3 ms·

The goal was to have an AI defeat humans at SC2. Mission accomplished. Was it fair when AlphaStar tried getting at an air unit with all its land units and piss

The goal was to have an AI defeat humans at SC2. Mission accomplished. Was it fair when AlphaStar tried getting at an air unit with all its land units and pissed away a lot of time doing so, while Mana proceeded to obliterate its base? What if a future AI solves cancer from scratch in a day? Are we going to find reasons to call this 'AI vs Mother Nature' match unfair? Silly hoomans don't need to look for reasons why AlphaStar was playing unfair. It's just a game. And machines will be better at it. They'll be better at everything. Make your peace with that. DeepMind has only just gotten started. Like with Go. AlphaGo was trained for years. AlphaZero was trained for days. The latter obliterated the former. DeepMind is reading this criticism, don't you know. What do you think AlphaStarZero will be doing in a few years time? Crushing AlphaStar 100 out of 100 times, with low APM. That's what.

dragontamer 8y ago
> The goal was to have an AI defeat humans at SC2. Hardly. They put APM caps in there for a reason. We all know that a real research team going all out with 100,000+ APM would have zerglings dodging tanks and other super-human shennanigans going on. But that's simply not interesting. Nominally, AlphaStar set a 300 APM cap on the match. Apparently its a 300 APM average across the match however, which means the cap wasn't really effective at preventing "superhuman" conditions. Hopefully the next demo will better address that shortcoming and lead to more interesting matches. I think the main lesson of the demo is that "APM" is a poor metric for what they're going for. A new metric needs to be developed. Other posters downthread were talking about "eAPM" (Effective APM). A formalized version of the concept may help.
- Rallerbabs 8y ago
  >>The goal was to have an AI defeat humans at SC2. >Hardly. After cracking Go, DeepMind stated that it was going after SC2. This is recorded history, so don't even think about deluding yourself into thinking otherwise.
muh_gradle 8y ago
Stop it. No one is saying anything that you're saying. People are just disappointed because they wanted superior strategy and relatively equal game mechanics. We saw strategy to a certain degree, but it was spoiled by the lack of the latter.
- Rallerbabs 8y ago
  >No one is saying anything that you're saying. That's because my thinking is vastly superior and that is exactly the reason why you should listen to me. The fact that I told you to do so will ensure that you won't.
  - muh_gradle 8y ago
    nah
    - Rallerbabs 8y ago
      Told ya.

dragontamer 8y ago

> The goal was to have an AI defeat humans at SC2. Hardly. They put APM caps in there for a reason. We all know that a real research team going all out with 100,000+ APM would have zerglings dodging tanks and other super-human shennanigans going on. But that's simply not interesting. Nominally, AlphaStar set a 300 APM cap on the match. Apparently its a 300 APM average across the match however, which means the cap wasn't really effective at preventing "superhuman" conditions. Hopefully the next demo will better address that shortcoming and lead to more interesting matches. I think the main lesson of the demo is that "APM" is a poor metric for what they're going for. A new metric needs to be developed. Other posters downthread were talking about "eAPM" (Effective APM). A formalized version of the concept may help.

Rallerbabs 8y ago
>>The goal was to have an AI defeat humans at SC2. >Hardly. After cracking Go, DeepMind stated that it was going after SC2. This is recorded history, so don't even think about deluding yourself into thinking otherwise.

muh_gradle 8y ago

Stop it. No one is saying anything that you're saying. People are just disappointed because they wanted superior strategy and relatively equal game mechanics. We saw strategy to a certain degree, but it was spoiled by the lack of the latter.

Rallerbabs 8y ago
>No one is saying anything that you're saying. That's because my thinking is vastly superior and that is exactly the reason why you should listen to me. The fact that I told you to do so will ensure that you won't.
- muh_gradle 8y ago
  nah
  - Rallerbabs 8y ago
    Told ya.