UCSC Students Creating Killer StarCraft Bot
Two PhD students are working on an AI that will kick your butt in StarCraft.
GamePro has an interesting article about two PhD students at UC Santa Cruz who are working on an artificial intelligence (AI) that could take down the very best StarCraft player in the world. The project is just a portion of the University's overall goal in exploring the intersection of artificial intelligence, art, and design.
“Our goal is to create compelling new forms of interactive art and entertainment that provide more deeply autonomous, generative and dynamic responses to interaction," said Michael Mateas, the associate professor behind the project. "A major thrust of this work is advanced A.I. for video games, including autonomous characters and interactive storytelling.”
Currently the bot plays Protoss exclusively, and so far manages to rack in a 20-percent win ratio when playing against humans. Ben Weber and Peter Mawhorter, the two students executing the AI project, said that StarCraft was the perfect choice for the AI because of the active community.
"In South Korea, hundreds of professional gamers actively participate in tournaments such as the MBC and OGN star leagues," Weber said. "This community generates a large amount of replays that are available for building bots. It also generates a large number of interesting StarCraft matches to watch. StarCraft is played all over the world and there are several active community websites. It's easy to find players interested in playing the EISBot and our bot has already played against players in over 30 countries."
Mawhorter said that they also chose StarCraft because of its complexity. "StarCraft has three distinct races and is a well-balanced game; in any particular match up in StarCraft, there are a large number of strategies that are valid," he said. "Despite being over 10 years old, StarCraft does not have a dominant strategy. StarCraft has an active meta-game in which popular strategies are constantly evolving. This factor adds to the challenge of building strong computer opponents."
To learn more about their StarCraft-playing AI, check out GamePro's article here. Below you'll find the AI fighting against a human player named "PandaBearGuy."
- 200GB Toshiba HDD Built for Automobiles
- Nvidia Custom-Made the 320M IGP for Apple
- ATI Catalyst 10.4 Preview Driver Now Available
- Deals for April 15: StarCraft 2 Collector's Edition
- Intel Light Peak ''Launching'' in 2011, Maybe 2012
- New Patriot SSDs Use Latest JMicron Controller
- Steve Jobs' Smoke and Mirrors on 13'' MBP's CPU
- Intel Reports Best. First. Quarter. Ever.
- Intel Sampling Nehalem-successor ''Sandy Bridge''
- Hacker: Microsoft More Secure Than Apple, Adobe
- AMD Making Money, Sells 6 Million ATI DX11 GPUs
- Sony Outsourcing Vaio Laptop Design to Others
- Apple Mods Intel Chipset for Auto GPU Switching
- No Surprises Here, But Crysis 2 Looks Best on PC
- Report: Apple Looking to AMD for Future Chips
- Why Apple Fans Hate Tech Reporters
- Did Steve Jobs Steal The iPad? Genius Inventor Alan Kay Reveals All
- Office 2010 Hits Release to Manufacturing Milestone





pHD students and they're only at a 20% win ratio? I hope they just started...
pHD students and they're only at a 20% win ratio? I hope they just started...
Considering it's playing against the world's best players it's a good start.
A bot that will Zerg rush me? How clever.
i take it that they never heard of a zergling rush
i take it that they never heard of a zergling rush
You got rushed if you are bad or take the risk and expanded, else the rushing guy will usually lag in econ.
Anyway, it will be a interesting project. Consider the complexity of the game, a 20% win is good. I wonder if the AI takes in considerations when a terran is shelling over a cliff and try to lure the toss army in for an ambush.
Why don't just add maphack to ai eh, it's must be more easier.
I wonder what the build in computer would score cause 20% seems kinda low. And this article doesn't specify whether the 20% win was against average (random) players or professional ones.
pHD students and they're only at a 20% win ratio? I hope they just started...
The implication from this statement shows you don't really know anything about PhD research in engineering/science-related fields. PhD students aren't necessarily geniuses... Their certainly smart, but doing research doesn't involve conquering everything you touch. Research involves dedicated work towards reasonable goals set by a thesis. These goals constantly evolve as the research matures. It's a learning process, not a process of conquest.
Considering it's playing against the world's best players it's a good start.
"It's easy to find players interested in playing the EISBot"
I missed the part where they were designated as the world's best players...
Try make a bot for yourself, then talk.
Anyway, what they're doing is certainly impressive. I hope that they won't resort to the kind of cheap cheats developers are using nowadays instead of real AI.
some of the things the bot is doing is outright unfair, no human can click that fast.
Of course it's not easy. A bot that plays against a computer, like in World of Warcraft is easier than a bot that plays against a HUMAN brain.
However, 20% is a just a number. We need #s of win and loses against what faction.
Also they said the bot is protoss exclusive, why is the video only shows Terran?
Actually, the Panda Bear Guy is a nickname for the critter that you see in that video. It gets blown up at :17. The AI is playing against the default Starcraft AI (see the video description).
Cute, but it's not going to be beating top players anytime soon.
some of the things the bot is doing is outright unfair, no human can click that fast.
