Why can't we ever have decent AI?


  • Please log in to reply
57 replies to this topic

#1 Slaughter

Slaughter

    Colonel

  • Admin
  • PipPipPipPipPipPip
  • 6,409 posts
  • Gender:Male
  • Location:Norway, Oslo

Posted 16 March 2006 - 01:13 AM

One thing that annoys me with most games in most categories is the lousy AI. Through my years of gaming, playing everything from FPS to TBS, I've been grumbling over shitforbrain AI. Do you think the reason is that it's too hard to program, or that the developers aren't focusing on it?

One thing I've seen in lots of FPS games through the years is their idea of "good AI". On easy level the accuracy of the enemy is very low. They hit nothing, and you can run circles around them. Then, as you increase difficulty, they get much better. In quite a few cases they go inhuman, and make headshots from distances that no human player could ever manage, NO MATTER HOW GOOD HE WAS. That is an example of shitforbrains AI.

Typical lousy tactical AI includes rushing you like most mutants and aliens does in UFO: Aftermath. There's supposed to be some AI behind it all, but they sure camouflage it well! Can it really be so hard to make the Reticulans cooperate a little? To have them approach you from two sides of a building at least?

I have two examples where the AI impressed me in my life. There might be more, but I sure can't remember them.

The first one was while playing the FPS game Far Cry (good game by the way). Fighting my way through a ruined temple, I had to take cover behind some stairs.While there I ducked up and down to try and fire at them. I expected the retards to rush me as usual, but instead I suddenly heard the unmistakable sound of a grenade being utilized. It was heading my way, so I rushed out from the cover. A wall of bullets made that my last mistake.

The second time that I can remember was while playing the almost finished UFO: Aftershock at ALTAR's office in Brno. My first encounter was with two Reticulans in a mission. I downed one of them before they could react, and the other ran away in panic. That's what I thought at least, until I rushed around the corner of an old train. 5 or 6 Reticulans were sitting there in ambush, and my team never really recovered from that.

So I ask you again; Is it really THAT hard to produce proper AI? I don't need stunning, but decent would be nice!


#2 NKF

NKF

    Mr. Badger in disguise

  • Site Staff
  • PipPipPipPipPip
  • 4,449 posts
  • Gender:Male
  • Location:In my mind

Posted 16 March 2006 - 04:33 AM

At first, the lack of AI was because making a good AI would take up too many CPU cycles and you want the computer to be able to think fast. Especially so for games where everything plays out in real-time. For turn based, you don't want the computer to get into a critical state or a deadlock or even a situation where it has to take into account far too many possibilities that it looks like the game's locked up (which happens to me a LOT with turned based games on my older PC, like Apocalypse and JA2).

These days we have more CPU cycles to go around, but even then, the only good AI I've seen involve scripted events - which doesn't count (such as in Homeworld II). Or if you play a game that appears to have good AI (like Halo), becomes incredibly predictable after a while once you spot what conditions trigger what actions. I can't remember what game it was, but there was one where the enemies would jump or move away if you did something like toss an explosive near it. This would seem smart at first, until you watch the same enemies jump off a ledge into an endless pit when they could've just moved in a different direction.

I'm a bit tired of enemies that are static with perhaps a small walk route that start moving towards you or opening fire on you the moment you get within range. There's just far too many of these.

- NKF
NKF - finally built a gaming PC in 2020 (though not any of that RGB nonsense). Can now play games up to '89!

#3 FullAuto

FullAuto

    Catching the next pimpmobile outta here!

  • Chief Editor
  • PipPipPipPipPip
  • 4,853 posts
  • Gender:Male
  • Location:UK

Posted 16 March 2006 - 06:32 AM

Amen.  When are we going to get enemies that have several different possible responses?  When are we going to get enemies who gang up, cover each other, toss grenades, perform flanking manoeuvers, circle round behind you, ambush you when you're NOT expecting it, run away and hide when outgunned and so on and so forth, and WHY aren't AI opponents doing this now?  I'm sure it's perfectly possible.

