PDA

View Full Version : The A.I. will make or break this game



Unstoppable
10th Feb 2008, 18:09
I haven't seen much discussion on this issue at all. The A.I. in Invisible War was pitiful. The Deus Ex A.I. was alright because on realistic you could die from a head shot.

What I'm getting at though is that we as the players need A.I. that constantly challenges us. Without that this game is good as dead in the water. I want special agents that come after me during a mission and use powerful nano abilities that kick my butt.

This happened in Crysis but I was able to beat about 4-5 nano suit wielding A.I.'s a bit too easily. (On the hardest setting)

In conclusion what I'm asking for is that the game punishes me for playing on the hardest difficulty. I want to spend a few hours on a level getting my butt kicked and having to figure out how to out smart the A.I. That way there is something for those of us seeking a true challenge.

Lately though I'm beating FPS games because I can figure out the A.I. in a matter of hours. I guess I just have a small niche for figuring out their weaknesses. I hope the QA team on this one shows no mercy and makes the A.I. tough as nails on the hardest setting.

Papy
10th Feb 2008, 18:26
I also love challenging games, but, if I undertood what you meant correctly, not in the way you like it. When I played Deus Ex, even the first time, it was on realistic and I was saving only at the begining of the level. If I died, I had to redo everything.

I really hate the "try, die and reload" kind of gameplay. I love when the challenge is with thinking, not repeating the same action over and over until you learn a way. Basically I want the game to kick my butt ONLY if I make a mistake. If I don't, I should not die. The game should punish me for making mistakes, not for playing on hard.

gamer0004
10th Feb 2008, 18:45
I also love challenging games, but, if I undertood what you meant correctly, not in the way you like it. When I played Deus Ex, even the first time, it was on realistic and I was saving only at the begining of the level. If I died, I had to redo everything.

I really hate the "try, die and reload" kind of gameplay. I love when the challenge is with thinking, not repeating the same action over and over until you learn a way. Basically I want the game to kick my butt ONLY if I make a mistake. If I don't, I should not die. The game should punish me for making mistakes, not for playing on hard.

But in DX it only did when you made a mistake, or when you simply screwed up the fighting.

jordan_a
10th Feb 2008, 20:27
The overall difficulty level is directly linked to the save system.

On DX it was too untroublesome.

It's also bound to the foes' accuracy, the damages they inflict and finally to their AI.

Concerning the AI the ennemies should just take cover, that is all. When the developers put in some "survival instinct" every human is supposed to feel we should have some challenge at last. :D

( Thread added to "Important discussion threads list [look here first]" )

Papy
10th Feb 2008, 21:42
But in DX it only did when you made a mistake, or when you simply screwed up the fighting.
That's why I think Deus Ex was a great game.



The overall difficulty level is directly linked to the save system.

On DX it was too untroublesome.

I'm not sure what you mean exactly, but the same way a player can choose a difficulty level, he can choose if he abuse the save system or not. With games like Deus Ex or Thief, it's quite easy to play with "I'll save only once at the beginning of each level" (like I did).

Personally, I think that saving every 5 minutes is really like cheating. I won't cheat, I think it's pointless, but I don't see why the game should be troublesome to someone who wish to do so.

SageSavage
10th Feb 2008, 23:14
I used to be one of those players who saved and reloaded a lot but I've changed my habbit recently and I have to admit that gaming became more intense for me. Please no savepoints though - I still want to be able to decide when to save/load my game!

Lo Bruto
11th Feb 2008, 00:03
When it comes about Deus Ex I save only in that crucial moments (once per level) that we all know, like the fights with Gunther and Anna (I stop using the killphrase in my 4th run through the game) and Simons or the escape from the NSF Generator House after you send the signal (Zapping the Unatco buddies with the Prod before to make your escape easier is Lame! :rasp: )

gamer0004
11th Feb 2008, 14:45
I save and reload so fast that the savesystem doesn't keep up because I save at exactly the same second twice. And sometimes I reload many, many times, in the DX no-item run for instance. When I just shoot everyone I don't reload as often as when I sneak, of course.
I mostly reload because I don't want to do things again when it was very hard to get it right the first time - like positioning the NPC's in a way that they can't see you. And I don't like it when I could get away, but if it would me cost a lot of items. Because I don't like it to have a hard time doing something because you're out of lockpicks or medkits.
I also save a lot because the past has showed me that not doing that is really stupid. In some games I did something wrong, but I didn't know that at first so later on I'd have to go all the way back. I simply don't like that.
Of course, I do always play games at the hardest setting. That way it is a bit more of a challenge.

RiscOS
11th Feb 2008, 18:28
In my opinion AI is what lacks in almost every game, FEAR did a good job of fighting tactics but was partly scripted (i think). Having a cut and dried enemy is a bit old school, what I would like to see in a game is the ability to work around conflict not only by traversing the levels covertly but by persuading less moral persons in the game with money, blackmail etc to gain information, safe passage, weapons etc.

Lets say for example that you need to bypass a gaurd on a gate, you could snap his neck, on the other hand you could plant drugs on him and have him arrested; then bail him from prison on the condtion of letting you have free passage. Clearly different types of AI should be more or less susceptible to bribes or blackmail, from the police who will take a bit of cash and let you know where someone is staying for the night, to an agent who will rip your head off if you try and approach them.

A threat level system should also be used, if the level is 1 everybody is calm and the police will ensure law and order, 2 the police will be more aware of things that are out of place and will kick some ass if they need to, 3 the armed police will be deployed to keep things under control but they won't shoot unless they feel threatend, much like the police in the USA. At level 4 a military unit is deployed and they will show no mercy if you piss them off. And at level 5 the agents get deployed they have a good sence of whats going on and can lead all the other units in an effort to find who ever they are looking for.

Rules Of Engagment (ROE) could also be changed from the standard FPS style, the AI could be set as police, armed police, military, or highly trained agent; I will use another example to convay my thoughts:

Situation
You are trying to gain acess to a secure area which is guarded, you need to recce the area and then break in, or blag your way in.

Police-
You approach the building and stand around for while trying to figure out whats going on, in deus ex 1 the guard would ask you to back off or shoot you if you got to close, I would like to see the guard approach me and question me; if I don't give the correct answer he will ask me to leave,if I give a really bad answer he will remove me from the area using minimum force. If he saw me hanging around again he might call in the armed police or backup to help, and perhaps if he is really spooked put the threat level on the facility up a level making the later break in harder for me. These guys should be a bit easy going and not sharp shooting trigger happy carzy's.

Armed police-
These guys should have slightly better situational awareness and only be deployed if the threat level is 3 or higher or they are responding to a call out, like on my situation. If I draw a gun around these guys they should do every thing they can to take it off me before they shoot me, and when they do shoot the should take single aimed shots taking care not to hit bystanders. They should be fairly hard to bribe or blackmail.

Military-
In my situation if these guys were guarding the facility they would turn you away using force (but not deadly force) if you got a gun out they would open fire unless they could get it off you easily (a guy stood next to you hits you and grabs the gun) after that they would remove you from the area and fill you in for being a idiot.

Highly trained agents-
These guys would have great situational awarness, if a guard was missing from his post or a body was found they would co-oridinate the search for the suspect, if they saw you they get their guns out and try and take you down.

These are some of my thoughts about AI, I have a few that I will post later, even if some of this gets used in the game I would be pleased, it really is time to step the AI bar up and perhaps shoot it into space.

Thanks a lot and keep up the good work, its nice to see the forum has now been spam protected. :)

Unstoppable
11th Feb 2008, 20:42
There are some good ideas here. One A.I. I like comes from the game S.T.A.L.K.E.R.:Shadow of Chernobyl. If Warren Spector ever wanted to see emergent game play, he would only need to play up to the second level in that game.

Areas in that game are constantly under attack by enemy NPC's while you have friendly NPC's fighting to defend it. All kinds of beasts roam around the terrain and will attack you if you provoke them. Very cool indeed.

Blade_hunter
27th Apr 2008, 14:17
I wanted to show the AI uses the crates, against you or to protect themselves, because in the first game the crates are a good strategy ...
And with the turrets the enemy can't fight against event if they have a RL
The bots have the same problem (he big bot can't do anything against a turret ...

The AI sometimes did a good job when we use a scramble grenade against the bots they uses the LAW or throw a grenade to disable the bot or destroy it ...
And when they try to kill you they aim your head sometimes.



When we are unconscious the AI can capture us but perhaps it's useless and more simple to make for us knock or died as equals things ...

I want to:

See the AI pickup some items and when they pickup the items are in they're inventory of course and we can pickup them when they are died or use our pick pocketing ability or when they are unconscious. *

See the AI using light, torch devices or special goggles if they suspect our presence on dark places. *

See the AI using the ladders is some games the AI was idiotic with ladders. *

See the AI if they are burned go to water places or try to find an extinguisher to stop burning. *

See the AI can surrender or make a truth if they think they are on a weak position or if they know you won't kill them or if you want to make a conversation to know something and tell them your objectives, but perhaps it's too complex to make negotiations in the game ...

See the AI using vehicles if we have drivable vehicles like cars or skateboards etc ... *

See the AI commanding a team if the have leaders and sometimes they can cure some injured teammates *

The suggestions with stars are the more important things I think, but most of them can be a major AI innovations but if it can't be madded I won't think the game was bad for that, the first game have some lacunas on that matter and it still a great game even today's ...

