PlanetCrap 6.0!
Front Page (ATOM) • Submission Bin (1) • ArchivesUsersLoginCreate Account
You are currently not logged in.
T O P I C
Adaptive AI in games?
August 7th 2001, 11:42 CEST by Ashiran

Adaptive AI seems to be a keyword in the selling of games recently. But when said games are actually played the AI usually displays the same old behaviour as we have seen a thousand times. Since Pacman games have come a long way in sound, gameplay and graphics. If we look at AI however that way is a relativly short one. Why?

The problem is that the step between precoding an AI and just setting basics or allowing it to grow is a very big one. Between precoded adaptive AI (like Max Payne) and real adaptive AI hovers a very thick barrier. The static nature of digital programming doesn't really allow for dynamic behaviour. It can be faked by defining all possible paths but it is not really non-static.

But even worse is that developers don't seem very eager to spend time with their precoded AIs. In most RTS skirmish games you can beat any AI after a couple of games cause they always do the same. And for most singleplayer modes scripting seems to be much preferred tool to add difficulty to the game rather then making an AI that makes enemies act intelligent or different.

Basically it comes down to fighting scripting in campaigns and humans online. Cause else there would be little challenge.

So where does this leave the future? Should developers add more thought to the coding of their AI? Or isn't this necessary cause everyone is online anyway? Wouldn't you love to be suprised not just once but multiple times?
C O M M E N T S
Home » Topic: Adaptive AI in games?

|«« - Previous Page - Next Page - »»|
#1 by "m0nty"
2001-08-07 11:43:10
paul.montgomery@delphigroup.com.au www.delphigroup.com.au
first post! ah, that feels SOOO good.
#2 by "MaverickUK"
2001-08-07 11:49:42
peter.bridger@tpg.co.uk http://www.thisstrife.com/
Quite a few questions raised there...

People liked the Grunt AI in Half-Life, because it seemed to minic human teamwork.
But don't be fooled into thinking that clever AI = difficult AI = fun game.
AI is one of the many tools in making a game, that enables the developers to make it fun. The bad guy AI Duke3D & Doom was terrible, but did that *really* make the game any less fun?

The creature AI Black & White is fantastic, but does it create one of the best games ever? To some, maybe so... but as the development of Black & White shows, to create something really groundbreaking, takes a lot of effort, and is creating complex AI *really* the holy grail of computer games?


--
Mav
\"What kinda shithole planet is this!?\"
#3 by "Ashiran"
2001-08-07 11:49:48
ashiran@ashrain.net wtf.couchcrew.com
I just found out that the AI programmers who worked on Black&White are among the best in the world.
It shows. It's more a simulation program then a game. Good job gang.

Not to mention that I'm still not sure if this is real adaptive AI.
#4 by "Ashiran"
2001-08-07 11:53:50
ashiran@ashrain.net wtf.couchcrew.com
and is creating complex AI *really* the holy grail of computer games?

It's my holy grail. :)

And besides there has never been a game that sported a clever AI = difficult AI. So who can tell if it would be fun or not?

I'm sure it beats fighting scripting. The campaign(which bites hard) in Cossacks for example is a nice example of that.
#5 by "Desiato"
2001-08-07 12:02:44
desiato_hotblack@hotmail.com http://www.spew2.com/
I have to jet soon, so briefly - AI in games will be implemented based on their "cost" to the processing time available in a game. 3D has gotten to the point where many functions dedicated to graphics are now pushed off of the CPU to a GPU. If this continues, there will be more time possible to do real on-the-fly AI calcs instead of worrying about weapon particle transforms. The real question is, in a typical FPS game, how much of the CPU is really available for this kind of overhead? Obviously, I don't design games - but I would like to hear some rough numbers from the people in the know that frequent this board.

How much time is typically available for the CPU to be burdened with a (possibly) cycle-sucking task?

Given examples that feature some kind of AI (please, no flamewars on how good "X's" game AI is) what kind of extrapolation can be made for future games given the increases in average CPU/GPU speed? (The more cycles pushed off to the GPU, theoretically the more CPU cycles can be dedicated to other tasks...)

I would love to see a game utilize a self-modifying ruleset with some Neural-Network learning heuristics for "fuzzy" actions, (such as - my computer AI character is near water, is it shallow - should I forge across - or should I bother to navigate around it? Note - no "bot paths" or other pre-computed external routes..) but the CPU cost would be staggering for something like that.

Just a few thoughts, I'll check back later!

Thanks for posting this topic morn!