  I'm pretty sure it hasn't already been done because A) It's not an immediately obvious improvement over other games, like better textures or the latest in facial hair mapping and B) I suspect lots of gamers would find opponents with anything more than vestigial AI far too challenging to fight, and merely label the game 'too hard'.  Lots of games these days hold your hand.  The days of the challenging difficulty level are mostly gone.

Still, I don't think the situation is going to get any better.  :D

Erfworld - the finest comic about turn-based gaming ever.


#4 Slaughter

Slaughter

    Colonel

  • Admin
  • PipPipPipPipPipPip
  • 6,409 posts
  • Gender:Male
  • Location:Norway, Oslo

Posted 16 March 2006 - 10:32 AM

 NKF, on 16th March 2006, 5:33am, said:

At first, the lack of AI was because making a good AI would take up too many CPU cycles and you want the computer to be able to think fast...
Yeah, that much is true. Still, with the processing power of today it should be possible. The problem with CPU utilization is just as much because of useless code as the need for power however. Optimising of code is foreign language to most developers these days. This is mostly due to their tight schedules I guess, but it is also a problem with their focus.

I mean, if you're making a game that will never appeal to the average "Halo roxors" crowd anyway, why focus on bump mapping or whatever it is called? (as FullAuto mentions) Strategy games should sacrifice a little graphics for AI, as serious strategy gamers are looking for beauty in challenge and not appearance.

@FullAuto: A) That is true. On the other hand, some strategy games aren't made to sell to the average crowd. Why can't they focus on features that WOULD intrigue old time fans of the genre. B) That is true, but there are two ways "around" this. First of all, you can scale the AI based on difficulty. If you're playing easy, the AI will make several mistakes, while on hard it will chew your arse. Second, we're back to the target audience. If you're making a turn-based game like Silent Storm, you're not looking to sell to those who would find it too challenging.

I agree that it will probably not happen anytime soon however. At least not in many games.


#5 Matri

Matri

    Official Love Nest Mascot

  • Members
  • PipPipPipPip
  • 587 posts
  • Gender:Male

Posted 16 March 2006 - 11:09 AM

 Slaughter, on 16th March 2006, 9:13am, said:

One thing I've seen in lots of FPS games through the years is their idea of "good AI". On easy level the accuracy of the enemy is very low. They hit nothing, and you can run circles around them. Then, as you increase difficulty, they get much better. In quite a few cases they go inhuman, and make headshots from distances that no human player could ever manage, NO MATTER HOW GOOD HE WAS. That is an example of shitforbrains AI.
I agree. A number of games these days use this method of "AI". On easy, they're almost passive. Sometimes I think the only reason they're not frozen in place is to give you the impression that they're "alive". On difficult, it's as if their stats have been pumped up to the maximum the storage variables will allow. Imagine a race where everyone supposedly has "duplicate" cars. Imagine finding out the AI's "duplicate" cars have twice the acceleration of yours, in addition to being solid as rocks. Meaning one nudge from them and you go flying off the track. On the other hand, one nudge from you and you go flying off the track.

 Slaughter, on 16th March 2006, 9:13am, said:

Can it really be so hard to make the Reticulans cooperate a little? To have them approach you from two sides of a building at least?
I know the Total War series employ flanking maneouveres, and Knights Of Honor does try it, but their timing is off if the terrain doesn't allow for it. It's becoming a lost art.

 Slaughter, on 16th March 2006, 9:13am, said:

The second time that I can remember was while playing the almost finished UFO: Aftershock at ALTAR's office in Brno. My first encounter was with two Reticulans in a mission. I downed one of them before they could react, and the other ran away in panic. That's what I thought at least, until I rushed around the corner of an old train. 5 or 6 Reticulans were sitting there in ambush, and my team never really recovered from that.
Han Solo, I presume? :D
"Never be normal!" - Ron Stoppable

MechQuest. Get your mecha RPG fix.
Star Citizen.