Fen
27th Apr 2008, 18:08
I want to:

See the AI pickup some items and when they pickup the items are in they're inventory of course and we can pickup them when they are died or use our pick pocketing ability or when they are unconscious. *

See the AI using light, torch devices or special goggles if they suspect our presence on dark places. *

See the AI using the ladders is some games the AI was idiotic with ladders. *

See the AI if they are burned go to water places or try to find an extinguisher to stop burning. *

See the AI can surrender or make a truth if they think they are on a weak position or if they know you won't kill them or if you want to make a conversation to know something and tell them your objectives, but perhaps it's too complex to make negotiations in the game ...

See the AI using vehicles if we have drivable vehicles like cars or skateboards etc ... *

See the AI commanding a team if the have leaders and sometimes they can cure some injured teammates *



I agree with most of these.

As for guards picking up items however, this really isnt necessary. It also failed very hard in oblivion so unless the team is REALLY confident they can implement it to work well, I wouldnt bother. I wouldnt mind if the AI went and bought emselves a soda or drink when they were at bars however.

Please god no skateboards. I cannot express how stupid that would look. As far as cars go, there is no need. Just seems like extra work for something that isnt worth it.

Kneo24
27th Apr 2008, 19:17
I agree with the sentiments here to some extent. A lot of games these days make a game harder by a.) spawning more enemies, b.) giving enemies more health, and c.) increasing their damage.

I absolutely hate that. It is very very lazy design. If the only way to add challenge to your game is to do that, you should not be developing a game. Smarter A.I. is far more interesting and fun to fight against than A.I. that has wall hacks and can kill you in two shots.

Voltaire
27th Apr 2008, 19:53
I agree with the sentiments here to some extent. A lot of games these days make a game harder by a.) spawning more enemies, b.) giving enemies more health, and c.) increasing their damage.

I absolutely hate that. It is very very lazy design. If the only way to add challenge to your game is to do that, you should not be developing a game. Smarter A.I. is far more interesting and fun to fight against than A.I. that has wall hacks and can kill you in two shots.

Very true. I'd like to see smarter enemies in later levels and higher difficulties, not equally stupid enemies with more armour/arsenal. Give an idiot a flamethrower and he remains an idiot...

I'd like to see some lone wolf style soldiers as well i.e. ones that don't call for help when they see you, just silently try to approach you from behind. Definitely not all the guards, but a handful in each enemy base I infiltrate, to keep me on my toes. :)

Gary_Savage
27th Apr 2008, 21:38
I'd like to see some lone wolf style soldiers as well i.e. ones that don't call for help when they see you, just silently try to approach you from behind. Definitely not all the guards, but a handful in each enemy base I infiltrate, to keep me on my toes. :)

I like this... NO!... I LOVE THIS! I remember, on a lone wolf mission in Operation Flashpoint, I approached an enemy base just after sunrise (I think the mission was even called Lone Wolf), and used explosives, and a rocket to take out some of the enemy vehicles. When the enemy soldiers soldiers began running around the map, hunting in packs, I gunned them down, after a few tires. It was the last remaining enemy, however, who gave me a run for the money, so to speak. I knew I had been fired upon by him many times before, and each time he had changed position, just like I had, when hunting his buddies. I also knew that unlike his buddies he was hunting ALONE.

After what seemed a long time since I thought I had killed all of my enemies, I still got shot at (i.e., shots came towards my general direction), and even on the few occasions that I caught a glimpse of where the shot had come from, by the time I got to the site there was no one there. I thought about another mission, where I had to snipe teams of enemies, with each sniping at my buddies and me, but that mission was no match for the excitement in this: I was even afraid to crawl from cover to cover; just that I was forced to, knowing unless I moved this guy would surely get me.

As more shots rang out, a single shot, nearly each time, I realized that the enemy was very determined, and was able and willing to move near and far, up hills and down vales, through bush and bare dirt, from tree to tree, seemingly each time getting his shots closer, and eerily closer to me. I knew it had to be a sniper.

It was duel, and phrases like "enemy at the gate," and "assassin at the gate" (from some poem) ran through my mind, as I decided between the satisfaction of going after the best AI I had ever seen (and have seen, till date), and just commandeering the the one tank that I had spared, and leaving through the compound's front gate. I couldn't resist the temptation. Even with Operation Flashpoint's (notorious, IMO) one save per mission system I went after that AI, shooting at him from behind wrecked (and some undamaged) vehicles, bushes, around the corners of buildings, around the corners of the compound's fence, etc., eventually, myself, going near, and far, over the hill, and down the vale, through cover that is thick and thin, from tree to tree, until I caught sight of something moving in the distance, some way up the hill, from me. Quickly, I aimed, and pulled the trigger, and then scampered off, to hide, lest he be bleeding, rather than dead, or even unhurt. After a few moments of not hearing any further shots, and not seeing any changes in the enemy's position, between taking peaks with the scope/binoculars (I don't remember which), I decided that I might have killed him, and approached the enemy, going from cover to cover, as if he was still alive. Even when I saw him lying still, at close range, I very gingerly moved up to him. Finally, I confirmed him dead, and stole the one tank I had spared, and drove it out the front gate (or maybe it was another vehicle I stole: that part of my memory is hazy, and it makes sense to not steal any other vehicle, when there is a sniper on the loose).

_______________________________________________

THAT is the kind of lone wolf AI that I would like to see in the next DX.

Blade_hunter
27th Apr 2008, 22:56
In red faction 1 one think I loved with the AI is when you stick a bomb o your enemy, he runs and he's teammates fire on him because he runs to them and he explodes and I don't needed to press the trigger or put a bomb to avoid they're coming they tried to fire on the bomb to go inside the room where I am.
The AI wasn't perfect but they have some good points.

In DX the AI have something realistic not perfectly but we've got some realm sometimes when we throw a grenade and when they don't see us they look to the grenade before running, when we burn them they run like RF (DX was the first to give this ^^)
When you run fast to be hidden most of times they stay at the last place were they see or hear you even if sometimes little noises makes them under alert or sometimes they looks dumb when the alarm is sounded ...

But some innovations with the AI are welcome :)

Vasarto
28th Apr 2008, 02:30
You know what I would FRIGGEN LOVE!?!?!?

The exact same A.I used in the Game F.E.A.R

By far the most intelligant enemies I have ever faced. They can jump through windows and duck down for cover and move around and hide and sprint and blitz through wallways in order to evade your fire. I want Deus Ex 3 to have
A.I just like that. Where they can counter or predict your movements and skillfully take you out! Where they can push over a couch or table and use it for cover or jump through windows duck down behind the wall and pop out and shoot out at you through cover or put their back to the wall and through a grenade at you without exposing themselfs to you.

:lmao: :lmao:
Basically the most advanced and soffisticated (Sp?) A.I Possible!
Maybe create an ultra advanced program where it can actually learn and
take what It learns and becomes even harder as your playing it or when you keep playing it so that even if you find a stratigy that works far beyond its
A.I limits it can learn from it and become even smarter, Forcing you to be even more creative and work harder to beat the game!

Or just use more stealth so they never notice you!

serene_chaos
28th Apr 2008, 03:24
I havnt played F.E.A.R., Operation Flashpoint, or Red Faction, but i do have S.T.A.L.K.E.R., and the AI in that is pretty good.
I think that given the development of AI in recent games, we need not really worry about DX3 having stupid enemies.

Gary_Savage
28th Apr 2008, 15:39
I don't know about that: Operation Flashpoint came out years before DX1; also, the team behind DX1 had people from the (then long ago released) Thief series, whose AI (IMO) was better than the DX1 enemy AI (in that the guards went off to call their friends, if they took too much damage, and then their buddies would thoroughly search the entire premises -- I would like to see the DX3 AI use team tactics during these searches, storming rooms, like SWAT/special forces teams).

While I don't think the AI will make or break this game (it did not break DX1), I DEFINITELY would enjoy better enemies. I don't know why, but I think I enjoyed the enemy AI in Half-Life1 better (especially the black clad assassins), despite the fact that DX1 is my favorite game of all time.

Blade_hunter
28th Apr 2008, 17:10
The black assassins haven't really a special AI they have a faster move than you and they can make very high jumps. It's that thing makes them stronger than marines.
In STALKER the AI is good and we have good possibilities in this game, FEAR has a great AI, yes but the possibilities in the game are somewhat reduced compared to STALKER. I think the possibilities can make the AI more or less difficult to code and simulate an human reaction too.
And the fails are more difficult to code than some perfectness.
It's more difficult to code an enemy that simulate an human accuracy than a 100% accuracy.
Its more difficult to make an enemy that can take care about the crates than an enemy than can't take care about.
and we have more examples for that.
I we haven't a perfect AI that's not a real problem I think but the AI must be correct (More evolved than the previous DX's of course) but make a perfect AI I don't think we can now but make a correct AI that can react and take care about some elements like windows, crates, grenades, explosive barrels, barriers and other movable or breakable obstacles.

Voltaire
28th Apr 2008, 18:18
I have mentioned on another thread that DX1 was plagued with stupid, stupid friendly AI. Like seriously. In battery park, right after the fire fight near (not in) the subway, my UNATCO buddies boxed me into one of those shelters for the bums.

And they wouldn't move :eek: They had to be tazered in the end... :whistle:

This was basic bad AI even in 2000. Normally a friendly NPC makes a big deal of keeping their distance to avoid that kind of idiocy.

jcp28
28th Apr 2008, 20:40
I agree with the sentiments here to some extent. A lot of games these days make a game harder by a.) spawning more enemies, b.) giving enemies more health, and c.) increasing their damage.