Desiato
#6 by "Spatula Man"
2001-08-07 12:06:01
llama@verbalchilli.com www.verbalchilli.com
MaverickUK:
is creating complex AI *really* the holy grail of computer games?


Well, I can't think of a game that has suffered because it had better AI. Duke was fun, but it was also at the limits of what was possible at the time. To remake a Duke-esque game now with similarly stupid AI wouldn't make sense to my mind. Which is probably why I thought Serious Sam was a bit pointless. Why try to return to our primitive ancestry?
#7 by "m0nty"
2001-08-07 12:11:39
paul.montgomery@delphigroup.com.au www.delphigroup.com.au
I agree with Maverick. B&W was a prime example of the tension between computer games being little lego sets in your sandbox which you used to build things out of your imagination and some raw materials, and on the other hand computer games having a narrative like a movie script. Deus Ex was another.

Note that having a narrative does not equate to being strictly linear - the AI in those games, such as it was, often served as a device to branch off into different plot streams. Thus your actions in interacting with the AI of the cow determined the direction of the narrative.

In B&W, the user themselves could decide whether AI was an adjunct to the main narrative, or the narrative was something you ignored as much as possible in favour of the sandbox approach. Most other games lean towards the narrative.

I think the nub of the problem Ashiran raises is that no computer game has yet truly been devised with an AI at the centre, but that even seemingly open-ended games are not adaptive, but merely multi-linear.

Whether this is a good thing or not will be proven in the marketplace. That's what capitalism is for, people!
#8 by "Ashiran"
2001-08-07 13:10:42
ashiran@ashrain.net wtf.couchcrew.com
#7 by m0nty
Whether this is a good thing or not will be proven in the marketplace. That's what capitalism is for, people!

Normally I would agree with you. However gamedevelopers don't seem to care about making a real adaptive AI, which is quite understandable seeing the amount of time and research goes into such a thing(trust me, I know). But when we look at the more scientific minds in the world, they are all trying to create the ultimate self-aware AI. Now this is where it goes wrong.

The natural progression(roughly) of AI could be as follows(according to me),

1)Precoded AI(If this happens then do that)
2)Precoded Adaptive AI(all behaviour is set, behaviour chosen is dynamic based on external input)
3)Real Adaptive AI(behaviour is created based on basic motivations and external input)
4)Learning AI(same as above but the AI can also increase the number of actions it is able to perform)
5)Selfaware AI(This is basicly a Learning AI that "evolved" into a selfaware entity)

As you can see gamedevelopers use 2 cause that is the easiest and the real AI scientists start at 4. So there is nobody who tries to solve 3 which is the perfect AI type for games. This has been the case for decades now and to be honest I don't see it changing really fast(atleast not until I'm done with my own project).

Note that the opinions on how to create an 4 are quite different. From fuzzy logic and neural network combos to stupid ideas like Mindpixel which is basicly nothing more than a program that mimics a real AI. This is what some people call a MEGAPARROT(thanks to Desiato :).

While I don't really see added value of 3 for FPS and the like, for RTS games and other strategy games this would be heaven.
#9 by "Ashiran"
2001-08-07 13:11:50
ashiran@ashrain.net wtf.couchcrew.com
Oops, no bold close
#10 by "Flamethrower"
2001-08-07 13:22:53
patch@evilemail.com www.Jesus-vs-Judas.com
We need far better AI in game. Far better. I coded fucking awesome AI for Quake. I should dig those files out.

Carmack, for example, refuses to use unused processor cycles on AI. You don't want low-end machines having different AI from high-end. I guess that's the fucking price you pay for having high-end : you have to suffer low-end cunt's woes.

I disagree. I think the more power you CAN throw at the AI you should.
#11 by "m0nty"
2001-08-07 13:33:51
paul.montgomery@delphigroup.com.au www.delphigroup.com.au
For all but the most experimental of games, your type 2 AI is all that's sufficient - and, perhaps more importantly, it would hurt the structure of the game if it was more evolved.

After all, do you want your Sim to suddenly decide they don't want to work any more, and bum around the house all day every day? Do you want Duke to come over all peaceful and lay down his Uzi because his AI suddenly developed a self-protection subroutine? Do you want Lara to quit her life of dungeon crawling to become a swimsuit model because she's sick of getting eaten by tigers? Wait, don't answer that...