#6 Kernel

Kernel

    Her Majesty's Imperial Guard

  • Members
  • PipPipPipPip
  • 125 posts

Posted 16 March 2006 - 11:27 AM

Some of the best AI I've seen so far was in the demo of FEAR. I threw a grenade, which knocked over some shelves blocking two enemy soldiers' path. I thought "Ha... your stuck now"... but alas I was wrong. One of the enemy soldiers decided to jump over a low wall while the other proceeded to crawl through a gap under the shelves that I had knocked over.

One of the main problem with creating realistic AI in modern PC games are the HUGE number of different scenario's it has to deal with and the huge number of decision it has to make.
Where's my nearest cover? What actually constitutes as cover? What way is the player moving? What weapon does the player have? Where are my allies? Is there anything between me and my cover? Is there anything between me and the player? How do I get to my cover? Will moving to my chosen cover place me into a corner?, What happens if another ally starts attacking the player, etc.
Just think of all the decisions you would have to make in a given situation, and remember those that you can thing of are probably only half of them. The rest would be subconscious decisions your mind would make when you carried out the action. A PC game AI doesn't have the luxury of a sub-consciousness.

With the increase of cpu power over the last few years creating AI that can handle these sorts of decisions has become easier. I recently read an article about AI that said to make an "intelligent AI" isn't that much of a problem any more; the real problem is what's known as artificial stupidity. The ability for the AI to make "stupid" mistakes, just as a human would is actually a lot harder than making an AI appear intelligent.
"Any rumors of a military coup by the Imperial Guard are greatly exaggerated."

Duct tape is like the force. It has a light side, a dark side, and it holds the universe together....
-- Carl Zwanzig

#7 Kret

Kret

    Lieutenant

  • Site Staff
  • PipPipPipPip
  • 759 posts
  • Gender:Male

Posted 16 March 2006 - 11:52 AM

All AI has to be scripted, as the behaviour of coputer controlled units depends on the actions of human controlled units. However, event scripting is one thing and behaviour scripting another. The former is a way for a game to cheat but is used normally in campaing settings while the latter is the one we most crave for improvement.

The "leaping to their doom rather than face a grenade" example given by NFK is found in HALO at least. This is probably because they never included in the script a "safe landing" check. But can't this be also considered part of the AI, after all, mistakes are part of intelligence.

AI is much more complex that just sheer computing power. And contrary to the suggestion, it can't really be scaled, just classified. An example, what would be the "smartest" action taken when a unit is under attack? A lot of queries have to be made depending on the depth of the game's engine, like: is the enemy visible? is the unit alone?unit "personality" (coward, berserker, etc...) is the damage taken severe? how wounded is the unit? how wounded is the enemy? nearest cover available, nearest allied units, weapon types of both the unit and the enemy, etc... Building a script that can properly evaluate all possible combinations of different conditions is hard, particulary when conditions are in constant change. The more factors, the more complex the script must be, and the higher the risk of it making odd decisions or even a virtual "multiple personality syndrome" with the unit constantly changing the choice made.

A curiosity for some that like to know about it, is that there's an application challange out there that several universities and freelancers take part in regarding the AI of a virtual "football" team. The basic restrictions is that all units have a visual limitation of a 90 degree angle and can only turn, move forward and I think "passing" the ball was allowed too (as far as I recall) The AI behaviour from that point on is for everyone to code. Well, in about 90% of cases, the sheer, "everyone chase the 'ball' and walk into the goal" strategy was the most effective of all. In other words, the best strategy available for computers is rushing.
There's always some dirt to dig out

#8 Accounting Troll

Accounting Troll

    Bridge troll

  • Admin
  • PipPipPipPipPip
  • 1,060 posts
  • Location:Wales

Posted 16 March 2006 - 01:21 PM

Matri, don't get me started on the Total War series AI.  The one that came with Rome Total War was absolutely pathetic.  It mostly consisted of suicide charges up a hill only to be decimated by my archers and javelin throwers :D

One problem with creating a good AI is that the computer has to calculate each one of its possible choices in each situation and make the optimal move based on its calculations without appreciably slowing doewn the game play.

