Slashdot Mirror
StarCraft AI Competition Announced
bgweber writes "The 2010 conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE 2010) will be hosting a StarCraft AI competition as part of the conference program. This competition enables academic researchers to evaluate their AI systems in a robust, commercial RTS environment. The competition will be held in the weeks leading up to the conference. The final matches will be held live at the conference with commentary. Exhibition matches will also be held between skilled human players and the top-performing bots."
Let's teach our AI systems how to do battle... against humans. Skynet anyone?
Perhaps a game not so dominated by rushing tactics would be a better choice of base game? It definitely seems an interesting idea, but there must be games better suited to an AI contest like this...
Human Advantages:
Imagined Prediction Advantage
Flexible Stategies
Arguably Faster Learning
AI Advantages:
Able to command all units at once
Usually More efficient w/ resources
Instant Macro management
Korean Advantages:
Superior Strategies
Advanced Prediction
Flexible Tactics
Arguably Faster Learning
Able to command all units at once
Usually More efficient w/ resources
Instant Macro management
Fixed that for you :D
The aiide conference web site has been Slashdotted... even though Slashdot didn't link to it. :-)
Most game AI's are not well designed, but not because they can't be. Most game AI's are built from the prespective that the player should be able to win, therefore Grandmaster level thinking is less desirable than preditable patterns that seem impossible to be till the player realizes they can be exploited.
Instead of an AI that can win at Starcraft, maybe they ought to try to build an AI that can finish Starcraft 2.
Apparently, that's a much greater challenge.
You are welcome on my lawn.
Depends where you look. Last month's KESPA ratings (the latest, at least on TLPD) put Jaedong at #1 and flash down at #6. In fact, the last time he wasn't #1 in that ranking was March.
The competition starts NEXT October (ie 2010). It's still 2009 check a calendar.
RTFA. There are four competitions, and on the only "complete" game, all AIs have the completeMapInformation flag in the Broodwar API disabled. Therefore, fog of war is on.
Obligatory Soundbite Catchphrase
My AI would design its base to be a rough representation of a naughty picture on the minimap. Human players would always lose as they just let the AI build away to see the picture get a higher resolution.
It must be very difficult if you cannot click the link under "rules"
#
Programs that attempt to cheat will be disqualified
1.
Bots must disable the perfect information flag in tournaments 1,2 and 4
A player could feel more satisfied if it plays against a computer with the same knowledge and resources as a human player, because then it would have to play more like a human. With such cheats, the player will feel annoyed that the computer always attack when he is the weakest, without real knowledge, or can attack with twise the units he know is the maximum at a given time.
Worst. Poem. Ever.
RTFA.
The competition is being held by Expressive Intelligence Studio at an AI conference. Blizzard has nothing to do with this, AFAIK.
Abaddon: An Xbox 360 Indie game
(I apologize in advance for the lack of paragraph spacing. Slashdot appears not to recognize the carriage return/line feed from this browser/computer?) Most games(I dare say almost all AAA titles) don't have anything resembling actual AI. Including AI is very very expensive computationally, it simply isn't feasible for most of the lower-end consumer users. To get around this, most games include a large variety of playbooks that define how the computer opponent should build, what to build, when to attack, etc... Sometimes there are minimal elements of AI, such as "if (terran) skip zergling rush". But, by and large, the AI is simply following a set of rules of when/what to build. If you switch the mode to "hard", most games simply ratchet up the minerals/second income for the computer, or remove fog of war (all Blizzard games do this). If you wish to experiment for yourself the 'ORTS' engine is a near replica of StarCraft but fully open-sourced. (http://www.cs.ualberta.ca/~mburo/orts/) I believe there are multiple AI examples included (there used to be) so you can foray into the challenges presented by real AIs; computers that actually adapt their playing style to your own. As a warning, the engine does not abstract away details to make it easier (eg: there are unit collisions, writing a script to mine a patch of minerals effectively suddenly became much much harder). Disclosure: I am not affiliated with the ORTS engine directly, but I did take a class in my undergrad doing game AI on it.
Thats why StarCraft 2 was delayed, they don't even feel like writing the code anymore and want people to do it for free.
