If DOTA is anything like league I'm not sure I agree completely. I think in league there's more 'future prediction' needed, i.e. the current state is less immediate than in star-craft. In star-craft you can quickly see who is winning but in league there are things such as pushing lanes to consider and knock-on effects from later back timings (I know dota doesn't have backing but it has couriers?).
While that all can be extrapolated from current state I think starcraft is much easier to go for immediate gains by destroying more supply/resource value of units and extrapolate from there.
Starcraft strategy has a lot more weird nuance to it.
I noticed this building in this position at this time and I haven't been attacked by X unit yet, so he's probably doing strategy Y. I better skip some unrelated building I was going to make, so I can have an extra unit Z in case he's doing that strategy. Then I'll place the units at a particular spot to try to trap him because that unit will be vulnerable in this other spot so he's unlikely to move through that spot.
A SC:BW bot will have the ability to be perfectly aware of every unit in vision at all times.
It wouldn't be a suprise if some research team could put out a bot achieving superhuman victories purely by out-microing an opponent with minimal strategic choices.
>It wouldn't be a suprise if some research team could put out a bot achieving superhuman victories purely by out-microing an opponent with minimal strategic choices.
Yeah they did pretty much that. But the problem is it's a very brute-force approach and violates some rules of the game.
They jam thousands of commands per second into the game, and give each unit its own rudimentary AI. The units basically just dance at maximum range, magically dodge hits, etc.
If they limit it to 600 actions per minute (10 keystrokes hitting the keyboard every second - still beyond the human mind but beyond human fingers) it becomes a much harder AI problem.
Yeah, for people unfamiliar with starcraft bw: Whereas in other strategy games you may be able to improve the effectiveness of a unit 2-3x by micromanaging your units perfectly, in bw microing certain units perfectly can improve their effectiveness by something like 100x.
In the case of certain unit matchups, say, zergling versus vulture, the vulture should be able to kill an infinite number of zerglings given that it is microed correctly. However, despite the zergling being useless against a vulture on paper, In a human game you just don't have enough time to babysit your vultures with everything else going on so you end up seeing zerglings being used against vultures somewhat cost effectively even at professional levels.
> It wouldn't be a suprise if some research team could put out a bot achieving superhuman victories purely by out-microing an opponent with minimal strategic choices.
The chess equivalent would be letting Deep Blue take 10 years to evaluate each move; it's not a very interesting system anymore since it isn't playing under normal rules (~90 minutes per turn).
Any "real" SC AI will have limitations on input, say 300 actions per minute. It'd be pretty interesting to see how few actions per minute an AI could use to defeat the top human players.
While that all can be extrapolated from current state I think starcraft is much easier to go for immediate gains by destroying more supply/resource value of units and extrapolate from there.