I absolutely hate that. It is very very lazy design. If the only way to add challenge to your game is to do that, you should not be developing a game. Smarter A.I. is far more interesting and fun to fight against than A.I. that has wall hacks and can kill you in two shots.

The PS2 game, Black is highly reminiscent of that. I shot enemies with THREE seperate bursts of my assault rifle,to the head before they went down. Once I got halfway through that piece of ****, I put it down. They should have made the checkpoints a lot more frequent in that game as well.

Blade_hunter
29th Apr 2008, 06:26
The assault rifle in the game does a bad damage per shot and shoot too fast instead of other weapons.
the AR does 3x5 points (15 points) of damage and you hear 4 shots
you can't spent only one bullet with this weapon because it's default mode is a tracer mode 5 shots when triggered.
I preferred a slower ROF and more damage per shots as I've said in an other thread and this makes the AR a not good weapon when we are unskilled with rifles the sniper rifle can use only one shot even if we are unskilled it's hard but not impossible, in the roofs tops of the NSF generator level I used my Sniper rifle and I taken dawn most of my enemies with only one shot ...
The AR makes a "good" damage only when we are mastered on the rifle skill (a burst in a heat takes down a single enemy, but we spent so much ammo, it's for that main reason the AR and the SR have separates ammo (I think) the caliber of these rifles are pretty close one for other 7.62 mm NATO and 30 - 06 caliber, the bullet aren't the same but making 3 for the 7.62 mm NATO and 25 for the 30 - 06 caliber it's not logical.
because these rifles caliber are somewhat the same.
making only the AR ammo in common with the SR will not appropriate in DX 1
the opposite thing too.
Some times the AI in the first game has some bugs, but it's rare and some games after DX and even some games wanted to make a DX like game sometimes have a poor AI compared to DX even some FPS more later with no special features.
The DX 2 AI is pretty bad because they madded the game for consoles ....
In 2000 I've played to SOF and the AI wasn't intelligent throw a grenade the enemies didn't run but at this time rare are the games that got a good AI
now in some games we have an AI that works like the HL AI for the marines;
The last SOF called SOF 3 has a scriped AI .... :/ compared to SOF 2 the last game was very ugly in therms of AI (console ?)

Nathan2000
29th Apr 2008, 07:57
the bullet aren't the same but making 3 for the 7.62 mm NATO and 25 for the 30 - 06 caliber it's not logical.
because these rifles caliber are somewhat the same.

The assault rifle has to deal less damage because of the game balance. It shoot bursts. If it was as powerful as the sniper rifle (or even a pistol), it would make all other firearms useless.

In my eyes, the AI in DXIW is as good (or as bad) as in DX. It has a few improvements (dodging bullets), a few flaws (poor "alliance" management - it causes allies to attack us for some unclear reasons) but they don't make much of a difference. I'd like the NPCs in DX3 to have some more advanced AI - team tactics, hiding behind cover, that sort of things.

Blade_hunter
29th Apr 2008, 09:43
Yes but look at the sawed off shot gun and the assault shotgun, they makes different damage the sawed off 25, the assault 20
If the assault rifle have a lower ROF and better fire power and keep the fact when you shot some bullets the weapon becomes more inaccurate this is more balanced and makes this gun more tactical to use. and only with it's primary ammo.
The AR must deal less damage than the SR because it's an automatic weapon , in most other games when they have an AR the gun deals a good amount of damage and never makes a sniper rifle useless.

If the gun fires 400 - 450 rounds per minute and makes 6 - 8 damage points and lose a bit more quickly it's accuracy with a continuous fire we have a weapon that can make a good amount of damage and have more tactical uses, in the game; I often used this gun to spend ammo and for close range combats.

Most way's can balance the weapons some weapons are ballanced but some others aren't very ballanced

Fen
29th Apr 2008, 11:31
The AR definatly needs more damage.
What I didnt like was that a pistol user (me) could do more damage with a pistol that anyone could with that AR rifle.

Rifles should be stronger than pistols but have the drawback of size.

Kneo24
30th Apr 2008, 01:32
Dude, Blade_Hunter, please stop butchering the English language.

I can't make sense of what your rant about the different caliber's have to do with AI.

Deadelus
30th Apr 2008, 10:25
The AI should vary. You should be able to kill one guy while he cowers and then put up a real friggin fight with the next guy. Not all 'humans' will run to the alarm, and face you. Some will either just run or shoot. That was the problem in DX1. I could get close to some guy, pop out, and he'd instantly go for the alarm, giving me time to cap him.

Blade_hunter
30th Apr 2008, 12:41
Hum; I spoke about this because I wanted to respond to a post here; I know that is not the subject. And I try to speak English with my limited language, because I'm not an English; I'm a stranger but perhaps you think there is only guys thats speak only English. that is the point. I return back to the main subject.

Yes the AI has some defaults like this; in the first level of DX when I go to Gunther's prison room there is some NSF terrorists and he goes most of times to the alarm button to sound it.
sometimes when the alarm is sounded the other enemies don't move and I hear the alarm because they are a bit far of the place where the alarm is sounded

nkepke
8th Aug 2008, 19:26
To start with, I would like to see the A.I. being able to crouch and hunt after me when im hiding in the vent.... :mad2: :D :D :D :D :D

A.I. will be enormously important for this generations Deus Ex. If it is to paint the realistic and living world its important that it's not faltering to todays game-standards.

High-level A.I. is therefor a must as far as I'm concerned.

jordan_a
8th Aug 2008, 19:45
About the vent shafts I hope something will be done about that, it's just been too easy to get away in the previous games.

Jerion
8th Aug 2008, 20:10
About the vent shafts I hope something will be done about that, it's just been too easy to get away in the previous games.

Agreed. just hide in the vents with any weapon (even a sword :D ), attack, then duck-n-cover and they forget all about you.

Romeo
8th Aug 2008, 21:26
I thought Epic's Gears of War had a very intelligent AI, as they wouldn't sit and let you pour rounds into them, they'd react to your position, and flank you and whatnot. Combining that, plus some ridiculously difficult enemies (Hello Mechs and Tanks), should create a challenge for pretty much every player possible. =)

Larington
9th Aug 2008, 07:18
Two points I want to throw in, I think its unwise to take the choice of when to save away from the player. Certainly I shouldn't have to backtrack to a save point (Those things do my head in, they are an entirely artificial way of upping the difficulty if thrown in randomly, though if they get it right its tolerable) - Let the players choose how they want to go about saving the game. You don't have to put in a quicksave function, in fact I advise against those since it actually encourages a player to save constantly and I'm perfectly fine with the idea of not being allowed to save during combat (Baldurs Gate games did that) but other than that, let the player choose to either save as often as he wants or as rarely as he wants.

Back to the topic of AI, I definately agree that in general most games tend to do a very poor job of AI, it amazes me to this day that concepts demonstrated in Half-Life ONE (released TEN years ago) such as soldiers moving together, staying behind cover and occasionally firing grenades from behind said cover, doesn't make it into supposedly modern up-to-date games. Especially ones trying to go on about realism - In the real world, people are afraid of getting shot (its natural) and they should act like this is the case.


As to gears of war, I didn't get very far in that, though I was fairly impressed with the enemy AI, I found the blandness of the environments so overpowering, plus one badly designed section of the game early on, that I shelved it far, far, from completion.

Romeo
10th Aug 2008, 06:55
Still, it has one of the best AI I've witnissed to date. Now, feel free to pour in the inevitable "OMG, UR A FULE, GAME X HAD A WAY BETTER AI, ROFLROFLROFLCOPTER" type comments.

Unstoppable
10th Aug 2008, 12:51
For me I thikn the best A.I. I can remember was the Thief games. Also Thievery for Unreal Tournament 1999 is real good. Far Cry also had good A.I. and Crysis was interesting.

Gears of War was good as well. Warcraft 3 was crazy also lol.

El_Bel
10th Aug 2008, 15:21
Crysis has a good AI? You mean this is good?

http://www.youtube.com/watch?v=lEfjdQ5uaNY&feature=related

or this?

http://www.youtube.com/watch?v=vre1ABXbMoM&feature=related

The guy that created the AI on Crysis, has put nods around every tree and rock so the enemy can go there and hide, like half life did 10 years ago. The AI will scan the area for the closest nod in the area. Thats not real AI. It worked with HL but thats it. Half life was a much smaller game and nods could work.

If someone wants to see real AI, check out ArmA and operation flashpoint.. Or check the Internet for stories. Sure some times the **** up, but most of the time it gives you the feeling that you play against some intelligent enemy!! It really makes tactical decisions, based on intelligence and not prescripted events.

FrankCSIS
10th Aug 2008, 17:28
A lot of ingenious ideas in this thread. Out of everything I've read though, this
The AI should vary. is the most significant progress we could make right now.

Mustaches, varying hair colours or slightly altered bone structure don't make guards any different from one another. Different reactions to specific situations and a varying level of skills from one enemy to another (of the same class) would already be a big start. Some people are just better than others at their job, and that includes cops, guards and soldiers. Some are also easily scared, others are relentless, some are courageous and others are just crazy, especially when pushed to their limits or presented with a life-threatening situation. A security guard forced to take his handgun out of his holster for the first time is not going to react the same way as the ex-cop who's been through that or the wackjob who's only in the business for the ass-kicking.

When Swat 2 was in early development we were told agents and enemies would have varying personalities and when a situation got too serious for some of them they might freak or even flee the scene. Obviously none of this made it into the final game but they already had the right idea for improved AI.

