What about AI: Wars?

https://www.codeofhonor.com/blog/the-making-of-warcraft-part-3 has a short section on how and why Warcraft's AI cheats. It ends with:

Because nobody wants to be the guy who inadvertently creates Skynet, thus being responsible for Terminator: Genisys.

RTS in vein of Total Annihilation has been all about multiplayer since StarCraft I think.
Just as the extreme focus on mutliplayer in FPS has made modern single-player FPS rather uninteresting, the same can be said about RTS games. Too much focus on multiplayer damages the single-player experience quite significantly.

Even the popular mutliplayer RTS games fail to offer something unique and interesting.
The only thing these games have added to the single-player is prettier graphics and more storyline or narrative.

Edit: there are exceptions of-course, like Grey Goo.
However its not as if we have that many modern RTS games to choose from.

Personally though, the AI in games like C&C etc have never really impressed me, even in the earlier games.
I still enjoy these games but to me they have never really been about the AI.

Hm, AI wars definitely gives you a very big number of very much different AI's to play against. It could very much be the best AI in RTS games when you turn all the options on (AI plots) but its very hard to put it in place since you as player start very much in a devastated position so much that the AI doesn't even notice you at first until you start wrecking its positions which makes it very angry :)

Let me just add that the field of AI in general, as in computer science, sadly hasn't progressed much since 1950s. We do get games with relatively better AI compared to the rest once in a while, but it is all thanks to dirty hacks and tricks all very specific to that game's ruleset. Imho we need some fundamental scientific breakthrough in AI research to push things to the next stage.

Sorta reminds me... i was wondering how to write an AI for a chess program, just off the top of my head. Then i thought how to do it, and i came up with a brute force method that is likely not too unlike what's really used.

Use a point system to determine how good/bad a move is. If a piece is in check or threatened, there's a negative value to it. If you threaten the other player's pieces it's a positive. Brute force through 1-2 moves ahead and choose the move set that gives you the highest score...

Then you expand on that. If a piece is threatened but is protected by another piece you ignore it, or if the piece that might take the threatened piece is worth more than what you lose, it's a positive score. Pieces protecting other pieces adds to the positive score as well, etc. Naturally the king in check is max negative and you have to improve that in the cheapest way possible (probably).

How much of this is the AI that's used in say the Atari chess that had 4k for a ROM, i am not sure.


Then i think to more customizable AI's. I played CRobots where you program your tank and give it whatever behavior you want. You move/scan/fire as appropriate based on the rules you give yourself, making perhaps the most complete AI i can think of. Some of the more hilarious AI's is one that only rams his opponent... that was fun :P

Then there's pseudo AI, more along the lines of FF12 with the Gambit system where it's rules from highest to lowest priority, and a condition to when it should be used (or is more important) and it acts based on that with a few simple rules. Excellent for making your team feel like it's actually doing stuff that's productive, but once all the rules run out they still stand there and won't explore on their own...

And then there's path finding... Uggg... I'm reminded of playing a game of starcraft where 3 other players were against me, so i went all out. Unfortunately pathfinding screwed me over. I sent 3 dozen marines and they ended up going single file, when they arrived the two bunkers just killed each marine in turn and didn't take a single point of damage. Had i been able to give better commands, or pause in order to prepare them as an onslaught rather than the single-file they ended up as, the game might have turned out a little better for me, but eventually i was screwed because i couldn't give fast enough commands to deal with huge airstrikes, while i had more than enough ghosts to do the job.

Stardock is trying to make a non cheating AI with Ashes of the Singularity by taking advantage of multi core processor.
http://steamcommunity.com/app/228880/discussions/1/490123938426647400/

The down side is you need a fairly advance system with quad core as the minimum requirements.

Good news is, they plan to bring it to GoG as stated in the Stardock Boarding news, so if they managed to pull it off, you would be able to buy it at GoG.
http://www.gog.com/forum/general/stardock_now_boarding_gogcom_3ccfe

Also this may be the first game to use Direct X 12, as well with its alternative (Vulkan) for Linux, as Oxide Games is part of the Khronos group, which is developing the next-gen Vulkan graphics API
http://www.ashesofthesingularity.com/game/faq

I think there are two things being confused here: The AI is cheating (i.e. it has no fog of war, and knows what you are doing), and the AI is not playing by the same rules as you do (e.g. I destroyed all his windtraps in Dune 2, and yet his turrets keep firing).

