1) Write some basic stuff, and then command it to do whatever you did, like a replay-bot
2) (i think this would work better) a mirror player, as in if you miss 3 inches to the right, so does it and such, then you can have tons of fun realizing your weak points, and trying to win, without losing
what do you (and everyone for that matter) think about that?|||you do realise this would be super easy to sploit.
that fact you know exactly what the computer is going to do is bad enough... but this would allow you to control it!
the problem here is that something tring to copy your everymove won't do as well as you. (have you ever tried to mirror someone you were playing in chess? didnt work did it.)
the only AI that came close to making something good out of a similar idea was the unreleased C&C tiberium sun skirmish AI. (i somehow got my hands on a copy unknowingly)
what it did was try to use your tactics. rather than mirror you. if you used a spcific type of unit it tried to use it next time. it leant to buld buildings you did. and it learnt when and how you attacked. it was truely awsome.
it even lernt how to pack 5 engineers into a subturainium APC and pop it up under my constructionyard, take it over and sell it...
it was probably the hardest AI i had ever played that wasnt conditioned to a specific map. and thats what made it great. It could adapt what i had learnt and use it in any map situation. ofc sometimes it had greater success than others... but it was very fun.
Dont ask me why this AI was removed from the official tiberium sun release... but im not kidding. it is possible... in fact it was possible 6 - 7 years ago!
i really dont knonw why this wasnt explored further.
|||What you could do is have an AI setup to your style of play. This I have seen done... it is hard but I think the outcome is worth it. For the most part, it would be pretty much a balance of how fast your style of play is and how you control your attacks and your defense.|||Darac|||If DFS gets his ideas into the AI for FA then it will allow for the things you guys mention quite easily.|||you do realize that you couldnt win and you would always draw when fighting a mirror ai? You launch a nuke, they will launch a nuke, if you have nuke defenses so will he. Imagine fighting a mirror, the mirror will always end up countering everything you do, if you throw a punch, they will throw a punch on the opposing hand to hit your punch. You cant win|||Darac|||dcbo89|||the thing is, with a true mirror ai, they would if you changed a unit direction, they would change unit direction in the opposite way(essentially the same way, but your left is their right type thing). Its essentially Newton's third law: "Every force has an equal and opposite reaction" You hit a wall, its going to hurt, you hit a mirror, its actually going to hurt more. So it is essentially impossible to win, you launch a strike force at them, they will launch the same one and you will both destroy each other ACU at the same time|||dcbo89|||Thats starting to get confusing, But I think that it could be pretty fun to vs myself.... Actually super annoying no doubt.|||Well, I see what he is aiming for.... essentially a game that doesn't end unless you've made a mistake.If you are able to kill his commander, you'll die. Which means your defence wasn't good enough. But if you sit back and turtle you won't win, which means your offense wasn't good enough. So you attack, and if it is repelled then your defense was good enough.
So the goal is to win (of course), and through winning you'll identify weaknesses in your game.|||There should in theory be a small lag, between the time it identifies what you are doing and then does it itself, and I can actually envision that lag widening over time and the thing eventually getting confused as it periodically fails to duplicate exactly what you are doing.
It would be an interesting experiment, but how would it deal with that T3 bot that doesn't make it to the transport on time due to some pathfinding bug that it encounters? Your transport is already fully loaded and on its way to the destination, while the AI's transport is still sitting there waiting for the T3 bot to arrive that will never arrive because that T3 bot got stuck on the mountain enroute and will be there until the end of the game.
I'm not sure that the actual implementation will turn out as efficient or as inefficient as the human player, and again, the human player will be given the occasional window of opportunity to gain the upper hand. I don't think a perfect mirroring of the human's every move would be realistically possible for an AI.
I mean, in the past, the AI's ships could apparently walk on land, yet I don't ever remember the human's ships being able to do so. Gee, in that case, the ships walking over the land might actually get there faster, whereas the ships that can only go through the water might get sunk by submarines enroute. What you actually witness from such an AI might be different than what is expected due to the bugs in the game if due to nothing else.