An experienced human player is more intuitive and he is capable of learning from experience, which gives him the edge in a strategy game.  A veteran X-Com player will know the value of getting a sniper onto a particular rooftop or the correct methodolgy for storming a landed Sectoid battleship without using psionic weapons.

Another problem is that game designers are under pressure to produce a game that will have strong sales, which means that it will have to appeal to the professional reviewers in gaming magazines and websites.  The professional reviewers rarely have more than a few hours to play the game before they write their review, so they tend to be won over by games that are easy to get into and have excellent graphics.  I seem to recall that some reviewers critisised UFO: Aftershock as being too difficult because the AI sometimes makes good choices ;)

#9 Matri

Matri

    Official Love Nest Mascot

  • Members
  • PipPipPipPip
  • 587 posts
  • Gender:Male

Posted 16 March 2006 - 09:38 PM

Hey, when you're on a timed assault, even human players will suicide charge up that hill.

There was one game with good AI that I still remember. Z by Bitmap Brothers. Not the newer Steel Soldiers, this is the original. I'm embarassed to say the AI successfully used decoy tactics on me. They grabbed my radar building while I was trying to hold off an attack on my heavy factory. Mission went downhill from there, meh. :D
"Never be normal!" - Ron Stoppable

MechQuest. Get your mecha RPG fix.
Star Citizen.

#10 Slaughter

Slaughter

    Colonel

  • Admin
  • PipPipPipPipPipPip
  • 6,409 posts
  • Gender:Male
  • Location:Norway, Oslo

Posted 16 March 2006 - 11:46 PM

The points you have posted about AI is true enough. AI is hard to program, because of all the factors that go into every choice, and it takes a lot of time (something developers rarely have enough of). There are however a few points that I find odd:

1: AI should be pretty similar for A LOT of games. Take a developer like Nival. They have developed something like 15 PC games, and several of them are very similar (Silent Storm, Silent Storm: Sentinels, Hammer & Sickle and Nightwatch are all small squad tactical turn-based games using more or less the same engine). This means they should be able recycle large parts of the AI code. And not just from those games! Pretty much all games of similar genres use fairly similar AI.

2: The AI should be able to learn. First of all, take a look at something like a spam filter. It learns slowly but surely. Developers could include AI that taught itself with their games. In the beginning it lacks combat experience, and will make stupid mistakes. As you play the game, it improves. That's how robots learn these days, so why shouldn't the AI. This would be VERY hard to develop, but OH would it be fun :D. And I'm not saying the AI should be a moron to begin with, just that it should improve through play.

Problem with this would be that all the good "training" would disappear during reformat / uninstall. This could be solved by allowing players to make a backup of the AI file, or ever better. What if all players had the option to upload their AI to the developer website. That way they could collect large amounts of AI data, and players could download the "AI profiles" they could make from this. Guess I'm talking something that is unlikely to happen, but it would be OH SO GOOD!

3: Why aren't there developers working solely with AI? Developing an RTS and outsourcing the AI could be a good way to do stuff. Or at least split the developer into specialized departments. Nival has more than 300 developers. Why not have one engine department, one AI department and so on?

Conclusion: I see your arguments, but why can't developers recycle AI code to a larger extent. ALTAR is developing their third more or less similar game now...


#11 FullAuto

FullAuto

    Catching the next pimpmobile outta here!

  • Chief Editor
  • PipPipPipPipPip
  • 4,853 posts
  • Gender:Male
  • Location:UK

Posted 17 March 2006 - 12:00 AM

I think it could be something as simple as the fact that improved (or just plain good) AI is not immediately apparent.  You can't show it in screenshots, unlike the latest in graphics.  So it's not only costly to develop, there's no immediate reward (unlike if you dropped a few hundred grand into the graphics budget, say).

I'd like to be proved wrong, though, and have someone come up with AI that can at least challenge players.

Erfworld - the finest comic about turn-based gaming ever.


