Let's compare to chess: when alphazero beat stockfish, a lot of chess analysts thought alphazero played beautifully and very different from the typical brute force AI. Youtube is full of game analysis of those games because GMs truly find those games beautiful and interesting.
Here it looks like the AI can get away with suboptimal decisions, bad decisions just because it can click simultaneously on 100+ units at a time anywhere on the map.
This feels like giving an average Joe an aimbot in a FPS game.
On January 27 2019 07:19 Xitah wrote: Not impressed. Why?
Let's compare to chess: when alphazero beat stockfish, a lot of chess analysts thought alphazero played beautifully and very different from the typical brute force AI. Youtube is full of game analysis of those games because GMs truly find those games beautiful and interesting.
Here it looks like the AI can get away with suboptimal decisions, bad decisions just because it can click simultaneously on 100+ units at a time anywhere on the map.
This feels like giving an average Joe an aimbot in a FPS game.
On January 25 2019 06:48 ArtyK wrote: I'd like to know epms and not worthless apms for this, considering the AI was capable of microing stalkers in 3 different places at once and win vs mass immortals.
I'm pretty sure apm = epm for the AI
I agree with this notion.
AI APM should be based on human EPM, not human APM.
Let me put this differently. After deep blue best Kasparov, brute force chess engines clearly surpassed humans but still it was the case that a human GM + engine could easily beat an engine because while the engine could see 20+ tactical combinations by sheer brute force, they could still make long term obvious mistakes in mid game. Alphazero seemingly does not so human GM + alphazero is not really better than just alphazero.
Here it's obvious that humans can made a lot of improvements to the game play so humans are not surpassed yet.
On January 25 2019 06:48 ArtyK wrote: I'd like to know epms and not worthless apms for this, considering the AI was capable of microing stalkers in 3 different places at once and win vs mass immortals.
I'm pretty sure apm = epm for the AI
I agree with this notion.
AI APM should be based on human EPM, not human APM.
What's the average pro's EPM?
usually about 200 to 240, with slight variations between races. Still, there is a fundamental difference if the A.I can access the control of units using background A.P.I. For instance, say you have an army of pheonixes or mutas, and you want to pulll the injured unit back to save its life or sometimes to save the entire fleet from a parasitic bomb. For humans, it's impossible to achieve in one click because these units are stacked. Best case scenario: ctrl+left click to select all units, find the one(if it's actually in first page), click it, click on somewhere else on the screen right click. These are already three clicks, plus three drags. practical case scenario, drag drop right click, drag drop right click, drag left click, drag right click Worst case scenario. drag drop right click, drag drop right click, drag drop right click, and everythign dies.
If you are an A.I with direct interface to select any units on screen, all you have to do is left click to select, drag right click, yeah there is 350ms latency, but all you have to do is two click and one drag.
On January 27 2019 11:10 Aegwynn wrote: So now do we have human elitists as well?
Nah we do not, we will not be surprised if A.I is shown to be mechanically superior to humans it's perfectly fine and expected. But what we want from Deepmind is to show that A.I can also exhibit some sorts of resourcefulness and wisdom in terms of tactics, to inspire human play, say the 25 probe-mining is inspiring, the 40 stalker 1000 apm micro is not.
If they just want to show an A.i that can beat humans, they could just make an automaton 2000 that does not hack and totally destory any human players with ease. That's like asking an Marathon elite to run against a car with GPS, the result is already written. There is a reason for them to implant a 350ms delay and to show that the A.I has lower average APM than human players even though they do not understand human APM very well. That is, through their neural network training on TPUs, the A.I can be quickly trained to be superior than human players even if they are mechanically equivalent.
I am not saying they should build a robot and entire real time image processing to precieve the game and to control the game the same way a human player would do, but it should be their goal to make something as equivalent and as fair as possible.
Otherwise, it makes no point to prove that a car can run faster than a man on racing tracks. It is much better to prove that an autonomous car can run faster than its human-driven version with equal weights on the racing track, the exact controls and interface sure are different, but this is much more fair.
Very nice showcase, but as it stands now the AI has the equivalent of an unlimited amount of hotkeys at its disposal. It doesn't blink back a bunch of stalkers, but rather decides which stalkers to blink within the bounds of it's APM. I think the mechanical brute forcing will not stop until the AI shows mistakes stemming from imperfect micro.
On January 27 2019 07:19 Xitah wrote: Not impressed. Why?
Let's compare to chess: when alphazero beat stockfish, a lot of chess analysts thought alphazero played beautifully and very different from the typical brute force AI. Youtube is full of game analysis of those games because GMs truly find those games beautiful and interesting.
