Game AI The Good The Bad And The Ugly.

I have been a gamer for a long time, But have not played that many games, in the last ten years maybe four titles
There was a comment in another post about Game AI in single player mode.
This started me thinking has Game AI progressed say as much as graphics or even hardware spec in the last ten Years. Some Gamers me being one of them, like to blame AI for a failed mission /level.
Its either to dam hard, or fxxing to Dumb a bit of a cop out I will admit.
One of the first Games where the AI impressed me is probably the oldest Sim I ever played
The original tank platoon on the Amiga over twenty years ago, So my question is has game AI progressed as much As all The other components That make up a Game ?

Generally I don't see what would make game routines better with hardware advancements- adding more transistors and gates in a microchip won't make the software seem more intelligent, the bottleneck is still a human being telling the software how to behave, rather than more hardware drawing more detailed objects with increased textures and effects with increased efficiency.

So imagine under the best scenario the programmer, who must predict ahead of time how a human player will behave in real time, and tailoring a set of pre-defined instructions to anticipate that. Imagine the extra challenge of making this behavior somewhat unpredictable at the same time, or deliberately programming in human-like errors so as not to seem like it's cheating or behaving unnaturally.

In some type of games, this is irrelevant (Tetris), others relatively easy because of the limited number of positions or possible paths within a flat maze (Ms. Pac-Man) or basically is more like an interactive movie on rails (Medal of Honor).

For an open, three dimensional world with freedom of movement in any direction that simmers crave- good luck.

I've yet to see a shooter where the AI knows basic gun safety. Never fails that one will step right in front of a firing gun.
I'm sure graphics have far outdistanced AI in development. Graphics sell games, while you never hear a commercial touting advanced AI.
Odd, since good AI can make a "good" game "great", when you actually encounter it. Furthermore, a game is good, IMO, if the difficulty levels change the AI behavior/intelligence, not just some kind of BS hitpoints/damage ratio.
So I think it's important, but underrated by most and hasn't kept up with graphics development.

If there were an elegant solution, you'd see more developers able to do it and less complaints about it.

A program like Steel Beasts can show you the challenge you're facing, the mission design tools are scripts in a natural language with Boolean operations rather than back end machine language, and you're still designing computer behavior against a human opponent who can think and react in real time, whereas your AI plans are just anticipating all of this at best without the same benefit.

Imagine being the coach of a sports team and you can't give your team instructions as the game is playing, but your game plan has to be written and submitted ahead of time, then once the game starts, you are entirely hands off while the other team still has coaching in real time. The more complex the game environment, the more problems you'll have with intelligent pathfinding and other types of complex behavior as well.

I'll cite racing sims for this one.... on some sims, they get the AI to a level wherein they react somewhat realistically to what the player is doing.... then on a later iteration of the same sim or on a different product but developed later in the year, the AI can be a step back and suddenly crash into the player for no reason. While AI progression could probably be coded better now with the advancements we have, nobody seems to bother with this. Heck, in Falcon, the wonky dynamic campaign is still its strong point over newer sims like DCS which is still "on the drawing board" as far as dynamic campaigns are concerned.

It maybe the case that developers have the software for advanced AI
But I think a game/sim would lose it entertainment value if it kicked its human opponents ass every time.

There have been precious few games that the AI really stood out as acting believable.

SWAT 4 was the gold standard for me. Wish it had widescreen resolutions that worked. The other is GTR2, those drivers were better than me.

That's not holding it back. You might as well be arguing that computers have reached human like intelligence or better.

The bottleneck is human programmers spending a lot of time and effort on something that doesn't pay off, like graphics, or in the same way as a gimmick like sex or violent content might. For corridor shooters and limited game environments, this is not as difficult as a game arena with open worlds and lots of complex characters.

And that's why MP games are so attractive from the point of view of developers, less time spend on something like coding computer behavior, leave it up to human players to provide the challenge, which is usually going to kick a computer's ass all day long (unless you're talking about certain games like chess), unless the computer is cheating or has unfair advantages.

I have no idea how to program AI.
But it must be incredible tedious even for the most simple of tasks with all the variation to cover
Watching something like FiFA 14 when you think about it incredible what AI can do.

I think a lot has to do with marketing. You can show the new! shiny! in screenshots or trailer video, but that won't showcase AI.

There's nothing that's revolutionary to be invented, because computers can't think on their own. The bottom line, the lowest common denominator is not hardware. It's the programmers, and they haven't discovered ways to make a computer 'smarter,' it can hold more storage, it can draw landscapes and animate characters better, but faster and more efficient processors do not think any more than a slow, obsolete one does. They can process more instructions at the same time, but in the end, they have no intelligence. Only the programmer does, and they are still the bottleneck with the same limitations they had decades ago. The AI is not really AI, all the intelligence comes from the person programming the game.

