8 ms·

Starcraft is a much larger, more complex, more freeform game than Dota 2. It's like Go compared to chess.

by dxhdr 9y ago

cli 9y ago
I disagree with this (I used to play Warcraft 3, and currently play Dota 2), but that's beside the point. The Dota 2 OpenAI is only set up for one mirror matchup (impossible in real games) involving one hero on each side, in one lane, and only for the first 10-ish minutes. This is maybe 1% of a real Dota game.
- kazagistar 9y ago
  I think you are both correct. Starcraft has a far larger space of verbs at any given moment, and many of them can impact each other, giving it one form of complexity, while Dota2 clearly has a much more complex set of units and abilities, leading to more possibilities total, even if the number of possible actions moment to moment are more limited. But yeah, the bot was a teeny little bit of the game, impressive as it was.
sametmax 9y ago
A lot of complexity in dota comes from the interactions between 10 players. Make it 10 ai having to communicate using chat. Make them pick and ban. Make all objects available. And then you'll have real complexity. With hundreds of unit x objects, jungling, roshan, cd and pick + ban, you can actually get at the sc level of complexity.
- Retric 9y ago
  It's 10 players vs ~200 bots. So, SC is still a much more complex space. DoTA has non player bots, but they are similar to SC buildings and follow very simple rules.
  - sametmax 9y ago
    You can't compare the complexity of a dota char with a unit. Sc units are very basics, they don't have xp, don't have a skill tree, don't have have 100 of possible objects and their skills don't vary so much in context.
    - Retric 9y ago
      This matters far less than you might think. Go has simpler peaces without movement and it's still much more complex than chess.
      - sametmax 9y ago
        Not in Dota. In dota combinations open the road for brand new moves. Some items tp, some regenerates, some cancel buff, some critics, some cleave, some cut trees, some slow, some stun, some give visions, etc. Now in SC, you have 3, 4 main builds for a given match up. You see the building, and you know where this is going. In Dota, depending of the 10 heroes, current money and objects combination, and player skills, you may expect one build or another. Also, a zergling or 10 zergling is pretty much the same the same to consider from the behavior point of view. The number doesn't matter that much, only the intensity of the effect. And a gling will alway do the same thing. Move. Attach. Burrow. The same unit in Dota can have a completly different role depending of the context. My guess is that an AI would give you a much bigger advantage in SC because they can make more APM than a human, strat or not, while on Dota at high level strat is more important on the long run.
        Retric 9y ago
        SC 2 has ~100's of viable strategy's per race as openers, but like chess openers are just that. As the game evolves you get complex iterations and risk reward situations. Bluffing research is very much a part of high level play as min maxing requires you to force the other player to waste resources in any way possible. For example larva are one of Zerg's most valuable resource and there are several ways of attacking that resource by killing units or simply forcing them to go more defensive.
      - 9y ago
eterm 9y ago
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.
- cosarara97 9y ago
  OpenAI was playing 1v1
- SerLava 9y ago
  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.
  - roenxi 9y ago
    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.
    - dxhdr 9y ago
      > 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.

Starcraft is a much larger, more complex, more freeform game than Dota 2. It's like Go compared to chess.

by dxhdr 9y ago

Starcraft is a much larger, more complex, more freeform game than Dota 2. It's like Go compared to chess.

cli 9y ago

I disagree with this (I used to play Warcraft 3, and currently play Dota 2), but that's beside the point. The Dota 2 OpenAI is only set up for one mirror matchup (impossible in real games) involving one hero on each side, in one lane, and only for the first 10-ish minutes. This is maybe 1% of a real Dota game.

kazagistar 9y ago
I think you are both correct. Starcraft has a far larger space of verbs at any given moment, and many of them can impact each other, giving it one form of complexity, while Dota2 clearly has a much more complex set of units and abilities, leading to more possibilities total, even if the number of possible actions moment to moment are more limited. But yeah, the bot was a teeny little bit of the game, impressive as it was.

sametmax 9y ago

A lot of complexity in dota comes from the interactions between 10 players. Make it 10 ai having to communicate using chat. Make them pick and ban. Make all objects available. And then you'll have real complexity. With hundreds of unit x objects, jungling, roshan, cd and pick + ban, you can actually get at the sc level of complexity.

Retric 9y ago
It's 10 players vs ~200 bots. So, SC is still a much more complex space. DoTA has non player bots, but they are similar to SC buildings and follow very simple rules.
- sametmax 9y ago
  You can't compare the complexity of a dota char with a unit. Sc units are very basics, they don't have xp, don't have a skill tree, don't have have 100 of possible objects and their skills don't vary so much in context.
  - Retric 9y ago
    This matters far less than you might think. Go has simpler peaces without movement and it's still much more complex than chess.
    - sametmax 9y ago
      Not in Dota. In dota combinations open the road for brand new moves. Some items tp, some regenerates, some cancel buff, some critics, some cleave, some cut trees, some slow, some stun, some give visions, etc. Now in SC, you have 3, 4 main builds for a given match up. You see the building, and you know where this is going. In Dota, depending of the 10 heroes, current money and objects combination, and player skills, you may expect one build or another. Also, a zergling or 10 zergling is pretty much the same the same to consider from the behavior point of view. The number doesn't matter that much, only the intensity of the effect. And a gling will alway do the same thing. Move. Attach. Burrow. The same unit in Dota can have a completly different role depending of the context. My guess is that an AI would give you a much bigger advantage in SC because they can make more APM than a human, strat or not, while on Dota at high level strat is more important on the long run.
      - Retric 9y ago
        SC 2 has ~100's of viable strategy's per race as openers, but like chess openers are just that. As the game evolves you get complex iterations and risk reward situations. Bluffing research is very much a part of high level play as min maxing requires you to force the other player to waste resources in any way possible. For example larva are one of Zerg's most valuable resource and there are several ways of attacking that resource by killing units or simply forcing them to go more defensive.
    - 9y ago

eterm 9y ago

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.

cosarara97 9y ago
OpenAI was playing 1v1
SerLava 9y ago
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.
- roenxi 9y ago
  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.
  - dxhdr 9y ago
    > 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.