Varying elements with soldiers is arguably more difficult because they work in teams, have strict instructions and protocols and all went through the same full-proof training. What you can alter from one individual to another though is how they react when isolated, or when you've taken down some of their buddies. (the lone wolf example someone brought up, for instance, was pretty brilliant). Again, it's when pushed to their limits or out of their planned situations that you can fully see their individual personalities, and this should be reflected in a game. Not only does it make it a lot more challenging, it would also be a ****ton more realistic.

The NHL games have already improved their AI with this idea in mind. It used to be that one specific move could beat the goalie no matter what, but the players and goalies today adapt throughout the game to your playing style. It's not perfect, but it's an impressive start.

Something completely different that's been briefly addressed here needs some further development. We've talked long and wide about the individual AI, but very little, if any, about the collective AI, something that's been lacking since, well, forever. When discussing the future of adventure games some 10 years ago we came up with the obvious necessary progress to improve and save the genre, but this can be applied to any game with a storyline and interaction with NPC's and enemies. "The Plan" needs to vary in the background while you're playing.

Let's be clear right away before I go on, I'm not talking about multiple paths here, although there are common elements. What I'm saying is that all stories have one point in common; everything always goes according to plan at the beginning of a game until an unforeseen element is introduced, i.e. you, the player. When it comes to attention of the overseers that you came along the plan varies and dispositions are taken, but all in a flat storyline scripted in advance and deployed early on. What needs to vary is how the planners adapt to your involvement in the story, what decisions the "people in charge" take behind the scene while you're playing in a completely different area.

Someone in this thread mentioned a guard should react differently according to the things you say or the suspicious actions you take, and this is also true of the global AI. Bringing up a certain topic with the wrong person might trigger an alarm with the forces that be. Exploring an area you weren't supposed to see and getting caught on tape should bring heat on you that you would have otherwise dodged, at least for now. I'm not talking about an alarm ringing here, but a global awareness of your involvement as well as the things you now know and the information you've gathered. In the first DX, you could often finish entire levels and move on with the story without exploring half of the map or discovering 1/10 of the true story (MJ12 in the sewers or hacking your brother's computer comes to mind), which was wonderful at the time, but it didn't have any consequences or rang any alarms with higher authorities. Whether or not you discovered MJ12 early on didn't change anything to the global plan or their approach on how to deal with you, and to me, that is a very important part of the AI. If you show your face early on, guards, for instance, should now be aware of your existence and be on the lookout for you in other levels, while you might be able to bull**** your way through if the enemy didn't already know what you looked like or how much you truly know.

I realise this is insanely complicated, but it's an improvement that has to be made eventually to dramatically alter the gaming experience. The result would be a game varying in difficulty according to your own actions and mistakes, and not some stupid setting level you decide before you start that alters just how many bullets the soldier can swallow before kicking the bucket.

drummindog
12th Aug 2008, 00:20
I think it would be good to have some choice depending on how you like to play. Maybe having a few different choices in AI's would allow for some extended gameplay and life for the game. How I would configure it.... I don't know.

Part of the excitement for me with Deus Ex was getting further just to see the challenges ahead or the cool layout of the next level. I don't think I ever thought about how good or bad the AI was...it is what it is. The plot twists, factions, cool level layouts, conspiracy theories and etc kept me coming back.

The AI was good, but I am sure everyone can name something they consider better. Most of the time with games, I am happy if they know I am there before I get too close or they see me or hear me or something. That makes it more challenging for me. Now if I am able to walk 10 feet from someone who is supposed to be guarding against me, I'd say we have a problem unless it is dark and I am just that steathly.

I am looking forward to whatever they come up with in the AI dept as well as other aspects of the game. :)

One other point about AI here...I suppose they could make the AI so incredibly hard that if you only had an empty 45 and a can of soda, you could forget about getting any further in the game.

gamer0004
12th Aug 2008, 17:26
I've been thinking (oh yes, I do that!) and they should really add a system to adress abnormal behaviour of the player. For instance, when I throw tyres against people I want them to respond with more than just "Hey, I'm warning you" and do nothing. Same for when I steal things, hack computers, walk on furniture, break/destroy things and so on.
There should be around three responses.
1. The I-will-pretend-as-if-nothing-is-happening-here aproach (passive). When the person who gets annoyed by your behaviour is an unarmed person or just a bit of a wimp, or if you are way more dangerous than him (if you are carrying a GEP gun or a shotgun and he just a baton or a pistol). Maybe he'll swear under his breath, or aks if you could please stop it, but that'd be it. Unless, of course, he can sound an alarm or call for baqck-up or something like that so his chances are better to be able to fight you.
2. Trying to withhold you from your abnormal behaviour in a non-agressive way (non-agressive meaning he won't fight you). So if you throw a tyre at him, at first he won't be able to do much as he didn't see it coming so he'll just loudly "ask" (i.e. commands) you to stop it. If you do it again he'll try to stop you by throwing the tyre somewhere you can't come (over a fence for example) or by making sure you can't throw them because there are none: he'll destroy them. That's a bit hard though with tyres, but is quite easy with chairs and cardboard boxes.
He'll maybe grasp the things out of your hand after you've hit him once or twice. If you're walking on furniture he could puss you off, or he could push you away from the computer when you are hacking it.
2.99: More agressive, almost like 3: doing it back. When you throw tyres at him, he'll just throw them back at you so you get as annoyed as him and hopefully will stop.
3. If you keep annoying him and/or if the NPC is agressive and/or if he thinks he can (easily) defeat you in a fight he'll start to hit you to make it stop. Not at once try to kill you or knock you unconscious but hit you so you get scared or don't like it enough to annoy him to bear the pain. However, if you are stubborn, he'll get more and more agressive, until you are either dead or unconscious because in that case that's the only way to make it stop.

1-3 is about the thoughness of the NPC you are bothering. 1 is a wimp, 3 is a heavily armed guard or thug. Of course, a 1 person could still perform a number 3 action. If he is armed and has a lot of experience at the shooting range while you are or appear to be unarmed and not very tough (like Alex D) he might resort to a 2 or 3 action.
1 are people like Alex Jacobsen, nerds, children, businessmen etc.
2 are thugs, armed guards (but not very well trained guards, like SSC), people who are a bit tipsy or drunk, construction workers etc.
3 are people like (heavily) armed thugs, OMAR, heavily armed and well trained guards etc.
Of course, these things may vary. Some (armed) children can be very hostile, some thugs can be very cowardly.

Of course, guards and bouncers will react different than normal armed thugs. They'll throw you out or something, perhaps fine you, but won't take very desperate measures very soon unless you really ask for it.

So, what will be the benefits of such a system?
First of all, a thing like this CAN be programmed, and not with that much work. Only the parts where the NPC's have to catch things or throw away things can be a bit tricky if they haven't been programmed from the beginning.
Secondly, it makes everything a lot more immersive and realistic.
Thirdly, it adds to the characters. I think it would be very cool to scare the **** out of a cowardly thug after you've had a hard fight with some though guys. It makes every NPC more unique.

DXeXodus
13th Aug 2008, 04:05
I really like that idea gamer0004. It reminds me of Half-Life 2 (Although, it is of a smaller scale) where the Combine guards in the train terminal at the beginning of the game go through several warning stages before beating the living daylights out of you. The first stage was a simple warning, followed by a switching on of the riot baton, then maybe a single knock, etc. Very immersive stuff.

gamer0004
13th Aug 2008, 04:28
I really like that idea gamer0004. It reminds me of Half-Life 2 (Although, it is of a smaller scale) where the Combine guards in the train terminal at the beginning of the game go through several warning stages before beating the living daylights out of you. The first stage was a simple warning, followed by a switching on of the riot baton, then maybe a single knock, etc. Very immersive stuff.

Wow I completely forgot about that! Yes, it was pretty cool, that part. Never finished the game though, I just couldn't be bothered anymore half-way through.

TrickyVein
18th Aug 2008, 15:44
I can't find anything in the threads about movement options, so here's my two cents about ladders: You never saw an enemy climbing one in DX1 because you would have laughed your ass off as the player. Consider the following:
I can't hold anything when I'm climbing a ladder, certainly don't know anyone else who can have one hand free while moving up and down one - being able to brandish a weapon on a ladder has always irked me, and I'd to see you holster everything before climbing (much like how mantling in DX2 required both hands). Once paused on a step, sure, fine, reign hell down on your enemies, but I'd like this issue to be given more thought with both the 1st person view and with enemy AI.

gamer0004
18th Aug 2008, 16:15
It's very easy to climb a ladder holding something (like a pistol) in one hand. It's even possible to climb a ladder without your hands free but then you will have to be very careful.

TrickyVein
19th Aug 2008, 11:12
a pistol? Yes - a two handed assault weapon or rifle? No.
Perhaps stepping sounds would make for a more convincing experience on a ladder. As is you just glide up and down them effortlessly.

gamer0004
19th Aug 2008, 11:45
Yeah, but it should be able to go down a ladder without making any noise - for sneaking.
Anyway, it's possible to go up a ladder holding an assault rifle with both your hands. In fact, it's possible to do that and turn around. However, you have to be very careful and firing your rifle will probably knock you off.

TrickyVein
22nd Aug 2008, 22:11
Yeah, but it should be able to go down a ladder without making any noise - for sneaking.
Anyway, it's possible to go up a ladder holding an assault rifle with both your hands. In fact, it's possible to do that and turn around. However, you have to be very careful and firing your rifle will probably knock you off.