#12 Kernel

Kernel

    Her Majesty's Imperial Guard

  • Members
  • PipPipPipPip
  • 125 posts

Posted 17 March 2006 - 09:01 AM

 Slaughter, on 16th March 2006, 11:46pm, said:

1: AI should be pretty similar for A LOT of games. Take a developer like Nival. They have developed something like 15 PC games, and several of them are very similar (Silent Storm, Silent Storm: Sentinels, Hammer & Sickle and Nightwatch are all small squad tactical turn-based games using more or less the same engine). This means they should be able recycle large parts of the AI code. And not just from those games! Pretty much all games of similar genres use fairly similar AI.

To an extent they do recycle a lot of their code, including the AI code. The problem is requiring new better AI and using OLD code is somewhat mutualy exclusive.

 Slaughter, on 16th March 2006, 11:46pm, said:

2: The AI should be able to learn. First of all, take a look at something like a spam filter. It learns slowly but surely. Developers could include AI that taught itself with their games. In the beginning it lacks combat experience, and will make stupid mistakes. As you play the game, it improves. That's how robots learn these days, so why shouldn't the AI. This would be VERY hard to develop, but OH would it be fun :D. And I'm not saying the AI should be a moron to begin with, just that it should improve through play.

Problem with this would be that all the good "training" would disappear during reformat / uninstall. This could be solved by allowing players to make a backup of the AI file, or ever better. What if all players had the option to upload their AI to the developer website. That way they could collect large amounts of AI data, and players could download the "AI profiles" they could make from this. Guess I'm talking something that is unlikely to happen, but it would be OH SO GOOD!

Remember it's as easy to learn the wrong way of doing something as it is the right way of doing something. Taking your spam filter example. I'm forever having to go into my junk mail folder and recover good e-mails because the spam filter has decided through learning that they are junk. So how do you tell the AI the way it's doing something may not be the best way? An example is path finding... If you have two routes to a destination, one long route and one short route, and the AI learns the long route and reaches it's destination. How do you then make the AI learn there's a shorter route when as far as it's concerened the route it's found is correct?

 Slaughter, on 16th March 2006, 11:46pm, said:

3: Why aren't there developers working solely with AI? Developing an RTS and outsourcing the AI could be a good way to do stuff. Or at least split the developer into specialized departments. Nival has more than 300 developers. Why not have one engine department, one AI department and so on?

There are developers working on 3rd party AI engines, there are even people working on AI hardware for add-in cards like a graphics card. These would mainly deal with things like path-finding etc. and are designed to take some of the AI work load off the CPU. For large game worlds where an entity must cross from one side of the world to another, path finding can actualy be quite an expensive task to do. Especialy when you want to make that path look realistic. There are numerious optimisations around but they tend to make the path look less realistic, or make the path less optimal in terms of length, etc.
"Any rumors of a military coup by the Imperial Guard are greatly exaggerated."

Duct tape is like the force. It has a light side, a dark side, and it holds the universe together....
-- Carl Zwanzig

#13 Kret

Kret

    Lieutenant

  • Site Staff
  • PipPipPipPip
  • 759 posts
  • Gender:Male

Posted 17 March 2006 - 09:13 AM

Quote

The AI should be able to learn
I believe that his was attempted with Apocalypse back then. Something about a "learning matrix". Can't say I saw much of it while playing...

Quote

there are even people working on AI hardware for add-in cards like a graphics card
This just sounds plain odd, but I wouldn't mind work being done in some sort of an AI co-processor.
There's always some dirt to dig out

#14 Kernel

Kernel

    Her Majesty's Imperial Guard

  • Members
  • PipPipPipPip
  • 125 posts

Posted 17 March 2006 - 09:24 AM

 Kret, on 17th March 2006, 9:13am, said:

This just sounds plain odd, but I wouldn't mind work being done in some sort of an AI co-processor.

That's basicaly what I meant. You plug the card into a PCI slot like you would a graphics card.
"Any rumors of a military coup by the Imperial Guard are greatly exaggerated."