They have game development kits, but these still don't make a computer smarter. They are pre-optimized to render games more easily or efficiently, but in the end, your opponent is a non thinking entity. You are a thinking entity. They can do better at some games that require storing and recalling vast libraries or pulling data from memory, or performing math operations, but the way your brain thinks is far and away more complicated, like you have things that can't be measured such as intuition, or 'peripheral vision', the ability to learn, the ability to think.

Did you try the method on widescreengaming, or is there something specific not right with that one, like multiple screens?
IIRC you have to set the config file to read only or something like that.
Agreed on the AI, it's pretty well done in that game, other than getting your team to step around you to set up on a door. At least they say something.
"You're in my way sir". LOL

http://www.widescreengaming.net/wiki/SWAT_4

Not a sim, but Galactic Civilizations II has some absolutely amazing AI, and the AI is actually the featured part of the game. Multiplayer was totally eschewed for the single player experience, though it certainly felt like a human opponent, and people could always upload their scores and metadata to Stardock's server. The coolest thing about it is that when players were able to beat the AI on the hardest level (which by the way still didn't cheat), their tactics were analyzed with the uploaded metadata and then incorporated into the AI in a free update.

Granted, this was set piece strategy and not fluid tactics, but this does beg the question: why don't more games allow people to script an AI in games or missions? I'm talking about creating a series of LUA files from green replacement pilots to aces with different levels of aggression that could be assigned to individual units/aircraft. Your really smart hobbyists would have a field day with minimal implementation time to the developer.

Advancing AI isn't about making the computer a better opponent, it's about making it more realistic. More stupid, if you prefer.

Yep, I concur. Humans are weird in their behavior.
It's hard to emulate quirks, fear, double thinking, OVER thinking, change of mind, recklessness...

Having a great AI would mean being ale to discern such behavior, not just "terminators inside".

EDIT: More specifically I think i't important to make AI coherent. If I'm leading a crack-elite team of Rainbow 6 operatives I expect the generic thug to have a worse aim,(higher bullet spread), lower reloading speed, longer reaction time and possibly being prone to fear...

Rogue spear for me will always be dear to the heart but I really hated how my teammates always felt a bit worse than the tangos we were hunting...

Whereas other scenarios may involve former military guys, with better performance.

Crysis is pretty good in that regard. The AI acts "scared", they generally know where to look for you, they react to swaying branches (even if you are cloaked) but down blind/cover fire and throw grenades well.

Perfect AI is easy, "dumb" AI is hard. Arma2 (haven't played 3 yet) was a step in the right direction as far as AI being able to "read an environment".

Agree on Crysis 2. I don't think a simple SP FPS gave me more excitement than that one. Well a shooter one. Portal 1 & 2 are still the best "Experience" for me... But that's another level outright.

Grand Prix Legends has some of the best driver AI I can think of.

Pfunk nailed 2 titles Swat 4 and indeed GTR2,and who can forget like Semmern says Grand prix legends..man thats one tough cookie the new Wings over flanders fields (WOFF) by OBD has got the finest AI ive ever seen in ANY game ive played,it really makes you work.

Except for John Surtees in the Honda... Get outta his way! Obligatory shout out for the AI in BoB2. The AI in WoFF is very close to as good, if not better, as well.

The vast majority of games that I've played do just that. It annoys me and this is why I hardly play any games on anything higher than "normal" difficulty. I've noticed with RTS games that putting the AI on "hard" or "very hard" difficulty simply means that the AI will get bonus resources and extra units. That's simply giving the AI a cheat. It's not making it more intelligent.

At least in the Thief games they can increase the guards' ability to see you, increase their persistence in searching for you, and so on. It's not just "their guns get bigger and yours get smaller" that many games fall back on.

I liked the AI in games like FEAR and NOLF. Monolith was pretty good at it, actually, AvP2 and TRON 2 were good as well.

I loved it when one of HARM's goons would shout "Whoa, bullets!" and dive for cover when you started firing.



The Jedi Master

SC: Chaos Theory had pretty good AI. The best part was that the entire map was always running and "alive", not just some section you were in after a checkpoint. Something you did early in the mission could be noticed...a missing guard, a closed/open door that wasn't before, a light shot out or even just turned off...they would wake up a knocked out guard as well. Anything that was off and it could set off all guards to be more alert, or equip armor and guns, lights etc. This is the main thing missing from the current gen of the SC games, since they rely on checkpoints.

Realistic AI (meaning actually behaving realistically) in open world games would inevitably ruin the story or destroy the intent of the game. For instance, in Grand Theft Auto, you wouldn't be able to do those things and get away with it, you would be a fugitive for the rest of your life until you were brought down, likewise, I've played a few games where you kill a villager or kill a guard sentry, so maybe like the locals react to what you did for a few minutes of game time, then reset back to their original state. Memories are short at best, or don't exist at all.

I don't care how evil some of the characters are in Grand Theft Auto, once they see that you're capable of carjacking and murdering grannies, slaying law enforcement, and shooting up the town, you're a huge liability, they simply wouldn't deal with you, or they'd kill you themselves. While there are no checkpoints, for the game world to be consistent within the context of the story (and playable as well, one versus millions wouldn't work too long as a game), there are always shortcuts made for practical reasons.