Here it looks like the AI can get away with suboptimal decisions, bad decisions just because it can click simultaneously on 100+ units at a time anywhere on the map.
This feels like giving an average Joe an aimbot in a FPS game.
That's a superficial and wrong analysis. You're missing many things the AI did right, and that isn't related to micro.
For example, the AI had an absolutely perfect understanding or ability to predict the outcome of a battle based on army sizes and unit composition. This shouldn't really come as a surprise, since it's something you would expect an AI to do really well after 200 years of accumulated playing-experience, but it's really an important point. Even if you reduce the micro-abilities of the AI even further, humans can't hope to ever be as good at judging the outcome of an engagement as an AI. You often see this reflected in commentators in professional matches, when a medium-sized to large battle ensues, and they can't tell who is gonna win it until the end.
And you also saw this in these games: there were several situations where TLO/Mana misjudged the winning chances of an engagement, and got crushed, both on defense and offense. The AI understood the army compositions better, and knew when to move in or when it had enough defense, where's the human player made a sub-optimal judgement.
Trying to reduce the outcome of this match to simply being about superior micro is plain ignorant.
On January 27 2019 15:23 Dumbledore wrote: The alphastar agents that went 10-0 had ability to see the map zoomed out all the time. Not fair at all. The camera based agent got slam dunked.
Humans evolve too, and Mana found a way to abuse the AI. That is expected to happen. He came into this match prepared.
Fact is that according to DeepMind, the camera-based agent is actually only slightly weaker than the non-camera based agent. You can't judge something from one game. If they had gone to a 10-game match, i would still expect the AI to win 6-7 out of 10 games.
Here's the strength-graphs of the camera-agent versus the camera-less agent.
On January 27 2019 20:31 akatama wrote: Very nice showcase, but as it stands now the AI has the equivalent of an unlimited amount of hotkeys at its disposal. It doesn't blink back a bunch of stalkers, but rather decides which stalkers to blink within the bounds of it's APM. I think the mechanical brute forcing will not stop until the AI shows mistakes stemming from imperfect micro.
Actually, according to the information DeepMind has released, the AI actually uses individual "clicks" (like a simulated mouse) or drag boxes just like a human.
Of course it's better and faster at it, but it doesn't have infinite hot-keys.
Just my thoughts from the point of view of a casual sc:bw/sc2 player. I take it that AS learned "how to play" by getting an initial seed of pro replays, thus probably picking up the initial concept of "mining is good" and "kill enemy stuff to win", and then evolved by playing countless games versus instances of itself, right? If so, I think that this is a major flaw in the eventual evolution of its decisionmaking ability. It´s more or less the same if I was watching some replays and played a tutorial with my friend, and then never took a glance outside the box but only played against him, over and over again, for, well... 200 years. The result is a tree of winning techniques (not necessarily strategies or refined builds) that proved to be good within the scope of the learning environment, but it doesn´t have too much of a similarity to the evolution of the PvP meta over the years, or even within a patch. It´s not trying to be a better gamer (c), but trying to reach higher win-% with what it´s doing. This if of course backed by an access to mechanics way beond human capabilites.
Well, so far so good. It´s still nice to see that AS figured out things like flanking, proxies, blink micro etc. more or less by itself. What it failed to learn about is any consistent form of reactionary play depending on opponents' actions, see manas prism harass. But why should it, as long as pushing hard with near-perfect macro at home while aiming for an economy lead still works? It´s been practicing against other versions of itself that (mostly) didn´t execute any major tech switches after all.
Also, I see its heavy emphasis on blink stalkers in one form or another as a result of AS "understanding" that perfect micro with them is freaking awesome, or rather: cost/return wise they are oftentimes more effective than other protoss compositions, given the fact that fueling hundreds of apm into microing them makes them a dozen times better over a-moving them, or just focus firing. I assume that in the history of ASvAS games, any AI instance that tried to play a conservative composition against 80-100% blink stalkers eventuelly got wrecked, so this tree of decisions eventually died, or rather fell behind too much in priority to be considered.
Basically, with unlimited eapm, SC2 would need an independent set of balance patches around the mechanical capabilities of AS, or at least AS-PvP would sooner or later evolve to a point where all it needs are nexi, gates, core, forge, probes and stalkers. kappa.
But: what would have happened if AS didn´t get to learn in an environment where it can spike eapm to 1500 and beyond, but only to a human level? Or even as low as maybe 50ish eapm? Obviously letting it learn at high eapm and then hardcapping it way below that for an actual match will distort the result big time. We will probably never know, but I guess that with different cost/return weights on units, we would have seen totally different preferred unit compositions. Maybe 11/11 soul train games.
