less interesting.
Well, whether or not that's true, it's well beyond current technology, and not just for reasons of computational limits. Computers just aren't really capable of humanlike reasoning; all they can really do is follow a set of rules exactly as they are written. And yes, we've figured out clever ways to design those rules so that computers can get better at certain tasks over time (e.g. Go, or robots walking, or many other more "modern" inventions) but that is very far removed from any kind of AI that can deal with a situation that can't really be defined well by some set of specific rules. In that sense, Starcraft strategy is pretty abstract, even if micromanagement on its own really isn't.
On September 18 2016 23:25 Endymion wrote: im pretty convinced that an AI with enough logic written into it could perfectly emulate humans. it just takes a brilliant architect for optimizing the massive amounts of processes that would be needed and a thorough unbderlying understanding of the subject material. emulating humans as whole is probably pretty far off, but emulating a perfect starcraft player could be done much sooner in my opinion. when it's written correctly i dont' think a human player will ever be able to beat an ai..
now whether it should be done is a whole other question.. personally, i think the fun and beauty of starcraft comes from the players' abilities to overcome stress and multiple points of attention on the map to form a cohesive gameplan, using their game knowledge as well as intuition. when you break it down into pure logic it becomes less interesting, imo, which is why i find less intense rts a lot less interesting.
Well, whether or not that's true, it's well beyond current technology, and not just for reasons of computational limits. Computers just aren't really capable of humanlike reasoning; all they can really do is follow a set of rules exactly as they are written. And yes, we've figured out clever ways to design those rules so that computers can get better at certain tasks over time (e.g. Go, or robots walking, or many other more "modern" inventions) but that is very far removed from any kind of AI that can deal with a situation that can't really be defined well by some set of specific rules. In that sense, Starcraft strategy is pretty abstract, even if micromanagement on its own really isn't.
Also, we don't want to handcraft logic and features in an AI, otherwise it becomes a 'bot'. We want a very generic algorithm where the computer sees rewards and high-level objectives, and then discovers strategies to maximize them. This is now possible to computers in spite of what was written above - such 'emergent', non-strictly algorithmic behaviour is accessible to large-scale neural networks such as the one used by DeepMind in AlphaGo. The math behind it is just extremely complex and not fully understood yet.
On September 13 2016 16:58 Endymion wrote: Very interesting, and strange... seems like a weird publicity move to choose broodwar given how little attention it gets from the mainstream nowadays, but still super cool. maybe it'll bring more people to bw when they realize that facebook is tinkering with it
So I have a better answer for you as to why, if you read that U Alberta link. Basically the reason is BWAPI and how Blizzard does not allow any similar software to modify SC2, per the EULA. So it's a legal issue.
This for sure, as well as in general, when training neural networks on pixels, you want as little of those as possible ( as it makes the statistical estimation process easier ), both in terms of resolution and FPS. So if the same game was available to play on an old school Nintendo versus a Playstation 4 HD, you'd want to use the former for AI training purposes. BW makes more sense by that logic too.
Actually, they're not doing this by image recognition. They hook up directly to the game and get the units information and input that way.
Good point but they probably ran out of time and/or didn't get their network to converge, you can see in their conclusion : 'There is ongoing work on convolutional networks based models that conserve the 2D geometry of the game (while embedding the discrete components of the state and actions).'
If they are going to run a convolutional network for strategy, they will probably end up adding pixels as inputs..
Any update on this? It is been almost 2 years. I know facebook have a bot cherrypi but is not machine learning. Do they have working agent that actually can play a game vs build in a.i.?
EPT
![[image loading]](https://www.coh2.org/images/Smileys/hijacked.gif)