"They confiscated everything, even the stuff we didn't steal!"
I imagine that a computer's ability to control units with instant reflexes and frame precision will make AI Starcraft a completely different game from anything we've seen. Watch some Tool Assisted Speedruns and see how the gameplay of a person playing frame by frame transcends that a skilled human playing normally. Games are designed, tested, or balanced with the expectation that a player cannot press a button thirty times a second, anticipate the frame in which a projectile which hit, or issue commands at ten map locations at once. Without these limitations, the game can be broken and become something the designers never intended.
I expect the contestants to abuse lots of bugs and glitches exploitable only with frame-perfect control. For example, there's a known bug when about 1% of the time, a dragoon shot will miss a moving SCV (with no high ground or cover). If something like this can be consistently reproduced, the game will warp. Another rare bug has units become stuck while moving past each other, causing them to dart at ridiculous speeds in a perpendicular direction. This is likely reproducible, and could become the main mode of unit movement in this contest. Even if this doesn't work, there's probably a way to move faster by issuing rapid commands in a way that takes advantage of animations, since Starcraft ground unit movement speed is not hardcoded but animation-based.
Even without bugs, an AI could dance around ranged units to be basically invulnerable to melee, or any slower unit with lower range. This will give Terran a powerful rush strategy (how ironic).
Imagine a game of Terran versus Protoss. The Terran builds a fast barracks, and sends two three marines at the Protoss base. By then, the protoss has two zealots, but they won't matter, since they'll never get a hit. The Terran player dances the marines to shoot the zealots while taking no damage by always moving whichever two marine are being chasing, while the third is free to fire. Even against an equal-sized zealot force, marines are slightly faster that zealots, so they can shoot and move with impunity. The marines can slowly make their way into the Protoss base and behind the mineral line where they'll slowly wear down the mining probes' health, even as the Protoss makes focussing on any one probe impossible. The Protoss might try to get a surround with probes, though I think the marines will escape. Even if the Protoss fends off this rush, the Terran can have vultures before opponents have dragoons and wreak havoc with them.
Since zerglings are faster than marines, I'm not sure if this strategy would work in TvZ, though we might still see some epic bunker rushes.
Note that I don't think this distorted gameplay is bad, just different from human play. I rather like the ridiculous perfection, timing, and bug-abuse of tool-assisted speedruns, and look forward to seeing what the contestants come up with. I would love for a contestant to find a strategy that completely breaks Starcraft as we know it and wins unopposed. However, I think those who expect the final matches to look like really polished high-level human Starcraft play might be disappointed.
Allowed html is displayed below the comment form when you write a comment. Only the the following are allowed on slashdot:
"Trolls they were, but filled with the evil will of their master: a fell race..." -- J.R.R. Tolkien on Olog-hai
If you like TA, check out Spring at http://springrts.com/
This is an open source fully 3d replica of TA. They've now built it to the point where it is a base engine that can host one of several mods - mostly based on TA style models and concepts, although a few are completely unique. AI's are plugins that can work over several mods if the author chooses so.
My favourite is the Complete Annihillation Mod - http://springrts.com/wiki/Complete_Annihilation
The "chicken" mode has a weak AI, but enough brute force (attack waves) to keep you on your toes.
They ought to release it with dedicated servers and LAN play and roll out Battlenet 2.0 when it's ready, if that is indeed the case.
You are welcome on my lawn.
Blizzard is basically trying to replicate all of the major features of a service like Steam, all in one go- and no, WoW expertise will only translate to the development slightly, if at all, since the Battle.net team is completely separate from the WoW team. It's not as if they just reassign all of their programmers (which likely don't even specialize in the kind of things a battle.net programmer would have to do), and put WoW expansions and updates on the back burner.
While there are undoubtedly other factors at play, I believe you are grossly overestimating the effects of those factors. Development time for a new service is more than sufficient a reason for a delay.
If I recall correctly, they were talking about Battle.net 2.0 features in a "we're going to have it but it's not implemented yet" way at this year's BlizzCon. That was three months ago. Considering the fact that is Blizzard we're talking about, a four or five month development time for anything is fast.
The Guide is definitive. Reality is frequently inaccurate.