Personally, I'm as much a fan of sf as the next geek, but even I am suspicious of AI ever advancing to 4, let alone 5. Just because a computer program pasees a Turing test doesn't make it alive, I don't care what your fancy-schmancy whitecoats say. Calling it a Holy Grail gives the false impression that it will inevitably be reached - I'd prefer to call it Shangri-La or El Dorado, as they were merely mental constructs.
#12 by "Spatula Man"
2001-08-07 13:36:56
llama@verbalchilli.com www.verbalchilli.com
Thanks for those definitions Ashiran. This has generally been a very informative thread, but, not being a coder, I was getting lost amongst some of the terminology.

at least not until I'm done with my own project


Would you elaborate?... or have you mentioned this already and I missed it ( - I've been away for a few days and couldn't face reading the hundreds of back posts)?

While I don't really see added value of 3 for FPS and the like, for RTS games and other strategy games this would be heaven.


Maybe not for FPS games as they stand at the moment, but this would be a blessing to the next generation of roleplaying/shooter genre-breakers like deus ex 2, etc.
#13 by "m0nty"
2001-08-07 14:17:33
paul.montgomery@delphigroup.com.au www.delphigroup.com.au
Ashiran (#8):
While I don't really see added value of 3 for FPS and the like, for RTS games and other strategy games this would be heaven.

I don't know about you, but I prefer my AI simple in a game setting. Take Tropico, for example, which mimics your number 3 AI. The fact that your little Tropicans have overly-simplistic motivations which you can manipulate transparently through your actions as a governor is the whole purpose of the game, not a flaw in AI design.

The function of games is to reduce some situation in the real world to a set of sharply defined parameters, within which you then find a logical way to achieve a stated goal and "win" (or at least find an optimal state). One of the greatest games ever made, Civilization, is the ultimate in applying simple rules to complex real-world systems and seeing how complex the imagined world can become. You can have your self-important pseudo-science tomes like The Physics of Immortality, but give me a game of Civ any day.

I would argue that 3 is not really much different to 2, and may even be the same. I know you haven't expanded on the concept much (feel free!), but it looks like a semantic difference between looking at an AI as a collection of behaviours (2) or a collection of motivations (3), both of which are different (but not wrong!) ways of looking at the same beast.

Of course, IANA AI expert, so I could just be talking out of my arse.
#14 by "Ashiran"
2001-08-07 14:40:08
ashiran@ashrain.net wtf.couchcrew.com
#10 by Flamethrower
Carmack, for example, refuses to use unused processor cycles on AI. You don't want low-end machines having different AI from high-end. I guess that's the fucking price you pay for having high-end : you have to suffer low-end cunt's woes.

This indeed is BS. More AI modules in effect means that more CPU time is needed.
Note: An AI module is an seperate AI entity(type 3) that can run next to and is independant from other AI modules. Take a game like Decent: Freespace, if there are 14 ships present in the level each of them is running an AI module. You can consider this as there were 14 humans present in the game. Each doing his/her own thing while responding to events in the game(input). So if you were to run this on a CPU it had to mimic the (simplefied)thoughtprocesses of 14 persons. You can imagine that this creates quite a drain on your system.

It is also important to realise the following. All fighters of Type X start out with an AI module suited for type X. So the code they all start with is the same(this is called base coding). During the game however they all recieve different input and therefore there code starts to become different from each other. An AI module that was attacked repeatedly will contain other code then one that just flew around a bit. If these two got in a fight together then the chance of winning would be slightly higher for the one that got into combat before. That one would already have some added behaviour regarding battles and would therefore be more effective. As you can see this is a simple example for letting AI units gain expierence.

You can set the actual CPU time that an AI spends in adding new behaviour to any number. So if you don't have a high end system you just lower that number and the resulting drain on your system. Ofcourse this will decrease the learning capacity but that is just for low end systems. If everybody would do it like Carmack says then everyone would have lowgrade AI regardless of what kind of system you have!

Pick between low-grade AI for everyone or low-grade AI for low-end systems and high-grade AI for high-end systems. Which one would you choose?
#15 by "Whisp"
2001-08-07 14:53:14
Pick between low-grade AI for everyone or low-grade AI for low-end systems and high-grade AI for high-end systems. Which one would you choose?


Low grade for everyone.  Can you imagine the bitching that would occur when players found out that not only does their game not look as pretty as the one their neighbor is playing, but that it is fundamentally different game as well?  That's what would happen if someone did this.  Plus, how could you balance the levels?  More complex AI is almost certainly (but not necessarily so of course) going to be more difficult to beat.  You'd have to go through game balancing for multiple configurations, making an already tedious task even worse.  I think Carmack's decision is perfectly rational, even if it disappoints you.

-Whisp
#16 by "Ashiran"
2001-08-07 14:55:18
ashiran@ashrain.net wtf.couchcrew.com
#12 by Spatula Man
Would you elaborate?... or have you mentioned this already and I missed it ( - I've been away for a few days and couldn't face reading the hundreds of back posts)?

