After watching Oriol & Dario on the Pylon Show, I now feel that my previous statement was a bit too strong. It’s true that Alphastar doesn’t scout as much, but it certainly does seem to scout in the early game. They showed an excellent example of Alphastar specifically checking for a lair, examples where it attempted to hide a dark shrine (unsuccessfully), and they spoke about ways in which it would explicitly scout for various building types. They did follow this up by saying that Alphastar is definitively not as good at scouting as it is at other things, and that its deficiencies at scouting caused it to underperform at other strategies. But my earlier claim of it not scouting buildings at all is certainly incorrect. It scouts, but not well, at least not in comparison to the other things it does at the pro level.
I think the AI just isn’t sure what to do with the information received from scouting. Unlike AlphaZero, AlphaStar doesn’t learn via self-play from scratch. It has to learn builds from human players, hinting at its inability to come up with good builds on its own, so it seems likely that AlphaStar also doesn’t know how to alter its build depending on scouted enemy buildings.
One thing I have noticed though from observing these games is that AlphaStar likes to over-produce probes/drones as if preempting early game raids from the enemy. It seems to work out quite well for AlphaStar; being able to mine on full capacity afterwards. Is there a good reason why pro gamers don’t do this?
It’s been two years since AlphaStar used anti-harassment worker oversaturation to great effect, and, as far as I’m aware, not a single progamer has tried it in a GSL match since then. This morning the last GSL match of 2020 marks the 1003rd GSL game since AlphaStar’s demonstration, and (to my knowledge) the only games where workers were overmade were when there was a significantly different reason for doing so (being contained, maynarding workers, executing a timing attack). The consensus among SC2 pros appears to be that AlphaStar was wrong to overmake workers as an anti-harassment measure, despite significant community discussion in early 2019 about its potential.
After watching Oriol & Dario on the Pylon Show, I now feel that my previous statement was a bit too strong. It’s true that Alphastar doesn’t scout as much, but it certainly does seem to scout in the early game. They showed an excellent example of Alphastar specifically checking for a lair, examples where it attempted to hide a dark shrine (unsuccessfully), and they spoke about ways in which it would explicitly scout for various building types. They did follow this up by saying that Alphastar is definitively not as good at scouting as it is at other things, and that its deficiencies at scouting caused it to underperform at other strategies. But my earlier claim of it not scouting buildings at all is certainly incorrect. It scouts, but not well, at least not in comparison to the other things it does at the pro level.
I think the AI just isn’t sure what to do with the information received from scouting. Unlike AlphaZero, AlphaStar doesn’t learn via self-play from scratch. It has to learn builds from human players, hinting at its inability to come up with good builds on its own, so it seems likely that AlphaStar also doesn’t know how to alter its build depending on scouted enemy buildings.
One thing I have noticed though from observing these games is that AlphaStar likes to over-produce probes/drones as if preempting early game raids from the enemy. It seems to work out quite well for AlphaStar; being able to mine on full capacity afterwards. Is there a good reason why pro gamers don’t do this?
It’s been two years since AlphaStar used anti-harassment worker oversaturation to great effect, and, as far as I’m aware, not a single progamer has tried it in a GSL match since then. This morning the last GSL match of 2020 marks the 1003rd GSL game since AlphaStar’s demonstration, and (to my knowledge) the only games where workers were overmade were when there was a significantly different reason for doing so (being contained, maynarding workers, executing a timing attack). The consensus among SC2 pros appears to be that AlphaStar was wrong to overmake workers as an anti-harassment measure, despite significant community discussion in early 2019 about its potential.