High quality AI to come with Forged Alliance?
This isn’t as interesting as it might be if high quality AI mods like Sorian’s AI didn’t exist, but it seems likely that the default AI will be receiving a substantial overhaul when Forged Alliance hits shelves. There is an E3 report over at shacknews which talks about this and other things related to the game. While the author is obviously quite biased against SupCom, he does make some good points that are worth reading.
Most interestingly, we get an admission from Jeremy Ables, GPG’s producer, that the default SupCom AI is “basically crap”. I couldn’t agree more. Anyone who has toyed around with the supposed “Supreme AI” included in the game knows how trivial it is to beat. Even moderately skilled players will have little trouble defeating several of these AIs.
According to the article, however, this stands to change with Forged Alliance. Supposedly, the AI has been “rewritten from the ground up”, to not only take up less processing power than previous iterations, but also be far more deadly. In addition, the new AI can apparently defeat the old one is less than 10 minutes. I remain cautiously skeptical, but hopefully, this new AI will prove more useful and challenging than the old.
July 13th, 2007 at 9:56 am
Definitely good to hear. I remember when I first grabbed the demo, Supreme AI beat me on Finn’s Revenge the FIRST time I ever played, then never again. What I hate about the AI is even if you let them sit there with half the map for 30 minutes, they won’t establish much of an attack force. They will sometimes go for nukes, and they usually shield up their base pretty well, but I’ve never seen them make any defenses that 20-30 SAB’s could crack. Not to mention they always just send their ACU out to its death even when they have the defenses/force to destroy your attacking force. I’ve never messed with any third-party AI’s, SupComTalk should do a brief review of one with a link and basic installation instructions.
July 13th, 2007 at 10:05 am
Baddox: That’s a good idea, and one I’ve been thinking about for a while. Look for it in the near future.
July 13th, 2007 at 10:48 am
That is VERY good news. Like Baddox, I picked up the demo. Having read everything possible about SupCom before hand (dont ask, I do it with about every game I buy) and having watched Last Commander Standing on Gamespot, I went with medium AI. I didn’t try to win the first time, I essentially just screwed around, but somehow, through screwing around, I had a Scathis and was bombarding targets. He knocked me out with tactical launchers (to my eternal shame)
The next time I played, I bomber rushed him. He died. I played on Hard AI and bomber rushed him. He died again. Now, when I play against AI, all I do is screw around, because I know that if I try, I can totally kick his ass.
July 13th, 2007 at 12:07 pm
The problem with all the AIs (even Sorians) is that while they do attack they only attack with a handful of units. When you reach T3 you don’t attack with 8 SABs, you attack with 30+, ditto for ships. If they new AI builds up armies and then pounds you with everything at the same time it will be much better.
July 13th, 2007 at 12:29 pm
The Sorian AI does not impress me. I’ve tried a few versions here and there and the only noticeable effects it had were less expansion, less mass use, and no more (or less) suicidal ACU runs. I guess it’s worth it just for that last bit, but I honestly felt the Sorian AI was easier because it was terrible at getting any decent amount of stuff built.
July 13th, 2007 at 1:11 pm
With the proper AI, the computer should have no problem beating any player. ANY player, pretty consistantly. A good AI should be able to micro everything at once, scout and adapt to information and change 100 build orders and move orders simultaniously. I’d really like to see AIs in FA range from extremely easy to near impossible, with other options in between.
July 13th, 2007 at 2:20 pm
Breezy, I definitely disagree that a good AI could beat ANY player. The complexity of the game means there’s no simple formula for winning, or even for countering a given tactic. Even chess programs can’t reliably beat chess masters, and chess is a 2d game with a set (and small) number of “units” and possible positions for them to move. The game Go, from what I hear, while also 2-dimensional, is so complex that a mediocre player can beat the best computer programs on a medium or large game board. That said, most RTS AI’s never give you “Holy cow, that was brilliant!” moments, but most other ones are at least challenging for a dozen or so games to a new player. Heck, even Age of Empires 2 Hardest AI could lay the smack down, until they ran out of gold and kept spamming skirmishers. But that’s off topic :-)
July 13th, 2007 at 2:44 pm
i find it ?ironic? that GPG has hyped a better AI and a new UI, two of the most downloaded mods that players had to make because GPG left them out of the first game. good job GPG! except not.
Sorian Rush felt more like a real person fighting me than the standard AI, but still refused to do anything other than spam t1 bots.