I'm working on a programming language that allows someone to code type 3 AI's with relative ease. I called it SilGol. Its based on a structure that seperates the code into 4 parts. A Body part which contains all the actions the AI can perform, a Memory part which contains stored data, a Soul part which contains the behaviour tree and finally a Awareness part which is responsible for processing all input recieved from the Body and responding to it as specified in the Soul with help of the Memory.

Note that this is all theory as of yet and subject to change while I continue to develop it.
#17 by "Ashiran"
2001-08-07 15:10:18
ashiran@ashrain.net wtf.couchcrew.com
#15 by Whisp
Can you imagine the bitching that would occur when players found out that not only does their game not look as pretty as the one their neighbor is playing, but that it is fundamentally different game as well? That's what would happen if someone did this. Plus, how could you balance the levels? More complex AI is almost certainly (but not necessarily so of course) going to be more difficult to beat.

You are correct ofcourse but the games difficulty shouldn't be directly connected to the system you have. If you have a better system you just can raise the difficulty bar higher then someone on a low-end system.
Also take note that after a certain amount of spended CPU time there is no significant added benifit when even more CPU time is used. So if even a low-end system can allocate that certain amount of CPU time for his AI modules then there is no difference between systems.

I think Carmack's decision is perfectly rational

While it may be perfectly rational its also a stop in AI development. If every developer adopts this additude then we will never see any progress in game AI.
#18 by "MaverickUK"
2001-08-07 15:13:24
peter.bridger@tpg.co.uk http://www.thisstrife.com/
The trouble with game AI, is that's it's a project in itself.

Developers can license a 3D engine (Unreal for example) which then talks to the DirectX layer, which in turn talks to the drivers, and finally the hardware.
AI on the other hand, needs to be written from scratch almost every time.

The Black & White (creature) AI was great, but that was one of the main focus's of the game. I personally don't believe the computer controlled GOD AI in B&W was up to scratch. Not whilst playing along side the AI creature.

The best most developers can hope for, is to refine a certain AI process for a very specific game purpose.
For example, the Thief games are some of the best games I've ever played. The AI of the guards was pretty good. But what *really* made it fun to go up against, was that it felt real..especially with the many 'Taffer' curses being shouted out.

My point is, AI is a great tool for creating a more convising world for the player. But like the polygon-ised world in games, it's mostly smoke and mirrors. The best we can hope for in the immediate future, is a well crafted illustion.

--
Mav
\"What kinda shithole planet is this!?\"
#19 by "Ashiran"
2001-08-07 15:20:17
ashiran@ashrain.net wtf.couchcrew.com
#13 by m0nty
I would argue that 3 is not really much different to 2, and may even be the same./quote] You are correct. Effectivly 2 and 3 are the same. They are both Adaptive AI after all. If you have a type 2 AI with 1 million precoded behaviourpaths then you indeed would be hard pressed to see the difference between a type 2 and 3. The point is that type 2 will never exceed that 1 million nor alter them. A type 3 can get an unlimited number of behaviour paths and he can alter/delete paths that have become obsolete or radically different. Not to mention that it would save an tremenous time in coding the AI. Type 2 requiring millions of lines for all those paths while a type 3 only would need some code to describe its basic motivations.
#20 by "Kayin"
2001-08-07 15:44:13
jeichena@digipen.edu
But when we look at the more scientific minds in the world, they are all trying to create the ultimate self-aware AI.


Have these brainies watched ANY movies at all? Terminator 2? Matrix? Hello? Destruction of the entire human race and all that?

Self-aware AI = not human controlled = ut oh.
#21 by "Kayin"
2001-08-07 15:46:14
jeichena@digipen.edu
btw not to be a jerk or anything...

but if you can't even use bold and quote tags correctly (or at least PREVIEW) how the hell do you code at any decent efficiency?
#22 by "Ashiran"
2001-08-07 15:54:47
ashiran@ashrain.net wtf.couchcrew.com
I never use preview. I think I should.
#23 by "asspennies"
2001-08-07 16:04:23
asspennies@counter-strike.net http://www.asspennies.org/
Kayin, in #20, wrote:
Have these brainies watched ANY movies at all? Terminator 2? Matrix? Hello? Destruction of the entire human race and all that?

That's sci-fi, just like all those movies about the horrors of cloning.  Meanwhile, you see what happens in the real world - the government sets up barriers and laws right away to make sure that not only do we not go to far, but nothing at all gets done.