When I for myself try to improve, I play people that are seeded somewhere between "a bit better" and "quite a bit better", because otherwise, I am unlikely to get a glimpse of my own flaws, and all I get out of a game is a tiny bit of mechanical routine. Obviously, it´s quite hard to find a human opponent that´s stronger than AS and able to play hundreds of thousands of games per hour, simultaneously. So what do you think would happen if AS wouldn´t play-to-learn against itself, but rather against conventional bots with a gazillion of apm, perfect micro and strats that are designed to crush it with random multi-front pushes, while itself being constrained to screen vision and, say, 200-250 eapm? And once it starts to consistently deal with it, improve the sparring bot? Would AS learn to be more of an e-serral than a blink-god? Or would AS just evolve into developing a pushing technique that is mostly unstoppable by this kind of play, thus basically ending up where it is right now?
On January 27 2019 07:19 Xitah wrote: Not impressed. Why?
Let's compare to chess: when alphazero beat stockfish, a lot of chess analysts thought alphazero played beautifully and very different from the typical brute force AI. Youtube is full of game analysis of those games because GMs truly find those games beautiful and interesting.
Here it looks like the AI can get away with suboptimal decisions, bad decisions just because it can click simultaneously on 100+ units at a time anywhere on the map.
This feels like giving an average Joe an aimbot in a FPS game.
fair enough but this is a lot harder than chess. This might take some more innovation and more work. This was a good first step, and I thought it played very similar to top level humans which is a big achievement itself. No bot ever in rts has ever played very human like. It made a lot of great decisions even minus the amazing and inhuman micro it had This game is a lot different and much harder than chess. Chess really is a simple game even compared to GO. I think deepminds biggest achievement was GO that was truly amazing. Playing within human limits and beating the top ranked korean SC 2 players would be an amazing achievement though too. Even a perfect AI in sc 2 though you would expect it to lose because some strats just lose to other strats.
On January 27 2019 07:19 Xitah wrote: Not impressed. Why?
Let's compare to chess: when alphazero beat stockfish, a lot of chess analysts thought alphazero played beautifully and very different from the typical brute force AI. Youtube is full of game analysis of those games because GMs truly find those games beautiful and interesting.
Here it looks like the AI can get away with suboptimal decisions, bad decisions just because it can click simultaneously on 100+ units at a time anywhere on the map.
This feels like giving an average Joe an aimbot in a FPS game.
That's a superficial and wrong analysis. You're missing many things the AI did right, and that isn't related to micro.
For example, the AI had an absolutely perfect understanding or ability to predict the outcome of a battle based on army sizes and unit composition. This shouldn't really come as a surprise, since it's something you would expect an AI to do really well after 200 years of accumulated playing-experience, but it's really an important point. Even if you reduce the micro-abilities of the AI even further, humans can't hope to ever be as good at judging the outcome of an engagement as an AI. You often see this reflected in commentators in professional matches, when a medium-sized to large battle ensues, and they can't tell who is gonna win it until the end.
And you also saw this in these games: there were several situations where TLO/Mana misjudged the winning chances of an engagement, and got crushed, both on defense and offense. The AI understood the army compositions better, and knew when to move in or when it had enough defense, where's the human player made a sub-optimal judgement.
Trying to reduce the outcome of this match to simply being about superior micro is plain ignorant.
predicting the result of a battle is actually very hard. When you run up a ramp you might only see part of the army. Sometimes you cant see the entire enemy army than your guessing if you can win or not just based on partial information. You need to then make educated guesses to try to fill this in. Personally I think brood war was harder because units were less clumped so it was harder to see all the enemy units since they were more spread out. In sc 2 units bunch together into tight balls easier to see. In brood war you would have huge sprawling groups of units or tank lines that you cant see all of them. You need intuition. It will be interesting to see the bot play TvT since I think tanks are one of the most interesting and strategic parts of the game. How does it manage and position its siege lines? There is no doubt it will learn though. This AI is very impressive even minus the cheating micro and map management. It did a lot of amazing things I have never seen bots to do in RTS games. I have watched and played a lot of bots though, and some understanding of how they work. Some of it is lost on average people
EPT![[image loading]](https://storage.googleapis.com/deepmind-live-cms/images/SCII-BlogPost-Fig10_02.width-1500.png)
![[image loading]](https://storage.googleapis.com/deepmind-live-cms/documents/Oriol-Fig-Anim-170809-Optimised-r03.gif)
