Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

DeepMind and Blizzard Open StarCraft II as an AI Research Environment

deepmind.com

Post reply on HN
Print view View original

161–170 of 283 posts

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#161

Post by jtolmar »

aerovistae wrote: post #41
ipnon wrote: post #2

Any predictions for how long it will take for an AI to win against the world's best player?

Awhile. This just isn't like Go or Chess. The gap from perfect information to imperfect information is quite a chasm, and from turn-based to real-time is even more vast. I play Age of Empires 2 semi-competitively, and I just can't imagine the research progress that would have to be made for a pro to lose to an APM-limited AI agent. So much of the game comes down to intuiting what your opponent is planning without bei…

> intuiting what your opponent isn't ready for

I haven't played AOE2 so I don't know if the mechanics are similar enough to translate, but my goal for my Starcraft bot is to do precisely this. If you can enumerate the possible builds (what's available when) and assess the matchups between builds, you can make this happen using some intuitive expansions on adversarial search.

> Your moves must be cohesive and planned and flow continuously without time to ponder, each connected to the last.

Recomputing the entire plan from the current state works in RTS too, but only if your decision-making takes every already-in-motion thing into account and has no internal discrepancies. That's a pretty big if; this sort of weakness accounts for a lot of bot weakness currently. Units spinning around due to slight changes in perceived state cause lots of wasted resources.

> Can an AI agent learn to appreciate the power of distraction?

Despite multitasking theoretically being one of the strengths of an AI, a lot of the current field can't handle more than one military situation at a time. In this year's SSCAIT a lot of bots completely fell apart when confronted with one of the top bots (Bereaver) doing reaver drops.

I'm not sure a bot can meaningfully learn distraction, but I'm not sure it's necessary - attacking on simultaneous fronts is optimal anyway. The army can only be so many places at once.

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#162

Post by bluetwo »

littlestymaar wrote: post #158
qub1t wrote: post #71

A lot of people here seem to be underestimating the difficulty of this problem. There are several incorrect comments saying that in SC1 AIs have already been able to beat professionals - right now they are nowhere near that level. Go is a discrete game where the game state is 100% known at all times. Starcraft is a continuous game and the game state is not 100% known at any given time. This alone makes it a much hard…

As a long time StarCraft fan I don't share your point of view : People usually refer to StarCraft as a strategy game but there's actually really little strategy involved : during the first weeks after a new map pool is released, the pro players explore different build orders that are strong on it. And after this period, when the meta-game has settled, the winner of a match (best of 3 or 5) is almost always the one wh…

Interesting. So would you say that there are two parts here, figuring out a general strategy for a new map and then maximizing execution?

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#163

Post by rryan »

littlestymaar wrote: post #158
qub1t wrote: post #71

A lot of people here seem to be underestimating the difficulty of this problem. There are several incorrect comments saying that in SC1 AIs have already been able to beat professionals - right now they are nowhere near that level. Go is a discrete game where the game state is 100% known at all times. Starcraft is a continuous game and the game state is not 100% known at any given time. This alone makes it a much hard…

As a long time StarCraft fan I don't share your point of view : People usually refer to StarCraft as a strategy game but there's actually really little strategy involved : during the first weeks after a new map pool is released, the pro players explore different build orders that are strong on it. And after this period, when the meta-game has settled, the winner of a match (best of 3 or 5) is almost always the one wh…

Do StarCraft AIs (the ones that are included with the game) cheat? (e.g. can they see past the fog of war?)

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#164

Post by TulliusCicero »

littlestymaar wrote: post #158
qub1t wrote: post #71

A lot of people here seem to be underestimating the difficulty of this problem. There are several incorrect comments saying that in SC1 AIs have already been able to beat professionals - right now they are nowhere near that level. Go is a discrete game where the game state is 100% known at all times. Starcraft is a continuous game and the game state is not 100% known at any given time. This alone makes it a much hard…

As a long time StarCraft fan I don't share your point of view : People usually refer to StarCraft as a strategy game but there's actually really little strategy involved : during the first weeks after a new map pool is released, the pro players explore different build orders that are strong on it. And after this period, when the meta-game has settled, the winner of a match (best of 3 or 5) is almost always the one wh…

As someone who probably has played at a higher level than you I disagree.

Yes, it's true that mechanics are a large determinant in who wins between people. But pro humans are not easily thrown off by odd or novel strategies or tactics. They can react to things that introduce small wrenches into their build order without serious issues. Players can even adapt to things they've never seen before. The issue with Starcraft is that the state space is so large that it will likely be hard to get an AI that can flexibly and intelligently react to unusual or bizarre things that mess with their build order, because the neural nets will have nothing to account for, say, a mid-game cannon rush, or whatever.

If it was simply a matter of computers taking humans things and executing them better, computers would already be better than humans at Starcraft (there have been plenty of AI competitions using Brood War), but they're not. Not even close.