I wouldn't worry about the little movies you watch coming true.  If we ever did come up with a self-aware AI, you can be 150% sure that it will never, ever be placed in a position where it has access to anything even remotely resembling a weapon.
#24 by "Kayin"
2001-08-07 16:10:38
jeichena@digipen.edu
^_^ just sarcasm kids.

for all seriousness, i think predictable AI is fun. classic gaming traps literally count out "3---2----1----POW!" and if you get hit, you know it's your fault, not the game's. the tension came from just looking at it and thinking... "holy hell... am i going to get thru this? *deep breath* ok here i go..."

but adaptive AI has some potiential if used in the right ways. what i don't want to see is "randomized" enemy behavior... i think a greater challenge lies in giving certain types of baddies a familiar way of doing things, but are adaptive within that scope.
#25 by "MaverickUK"
2001-08-07 16:12:49
peter.bridger@tpg.co.uk http://www.thisstrife.com/
asspennies (#23):
Kayin, in #20, wrote:

Have these brainies watched ANY movies at all? Terminator 2? Matrix? Hello? Destruction of the entire human race and all that?

That's sci-fi, just like all those movies about the horrors of cloning. Meanwhile, you see what happens in the real world - the government sets up barriers and laws right away to make sure that not only do we not go to far, but nothing at all gets done.

I wouldn't worry about the little movies you watch coming true. If we ever did come up with a self-aware AI, you can be 150% sure that it will never, ever be placed in a position where it has access to anything even remotely resembling a weapon.


That "Don't have AI, it will take over the world" argument never holds up. If one day we had an *evil* AI that ruled the world, and was hyper intelligent. Don't you think *evil* AI version 1 would have been a suggestion were more powerful versions of this software where heading?

Big changes take a long long time to come into effect...

--
Mav
\"What kinda shithole planet is this!?\"
#26 by "Kayin"
2001-08-07 16:13:46
jeichena@digipen.edu
If we ever did come up with a self-aware AI, you can be 150% sure that it will never, ever be placed in a position where it has access to anything even remotely resembling a weapon.


yes, but would it have access to our porn?

now there's a matter of national security.

and what if an adaptive AI found a way around these "barriers"? i know it's jumping off a tall bridge but still...
#27 by "Kayin"
2001-08-07 16:16:26
jeichena@digipen.edu
Big changes take a long long time to come into effect...


In human terms. Computers run much faster... an adaptive AI may very well evolve much faster than what us meat based organisms are used to...
#28 by "Ashiran"
2001-08-07 16:21:40
ashiran@ashrain.net wtf.couchcrew.com
You program its basic code. So if you screw that up it will be a humans fault if an AI becomes evil.

Programming a behaviour that goes like "Kill all humans" can have such effect.
#29 by "Kayin"
2001-08-07 16:37:17
jeichena@digipen.edu
as for the reason why AI hasn't changed over all these years while sound and graphics have...

have you ever wondered it's because the general populace would be outright overwhelmed by a truly adaptive AI? i watch my cousins reach for code books five minutes after getting stuck at a puzzle... a truly adaptive AI would probably kick their ass so hard it'd get returned to EB in a heartbeat.

(most) people like to win. (a lot) an adaptive AI would make winning be a lot of work.
#30 by "MaverickUK"
2001-08-07 16:53:29
peter.bridger@tpg.co.uk http://www.thisstrife.com/
Kayin (#29):
as for the reason why AI hasn't changed over all these years while sound and graphics have...

have you ever wondered it's because the general populace would be outright overwhelmed by a truly adaptive AI? i watch my cousins reach for code books five minutes after getting stuck at a puzzle... a truly adaptive AI would probably kick their ass so hard it'd get returned to EB in a heartbeat.

(most) people like to win. (a lot) an adaptive AI would make winning be a lot of work.


That's the thing, adaptive/advanced AI should just be a tool to make the game more fun.
The Sims (for some reason) is good fun. This is a game based around imitetting humans, using clever AI.

AI doesn't and shouldn't be something made to be pitted against the player, in a game of pure logic or reaction times, because this is were  good computer AI is very difficult to beat, and normally not much fun.

--
Mav
\"What kinda shithole planet is this!?\"
#31 by "MaverickUK"
2001-08-07 16:55:58
peter.bridger@tpg.co.uk http://www.thisstrife.com/
E.g. Quake 3 bots on high difficuties, just get more accurate with the rail gun... so the AI is better at playing Q3 than the human, but this doesn't make the game much fun.

UT bots talk out loud, and generally play in a more 'human' way. In my mind, UT is a much better game for this, because the AI helps it to become a more fun game.


