As impressive as this is, APM matters a lot in an RTS such as Starcraft and computers have a massive advantage here. I'd like to see them tackle an RTS like Civ where APM doesn't matter whatsoever, only decision making.
Grandmaster level in StarCraft II using multi-agent reinforcement learning
91–100 of 321 posts
Re: Grandmaster level in StarCraft II using multi-agent reinforcement learning
#92Has anyone read the actual paper ? This summary really makes it look like "mission accomplished", but this was much much more interesting than that. We saw AI do "obviously stupid things", and we also saw them improve a lot in the middle of the trial, as many youtubers showed. AI was also much more interesting when playing the protoss race, and really felt like it was responding to the opponents actions, and the othe…
> Actually not a single game made pro player realize something new about the game.
People are now over saturating their mineral line (making more probes than before), so I don't think that's true.
Re: Grandmaster level in StarCraft II using multi-agent reinforcement learning
#93Earlier quoted context omitted.
> The speed at which AlphaStar overtook previous AIs seems to me to be nearly unprecedented in AI research. Is it not simply the case that, before AlphaStar, very little money and effort was being put into developing AIs for Starcraft 2?
Starcraft is an incredibly complex game. 10^26 possible moves at any point (you can click/drag anywhere on the screen, pressing a keyboard button as you do so), imperfect information, real-time constraints, etc.
Yes it is, but you can prune the decision space significantly by discarding bogus choices, or more concretely by modelling higher level goals and then compiling them into actual moves.
Re: Grandmaster level in StarCraft II using multi-agent reinforcement learning
#94What do you guys think about having the ai communicate to a person and having them execute the moves?
Re: Grandmaster level in StarCraft II using multi-agent reinforcement learning
#95Earlier quoted context omitted.
When Watson won Jeopardy, its lead designer thought a 5 ms reaction time was fair because one time a human anticipating the buzzer light got 2 ms. Of course to any reasonable person a consistent 5 ms is obviously vastly superior to once in a blue moon lottery-winning luck. Sounds like it's the same deal here, where the designers made some effort to level the interface playing field, but also left in some advantages l…
Yes, the point of the entire project was for the system to see if it can win. If we have reached the point where to even risk a loss requires implementing artificial restraints on the AI...well...
They aren't really artificial restraints though, more like artificial advantages without them. The interface delay challenges are pretty relevant to making things fair in almost ANY competition. The physical part of the interface is an essential part of the game here.
You aren't really testing the thing you really want to test otherwise.
You could probably play tennis pretty well against Nadal if you made him wear goggles that gave him a quarter second delay in response time, but that's not really a fair test.
All other things being equal, if you match up two chess players, and one can just say or think the position instead of having to physically move the piece before they hit the clock, that advantage will accumulate over time.
Re: Grandmaster level in StarCraft II using multi-agent reinforcement learning
#96Earlier quoted context omitted.
The question isn't whether the AI still has advantages, but whether it's an unfair one. The current restrictions were implemented with input from TLO, and seem to work quite nicely. One could, of course, add more restrictions (like not noticing things on the minimap all the time...), but that's not what's it about anymore. At this point, micro is comparable to humans, and we can start to compare macro and strategy. F…
> a well executed strategy with few errors often wins, even if it wasn't optimal I think this is hits the chord with why a lot of people are talking about this bot as cheating more than they are being outplayed. It's like fighting an aimbot in an FPS. They arent outsmarting you, but the mechanical consistency is inhuman and you are unable to force mistakes. At best itll feel like playing someone on their best day. Th…
The issue with limiting it past what humans can do is that in the self play training they'd never encounter what humans could do. The benefits of trying to make the limitations somewhat human-like is that the self play optimizes around roughly what it will encounter when it goes against humans
Re: Grandmaster level in StarCraft II using multi-agent reinforcement learning
#97There has always been the issue of interface when playing videogames AI vs human. Either give the human a brain-computer interface or give the AI a mouse, keyboard, monitor, robot hands and a camera. Anything else seems inherently unfair.
Depends on what you're trying to test. You could just as easily argue that _not_ allowing the AI to make use of its natural APM and interface advantages is unfair, and that it should be allowed to play without restrictions in order to demonstrate its full potential.
AlphaGo had an "unfair" advantage in its games against Lee Sedol in that it was able to "think" far faster than you could reasonably expect a human to, and could therefore evaluate many, many more moves than a human player possibly could have. If you artificially limited the speed at which AlphaGo was able to "think" to match human limitations, would it have been able to win against Lee? I'd argue probably not.
Similarly, here we have an AI that's easily able to crush top SC players if it plays with an uncapped APM and no camera limitations, but only reaches grand master level when you impose artificial restrictions that force it to behave more like a human. The question of which configuration is more "fair" is kinda arbitrary; it depends on what your goal is.
Re: Grandmaster level in StarCraft II using multi-agent reinforcement learning
#98This is a really interesting one to digest. As with previous announcements about AlphaStar, much of the feedback (here and elsewhere) is about the fundamental challenge of assessing human vs. machine in an RTS. These points are very valid - stepping back however, this still feels like a pretty incredible accomplishment. I'm a gold league SC2 player, so maybe in the 30th-50th percentile. Three years ago, when DeepMind…
> The speed at which AlphaStar overtook previous AIs seems to me to be nearly unprecedented in AI research
In pretty much any field, top performing humans are at the physical limitation level, you will not see any sort of breakthrough, just incremental improvement.
Machines on the other side, can be scaled arbitrarily. Once you've built a small crane, you can build even increasing ones, it's just a function of money and interest.
Some say that intelligence is not like that, that it can't be scaled arbitrarily. But the burden of proof is on them, especially given the consequences if it can.
Re: Grandmaster level in StarCraft II using multi-agent reinforcement learning
#99Earlier quoted context omitted.
Or even a turn based game, where click speed does not matter...
If Google could tackle Civilizations god-aweful AI next, I'd be over the moon.
Do you mean Google should write a better AI than the one in Civilization?
Interesting. But also consider this: in casual strategy videogames -- actually, strike "strategy" and just consider videogames -- most players don't want a really hard opponent. A computer opponent that is really very hard to beat is not what we want, because that'd be frustrating and many of us play videogames (yes, even strategy games!) to unwind; we want the illusion of challenge, an opponent that is challenging to beat but within the possibilities of every person who buys the game.
Which, by the way, is also the case with Starcraft II. Most people who bought it aren't tournament players. They expect a challenge, but not a really hard challenge. Game difficulty is all about perception ;)
PS: I shamefully confess I reloaded my X-COM (DOS!) game every time I lost one of the soldiers I was emotionally attached to. I don't like losing! :P
Re: Grandmaster level in StarCraft II using multi-agent reinforcement learning
#100Why they wouldn't solve more impactful problems rather than playing games?
Because that's all they're capable of doing with it. Games can be easily repeated, have clear outcomes. There's no need for nuanced thinking or actual problem-solving. Just approximate the current scenario against previously successful strategies and pursue that strategy.
A real challenge would be to invent a new game, have both a human and an "AI" read the rules for the first time, and then compete in their first game together. Would the "AI" be able to formulate a winning strategy without historical data? If not, then it's not "AI."