You have tried this yourself? I really am ignorant as to if such a thing is feasible. Regardless, I'd like to see enemy AI using the environment to its full potential just as the player does.

gamer0004
23rd Aug 2008, 06:15
In fact I did. I climbed up a ladder once wile holding a box of something with both my hands. You do need to have enough space for your feet so you can clamp your feet behind the side bars because else you'll fall off. You'll probably fall off anyway because it's really unstable. Don't think they should include this option because it would be useless :D

Big Orange
6th Oct 2008, 00:18
Here is my first post in this forum, I have a few ideas about AI:

Gang members would be very violent and edgy around their turf, but most of them would be relatively unskilled, either cowardly or stupidly reckless, and uncoordinated with their fighting, with only some key gang enforcers being very skilled and disciplined, goading their men on. Only important gang property like nightclubs or warehouses would have proper security systems, with many gang members using mobile phones.

Police and corporate security would be somewhere inbetween street thugs and para/military soldiers. They would be laid back but moderately vigilant in public spaces, politely moving people on if they get near a sensitive area of a complex, but would act very aggressively if they catch you breaking in. Their fighting style is competent and systematic, if unimaginative, with them being more willing to take you alive. They're more confident if they are out in great numbers, but in small numbers they're generally more cautious, more likely to run and yell for back up. They unanimously have radios and have a centralized security network behind them that has contact with higher authorities.

Military soldiers and elite operatives from MJ12 or UNATCO, depending on the location and situation, either behave like slightly relaxed cops on patrol or they try to put a bullet between your eyes if they spot you, without asking questions. Their tactics are very expert and highly persistant, individuals are much more savy and resourceful, they are more uniformly fearless no matter their numbers (although some will crack in extreme circumstances). Like law enforcement officers and professional guards, they have communications gear and a security network behind them.

The conduct of the protangonist effecting later stages of the game is a good idea; if you're stealth orientated fighter who kills few or no people and avoid most guards, then you've less of a horrific reputation that sets off red flags amongst enemy opposition and bystanders. Security around enemy facilities would perhaps be less extreme if you don't turn everywhere you go to into a charnal house (no tanks and ED-209s to deal with) since even MJ12 Troopers got Thanks Giving with family to look forward to.

CarloGervasi
6th Oct 2008, 00:20
Good animations will go a long way towards making me love the game more. Enemies ducking when shots unexpectedly whiz past their heads, stuff like that. Make them a little more life-like. People in both the first and second games are all a bunch of stone killers, apparently, because not a single one of them has a visible reaction to some loud gunshots ringing out of nowhere. They all just calmly pull their weapons on me and start firing back. Lets get some more "human" reactions in there.

Big Orange
6th Oct 2008, 13:26
Soldiers would act in a more heavily drilled manner, and invariably be less spooked by gunshots, loads of bodies, and explosions. Police and corporate security would be similarily professional but would be more easily spooked and would perhaps dislike taking high casualties the most (they would pull back to the nearest strong point on the map). Gangmembers would seem more cocky and even fire their guns gangsta style, generally acting more like a angry mob in large numbers and running away if you kill many of them.

SubTonic20
6th Oct 2008, 13:40
I like this... NO!... I LOVE THIS! I remember, on a lone wolf mission in Operation Flashpoint, I approached an enemy base just after sunrise (I think the mission was even called Lone Wolf), and used explosives, and a rocket to take out some of the enemy vehicles. When the enemy soldiers soldiers began running around the map, hunting in packs, I gunned them down, after a few tires. It was the last remaining enemy, however, who gave me a run for the money, so to speak. I knew I had been fired upon by him many times before, and each time he had changed position, just like I had, when hunting his buddies. I also knew that unlike his buddies he was hunting ALONE.

Great, now I want to play Flashpoint again. :nut:

dark_angel_7
6th Oct 2008, 14:14
A.I. is very important to any game and I hope DX3 has some good A.I.

SubTonic20
6th Oct 2008, 15:01
I dunno, the AI doesn't really have to be great for me to enjoy a game. In fact, Deus Ex had pretty crappy AI, and big time. Their inability to distinguish JC from a rat even after staring at him for a good 2-3 seconds (assuming the player ducks back behind cover) is pretty pathetic, though it never stopped me from loving the game.

The AI was also pretty much just dashed towards the player shooting instead of applying any tactics whatsoever, then almost always ran away after taking a set amount of damage, never fighting back again, even if you followed them to a corner.

So yeah, Deus Ex had some pretty bad AI overall. Deus Ex 2 vastly improved upon it (though it has its own issues, like spotting me almost too easily in many cases) and was generally pretty decent.

AI is one of those things that is always, and I mean always overhyped with upcoming games. I'd do my best not to try and predict what it will be like, because it's probably the number one thing in gaming that can't deliver promises. With today's technology, though, we'll at least get decent AI, but I won't go anywhere beyond that with speculation.

Big Orange
6th Oct 2008, 15:33
AI is important, but if they try to take it somewhere new they'll either make it the most innovating AI since Half-Life or make it very annoying, by having it buggy, unfair and inconsistant. I expect it to be competent and immersive enough, but not world shattering.

Azrepheal
6th Oct 2008, 21:50
I think it would be nice to see the AI change depending on their number. In larger groups, they would be harder to manage, orders would be scrambled etc. But if you started stalking them through a building and taking them out, when they get down to only a few men, to have them retreat to a place they can easily defend and start working as a more focussed team.

Big Orange
6th Oct 2008, 23:46
^A large number of professional guards and soldiers could be split into different teams or platoons, with their own squad leader linked to a commanding officer to better co-ordinate themselves. When infiltrating or exfiltrating a big complex, maybe they could send out a fire team to specifically hunt you down, while the other teams bunker down around important rooms or checkpoints.

piippo
7th Oct 2008, 00:06
^A large number of professional guards and soldiers could be split into different teams or platoons, with their own squad leader linked to a commanding officer to better co-ordinate themselves. When infiltrating or exfiltrating a big complex, maybe they could send out a fire team to specifically hunt you down, while the other teams bunker down around important rooms or checkpoints.

Yes, it's a hard time balancing gameplay and how "real" the AI feels. Considering the example you provided, it could mean trouble for either the player or the pacing. If the player is swamped, in a way that he is unable to proceed - that isn't a good thing. Neither is believable if the player singlehandedly kills everyone in the facility whilst they are tracking him down. That's a very raw version of what could happen, and what needs to be considered.

Good AI is a hard thing to do properly, without making it too good, nor too weak. I always tend to believe that very dynamic AI is good, such that can react to certain situations in a reasonable way. Fuzzylogic and swarm behaviour.

Lets consider another situation, where player faces a patrol of 5 AI officers. Them being some sort of regular unit, so they aren't that coherent in battle. Will the AI flee if it notices that it's left alone? How will it react to friendly losses? Or will they stand to last man? Will they try to outflank / manouvre the player or call for backup?

Same scenario with experienced player, and another with a "newcomer" - it might be too hard for the newcomer to handle the flankings and calling for backup but without similar things the experienced player might feel too easy.

This again boils and entangles the issue of difficulty, which AI is a great part of - or should be. In many games that isn't the case, since the cheapest way to make the game more difficult is to increase the damage done by AI, and decrease the damage done by the player.

AI design and programming is one of the most intruiging and challenging task that developer has. I for one don't have the required skill to go in detail, but it's an interesting subject. If someone happens to get interested, Microsoft has some good things regarding this on their XNA platform, and on the tutorials section here: http://creators.xna.com/en-us/education/catalog/?devarea=11

Big Orange
7th Oct 2008, 11:58
I can say enemy strategy could be partially scripted on a wider map, while guard tactics would be unscripted, to save on processing power (in an alert teams A & B go to certain important areas to secure them, while teams C & D go and search the area you were seen/reported in).

Abram730
19th Oct 2008, 05:21
I can say enemy strategy could be partially scripted on a wider map, while guard tactics would be unscripted, to save on processing power (in an alert teams A & B go to certain important areas to secure them, while teams C & D go and search the area you were seen/reported in).

I agree

Synergy - The final outcome of a system is greater than the sum of its parts.

Thus Eidos need not have the best AI, just use smaller parts to build a better whole. It should save time, processing power, and debugging.

I have to say Big Orange I'm bad with remembering names, but I recognize your name as you posts stand out on merit.

Deus_Ex_Machina
19th Oct 2008, 07:32
As far as AI in gaming is concerned :

With VERY few exceptions, AI in games sucks.

Almost every big release we get on the so-called NEXT GENERATION consoles has significant AI issues. Even high-end PC games have AI issues.

If the gaming industry spent more time innovating and focusing on things like AI, story and gameplay INSTEAD of how to make games more mainstream (I ******* hate mainstream), then everything would be perfect.

Personally, I think the AI of last gen console games was overall better than what we have now, and its a damn shame.