--
Mav
\"What kinda shithole planet is this!?\"
#32 by "Kayin"
2001-08-07 17:02:13
jeichena@digipen.edu
good points Mav.

i think that's what i was trying to hit on. Adaptive AI shouldn't be harder, it should just be more immersive. make the characters feel like a drunken shopkeeper, or a ditzy blonde in battle armor, voicing their opinions that are generated on the fly instead of scripted in a txt file somewhere.

of course, something like that would have to be done well, or you're opening yourself up to "all your base" all over again... :P
#33 by "Kayin"
2001-08-07 17:03:10
jeichena@digipen.edu
btw the sims rocks. i have my own little saber marionette family.
customizability wins the day. ^_^
#34 by "jafd"
2001-08-07 17:33:59
kallisti@hell.com http://jafd.isfuckingbrilliant.com
Pick between low-grade AI for everyone or low-grade AI for low-end systems and high-grade AI for high-end systems. Which one would you choose?

The latter, most assuredly. Barring technical impossibilities. You say you want an easier game? WTF, do you cheat at solitaire, as well?

While it may be perfectly rational its also a stop in AI development. If every developer adopts this additude then we will never see any progress in game AI.

Not necessarily a stop, but certainly a slowdown. But this is a cultural issue, even more than a techinical one. It will be quite a little while before the numbers of people interested in a greater challenge are even close to the numbers of people interested in greater graphics. With graphics tech getting as advanced as it is, though, I tend to think that AI development will get quite a boost, more from additional, faster CPU time being available, than from any popular pressure. But it will come, after all, what else is there to work on?

If we ever did come up with a self-aware AI

If an AI is dumb/ignorant enough to let people know that it is self-aware, it isn't terribly intelligent. I think the only rational fear in this area is that a very intelligent AI could be created, without any humans noticing that it is self-aware. I tend to think this is unlikely, but I don't think it is a possibility that should simply be written off. Of course, nor should it be something to be overly concerned about. Wow, that sounds like familiar advice.

UT bots talk out loud, and generally play in a more 'human' way. In my mind, UT is a much better game for this, because the AI helps it to become a more fun game.

I do enjoy the botplay in UT, but I am upset at the way the bots don't get any smarter past a certain point, they just play faster and aim better. I find this rather noticeable, since I find my skills to be not challenged enough by their behaviour after a certain, but challenged too much by their speed, which is the only option available to me now to increase the difficulty. Personally, I'm more interested in exercising my cognitive abilities, than my wrist and the neural pathways leading to it. :( I suppose that means I need more practice. :>)

Given the state of the art, though, it is quite satisfactory. (Especially since I paid twenty bucks for the silly thing.) Most people probably never get good enough at a shooter to notice. Now, how long is it going to be before we end up with "AI accelerator cards"? (I mean, for the computer, not schoolchildren's heads.)

have you ever wondered it's because the general populace would be outright overwhelmed by a truly adaptive AI? i watch my cousins reach for code books five minutes after getting stuck at a puzzle... a truly adaptive AI would probably kick their ass so hard it'd get returned to EB in a heartbeat.

Well, hopefully the goshdarned thing will include an intensity slider and an off switch. I can see the debates now, unlike limited saves, this one could actually lead to bloodshed.

Cool.
#35 by "Anonymous"
2001-08-07 18:56:21
How many games have a feature like SWAT 3 where you can edit a config file to beef up the AI (different settings for your teammates and NPCs) at the expense of performance?  Didn't the Black Isle games have this for path-finding AI (I'm pretty sure Baldur's Gate at least did, I'd assume the other games kept the feature)?  SWAT 3 also gives most of the NPC's different AI settings (less to more hostile) with each play of a map.

Kayin, the thing about an adaptive AI that wouldn't make it kick your ass is that it can adapt to make the game harder or easier.  I don't think anyone would want an adaptive AI that would just keep making the game harder for the player to succeed.  I think the point of an adaptive AI would be to make sure the player is constantly challenged, but never frustrated.
#36 by "Warren Marshall"
2001-08-07 18:56:41
warren@epicgames.com www.epicgames.com
I do enjoy the botplay in UT, but I am upset at the way the bots don't get any smarter past a certain point, they just play faster and aim better. I find this rather noticeable, since I find my skills to be not challenged enough by their behaviour after a certain, but challenged too much by their speed, which is the only option available to me now to increase the difficulty. Personally, I'm more interested in exercising my cognitive abilities, than my wrist and the neural pathways leading to it. :( I suppose that means I need more practice. :>)

Well, it's an FPS ... I think there's only so much they can do.  They run faster, aim better, play better as a team, defend the base better, etc ... they do all the things that a player in an FPS should do.