July 13th, 2007 at 3:32 pm
Beating it in less than 10 minutes is an ABSOLUTELY WORTHLESS STATEMENT. Anyone can change a couple values and create an AI that crushes the stock AI in less than 10 minutes. Ugh.
July 13th, 2007 at 3:56 pm
The thing with AI’s is it’s useless to make an AI specifically to beat another AI–that’s easy, since you pretty much tailer your AI to perfectly beat the techniques of another AI. GOOD AI has to be able to adapt, especially when facing humans. I’m not a programmer, but it would seem like a good AI for SupCom would have to understand the given map, and prioritize certain spots on the map as the places most necessary to take over and then defend.
July 13th, 2007 at 6:45 pm
One thing I’d love to see is an AI that can pull off sneaky maneuvers. Just imagine a tacsniping AI, or an AI that does an engineer drop at 5:00 into the game, or one that builds a Soul Ripper, a screen of T1 bombers, and strat bombers, and perfectly coordinates their arrival times so that not a single SAM targets the strat bombers (unless microed).
And as mentioned, an AI could theoretically out-econ any human player with superior micro (think PG PG FAB on repeat, building fabs with no fear of fabtard), not to mention dancing its ACU/LABs to near invulnerability. Although, come to think of it, since we’re getting build templates, we’ll also get to easily lay down PG-PG-Fab templates ourselves.
July 13th, 2007 at 8:18 pm
“quote: a good AI for SupCom would have to understand the given map, and prioritize certain spots on the map as the places most necessary to take over and then defend.”
this is allready done by the map maker now if the map maker stuffs this up then well the AI will just play poorly…
go onto the vault and grab fields of isis AI inhanced map and play the AI on that… the AI plays a lot better…
as for the reviewer he needs to be SHOT… he has a point we where promised so much and delivered so little… but to constantly bag out on it wont help the problem… and to end the review with
my list of complains is getting shorter with what i see in FA - or something simlar
i think is just a RUDE and offensive comment to make… personaly if i was on GPG Marketing team i would boycot that review company from reciving any futher new release stuff…
and HONEST review is fine… but a nasty one is just spitefull
July 14th, 2007 at 12:56 am
I fail to see why anyone would even want to play against AI?
July 14th, 2007 at 2:12 am
I fail to see why anyone would want to make a smurf
July 14th, 2007 at 2:47 am
There’s a reason it’s called “fail.” J/K
July 14th, 2007 at 6:19 am
Unconquerable: I have no idea, probably because they’re faggot newbs that don’t know how to have fun making smurfs and earning points on the ladder.
July 14th, 2007 at 12:12 pm
If an AI was actually intelligent, I’d definitely use them at least for a small amount of testing things. Maybe GPG should just super super study top player replays and make AI perform exactly like them… in any variety of situations. nah, thats silly. wow i’m tired
July 14th, 2007 at 12:56 pm
Unconquerable said:
“I fail to see why anyone would even want to play against AI?”
how about, my router is a POS(blocks one of the SupCom ports), so I have to wait until i am the only person in my apartment and steal teh modem in order to reliably connect to games. So I cant always play online. It is hassle-free to set up an AI game.
there are still people with slow/unreliable internets. The AI is arguably less fun to play but all you need is the game to do it.
July 14th, 2007 at 1:05 pm
Most people that purchase the game will only ever play against the AI, and even people that play online sometimes need a competent ai to test things with. A tough AI goes a long way to make supreme commander a more complete and polished product. For most people that play online it makes little difference, but at the very least it will improve preception of the game from outside influences, and help people that don’t play online for whatever reason.
This being said, while the supreme ai is bad, most strategy games have an AI that is incredibly weak, doesn’t adapt, and is easily defeated, so the complaints with supreme commanders AI, while true, is by no means unique in the world of RTS.
I believe its entirely possible to create an AI that can regularly defeat human players. Even top players don’t grow their economy with perfect efficiency, can’t dance their units to perfection, or can stage an attack on multiple fronts with perfect precision.
July 14th, 2007 at 6:47 pm
There is no AI in the sense of real intelligence. It’s just a bunch of if-else statements that attempt to make it play the game. Making something like that adapt is very difficult (if not impossible), so it plays the middle, always going for equal land, air, and naval (in theory) with proper (and exactly the same) defense at each key point it manages to capture. GPG attempted to make different types of AI, but in the end I’ve seen zero difference between them — the rush AI sends about six light bots before you can blow it up and the tech AI does the same. The best thing they could do for the AI is to stop it from tech EVERYTHING all the time and build some damn units. A little resource cheating wouldn’t hurt either.
