I thought this was already happening. Right after AlphaGo beat Lee, I remember hearing about it. Did they give up on having their AI playing SC2? I wondered if that would work, since it seemed to take turns in Go at the same speed as a normal player, I wondered if it was trying to compute the most likely winning move each turn and the late game implications of those moves. If it tried that in a fast paced game how it would deal with the speed. It obviously would need to develop a pattern of pre-baked strategies that would win it the game. Would it play the same build every round or would it realize that changing things up each match wins it more games?
DeepMind and Blizzard Open StarCraft II as an AI Research Environment
271–280 of 283 posts
Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment
#272Novice here: I really want to try this Starcraft API but I don't know how to start. I believe this uses more reinforcement learning and agent-based models (which honestly I am not familiar with yet) What are good papers to get started on this?
Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment
#273Earlier quoted context omitted.
In what universe is taking 10,000 actions per minute an easier problem for a neural net than 100 such actions?
"In what universe is taking 10,000 actions per minute an easier problem for a neural net than 100 such actions?" StarCraft is precisely such universe. If you could micromanage units perfectly, you can do some amazing tricks. Here's an example of what I'm talking about: https://www.youtube.com/watch?v=IKVFZ28ybQs
That's a rules engine, designed to do basically one specific thing. It was told how to micro. In that context APM is a meaningless constraint. May as well ask how many times you can print a message in a for-loop per second. yes...quite a lot, and the computer is unfazed by the workload.
This is a different type of bot we're talking about here. A neural net could not learn to work with unlimited apm more easily than limited apm. That just doesn't make sense. That's like saying it's easier to compute 1000 hashes in a second than it is to compute one hash in a second.
Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment
#274It's not like this is going to create fantastic AI.
Keep in mind there's been an amateur AI project for broodwar for almost 7 years now. Even after such a long learning period, the games are very primitive, and the AI's still couldn't pose a threat to even a beginner human player. Sometimes the games take hours. Trying to build strategy and decision making into an AI is incredibly complicated. There have been teams working at the SSCAIT for many years now, and the product is still fairly primitive.
So what CA did was instead write up a simpler AI that mimics strategy and decision making. We all know it's not great, but I'd be really skeptical that 3rd parties would magically create an AI that can think strategically.
Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment
#275Earlier quoted context omitted.
> the APMs for humans are obviously averaged over a minute This is wrong. I don't know the exact number, but APM is averaged over around a second. I suspect this is done because APM is a more meaningful compared to APS, for humans at least. Here is a graph from Scelight that high lights this: https://goo.gl/photos/9cjNxDwWoB1pmWkg9
You still end up with a computer that can perform 3 actions all within the first millisecond of a second and still end up with human-like 180 APM (3 APS), even though no human could replicate what it just did.
How is that any different than someone who goes 100MPH and then 20MPH and claims to have not violated a 65MPH rule because the average speed was 60MPH for the trip?
The units of the measurement do not dictate how the measurement is made.
Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment
#276When I used to play a lot of StarCraft, and then later with Total Annihilation, I wished for the ability to customize the AI.
So then BWAPI came along ... and ... AI is hard. The best SCBW bots are still pretty pathetic compared to a human player, never mind an expert human player.
Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment
#277"so agents must interact with the game within limits of human dexterity in terms of “Actions Per Minute”." I am really glad they are limiting APM because otherwise things just get stupid.
Currently, though, APM isn't a real advantage for AIs. They're still too stupid.
Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment
#278Earlier quoted context omitted.
I know this isn't the exact same as the article, but when genetic algorithms were introduced to solve for build orders, the "seven roach rush" was in vogue, something that was unexpected at the time and "discovered" using GA. I think there is a space for finding strategies that have more leeway in execution and thus are more suitable for humans to pilot rather than have machine level micro.
Part of the problem SC2 units are not very balanced and each patch tries to make them "more balanced". SC2 design settled on making unique units at cost of balance. Roaches in fact are quite overpowered, with very fast regen and quick ranged attack(they move as fast as hydralisks). A versatile and low-cost unit(cheaper than a Hydralisk). And the reason they're so powerful, SCII units of other races in general are mor…
SC2 is very balanced when you look at the total game across a range of skill levels. Each unit is balanced around costs, attention, utility, requirements, and other units. Ex queens are awesome early game as they hit air, don't take gas or larva, and have high HP, but late game it's all about healing ultralisks plus anti air.
Roaches take a lot of supply, are armored so take more dammage from tanks and can't hit air which is their counter. They are also ranged so they don't share dammage upgrades with lings or ultralisks. Late game players will sacrifice them for more useful units and they need a larger investment in overlords.
PS: If you watch some high level games units generally have their time to shine as part of individual games progression.
Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment
#279Earlier quoted context omitted.
> It still uses a Monte-Carlo Tree Search to get to the level where it can beat human pro players. I'm not sure this was true of the first AlphaGo, probably wasn't true of the Sedol AlphaGo, and definitely isn't true of Master.
Do you have an article about this? I never heard that AlphaGo moved away from MCTS.
It uses MCTS, but that's not the same thing as the claim, now is it? If you look at the win rates in the AG paper for the NN vs MCTS+NN and then consider the performance curve, use of a single TPU, crushing superiority of Master's flawless 60 blitz matches and Ke Jie matches despite very fast moves, the released self-play matches, and comparing with FB's Dark Forest, it's clear that the AG NN all on its own, without any MCTS, is a truly formidable player that would likely crush many pros, although I don't know if it would reach Sedol or Ke Jie levels of play.
Re: DeepMind and Blizzard Open StarCraft II as an AI Research Environment
#280Earlier quoted context omitted.
I wonder if it's come too late. Couldn't it be the opposite? Blizzard was willing to do this release exactly because SC2 is dead?
Why would popularity be a detriment to this API?
They released a headless Linux port you can download for free. This wouldn't be something they'd contemplate for a game in its prime.