But like you say, after a certain point, there's not much they can do to get "better".  Do want them to start challenging you with trivia questions or something?  :P
#37 by "jafd"
2001-08-07 19:10:09
kallisti@hell.com http://jafd.isfuckingbrilliant.com
But like you say, after a certain point, there's not much they can do to get "better".  Do want them to start challenging you with trivia questions or something?  :P


What I would really like is for their to be more up-front configurabilty. Say, you can wind up their speed and aim really high, but make them dog-stupid when it comes to getting weapons, or make them ultra-clever tactically, but not quite as johnny-on-the-spot with the aiming.

In UT, I just got to the point where I could outmaneuver the bots on any map that I knew fairly well, predicting when and where they would be presented no greatly significant challenge, but whenever they came on screen, I was pale pink mist. Er, well, green, whatever. Now, I'm the first person to stand up and say, "practice more, punk", but it would be nice if there were more tuning options available to me other than "fast" or "faster."
#38 by "szcx codemonkey"
2001-08-07 19:16:11
But like you say, after a certain point, there's not much they can do to get "better". Do want them to start challenging you with trivia questions or something? :P

Just what is the capital of Venezuala?
#39 by "Warren Marshall"
2001-08-07 19:35:29
warren@epicgames.com www.epicgames.com
What I would really like is for their to be more up-front configurabilty. Say, you can wind up their speed and aim really high, but make them dog-stupid when it comes to getting weapons, or make them ultra-clever tactically, but not quite as johnny-on-the-spot with the aiming.

I've never done it myself, but apparently there IS a bot configuration file where you can create customized bots.  Create one that prefers sniping, one that runs fast, etc.
#40 by "jafd"
2001-08-07 19:41:50
kallisti@hell.com http://jafd.isfuckingbrilliant.com
Super! Now I just have to bitch some more about the G(imped)UI.
I'm still looking to configure the in-game browser to only show me populated DM servers with all Relics enabled and Translocator disabled. Why play anything else...
#41 by "Dumdeedum"
2001-08-07 20:02:26
Note that the opinions on how to create an 4 are quite different. From fuzzy logic and neural network combos to stupid ideas like Mindpixel which is basicly nothing more than a program that mimics a real AI. This is what some people call a MEGAPARROT(thanks to Desiato :).


I think one of the best uses you could use something like the mindpixel system for is in adventure type games, being able to type in any question you want would be much better than the question a, b or c method.  It may only mimic intelligence but since games only mimic everything else it would suffice.
#42 by "Gomez"
2001-08-07 20:21:29
freaparn@hotmail.com
While the AI in B&W may have been leaps and bounds ahead of the "if A, then B" of previous work, I still found it to be a glorified tamagotchi. Understand I say this as someone who knows nothing of programming AI, but as a gamer. It may be an impressive advance over past work, but since it was touted as "almost humanlike", I have to call it a massive disappointment. I certainly wasn't impressed by a creature which prefers eating it's own shit over cattle, and chooses starving to death over eating fish. When the damn thing still the same stupid mistakes by the halfway point in the game, despite all my coddling and punishing, there's something of a letdown for this so-called amazing AI.

Hype is a very bad thing.
#43 by "jafd"
2001-08-07 21:40:54
kallisti@hell.com http://jafd.isfuckingbrilliant.com
When the damn thing still the same stupid mistakes by the halfway point in the game, despite all my coddling and punishing

Yeah, it is an advanced toy. /snicker
#44 by "Darren Coleman"
2001-08-07 21:57:03
durzel@barrysworld.com http://www.superficial.net
AI in FPS can more often than not simply result in computer controlled enemies reacting quicker and shooting more accurately which, beyond a certain level, actually has the effect of rendering the AI less human like.  Some of the most accurate AI I have seen to date has been that of the Gladiator bot for Quake 2, and more recently that of the Quake 3 Arena bots (unsurprising since they were created by the same person).
#45 by "Ergo"
2001-08-07 22:01:47
stu@dsl-only.net
more recently that of the Quake 3 Arena bots (unsurprising since they were created by the same person).


Well, if you set the bot AI high enough, you'll notice it cheats. It ALWAYS knows where you are, even through walls.
#46 by "jafd"
2001-08-07 22:11:20
kallisti@hell.com http://jafd.isfuckingbrilliant.com
It ALWAYS knows where you are, even through walls.