Which is why it's not really an issue in open world games.

In other games, though...



The Jedi Master

Now there is a man who knows good AI when he sees it!

I absolutely agree with you on GalCiv2. I had some of my most amazing strategy gaming experiences with that game, the AI was very well done. It had an option in the game settings "Allow AI to use max CPU" which was only recommended for fast CPUs but would unleash the full complexity of the AI. I still have that game installed on both my desktop and laptop. I really hope GalCiv3 lives up to the legacy of its predecessor!

I can think of a few examples of both good and bad AI. The AI in BoB2:WOV with the BDG patches really blew me away when I first encountered it. Really smart, almost human like behaviour. Likewise the AI in the original IL-2 series (all the way up to 1946) was never terribly impressive. One of the headline features of WOFF is they've overhauled the AI from Phase 3 and made it much less predictable and human like in its behaviour.

Way back in 2009 when Empire Total War was released to howls of derision because the AI was basically nerfed to the point of standing passively in front of you whilst your own forces blasted them away I remember reading a really good article about AI development. Basically the point was there are umpteen modellers, texture artists, animators, etc etc but comparitively few AI programmers. AI programming takes a great deal of skill and ingenuity on the part of the programmer, yet the results aren't readily apparent in promotional screenshots or Youtube trailers... Then of course as somebody above pointed out the current trend of Call of Duty clones and corridor shooters doesn't exactly require sophisticated AI, and nor will the intended customer appreciate it. Hence developers blow the biggest part of their budget on the high impact stuff like graphics and neglect stuff like AI. Which in turn creates the kind of climate where quality AI programmers are neither particularly sought after nor nurtured in the contemporary video game industry.

Some stuff to remember regarding AI. First is that there's a limited amount of resources that the computer has and when you're dealing with a lot of units you can only spend so much time on each one (and that's with what's left after graphics and other systems). There are tricks you can do to help with that, like only having the lead AI of a group doing the heavy thinking and passing it along to the rest, but it adds up. Generally I try to avoid having the AIs doing much every frame (some things need to be) and then try to work it so I don't have too many AIs doing complex thinking in the same frame.

Also there are only a few common things that translate well between games. Path finding is the best example I can think of. It’s not like graphics engines where, while there are variations, you can get one which will work for a number of different games. Even games of the same genre can have really different requirements. For example a fast mover game isn't going to need the same level of ground combat simulation fidelity as a slow mover. Sure, the slow mover's code could probably be used in the fast mover's but you're covering a lot more area and so will have a lot more units and there go your resources. Most games try to do or focus on something to differentiate themselves from others and the AI code has to take that into account.

The broader the AI needed means the less time there is to add depth. For the flight sims I’ve worked on along with dogfighting logic (taking into account group tactics) some of the other stuff you have to have logic for is CAP, SEAD, SAR, strike packages, escorts, refueling, takeoff and landing procedures, all of which has to take into account the interactions of a group of planes, and this is just dealing with air AI. Then you need to code the wingman logic which can’t really “talk” with the Player much and determine the interface so the Play can talk with them. Also there’s adding the hooks for all the radio calls that need to be made at the appropriate time and figuring out what data needs to be passed for multiplayer.

Another factor which has been mentioned is that often you have to spend extra time writing code to limit the AI’s knowledge. It doesn’t take much code for an AI to find someone in a particular radius. It’s another to take into account line of sight, blind spots and other things like that.

If you want to throw in a dynamic campaign then that’s probably coming close to adding the complete logic for an RTS game. That’s not something I’d want to try to do on the first version of a game. If we had done another Jane’s title then that’s what I would have been working on. (F-15 base game, F/A-18 multiplayer and convert to Navy based, ??? dynamic).

Also the less predictable an AI is the harder it is to debug. You only see the bug after the fact and you have to get used to looking for patterns so you can figure out where to look and add debugging aids.

Then there’s things like trying to make your code adaptable to different play styles and a few other bits that just ran out of my head but I’ve probably rambled enough.

Elf

I think the problem is really just that the resources to make really good AI aren't provided, for the reasons mentioned, but the industry as a whole seems "ok" with that.

There's no "feels like you're playing against a human" pledge that devs have signed to indicate they will spend less time on flashy effects (that frankly all the games have and we've seen before) and more on that kind of logic that makes the game more entertaining to play.

I would LOVE an AI slider in games that lets you balance the AI with the graphics so after you OOOH and AAHHHH over a game for the first few hours you could lower that and kick up the AI so that it's still worth playing years later once your PC is faster and you can then increase both AI AND graphics.



The Jedi Master

I Wonder do the military use Advanced AI in there super computers I can think of a lot of military Applications for same. think Deep blue it beat a grand master in chess.

I believe that the military would use petaflops computers for modeling and statistics.

I would agree.
Seemingly the Chinese have a supercomputer capable of 54.9 Petaflops.
Bet it has ant got a good graphics card though. LoL