10 ms·

50% ?! Wow.. I’d imagine going against any decent human player that would be more like 10% success rate.. considering by the time workers go to other camp and r

50% ?! Wow.. I’d imagine going against any decent human player that would be more like 10% success rate.. considering by the time workers go to other camp and rushed in unison the other player would have +1-2 workers and would have to really suck at micro to lose that battle.

omni 8y ago
You're right, which means DeepMind must have substantially better micro than the Insane AI. Not sure how impressive that is, but I do remember the stock AI does register a ludicrous APM (600+) in replay mode.
- keerthiko 8y ago
  Good micro has a lot of room for tactics and outplaying where there's a lot more to it than having a high apm, so it's not really a surprise that Deep Mind can outplay the AI, even with a lower APM. Evenly matched armies looking for a micro-based edge turns basically into a fighting game like Street Fighter, where it's about reading your opponent, baiting them into a mistake, pre-empting them, taking decisive action, crisp reflexes/timing, doing something unexpected or unorthodox to get an advantage at a huge risk, etc.
larkeith 8y ago
> would have to really suck at micro to lose that battle I'm not sure there would need to be that large a skill difference, relatively speaking - worker micro has a really high skill ceiling, and players start with 12 workers instead of 6 nowadays, so 2 extra is a much smaller difference.
lanius 8y ago
Worker rushing is pretty much a guaranteed losing strategy. One progamer who did it in competition was criticized for being disrespectful toward his opponent and viewers and lost his tournament seed as a result.
- cm2012 8y ago
  Which player? I've followed SC2 for a while and never saw that.
  - lanius 8y ago
    NaNiwa in the 2011 GSL Blizzard Cup https://liquipedia.net/starcraft2/NaNiwa%27s_Blizzard_Cup_Controversy https://liquipedia.net/starcraft2/NaNiwa%27s_Blizzard_Cup_Co...
    - cm2012 8y ago
      Thanks for sharing!
    - Nuzzerino 8y ago
      Here's the video for anyone interested. Looks like Naniwa didn't put any effort into making the rush effective. https://www.youtube.com/watch?v=5roOPKm5lOw https://www.youtube.com/watch?v=5roOPKm5lOw
      - treis 8y ago
        The announcers said that the game didn't matter. I don't think Naniwa was trying to win, just get the game over with.
    - Bootvis 8y ago
      If I understand the article correctly the verdict would have been different if he had won the match or at least an interesting fight had resulted. So, I'd guess that if an AI plays this strategy well enough (somehow) it would not be a problem.
    - empath75 8y ago
      Yeah I remember that. That was just seen as him throwing the game, not a sincere attempt to win. He had a reputation for throwing temper tantrums.
- 8y ago
thefreeman 8y ago
I think you are severely underestimating the APM and micro ability of these bots. https://www.youtube.com/watch?v=udIA6uvWS2Y https://www.youtube.com/watch?v=udIA6uvWS2Y I know they have said they will be limiting the maximum APM of the bot to around 180 for the final bot but they may not have implemented it at the time this was happening. Edit: another example https://www.youtube.com/watch?v=IKVFZ28ybQs https://www.youtube.com/watch?v=IKVFZ28ybQs
- friendlybus 8y ago
  The second example is likely a 'cheat' because the splash damage would be applied the same frame as the shot hits the centre zergling. For an AI to dodge that it would have to have knowledge of which zergling was going to be hit before it was fired upon. The video creator likely used map triggers to simulate the effect.
  - EForEndeavour 8y ago
    Here's the original mention of that AI in a teamliquid forum thread started in 2011: https://www.teamliquid.net/forum/starcraft-2/210057-automaton-2000-micro-bot https://www.teamliquid.net/forum/starcraft-2/210057-automato... As far as I can tell, the AI must be cheating, at the very least to obtain map vision beyond what is available in a match. In order for the zerg AI to observe each tank's turret rotation and predict which zergling it's about to fire on, zerglings have a sight radius of 8, smaller than the tank's 11. In a fair match, the zerglings would be fired on before getting a chance to see the tanks. Even as the group of zerglings gets close enough to see the first tanks, other tanks start firing from beyond the zerglings' vision, so I'm not convinced this AI could pull this off in an actual match.
    - a1369209993 8y ago
      In the linked video around 1:07, you can see that they already have vision on the tanks when the first one rotates to fire. This could be the result of a cheat, but SC2 does include NPC watchtower buildings that can be captured to grant vision over a large area of the map (you could also send flyers to scout). From there it is (or at least can) 'just' ridiculously fast and precise angle-tracking and micro, so this is a situational tactic at worst.
ddeokbokki 8y ago
Pretty sure Insane AI plays better than most bronze-silver-gold league human players.