Yeah, I don't like that. I can deal with a bot with inhuman reflexes, that's fine and dandy, getting smacked around will improve skill. But a clairvoyant opponent, that's just some bad, smelly ju-ju.
It is exciting to think of all the room for advancement there is in this field.
#47 by "Narcopolo"
2001-08-07 22:41:47
Some AI thoughts.  I think Desiato's right, and with the GeForce 3 and up we'll see a lot more CPU cycles freed.  This should result in sound and AI getting a lot more attention.  If 3D sound acceleration becomes required, and since there are perfectly good 3D sound cards for $40, I don't see why not, then that means more room for AI.

However, I don't really see why AI can't get more attention right away.  Think about RollerCoaster Tycoon, which has hundreds of little people running around your park, each with their own motviations.  The system requirement for that game is a P90 if I remember right.  The graphics aren't very demanding, but we're starting to see a real pulling away from the requirements of 2D vs 3D games.   The Sims was mentioned as well, and that AI is state of the art, while not very system demanding.  It also bears repeating that those are the two best selling PC games of the last couple years. Publishers certainly will notice.

As soon as 3D games all start becoming GPU dependent, we'll see a renaissance in AI in 3D games, which is another question.   Half Life was revolutionary to me for trying to emulate several senses with its AI.  Like the guys with the head crabs on them who smelled your trail and followed you, or the big arm that heard your footsteps/felt your vibrations.   The marines were interesting, but I think that there's a lot of room for experimentation with how the environment reacts to you, that wouldn't necessarily take a lot of cycles.

Last AI bit.  In Q3A, Xaero on Nightmare.  Final boss, final level of the single player.  Shoot him from across the arena when he has his back turned.  He will, every time, jump up in the air, spin around, and rail you before he lands back down.  Every time my reaction was the same.  Fucking bullshit!
#48 by "Apache"
2001-08-07 22:57:14
>>But like you say, after a certain point, there's not much they can do to get "better". Do want them to start challenging you with trivia questions or something? :P<<

LOL! that would make a great penny arcade comic
#49 by "__try"
2001-08-07 23:17:54
Hmm,
There is one aspect to a type3 that nobody has mentioned; QA and balancing.  I bet that well-coded learning AI will find every damned exploit in the game, given enough time.  Of course exploits outside of the game are different, but if you were to run an adaptive AI through a game enough times it could find a lot of exploits quickly.  Bunny jumping, for example, stands out as an expoit that an adaptive AI would start on and stay on.  

Similarly, if it had enough experiences, I bet that it would tend to expoit any bounding box errors by defending itself in strange positions, for example.  Things of that nature; run the AIs overnight, see what they are ALL doing, then rebalance the game...
#50 by "David"
2001-08-07 23:36:49
david@snowdesign.com http://www.snowdesign.com
Ergo,

Ever played a truly good dueller in Q3A? They have *amazing* prediction skills. I remember seeing a T2 duel with 'scoob' from the pov of the poor fellow he was smacking around. He'd run down one corridor, and BOOM a rocket explodes right as he's approaching one door, and scoob is nowhere to be found. Five seconds later, he runs to another door, BOOM another rocket explodes and you see scoob tooling away across the room... he had to shoot those rockets three or four seconds before the opponent got there, and each time if this guy hadn't had stopped he'd have eaten them. IMO the prediction isn't so bad.

What I DO dislike is hitting Anarki with a rocket and he rails me perfectly in the split-second after I hit him, while he's flying at 80 mph. I mean, I can see it if he gets a bead once he hits a wall - I've seen that in human player duels - but not *right* after he gets nailed.

Oh yeah and another thing - UT's bots shine compared to Q3A. Much more variety in the behavior and more human-like tactics. Now if only Tamerlane and Loque weren't so damn accurate with that damn sniper rifle. (Tamerlane put a bullet through Snowman's head!) :D

David
C O M M E N T S
Home » Topic: Adaptive AI in games?

|«« - Previous Page - Next Page - »»|
P O S T   A   C O M M E N T

You need to be logged in to post a comment here. If you don't have an account yet, you can create one here. Registration is free.
C R A P T A G S
Simple formatting: [b]bold[/b], [i]italic[/i], [u]underline[/u]
Web Links: [url=www.mans.de]Cool Site[/url], [url]www.mans.de[/url]
Email Links: [email=some@email.com]Email me[/email], [email]some@email.com[/email]
Simple formatting: Quoted text: [quote]Yadda yadda[/quote]
Front Page (ATOM) • Submission Bin (1) • ArchivesUsersLoginCreate Account
You are currently not logged in.
There are currently 0 people browsing this site. [Details]