Yeah if they are going to make an AI thats going to beat the best player in the world then it should be based on skill not because a human player couldnt possibly click or move the mouse as fast as a computer. A computer AI will be able to micromanage at any speed. The player should have the option to play it in slow motion if he so chooses to level the playing field.
Considering it's playing against the world's best players it's a good start.
Nowhere in the article does it say they've played against the best, it says they've played against players in 30 different countries, that is all.
This is for the contest EIS at UCSC is hosting.
http://eis.ucsc.edu/StarCraftAICompetition
There are lots of entries that deserve stories as well. It will be more interesting to see who wins.
The implication from this statement shows you don't really know anything about PhD research in engineering/science-related fields. PhD students aren't necessarily geniuses... Their certainly smart, but doing research doesn't involve conquering everything you touch. Research involves dedicated work towards reasonable goals set by a thesis. These goals constantly evolve as the research matures. It's a learning process, not a process of conquest.
EDIT: They're* God, my grammar is awful.
Nowhere in the article does it say they've played against the best, it says they've played against players in 30 different countries, that is all.
yeah, and 30 countries implies at least 29 countries that are not Korea, hence not the best
Hmmm, I'm neither a PhD student, nor artificially intelligent, but I very much doubt that I would achieve a 20% win ratio in online StarCraft. :-)
I would be interested to know if they are trying to write an algorithm specifically tailored to be effective at StarCraft, or whether they are writing a more general purpose learning algorithm and simply testing its effectiveness on StarCraft.
I'm actually impressed.
But, is every single unit individually micromanaged? Something that would be generally impossible for a human to do when controlling dozens of units?
I suck at RTS games because I can't control any more than 2 units at a time effectively...
Keep up the good work.
Seems like an interesting AI idea. I seen the one with the Terran but it just looked as if the A.I. was Group attacking within the Group which is very useful to do if you can mirco that far down.
What seems interesting the most out of this, is. If they start in on Making different and better A.I. system that allow reduce these mirco-management , then will say on the players side of features, will there be ways to incorporate a way to deal with such advanced A.I. ?
Very interesting though. I dont play StarCraft but still.
I have a feeling the war is coming to an end, it will be soon a GG for Korea. What's next finger augmentation for StarCraft II?
Lol I can only imagine when it's 100% win ratio.
Yeah if they are going to make an AI thats going to beat the best player in the world then it should be based on skill not because a human player couldnt possibly click or move the mouse as fast as a computer.
There are some RTS players that would rival a computer in click speed.
Remember that current generation AI has an inherent disadvantage that the human does not: It can only do what it is programmed to do with some limited "learning" capacity (which is also programmed so I suppose it's not even worth mentioning). The human can randomise strategies far more easily because unlike software we can think outside the box. It takes time for the AI to adapt and it will never predict exactly what the human player is going to do with perfect accuracy even though it may be able to respond to a perceived threat with very little delay. Computers are perfectly logical, humans are not.
Pure speed does not necessarily overcome skill. Existing AI is also very fast, yet has Buckley's chance of beating a seasoned player in any game in any genre. The gap will continue to close but until we reach the Singularity it will not be a perfect match.
Ok some of the people posting have never played in Korea obviously. Back when I lived in the USA I thought I was pretty bad a$$ at starcraft. Very few people could beat me, then I took a job in Korea and holy sh!t. The average 12yr old over here can kick my a$$ so completely that I feel like a noob. These guys have memorized every keyboard command and shortcut imaginable. They click on units with one hand and hit keyboard keys with another at lightning speeds, its almost impossible to follow.
Skynet
To the folks saying no human can click that fast you've clearly never seen the replays of folks getting 600+ average apm. Granted a lot of that is spam but if you watch the replays carefully on slow motion you see so much logic during battles it's scary.
S Korea actually has a dedicated TV channel where all they show is SC replays, so I've heard, I've never been there. It's amazing that such an old game has such huge following still.
To the folks saying no human can click that fast you've clearly never seen the replays of folks getting 600+ average apm. Granted a lot of that is spam but if you watch the replays carefully on slow motion you see so much logic during battles it's scary.S Korea actually has a dedicated TV channel where all they show is SC replays, so I've heard, I've never been there. It's amazing that such an old game has such huge following still.
RLY??? 600 apm is 10 clicks in 1 second. I don't think it's possible with just clicks though. You have to move the mouse around to get in different positions during that time. So probably they use the hotkeys along with the clicks, and many of those clicks are not needed like clicking on the same ground couple of times, redundantly. If a player with that way high apm can optimize what they do with them it would be HORRIFIC to play against them. I can usually do 4 clicks a second (optimized) to move units around, click unit, move, click unit, move. That is not really 400 apm avarage because I don't do that for most of the time. OR it could be that the high apmrs just smash the keyboard with random buttons lol.
Yeah now I want to move to Korea, but we should really create a bigger threat to them, making a USA team and of course have a TV channel with SC2 replays lol - SC1 is oooooold and agin lol.