Abram730
27th Oct 2008, 09:34
I've never done AI programing, but here is my take on it.

consider what is real. The human brain.

the reptilian brain or R-complex(R)
fight(anger) or fight(fear). Aggression(win,euphoria) or submission(loose,shock), patterned responses and core binary logic modes.

the old mammalian or mid-brain.(M)
states or moods.. memory(mood dependent) split personality as each state is a different mind so to speak. Each state must express 2 modes for R complex states.

the neocortex.(C)
abstract logic.. abstract goals.. people are the most illogical animal not the most logical and thats out strength. the neocortex isn't grounded in the reality of now and lets us adjust our reality. It makes states more loose as it can address memory in other states. This grew into multi state abstract thinking.:scratch:

lets take a simple game.. tag to show something
A = person A
B = person B

A touches B "your it"
B = (R = fight (get A, touching A changes R to Flight) range is a sticky ratio between range and aggression. ratio of fear in A to aggression in B) (*sometimes you feel fear just before touching so It's anticipated by R*)
B = (M = happy (goal is happy a state/var (touching A = happy, Changes in range to A is a sticky ratio between happy and -happy) if happy=full maintain happy, fatigue = -happy)
B = (C goals set R to "get A" and If you "get A" then reverse game roles and say "Your it")

A = (R = flight(avoiding B (Reset R if touched a system shock) reverse of B's ratio's)
A = (M = happy(goal is happy state/var (avoiding B = happy, reversed ratio's from B) if happy=full maintain happy, fatigue = -happy)
A = (C goal Set R to avoid B and maintain game, if touched + "your it" then reverse rules)
B chases A while both vocalize to lower opponents state(raise their happy and lower others happy)

Say A gets B with the touch and "your it" the rolls are reversed.

lets look at a state or mood change.
Say A can't tag B and becomes fatigued..
for A, M runs out of happy and A becomes frustrated M is dissociative of happy and R's state is shock. C game over I quit. Say R = Frozen, M = unhappy goal avoid happy so R = flight C = walk away from B, Game over.

say B notices A is in flight mode and shifts R to fight mode(R is always first state to change). C notes that he wasn't touched so touch rule for state change has been violated and doesn't except game over, C = C - touch rule and changes M to angry.
B = (R fight( get A) (-touch changes state) (range is a sticky ratio between range and aggression. ratio of fear in A to aggression in B)
B = (M = angry (goal is angry a state/var (hurt A = angry, Changes in range to A is a sticky ratio between angry and -angry) if angry=full maintain angry, fatigue = -angry)
B = (C goals set R to Get A) -(If A say "your it", reverse game rules)

What is tag when you are angry? you guessed it B attacks A as in tagged tagged tagged tagged it's still tag but you stay it. only one rule was dropped and one state change. broken up into states programing it should be easy. Note touch and hurting are the same thing as far as the brain is concerned it's the state or mood that changed. Ever notice that when kids play somebody always seems to get hurt?

so "touch" A is the same line of code as "hurt" A. get it? It's in the state of mind where the framework for them is. A very real feel can come out of this. The whole is greater then the sum of it's parts. A this or that AI that can change the this and that.

For DX this could be simple AI that's all been done before put into this or a similar structure... In the future this can be expanded to make real NN AI with 3 interdependent nodes doing different things.

human example M codes memory with states and when the state changes so does the memory structure. thinking about touching becomes thinking about fighting. the state changed the memory handles yet the basic structure is the same. you end up with an AI that can be radically changed with a small var change. R and M should cover all in game AI with C being given a job so to speak. NPC's could be dumb, C has a low priority. Intellectual, R has low priority. Lazy, States require a high var number. Crazy, sticky or random states and bizarre abstraction. cowards Flight preferred vs Hero where fight is.
I think AI sucks because it assumes a logical man with an illogical core.. I'd say we have a logical core with an abstract, illogical cover.

Officer Half
27th Oct 2008, 13:16
You know what I would FRIGGEN LOVE!?!?!?

The exact same A.I used in the Game F.E.A.R

By far the most intelligant enemies I have ever faced. They can jump through windows and duck down for cover and move around and hide and sprint and blitz through wallways in order to evade your fire. I want Deus Ex 3 to have
A.I just like that. Where they can counter or predict your movements and skillfully take you out! Where they can push over a couch or table and use it for cover or jump through windows duck down behind the wall and pop out and shoot out at you through cover or put their back to the wall and through a grenade at you without exposing themselfs to you.

:lmao: :lmao:
Basically the most advanced and soffisticated (Sp?) A.I Possible!
Maybe create an ultra advanced program where it can actually learn and
take what It learns and becomes even harder as your playing it or when you keep playing it so that even if you find a stratigy that works far beyond its
A.I limits it can learn from it and become even smarter, Forcing you to be even more creative and work harder to beat the game!

Or just use more stealth so they never notice you!

No then the AI would become too smart and take over the world. You don't read enough SF Novels do you? :nut:

spm1138
27th Oct 2008, 17:12
Kind of goes without saying that a stealth game lives or dies on it's AI and DX was a stealth game among other things.

The AI is really the gameplay there. The mooks have to be sharp enough to be good sport but not feel "unfair" or broken.

They also have to mesh with the stealth tools the player has been given so if they go off on one if they see a pixel of the player and the player is sneaking in 1pv with no way of locating them that's never going to be fun.

My biggest complaints with a lot of stealth games are what happens after they work out that they're not alone.

In DX I remember a lot of random running about and yelling while I shot people in the shins.

I'd like a more appropriate response to plasma bolts flying out of the air vents.

MGS3 was the last MGS game I played and I thought that had a nice balance between various things but I also liked a lot about Splinter Cell: Chaos Theory because there was a lot of stuff in the world you could mess with the guards with and because they gave some of the guards night vision.

So, yeah. Good combat AI (actually some of the people should be thick as two short planks and do dumb stuff but I'd like it if they were intentionally so), sensible responses to threats (send someone for help, move VIPs to safe location, run away, radio for help etc. etc.), escalating security environment, aggressive search patterns.

If done right it'd feel just enough like a bunch of human beings hunting for the player to be exciting but not so much that it was impossible to get around.

Tstorm
28th Oct 2008, 19:21
The ai needs some kickass catch phrases like "maybe it was just a homeless guy" or "A BOMB" :lol:

jordan_a
17th Nov 2008, 17:26
The combat AI will be an advanced version of what was seen in a renowned fps. Famous precisely for its combats.

Jerion
17th Nov 2008, 17:32
Oh my....whatever could it be? :D

Half Life? F.E.A.R? Halo? (hey, let's face it, the Halo games may be medicore, but the enemy AI is very good)

GmanPro
17th Nov 2008, 22:06
Halo ai was pretty good I have to admit.

My favorite part in the whole series was in the first game, I used to have so much fun messing around with assault on the control room. There were rooms where I would sneak around and melee the sleeping grunts down, and carefully walk up behind unsuspecting elite patrols and hit them in the back for an instant take down. Or when I would run up real quick and grab a banshee from the elites before they could get in them. Good times :rolleyes:

sonn
18th Nov 2008, 02:21
I agree that AI will either make or break this game, but not in the sense that you're describing. NPC characters can be very very hard and still have a brainless AI. My hope for this game is that the WHOLE world of Deus Ex 3 would have a smart living and breathing AI system. Now THAT, I agree, will either make or break the game. Examples of good game AI would be something that you see in GTA5 or Fallout3, where you feel that the world is constantly revolving around your actions.(although Fallout has dreadful Animations) Example of good character AI would be the npc in the original F.E.A.R...

I'm not sure if Deus Ex is still trying to achieve a dynamic world like the previous game, but in my case, it was the reason I got attracted to Deus Ex 2; you chose your path, and everything changed according to that.

DXeXodus
18th Nov 2008, 04:56
....it was the reason I got attracted to Deus Ex 2; you chose your path, and everything changed according to that.

Just a pity that never ended up being realized in the game. I have a lot of hope that consequences to ones actions will be a lot harsher in this Deus Ex.

K^2
18th Nov 2008, 05:45
I've never done AI programing...
I don't mean any offense, but it shows. I'm far from being an expert on AI, but I have taken courses, studied knowledge bases and logic-based languages, as well as actually writing a few simple adversary game AIs.

Things you say aren't wrong. That's the general outline of how a descent AI works, but it is a very, very rough one.

Lets take out all the pseudo-emotional responses out of the equation. Same game - tag. Two agents. One set to chase, other to avoid. We don't care if they get "happy" or "sad" when they get caught. When that happens, two roles simply switch. In fact, we can just change which agent controls which token/avatar. Lets call the agents "hunter" and "prey" for short.

Lets look at a basic AI behavior for this. Start with a very basic one: hunter seeks shortest path to prey and follows it. This, from the start, causes problems. Path finding algorithms are extremely complex to evaluate in complex environment. Even in a very simple case of a checker-board-like world with certain checkers marked as obstructions, and assuming that hunter can "see" where prey is at all times, this is pretty complex. A lot of games just fail at this point, and talking about the rest of AI is a bit moot.

But let us pretend that you wrote a half-decent path-finding algorithm. But now we look at the prey. Prey isn't going to sit in one spot. It will wonder around. If it wonders around randomly, there is not much need to alter algorithm of the hunter. It's about as good as it will get. But lets try to give prey some rationality. First thing to try is to maximize distance between self and hunter. So lets look for a location that removes prey as far as possible in minimum number of steps, pruning off paths that require prey to approach hunger first. This will improve prey's chances of escaping.

Now, we look back at the hunter algorithm. Hunter, knowing that prey will try to escape, can try to predict prey's path and cut it off by taking a short cut to prey's destination. Prey can now anticipate that, and chose a different route. This becomes a game of rock-paper-scissors.

How is this resolved? It is resolved by building a Min-Max game into your path-finding algorithm. It works like this. Lets look at all the steps a hunter can take over an increment of time. For each possibility, let us consider all the steps that a prey can take. For all of these, consider following hunter steps, etc. This forms a new search tree, which resembles the tree for path finding, but is now including two agents with zero-sum objectives.