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#165

Post by TulliusCicero »

rryan wrote: post #163
littlestymaar wrote: post #158

Earlier quoted context omitted.

As a long time StarCraft fan I don't share your point of view : People usually refer to StarCraft as a strategy game but there's actually really little strategy involved : during the first weeks after a new map pool is released, the pro players explore different build orders that are strong on it. And after this period, when the meta-game has settled, the winner of a match (best of 3 or 5) is almost always the one wh…

Do StarCraft AIs (the ones that are included with the game) cheat? (e.g. can they see past the fog of war?)

Last I checked most of them don't cheat, but a few do (I think they're labeled as cheating though?).

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#166

Post by TulliusCicero »

haeffin wrote: post #118
qub1t wrote: post #71

A lot of people here seem to be underestimating the difficulty of this problem. There are several incorrect comments saying that in SC1 AIs have already been able to beat professionals - right now they are nowhere near that level. Go is a discrete game where the game state is 100% known at all times. Starcraft is a continuous game and the game state is not 100% known at any given time. This alone makes it a much hard…

I don't know if I would label SC2 as continuous. I don't think anything happens to the game state at a finer granularity than tick level. So to me it seems that it's also discrete (but with the state changing 44.8 a second at default speed). I agree though that this looks more challenging for ML methods. I haven't looked at if they limit the rate of commands that the AI can issue, otherwise this will be something tha…

It's not literally continuous, but it is real-time rather than turn-based, and positions of units are essentially floats rather than (a small range of) ints. That makes it effectively continuous (too large to just generate a tree of all possible actions and then prune).

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#167

Post by keerthiko »

cjbprime wrote: post #76
sidusknight wrote: post #34

Earlier quoted context omitted.

Your definition of macro is a lot more encompassing than how I see it used in SC2. It normally refers to just spending resources properly (constantly producing units and workers, setting up infrastructure, etc).

Yeah, interesting! I agree I don't usually hear macro used that way. But if someone was microing a battle and as a result didn't look at the minimap and see a drop arriving at their base, I can totally see concluding that this player is bad at macro as a result -- macro is referring to there being a macro cycle of tasks you have to perform all the time whether you want to or not, and non-production tasks like checkin…

Exactly! Macro is everything "big picture" -- being aware of the overall state of the game, managing all resources involved (including attention, APM, time, mindspace), and grand decision-making ("I'm up against X overall strategy. How should I respond? What are the weaknesses in my approach? What should I be looking for that would exploit those weaknesses? How can I address those situations?")

Execution of the answers to those thoughts is in the form of tactics and micro, the other two aspects of macro. If you're following current SC2 pro meta, a "strong macro player" however has more right answers to most of those questions (Stats, Innovation) and that's their strength, versus a "strong tactical player" (TY, sOs) or a "strong micro-based player" (ByuN, herO).

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#168

Post by Impossible »

steinystein wrote: post #153
Impossible wrote: post #150

Earlier quoted context omitted.

Minor nitpick, video games running on digital computers are by definition still discrete even if they feel continuous. Networked multiplayer wouldn't be possible in RTS games if that wasn't the case. The granularity of unit positions and turns in Starcraft obviously leads to a much larger state space, so I get what you're saying, for AI its effectively continuous.

For a game of go the entire game state is known to each player. That's the diff. For vidya games state is hidden to the player if the player cannot 'see' it. Therefore u wrong fam.

I didn't say anything about hidden information. That clearly makes SC more challenging than Go, as it requires the AI to build some kind of mental model of possible player states from limited information.

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#169

Post by TulliusCicero »

obastani wrote: post #14
ktRolster wrote: post #3

When Watson won at Jeopardy, one of its prime advantages was the faster reaction time at pushing the buzzer. The fairness of that has already been hashed out elsewhere, but..... We already know that computers can have superior micro and beat humans at Starcraft through that(1). Is DeepMind going to win by giving themselves a micro advantage that is beyond what reasonable humans can do? (1) https://www.youtube.com/wat…

My understanding is that in a full match, AIs still have no hope against humans, since even though they can crush humans at micro, their macro is still abysmal [1]. I'm not aware of a match where any AI has beat a pro human player at Starcraft -- I'd be interested in learning otherwise! [1] http://spectrum.ieee.org/automaton/robotics/artificial-intel...

It's not that their macro is abysmal (macro in Starcraft refers to the mechanics of managing production and economy), it's that their strategy and tactics are real bad.

Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment

#170

Post by branja »

mulmen wrote: post #149
branja wrote: post #122

Earlier quoted context omitted.

ideally they'd train it on real keypresses rather than actions

Why would that be ideal? Wouldn't that just make ML at the strategy layer harder without doing anything to make the discoveries more valuable?

to emulate human handicaps at the interface layer. I didn't say it would be free
Post reply on HN