EPTBoxeR: "AlphaGo won't beat humans in StarCraft" - Page 7
| Forum Index > SC2 General |
|
Dianchie
Canada10 Posts
| ||
|
Liquid`Nazgul
22427 Posts
| ||
|
ETisME
12631 Posts
Google is choosing the game that has most exposure, big community members of cause have to jump to make big claims | ||
|
Pwere
Canada1557 Posts
From a purely theoretical point of view, SC is not an interesting challenge for an AI programmer/designer, so I doubt the resources will be dedicated to this for a while, and by then it won't be all that impressive. If they plan to adapt a general AI to play Starcraft, then that is a different challenge, but the outcome still comes down to where they draw the mechanical/perceptual line. | ||
|
Cluster__
United States328 Posts
| ||
|
Liquid`Nazgul
22427 Posts
On March 13 2016 10:45 Pwere wrote: Discussing a Starcraft AI is barely better than discussing a CS:GO AI. You first need to limit the huge mechanical advantage before you can discuss what is and isn't possible. From a purely theoretical point of view, SC is not an interesting challenge for an AI programmer/designer, so I doubt the resources will be dedicated to this for a while, and by then it won't be all that impressive. If they plan to adapt a general AI to play Starcraft, then that is a different challenge, but the outcome still comes down to where they draw the mechanical/perceptual line. Totally agree. It doesn't make any sense to beat top pros with limited strategy and perfect execution and then brag about it from an AI perspective. A waste of time in terms of prestige. | ||
|
necrosexy
451 Posts
On March 13 2016 09:35 thePunGun wrote: What you don't see is, when an AI scouts it knows instantly the tech, which type of units, their amount(including the current worker count) and what kind of strategies are possible. Whereas a human is 1. not able to identify the type and quantity of units and 2. most likely will not have a database of every strategy ever up to this point, the timings of those and the correct counter measures. An AI has no center of attention like we humans do, it does not need so called awareness like us humans. It sees what is and what is not in an instant and does not question itself or its decisions. I know the AI will determine all the possible build orders from a scout. But scouting ends, so the AI's determination of what the human can be true or false (human deceptivness). You seem to be implying the AI will always have scouting information at every point in the game. | ||
|
DuckloadBlackra
225 Posts
On March 13 2016 10:57 Liquid`Nazgul wrote: Totally agree. It doesn't make any sense to beat top pros with limited strategy and perfect execution and then brag about it from an AI perspective. A waste of time in terms of prestige. Exactly. | ||
|
Kyir
United States1047 Posts
| ||
|
sc2chronic
United States777 Posts
On March 13 2016 09:12 Scarlett` wrote: you're : not talking about the same game assuming the ai has full map vision assuming the game is about 1 fight and who has a better army wins ![[image loading]](http://msnsmileys.net/R/ROFL/rofl-3e.gif) | ||
|
Chocolate
United States2350 Posts
On March 13 2016 11:48 Kyir wrote: Since AlphaGo was specifically designed for Go, saying it won't win at Starcraft is probably a safe assumption. They mean the algorithm / procedure behind the AI, not the specifically trained AI for go | ||
|
stevorino
957 Posts
| ||
Legionnaire
Australia4514 Posts
- Perfect mining at every base. Which humans dont do because we can't. - Mass micro of scout units. Constantly running around the edge of the base etc? Half the time humans dont even look as they are microing other things. - While perfect resource spending. So many games i finish a battle and i have 1k+. - While perfect micro, at both an early game, and late game mass army stage? Seriously. - Worried about mine drops (sc2)? Perfect runaways. Constant mining at mineral patches that are outside of the range of the mine. Perfect timing of suiciding a probe to allow the others to mine for 30 seconds. There are just so many instances where AI would dominate at a level far in excess of human endevour. AI's true advantage is just any unit with range, which is where the heaviest micro element comes into play during battles. But even so, on a simple level, just utilizing a perfect number of stalker attacks on a single target without wasting unneeded firepower. Every single battle from the start of the game would add a 20%+ benefit to the AI. Which would then keep snowballing out of control. Perhaps the perfect strategy Boxer was talking about would be his rax first bunker rush. I'd pick terran, or toss as an AI. The link nazgul posted shows why. Range micro ftw. Rines or Bink stalkers. | ||
|
The_Masked_Shrimp
425 Posts
They don't want the machine to win with prestige, that's totally out of subject for the target. They will first try to make an AI that can beat a pro, even if it means it knows what units you build and can see through fog of war. Then they will increment on it and make it more humanlike one step at a time. | ||
|