omni 8y ago

You're right, which means DeepMind must have substantially better micro than the Insane AI. Not sure how impressive that is, but I do remember the stock AI does register a ludicrous APM (600+) in replay mode.

keerthiko 8y ago
Good micro has a lot of room for tactics and outplaying where there's a lot more to it than having a high apm, so it's not really a surprise that Deep Mind can outplay the AI, even with a lower APM. Evenly matched armies looking for a micro-based edge turns basically into a fighting game like Street Fighter, where it's about reading your opponent, baiting them into a mistake, pre-empting them, taking decisive action, crisp reflexes/timing, doing something unexpected or unorthodox to get an advantage at a huge risk, etc.

larkeith 8y ago

> would have to really suck at micro to lose that battle I'm not sure there would need to be that large a skill difference, relatively speaking - worker micro has a really high skill ceiling, and players start with 12 workers instead of 6 nowadays, so 2 extra is a much smaller difference.

lanius 8y ago

Worker rushing is pretty much a guaranteed losing strategy. One progamer who did it in competition was criticized for being disrespectful toward his opponent and viewers and lost his tournament seed as a result.

cm2012 8y ago
Which player? I've followed SC2 for a while and never saw that.
- lanius 8y ago
  NaNiwa in the 2011 GSL Blizzard Cup https://liquipedia.net/starcraft2/NaNiwa%27s_Blizzard_Cup_Controversy https://liquipedia.net/starcraft2/NaNiwa%27s_Blizzard_Cup_Co...
  - cm2012 8y ago
    Thanks for sharing!
  - Nuzzerino 8y ago
    Here's the video for anyone interested. Looks like Naniwa didn't put any effort into making the rush effective. https://www.youtube.com/watch?v=5roOPKm5lOw https://www.youtube.com/watch?v=5roOPKm5lOw
    - treis 8y ago
      The announcers said that the game didn't matter. I don't think Naniwa was trying to win, just get the game over with.
  - Bootvis 8y ago
    If I understand the article correctly the verdict would have been different if he had won the match or at least an interesting fight had resulted. So, I'd guess that if an AI plays this strategy well enough (somehow) it would not be a problem.
  - empath75 8y ago
    Yeah I remember that. That was just seen as him throwing the game, not a sincere attempt to win. He had a reputation for throwing temper tantrums.
8y ago

thefreeman 8y ago

I think you are severely underestimating the APM and micro ability of these bots. https://www.youtube.com/watch?v=udIA6uvWS2Y https://www.youtube.com/watch?v=udIA6uvWS2Y I know they have said they will be limiting the maximum APM of the bot to around 180 for the final bot but they may not have implemented it at the time this was happening. Edit: another example https://www.youtube.com/watch?v=IKVFZ28ybQs https://www.youtube.com/watch?v=IKVFZ28ybQs

friendlybus 8y ago
The second example is likely a 'cheat' because the splash damage would be applied the same frame as the shot hits the centre zergling. For an AI to dodge that it would have to have knowledge of which zergling was going to be hit before it was fired upon. The video creator likely used map triggers to simulate the effect.
- EForEndeavour 8y ago
  Here's the original mention of that AI in a teamliquid forum thread started in 2011: https://www.teamliquid.net/forum/starcraft-2/210057-automaton-2000-micro-bot https://www.teamliquid.net/forum/starcraft-2/210057-automato... As far as I can tell, the AI must be cheating, at the very least to obtain map vision beyond what is available in a match. In order for the zerg AI to observe each tank's turret rotation and predict which zergling it's about to fire on, zerglings have a sight radius of 8, smaller than the tank's 11. In a fair match, the zerglings would be fired on before getting a chance to see the tanks. Even as the group of zerglings gets close enough to see the first tanks, other tanks start firing from beyond the zerglings' vision, so I'm not convinced this AI could pull this off in an actual match.
  - a1369209993 8y ago
    In the linked video around 1:07, you can see that they already have vision on the tanks when the first one rotates to fire. This could be the result of a cheat, but SC2 does include NPC watchtower buildings that can be captured to grant vision over a large area of the map (you could also send flyers to scout). From there it is (or at least can) 'just' ridiculously fast and precise angle-tracking and micro, so this is a situational tactic at worst.

ddeokbokki 8y ago

Pretty sure Insane AI plays better than most bronze-silver-gold league human players.