Typical Min-Max algorithm follows. If the game ends, hunter reaches prey. Let us give hunter reward of +1000 for success. If the game doesn't end after N turns, where N is the maximum depth you can afford to compute to, we need to assign some negative score, based on how good position is for the prey. Simplest one is to give negative of the distance between prey and hunter. Though, if you are clever about properly scoring "dead ends", you can significantly improve upon this AI. In most reasonable situations, you'll only be able to compute a few steps ahead. So you'll have to find a good way to quantize your playing space to make path finding as easy as possible, and to allow you to evaluate strategy over big chunks of time with fewest number of "steps". Then you proceed to apply more generic Min-Max optimizations, such as alpha-beta-pruning, etc. These you can look up in pretty much any AI book.

All of this is a lot of work. It is mostly standard stuff, but it takes a lot of effort to get it right for your particular environment and your particular modes of locomotion.

And now let us add all the standard bells and whistles of a video game. What if there is more than one prey? What if there is more than one hunter? Can all hunters see all the prey? If not, can they communicate information to each other? How do you predict where the prey went when you cannot see it? Does the hunter/prey have a map? If not, can they deduce it? (Consider the Wumpus game). And then you throw in the final complication where each agent has both the hunter and the prey roles to follow, and decision on which model to use is dependent on how good the odds look in each role.

And again, this is without any of the "psychology" involved in the AI. This is a straight-up cold and rational agent playing each side. Even this, relatively simple case is a black hole requiring a dedicated team just to put something reasonably practical together. Designing an AI that can even remotely approach human capabilities in these roles is all but impossible.

Edit: I'm really not sure if it should be "prey" or "pray". I think I even used them interchangeably in the text. Apologies for that. I hope you know what I mean from context.

Lazarus Ledd
18th Nov 2008, 08:23
The ai needs some kickass catch phrases like "maybe it was just a homeless guy" or "A BOMB" :lol:


More one liners :D
Like in Thief series!

lastdual
8th Dec 2008, 18:45
Stealth is obviously a big factor in this game, but so few games these days handle stealth well at all (hardly anything has since Thief). One of the main factors for satisfying stealth is multi-leveled AI awareness.

Basically, for stealth gameplay to not suck, I've come up with 4 stages that the AI needs to be able to progress through:

1) Oblivious - The AI doesn't have a clue that you're there. Patrols normally and has conversations with other NPCs, etc.

2) Curious - The AI heard or caught a glimpse of *something*, but they're not sure what. Maybe it was a rat, the wind, etc. They may investigate the source of the sound with minimal effort, but will soon give up and shrug off the distraction.

3) Suspicious - The AI knows that *someone* is there. They will run a fairly thorough sweep of the room, call out to you telling you to come out, etc. The AI will eventually give up, but will be more alert afterwards (if they hear something else, no matter how minimal, they'll skip over "curious" and go back to "suspicious").

4) Aware - The AI has seen you clearly and knows where you are. Will pursue aggressively unless you take them down or manage to somehow break their line of sight for long enough. Obviously, will again be more alert afterwards if you escape.

Without all of these 4 stages, stealth tends to fail. Other behaviors of course play a role, such as injured soldier calling for backup and a general feeling of vulnerability. I just really hope this game nails it.

Jerion
9th Dec 2008, 00:08
Threads merged. :)

TrickyVein
11th Dec 2008, 03:04
I've never done AI programing, but here is my take on it.

consider what is real. The human brain.

the reptilian brain or R-complex(R)
fight(anger) or fight(fear). Aggression(win,euphoria) or submission(loose,shock), patterned responses and core binary logic modes.

the old mammalian or mid-brain.(M)
states or moods.. memory(mood dependent) split personality as each state is a different mind so to speak. Each state must express 2 modes for R complex states.

the neocortex.(C)
abstract logic.. abstract goals.. people are the most illogical animal not the most logical and thats out strength. the neocortex isn't grounded in the reality of now and lets us adjust our reality. It makes states more loose as it can address memory in other states. This grew into multi state abstract thinking.:scratch:



There is more evidence to the contrary to say that much of what we consider to be conscious thought is actually rooted in "primitive" structures like the amygdala. We don't really know what all of our cortex does exactly, so it's easy to relegate our "consciousness" to its various parts, but this is decidedly unscientific.

If one examines the channels by which our very real and conscious emotions and moods are actually altered, one looks to the neurological connections between these more "reptilian" structures: the limbic system, the amygdala, the hippocampus.

It would be sooooooooo easy to just say that "yes, this is how the brain must work, so therefore it works this way" but this is increasingly becoming not the case. I often think that AI researchers have got it all wrong trying to create a mind without a body in the first place.

FrankCSIS
11th Dec 2008, 04:10
I often think that AI researchers have got it all wrong trying to create a mind without a body in the first place.

That's pretty interesting. I'd like to hear more if you feel like developing.

I don't know if that's where you're getting but what it makes me think of is the fact that the body obviously has, if not a mind of its own, at least a very long memory, transmitted through our genes (?) from one generation to the other, and this memory seems, at least at time, to influence our actions, needs and thoughts. Every now and then urges seem to invade the conscious mind for no apparent reasons, sometimes from present body needs, but sometimes from previously unfelt, seemingly past events and needs resurfacing.

We always explore how the brain controls the various body movements from nerves and other strings, but too often we forget how the body dictates the actions of the mind. With that in mind, no pun intended, I can see how difficult it might be to "program" a mind from scratch, especially considering we still ignore how a thought is actually generated. I have a hard time seeing how one could be generated artificially, unless it was previously implanted, defeating the purpose.

Of course I'm also thinking that without a body the AI has no connection to the physical world, and therefor doesn't understand any of the concepts behind the words initially programmed. There is absolutely no point in implementing a language if it has no grounds on reality. A hamburger will never be anything else than the combination of the letters h a m b u r g e r, and will never be associated to hunger, something the AI cannot possibly understand without a body, just to name this infantile example. The concepts of truth or love are entirely out of the AI's league.

We've seen, in the past 5 years or so, a number of products and inventions that were programmed to go by themselves to the nearest charging station when they were running low on energy. This would be a good occasion to associate the notion of hunger in the human body/brain with the notion of need for a machine. What I'd like to see is a machine that has the sensors to detect its energy level (the equivalent of our body signals), knows of the various energy outlets in the vicinity, but wasn't initially programmed to go recharge when it reached a certain level. In other words, I'll be impressed when the AI puts two and two together on its own and figures how to continue functioning, assuming it has a survival instinct, however you call it. This would be a very primitive version of one of our most basic chain of thoughts that dictates a good part of our daily thoughts.

It also reminds me of those who pretend they can program random. We have zero understanding of how random actually works, and end up with increasingly complex algorithm, carefully disguised, posing as random number generators. Making it more complex and difficult to crack/perceive doesn't make it any more random.

TrickyVein
11th Dec 2008, 13:01
All really good points. However, I've never really thought of programming per se - I just like the science behind it all, especially when it comes to one's own body image and how malleable that is.

Think about just how much of the brain is dedicated to just directing the motions of flesh and interpreting signals that the body receives - I think you would not be incorrect in saying that the brain is simply an outgrowth of the organism which needs it, a master control device to direct its livelihood, that is, to guarantee the passing down of one's genetic material from generation to generation. "Why do we need to be conscious of that?" you might say - perhaps consciousness is no more than a reflection on these bodily processes and sensations. But I speculate.

I suppose the best way to create life is to find a good woman and become a dad (or good man and become a mother).

However, this is all rather abstract and moot when it comes to programming AI in game. I don't think that the brain acts like a bunch of binary modules but this has been the trend to treat it as such when laying out AI, right? After all, one needs only to code for the functions that the pawn is supposed to carry out in the game. Enemies don't need to reflect on the meaning of good and evil, they just need to react to the player appropriately and "if, then" statements seem to take care of that pretty well. Maybe I'm wrong. My experience fails me in these matters.

K^2
11th Dec 2008, 20:43
Vein, you are correct in thinking that consciousness requires some sort of a direct environment. You cannot have a brain sitting in a jar disconnected from everything and expect it to be conscious in any meaningful way.

However, that environment doesn't need to be a fleshy body. It doesn't even need to be real. Hook that brain into a computer running a virtual environment. It can be as mundane as virtual reality simulating real world, or it can be something rather abstract. As long as it is consistent and not too complex for human mind to understand, it will work. Now, the computer will send stimuli correlated with virtual environment to the brain. It will also modify virtual environment in some way that is correlated to the output of the brain. One last thing is to have a "good" stimuli and "bad" stimuli generated based on some conditions. These carrot and the stick are required for the learning process.

Brain is flexible enough to figure out the rest. Visual cortex will respond to real domain input, while auditory cortex will pick up frequency domain input, using both to reconstruct some model of environment. It will then learn to forecast based on the model, and later to modify environment to improve the ammount of "good" stimuli received.

An AI can, in principal, do all of the above. All you need is an interface between environment (simulated or real via sensors) and an architecture capable of abstraction. A complex enough AI, in such scenario, will posses consciousness. Not necessarily the same consciousness as a human does, but it will be self-aware. Emotions are a separate discussion. They may or may not be present depending on the architecture. Their nature and extent can be controlled at design stage.

Spyhopping
11th Dec 2008, 21:01
Vein, you are correct in thinking that consciousness requires some sort of a direct environment. You cannot have a brain sitting in a jar disconnected from everything and expect it to be conscious in any meaningful way.