MyLovelyLurker
France756 Posts
An AI victory in that context would arguably be much more meaningful. | ||
|
EngrishTeacher
Canada1109 Posts
Reading a few pages of replies, my view that perfect micro/macro will beat "strategy" and "overall awareness" every single time has been cemented. Exactly what are the main difficulties encountered by AI nowadays? Imagine this: Let's suppose someone creates an AI that opens CC first or rax-CC every game, then macros and micros perfectly of course. No deep strategic awareness or even variations in BO are needed, the AI just need to CONSTANTLY poke and pressure with bio, especially after medivacs are out on the field. In its very essence, SC is a game of economy, and as long as the AI maintains a 1 base deficit or less depending on the matchup, I just don't see how the human pro can keep trading efficiently vs. perfect mechanics. Surely, eventually the army value lost will favor the AI in the extreme? Could someone link me to some past AI vs. Pro games? I'd be really interested in seeing how the AIs are currently losing. | ||
|
MyLovelyLurker
France756 Posts
On March 13 2016 12:41 EngrishTeacher wrote: I've followed SC since the early OGN BW days, yet I've never understood why the SC AIs (both BW and SC2) are still so terrible vs. a real human opponent. Reading a few pages of replies, my view that perfect micro/macro will beat "strategy" and "overall awareness" every single time has been cemented. Exactly what are the main difficulties encountered by AI nowadays? Imagine this: Let's suppose someone creates an AI that opens CC first or rax-CC every game, then macros and micros perfectly of course. No deep strategic awareness or even variations in BO are needed, the AI just need to CONSTANTLY poke and pressure with bio, especially after medivacs are out on the field. In its very essence, SC is a game of economy, and as long as the AI maintains a 1 base deficit or less depending on the matchup, I just don't see how the human pro can keep trading efficiently vs. perfect mechanics. Surely, eventually the army value lost will favor the AI in the extreme? Could someone link me to some past AI vs. Pro games? I'd be really interested in seeing how the AIs are currently losing. This strategy goes away if the AI learns by playing against itself. It's that process which is interesting, much more so than creating an ever-winning, 6-pool + 1000 APM bot. | ||
|
fishjie
United States1519 Posts
I'm going to say yes. AI has been drastically improving by leaps and bounds. In the machine learning world, the switch from statistical methods of machine learning to neural nets (which are vaguely modelled on our limited understanding of our human brain) is happening. Often times due to lack of training data in problem fields such as natural language processing, statistical methods are used to generate the training data. Either way, the results speak for themselves. Image search in google photos (without any tagging) is disturbingly good. Self driving cars. Watson trashing Jennings. Alpha go destroying the world's best player. So on and so forth. Be afraid. Mass unemployment is just one of the dangers of this runaway AI. | ||
|
Legionnaire
Australia4514 Posts
On March 13 2016 12:41 EngrishTeacher wrote: I've followed SC since the early OGN BW days, yet I've never understood why the SC AIs (both BW and SC2) are still so terrible vs. a real human opponent. What humans like is something that plays; - a fraction above their current ability. (So you get a challenge and feel like you've done something worthwhile when you win) - is humanlike (can be unpredictable, yet it will still reacts to what you are doing) so that it gives you a challenge. - Yet its also something that is beatable, else whats the point of playing then? (Think TA:Escalation where hard comp mode has infinite resources and if you dont kill it with a commander rush you die after 5 mins). Yet from a companies prospective, this is so damn hard and expensive to do. It's far easier to make 'hard AI' by simply cheating and giving them map awareness or money or something. (Think Civilisation 4. They add difficulty by just making the AI build faster, research faster, and start with more units.) This takes a day of effort to make instead of weeks/months of design/ build time by a team of programmers. Even if a company goes to all of that effort to make good AI. Someone can still find a flaw in the AI which means you will always win. (scv rush and attack the enemy CC, then run away and have all the scvs follow you etc etc) It is a fine line for companies to walk. But i do agree, i wish they had something better for good players to play against. Considering the expense of making real AI. Combined with the fact most players are bad and would prefer to have a 20min BGH no rush game so they can build up and move out at their own pace. You can see why game dev goes the way it does. The top 5% will never really be happy with any AI that can be built. Besides, that's what multiplayer is for  | ||
|
EvilTeletubby
Baltimore, USA22258 Posts
| ||
| ||