Duct tape is like the force. It has a light side, a dark side, and it holds the universe together....
-- Carl Zwanzig

#15 Bomb Bloke

Bomb Bloke

    The Smily Admin

  • Admin
  • PipPipPipPipPip
  • 4,625 posts
  • Gender:Male
  • Location:Tasmania (AU)

Posted 17 March 2006 - 12:50 PM

My experiance of modern gaming pretty much stops at Aftermath, but I didn't think the AI was too bad in that. Once a Reticulan spotted my soldiers, and hid behind a wall. All the other aliens in the area joined him, and the first soldier to stick his head through got blasted.

Sure, it's an obvious ambush, but in order for the computer to know where I am he had to risk me spotting his unit.

TFTD apparently had better AI then UFO. I once spotted an Aquatoid, who rather then stay and fight, decided to run around his small craft and come up behind my trooper. Would have worked a treat if I hadn't decided to follow him; nearly did a full lap before he cottoned on and came back after me...

Magic Carpet is a more obscure title, but I quite lacked the AI in that, because it plays much like a human would. The enemy carpeters gang up and fight against each other, not just you; they tend to either go for whoever has the most or the least power (it's easy to wipe a small force off the map, but if you leave the big fish to the opposing sides you risk losing your slice of their considerable amounts of mana). But more notably is the way they fight, blasting other players whenever they get the chance without mercy, and running away to safety whenever they're taking a beating. It's not complex, but it's how a normal person would play.
BB's X-Com Projects Page - X-Com Games At GamersGate
You're just jealous 'cause the voices only talk to me :P
We love Tammy! :)

#16 Strong Bob

Strong Bob

    Sailor Jupiter is hot

  • Members
  • PipPipPipPip
  • 567 posts

Posted 17 March 2006 - 01:43 PM

I've never had a problem with AI, before... Mainly because, as Fulllauto pointed out, it's difficult to notice.

Halo and Doom 3 are the only FPS games I play right now. And of course I see AI crap outs all over the place. Halo pulls it off rather well, in my opinion, I just think the enemies strike me as kinda... "deaf". Doom 3 manages to hide the quality of it's AI by never really providing any "thinking" environment for the NPCs. (Although they seem to have a hell of a time chasing me whenever I ran up ladders.)

Real time strategy... I discussed this with someone a very long time ago, and we came to the conclusion that a real time strategy game with a realistic "human-like" response and skill would be significantly more trouble than it is worth. Reason being... In real-time, the computer takes actions into account with a blink of an eye, whereas a human would need several minutes (Perhaps even hours) to take in the same information. (Reason being, the computer is operating an entire environment, along with each individual unit's AI. It is therefore computing the location of every single unit, stats, conditions, modifying specials, along with the environment, each strategizing computer opponent, and the flora/fauna should it apply.) As such, for the computer to make human-like choices, complete with mistakes, it would take a strong AI to do it all. Not only that, how can it account for the time it takes to drag a mouse over to a building and click? In StarCraft, I wager it takes me about 3 minutes to move to each building and make a unit out of each and every structure on the Terran side. A computer would do the exact same within the blink of an eye, buying precious seconds. Now build this up over time... The computer has a generalized edge in speed. It will build entire armies in a fraction of the time it takes any fleshy individual. And even if you made it so that the computer has a delay between making units, it creates the unrealistic feel, as each interval of time could easily be identified and exploited like any other AI bug.

Turn based strategy appears to be the only plausible genre for a good AI. Not nearly as hard, and the Computer AI needs not deal with a constantly changing set of events. Just hit the "End Turn" button and the AI deals with the information presented right there.

The best Turn Based AI I've ever seen were in Master of Orion II and III. Followed very closely by Civilization II. The worst I've ever seen was X-COM Apocalypse (And it's also the reason why I only play real-time on that one.).

Master of Orion seems to handle AI like a pro. In II, the interface was easy, the amount of races around was rather small, and generally, the ENTIRE game was basically played off the AI, making it the prime feature of the game. III does it to a significantly better degree, as it incorporates dozens upon dozens of other game aspects and still manages to pull off a decent AI. (It's one and only flaw in that aspect is it's sorely underdeveloped combat interface.)