July 15th, 2007 at 1:53 am
If-else statements are a great AI system. It’s what human intelligence would do. If your opponent is building air, respond with anti air. GPG can just have AI collect information, and store it in a system short term per game somewhat like how webstats were, but with a sight radius for each unit to detect what the human player is building. For instance, an AI scout might fly over a base and count 16 Bombers, with no interceptors or other anti air, In response, this could trigger the AI to build interceptors to destroy the bomber fleet. But of course, a real AI would be much more complex than this. There would be very large chains of if-elses attached to data collection. It could include triggers such as… IF commander is taking damage AND there are > (1) gunship within (20) units of distance, THEN commander does little circle dances AND triggers another trigger, which triggers a series of triggers to deal with the gunships… etc etc.
AI is an intricate If-else statement trigger setup. But I think with a good one, they should be able to play near flawlessly. With good scouting, intricate knowledge of the game, and better than human reaction times, how could anyone lose? (those are the three things top players have, right? i swear they even have better than human reaction times)
July 15th, 2007 at 11:40 pm
Actually a big tree of if/then is a finite state machine or expert system, not a true AI. A finite state machine will always respond to the same stimulus the same way. While it is conceivable to anticipate every single possible state that a Sup Com game may reach, it isnt really practical, so bands of states become your conditionals.
Fuzzy state logic is the next step, which is best thought of as a reverse process. The ‘AI’ has a number of possible strategies to different solutions, and it matches them against the situations, weighting them based on how close they match up. It then randomly chooses based on that weighting.
The best ‘AI’s use a mix of the various approaches to artificial decision making.
But game AI’s have to deliver without requiring a super computer - vexing isnt it.
July 16th, 2007 at 12:07 am
the problem with massive chains of if then statements is the amount of processing power required. deep blue, the first truly successful [i]chess[/i] playing computer, was a friken huge supercomp. it needed to be to evaluate the statements it was running through. an if then statement tree for sup com on the same caliber would probably require a hundred times the power. the problem with computers is not their ability to micro. a perfect economy only goes so far. what really counts on the battlefield is creativity. attacking your opponent where he least expects. now I’m not the greatest player, in fact i loose most of the games i play because I’m not too good at managing my economy. but i can whip the comp because i know how to go around the main defenses and hit the targets that really count. all the best player know how to do this as well. you never see a computer send bombers around the back of a base, so why build sam-sites back there? and all that saved mass can be put back into your econ. while playing a person though, if you don’t have a few scouts patrolling 360 around your base, you will get a nasty surprise coming in the back door. to make a comp smart enough to do that, yet efficient enough to run on a pc and not a supercomp, well that would be hard
July 30th, 2007 at 11:44 am
Unconquerable said:
“I fail to see why anyone would even want to play against AI?”
I only play offline against AI. Too much shit going on online with human players. I just can’t be bothered.
July 31st, 2007 at 12:27 am
I only play against AI as well, comp stomps with my buddies. If playing against people, then it is personal. When playing with friends against a common ally, it is fun.
I wish the AI was more configurable.
July 31st, 2007 at 11:24 am
The AI sux but it will get the noobs started. I dont think it was ever intended for serious challenge for high end players.
July 22nd, 2008 at 1:56 pm
The AI in forge allaince is actulyl worse then AI in superem commander, I asked the forum mods in game about it and they said they did not improve the AI, if you can easly tell that the new AI id dumb down actully, becuase if you put in stragic points on a new map that u creat or even one thats already made, the New ai will not use the startic points example AI will not ferry large amounts of trops from one location to anthere to do an attack, the AI in orginal will and can if you set up the commands on map.
Ai has been repotredly down sized.
July 22nd, 2008 at 1:56 pm
The AI in forge allaince is actully worse then AI in superem commander, I asked the forum mods in game about it and they said they did not improve the AI, if you can easly tell that the new AI id dumb down actully, becuase if you put in stragic points on a new map that u creat or even one thats already made, the New ai will not use the startic points example AI will not ferry large amounts of trops from one location to anthere to do an attack, the AI in orginal will and can if you set up the commands on map.
Ai has been repotredly down sized.