Why do these thing happen: First of all: do you enjoy playing chess against the computer? It is just too strong, and you will make a mistake first, then then you are done for. Unlike what most people think, players don't want "better AI", they want a more fun opponent to play against (and online playing can provide that).

Then, you only notice the unfair playing field after a few hours of play. Some less sophisticated players may not even notice it! So developers prefer to divert resources to other areas of the game that they consider more important: more units, more varied maps, better user interface, better art... and then hack and cut corners in the later game to try to hide the AI's inefficiency.

AI is hard and complex. It can only be done after the game's rules are stabilized, and are often times limited by the way the engine works. So it is often times rushed through production.

Ah, it sounds very cool. I don't know how fun it would be, but sounds cool. You do find yourself often in the part of the predator.
However, like all other spiders, that one is a very specialized animal. It just specializes in very unusual stuff.

Also, you would require a very large number of plays for a system to learn the right patterns. Maybe if everyone would upload their match to a central server, and every few months a new, better AI would be available as an update. I think that would be cool too: a Borg AI. :-)

I love the idea behind systems like this (I believe Dragon Age had something similar) but usually they are far too basic to get any useful behaviour from. I almost always ended up (no matter how carefully I tried) with commands that would never be reached because the condition for the one above them was always fulfilled or no fall back so once all the specific situations were dealt with they wouldn't just attack an enemy and would wait for another situation before acting...

While reading this thread I've tried to come up with examples of good AI, but it's really difficult. Most games I've played (RTS or RPG) the way the AI attacks you is usually inefficient in some way and it's rare they do anything believable. And most of the time you cam figure out the basic rules and then work around them to win (unless they cheat heavily)

I don't know the specifics of FF XII too well as I've only played a bit of it (hoping they release it on PC eventually so I can play it properly) but my experience with other systems is the same. You either need to buy or earn (by levelling up or whatever) the extra slots and abilities to use to customise your parties AI and at the start it's more harm then help as you've got too few choices to make anything sensible.

Plus as you've listed there is generally a 'best' way that achieves all the various support and attacking objectives you might have and 9 times out of 10 you will go straight to that, so why bother having the system in the first place when you could just have three (decently made) AI options for Attack, Defend and Support or something similar...

But that being said, I hope more games use a similar system and it improves because I do like being able to tweak the abilities of my allies and be able to set them up so I don't need to choose every single one of their actions to fight effectively.

Dunno, i do believe that something like that happens with the Forza games for the Xbox one, they analyze your driving style send it to the "cloud" and supposedly adapts to you, now, how well this works i don't know but if it does work well it's actually a neat feature that i would like to see in more diverse games.

Yeah you start with like 2 slots, you can have up to 12 slots at the end. You get pretty much all of them through purchasing licenses.

Although i've listed off what generally the setup should be, that doesn't mean that's the best way it should be used. You can configure to use items, or magic only, or to try and keep someone blind so they use unseen sight...

Although there's plenty of ways to totally fail. If MP < 100% then charge is probably going to try and charge over and over again for nearly an hour and still might not get you to 100% due to charge.

One of the things i tended to do as a higher priority was if an opponent had 100% health, i'd have Vaan steal from them (the other two would attack). If no one had health at 100% then Vaan would begin fighting.

You could also make one person purely support and healing while the other two attack (which i haven't really done much).

Perhaps the largest problem is later when you try to go up the tower (past level 60? I forget, but it's an optional end-game objective) where your standard loadout doesn't work due to the difficulty and requires you to totally retool your group.

The reason probably mostly is that it gives you a chance to customize how your companions act (although badly designed gambits will just waste resources, time or do nothing useful). Still it makes for a decent example of how to make your own AI, so long as you can pre-write what casting, attacking, defending, using items etc all do.
I've heard some of the driving games where they just implement a rubber-band effect, so no matter how good you drive or how good your car is the other AI cars will keep going past you. Then they also have the same effect where if they get past a certain point they will wait for you rather than going to the end of the race, making the whole exercise pointless.

I'm not really sure how much you can analyze a person's play style. Either you drive faster or you don't. Either you handle the roads better or you don't...

Although they could save the time and running of a player's race and use that as an AI instead, although how well that works i'm not sure since it's a fixed play and not dynamic...

Oh, so someone already does it. Thanks for letting me know. There goes the patent, I guess. :-/