Still, Sorian seems to think that some of this is possible if GPG actually gets the bugs and glitches out of the game sometime in the future, but so far, such a thing hasn't happened yet. And, assuming that such a thing will happen in Forged Alliance is basically the same as assuming that such things would have already been possible for SupCom or should have already been included in SupCom. It didn't happen for SupCom and it might not happen either for Forged Alliance. Only time will tell if the promises of Forged Alliance equal the reality, just as time told us that the realities of SupCom didn't exactly equal the promises of SupCom.
And, as with any AI implementation I have ever seen throughout AI history, there are just some things that a human can do that an AI will never be able to do or isn't even allowed to do. I see it over and over again, in all of these various RTS games and skirmish mode games, instances where the AI isn't even allowed to attempt what the human can actually do. There are features in these games that only a human is allowed to use. Until recently, the SupCom AI wasn't even allowed to use transports or couldn't use transports. Only the human could successfully use the transports. How many other features in Supcom are still that way, for humans only?|||MeDDish|||It sounds like a really good idea, but it's very weak. You could build a tank army and the A.I. would follow and they would never attack unless you do. And inevitable, the A.I. would always be a step behind waiting on your move.
You could move an army into its base (not firing) and position your troops to destroy the base. Use your other armies to block his army from moving into your base then open fire.
But what you probably want to say is, a great A.I., like a chess player, should be able to see several moves ahead and take advantage of weaknesses and toy with inexperienced players, never scouting but always knowing how, where and what that pathetic little human player will try, avoiding direct confrontations while delivering devastating blows.
So then you need the Genghis Khan A.I.|||did you read both my ideas? how about a literal mirror? So your commands happen instantly|||I'm going to SAVE you a lot of trouble. You can TEST your theory by asking a friend to play with you and mimic your every move. I'm not sure that they'll be able to see what units you are building until the pop out of the factory, but you sure can send the same units after each other for a clash.
Or you could try this with a game of checkers or chess... but inevitable, you'll see the flaw in the plan. The loser always dies first... or in the end, only one will be left standing.
A monkey bot came out of the sea heading towards my camp. I was able to able to scrape together 3 monkey bots and ordered them to attack him. As I watch this mega battle take place (and even with some coaching), he was able to take out two of my units with his flamethrower (one was about 80% prior) before my third destroyed him. Why mines didn't use their main guns right away (and I tried to coach them) is beyond me and that's why I lost.
That said, there's no guarantee that a unit will fight evenly besides the many thousands of calculations that go on each second... a evenly match battle will come down to whoever has the last drop of meter life left.
Have you read about Genghis Khan battle tactics?|||sorry for the spaz.
I really meant, that is there a way, to actually make 2 of you on different teams. So you control both, by useing one. Or is that out of reach?|||I think the idea is that any command you make is translated to two players rather than one. Unfortunately this would not work because structures are always oriented in one direction. If you build a T3 artillery, and you are at the top then it will not have to turn to file, while the bottom would. If you think just put it on a map left to right, true, but units are always built facing left, so they would have to turn around to work on the right.|||The simply copying issues will not occur in SupCom. Since the AI only knows about what it sees, you can deny it intelligence and its got nothing to copy from. You can abuse this to pull a fast one on it.|||I meant something where you control both sides, nothing to do with intel or anything, sure it would be a little slow, but don't you get my point? Like just a mirror. Does a mirror have an entity on the other side copying you? No.
Ofc if this is impossible...|||Neural network would be worth it despite the filespace. I was working on one, it was very rudimentary and not for this game, not for any infact- simply a basic learning AI.
NN would be only as hard as the best player on the computer. Therefor, it is mostly auto adjusted to that person, set by themselves.
No comments:
Post a Comment