X-COM Apocalypse? Heh, I felt like I was heading up a squadron of stooges. Whenever a panic attack hit my men (And sometimes not even that...) I found them randomly running RIGHT UP to an Anthropod, kneeling, standing up again, then run around them to the rear (Staying adjacent the whole time) turning towards the Anthropod, then stepping back. Another such occasion (As a means of "reaction fire") saw a soldier running around in circles in a small room after an alien walked by a door. (Wouldn't it have been more likely to just run the other direction? It was almost like dealing with n00bs back in the days of XCAS...)

Meh, just felt like putting in my two-cents. Honestly? I think the solution to "bad AI" is simply multiplayer gaming. That way you get the real human intelligence right there.
Strong Bob's random Star Trek quote of the year:

*Scans rock* "It's dead, Jim." - McCoy

"Very funny, Bones." - Kirk

#17 Accounting Troll

Accounting Troll

    Bridge troll

  • Admin
  • PipPipPipPipPip
  • 1,060 posts
  • Location:Wales

Posted 17 March 2006 - 02:01 PM

Slaughter, do you really want an AI capable of learning?  Haven't you ever seen the Terminator films?  Or the Matrix?

Seriously though, as I recall, the instruction manual for X-Com Acopalypse promised that not only would the AI be capable of learning, but you would be able to swap a particular file with other players thus getting an AI that had developed tactics against them that you would never have encountered before.  I mostly seemed to get humanoid wave attacks...

#18 Gimli

Gimli

    Dances with Mutons

  • Retired Staff
  • PipPipPipPipPip
  • 1,036 posts
  • Gender:Male
  • Location:Croatia

Posted 17 March 2006 - 02:26 PM

Slaughter said:

Problem with this would be that all the good "training" would disappear during reformat / uninstall. This could be solved by allowing players to make a backup of the AI file, or ever better. What if all players had the option to upload their AI to the developer website. That way they could collect large amounts of AI data, and players could download the "AI profiles" they could make from this. Guess I'm talking something that is unlikely to happen, but it would be OH SO GOOD!

Have you been reading my mind recently? :D This is almost exactly what I had in mind. That either the player could choose from these online files, and maybe even that the developer checks some of the most popular files and then puts them in the next patch for the game. To take it a step further, it would be really nice if the player was able to make his own scripts for AI. With proper motivation, the players might compete in the field of making good AI. I don't know if this would be possible. Of course, I don't see this happening very soon...

#19 Crazy Gringo

Crazy Gringo

    War is hell and war never changes therefore hell never changes

  • Site Staff
  • PipPipPipPip
  • 512 posts
  • Gender:Male
  • Location:Albertslund, Denmark

Posted 17 March 2006 - 04:56 PM

Bad AI is all around....And setting the AI to behave like a 'human' can be really tiresome....And only if you can adjust the AI in the game you are playing. I remember in BF1942 that you can adjust how much CPU power the AI should have. Set it at 5% and the AI is nobrainers not know what to do when it is your team mates. Set it higher and you get the 'super human ability' problem.
AI in games needs to be able to adapt, improvise and overcome even if it means failure sometimes....And most of all no cheating AI!!! Even if it is 'supposed' to improve gameplay.
Words to live by from Sergent Gunnery Highway: ' You adapt, improvise and overcome!'

'To kill a dream it is not enough to kill the dreamer', -Quote by Crazy Gringo until proven otherwise.

Mental Father of Air Morelman and Air Wargot!

Official StrategyCore UFO: Afterlight previewer. Preview here!

Has joined the league of 'The Crazy 8800s'

#20 Matri

Matri

    Official Love Nest Mascot

  • Members
  • PipPipPipPip
  • 587 posts
  • Gender:Male

Posted 17 March 2006 - 09:55 PM

 Strong Bob, on 17th March 2006, 9:43pm, said:

Meh, just felt like putting in my two-cents. Honestly? I think the solution to "bad AI" is simply multiplayer gaming. That way you get the real human intelligence right there.

No. sometimes I get n00bs with the same level of "bad AI". The only way to tell sometimes is that they whine creatively. :mad:
"Never be normal!" - Ron Stoppable

MechQuest. Get your mecha RPG fix.
Star Citizen.




0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users