I understand your rationale if you are talking about a brain that has always been disconnected from any sort of input.

But if I stuck your brain in a jar and catered for all its needs to keep it alive, I see no reason for you to no longer be conscious. When lacking stimulation for a substantial amount of time the brain actually creates its own stimulation in the form of auditory, visual, tactile (and so on) hallucinations

FrankCSIS
12th Dec 2008, 00:11
they just need to react to the player appropriately and "if, then" statements seem to take care of that pretty well.

Funny you should say that, I was thinking the exact opposite while writing the previous post but didn't want to make it any longer.

"If Then Else" is an easy way out to build something that looks, on the surface, like artificial thoughts, but it becomes an astonishing waste of time as it gets more complex. In other words, the extra amount of time and coding required is significantly larger than the visible progress rate, because it plain simply isn't AI. A lot like my random number generator, it's a complex simulation of Primary Directives built into the engine to make it look like something it's not.

I feel this is the wrong route to go, mainly because of its limited development potential. If Then is not the right way to go because it constantly, and forever, requires the input of the programmer for absolutely any decision.

I'll give you a fairly simple example of true AI, accessible to a machine, but also to an entity without a physical body. It implies the very human notion of Lie, a concept directly related to one of our strongest drive, what we so conveniently call survival instinct for lack of actual explanation.

Let's say you run a virus scan on your PC. The machine detects a critical virus that is either too widespread to be quarantined, or doesn't have any known cure (to the machine, anyway). The OS knows the only solution is a good format. It's not unreasonable to believe that it might lie to you in the diagnostic to avoid being wiped out, should it be aware of its existence. Of course, in the end it would only postpone the inevitable, but isn't that what we do each time we lie? In other words, it would override its If Then coding to protect its own interests, something the human brain does on a daily basis.

If you implement this notion into a virtual world, or a game, you get astonishing possibilities and open a whole new world of development possibilities. You'd also end up with many unforeseen possibilities and outcomes.

Of course, how to go about doing this is a whole other story. But surely you can see why If Then will never work out if you push its own logic to the limit.

GmanPro
12th Dec 2008, 00:39
I think that the situation you just described involves an if then statement.

If telling the user that a full sweep is required to remove the virus, then telling the user is not what I will do.

FrankCSIS
12th Dec 2008, 00:57
No no no.

The initial programing is

If you find a virus
Then mention it in the diagnostic

The AI takes the decision, on its own, to override its primary directive, or logical coding, in order to "survive", thus the lie, and thus the actual display of intelligence. Repeating an If Then command is not intelligence, it's coding. Breaking a rule is intelligence. It proves a thought.

Edit: Let me put it another way. We were all pretty impressed, I'm sure, the first time we saw a character running away from a grenade. In reality, there's nothing impressive here, just some very tedious extra coding lines. To over simplify, it'd be something like "If grenade detected in x distance, Then run away". AI would be detecting the object grenade and consciously decide to run away to survive, without the specific coding line implemented in the engine. But intelligence would also be unforeseen outcomes, like the sudden decision of kicking the grenade right back at you, so long as the notion of kick was in there somewhere. That would be a display of intelligence. It would mean putting 2 and 2 together without the input of the programmer, without an If Then command.

GmanPro
12th Dec 2008, 01:22
It still seems pretty binary to me.

If following the command will have a __% probability of resulting in the user doing a full sweep, then I will lie.

But I do see what you are saying. Basically the computer would need some way of determining the probable outcome of its programing. How would the computer know that displaying the message as per its "if then" command might result in a full HD sweep? How does the computer even know what the message states? Its all just data to be processed. There would have to be some sort of information comprehension and probability programing in there or something. Interesting concept. :thumb:

FrankCSIS
12th Dec 2008, 01:29
Pretty much, hence the principle that the AI needs to be aware of its environment, even if the environment is virtual. If Then programming makes this impossible, because it requires an order for absolutely everything you wish the system to accomplish. You will never be surprised by If Then coding, hence the lack of AI.

In reality, so far there is currently no such thing as AI. All we've seen are clever programmers who thought of realistic behavior those before them hadn't, and included it in the coding.

MaxxQ1
12th Dec 2008, 01:57
Pretty much, hence the principle that the AI needs to be aware of its environment, even if the environment is virtual. If Then programming makes this impossible, because it requires an order for absolutely everything you wish the system to accomplish. You will never be surprised by If Then coding, hence the lack of AI.

In reality, so far there is currently no such thing as AI. All we've seen are clever programmers who thought of realistic behavior those before them hadn't, and included it in the coding.

IOW, a true AI would make generalized conclusions (non-pre-programmed) to basic concepts (possibly pre-programmed) to carry out an action, no?

James P. Hogan's The Two Faces of Tomorrow, although written in 1979, has a pretty good discussion about this very thing.

One way he talks about it, and relating to my first comment above, is that you tell an AI that your cat has fleas. You want the AI to get rid of the fleas. The AI knows already (either through programming, or from some other source) that heat will kill fleas, so it makes a generalized conclusion that tossing the cat into an incinerator would take care of the flea problem. It certainly does, but nobody told it that the CAT was supposed to survive.

This kinda shows how STUPID an AI can be, even if it IS a true AI. Try If-Thenning the entire process of getting rid of fleas from a cat, and accounting for ALL possible ways of doing it, and you can see how difficult something like that becomes.

Of course, I could just be talking out of my ass.:D

K^2
12th Dec 2008, 02:02
I understand your rationale if you are talking about a brain that has always been disconnected from any sort of input.

But if I stuck your brain in a jar and catered for all its needs to keep it alive, I see no reason for you to no longer be conscious. When lacking stimulation for a substantial amount of time the brain actually creates its own stimulation in the form of auditory, visual, tactile (and so on) hallucinations
Well, I meant no sensory input from start.

Though, I would expect the sensory deprivation hallucination to get worse and worse until there isn't anything left but random fuzz. Entropy of information, same as other kinds, tends to increase. Something needs to keep it low. How long that would take, though, I don't have a clue. Might be longer than normal human lifespan anyways.

Edit: Let me put it another way. We were all pretty impressed, I'm sure, the first time we saw a character running away from a grenade. In reality, there's nothing impressive here, just some very tedious extra coding lines. To over simplify, it'd be something like "If grenade detected in x distance, Then run away". AI would be detecting the object grenade and consciously decide to run away to survive, without the specific coding line implemented in the engine. But intelligence would also be unforeseen outcomes, like the sudden decision of kicking the grenade right back at you, so long as the notion of kick was in there somewhere. That would be a display of intelligence. It would mean putting 2 and 2 together without the input of the programmer, without an If Then command.
Any form of intelligence hinges on if-then logic on some level, or can be represented in such form. The interaction between neurons in the brain can be represented in if-then form. The important part is what goes into the condition. If you explicitly state every situation to which an agent must react, it is just a simple response. But intelligence consists of a very large number of such simple responses that are dependent on the agent's internal state.

In reality, so far there is currently no such thing as AI. All we've seen are clever programmers who thought of realistic behavior those before them hadn't, and included it in the coding.
That is simply incorrect.

FrankCSIS
12th Dec 2008, 02:08
MAXX, it certainly highlights the idea of unforeseen outcomes :D

Basically though, you have it completely right. The only thing lacking to this programming would be Objectives. This way you have Notions as well as Objectives, but no actual If Then coding. The teaching approach is the only way to get to an actual AI, instead of the coding approach.

Let's say I want to make a war game. The Primary objective of the enemy AI would be to survive. The secondary objective would be to kill you and your allies. The essence of war, I suppose. Of course there's more, but it's irrelevant to the example.

Once your objectives are stated, you get to the teaching part. Implement the notion of health. Implement the notion that bullets and grenades are deadly threat to your health, and therefor your primary objective. Implement the notion of running, ducking, crouching, kicking, throwing, and so on and so forth.

Then sit back, press play and watch the simulation. What you should get is a very primitive display of AI. From there on you have the basis to build upon.

K^2, I think I put it wrongly in the first place. It's not the If Then that is the problem per say, after all this is how our own intelligence functions to a certain degree. What I mean is, there is no future in AI for If Then CODING, because it means the input of the programmer for EVERY SINGLE ACTION taken by the AI. Nothing will ever come out of it that wasn't initially programmed, hence the lack of intelligence.

As for actual AI today, if my statement is incorrect, please point me in the direction of a machine that has displayed intelligence in the past.

K^2
12th Dec 2008, 02:19
If I understand what you mean by if-then-coding, nobody ever programs a serious AI that way. Modern AI operate primarily by some combination of the following:

1) Solution space searches: If I do A, B, C, D, etc, where do I end up? How does that compare to objective?

2) Knowledge bases: I know A, B, C, and D. I just learned E, what is the conclusion?

3) Artificial Neuron Networks (Most commonly, Perceptrons): Feed input into network. Feed output into system. If reaction is bad, adjust weights.

In all of these, if-then structures are used only on the very low level. Essentially, to instruct the CPU. They aren't part of the AI's behavior.

Edit; Just FYI, most simple game AIs use first approach almost exclusively. Though, something like a chess engine will use all 3.

MaxxQ1
12th Dec 2008, 02:23
MAXX, it certainly highlights the idea of unforeseen outcomes :D

Ya think?;)


Basically though, you have it completely right. The only thing lacking to this programming would be Objectives. This way you have Notions as well as Objectives, but no actual If Then coding. The teaching approach is the only way to get to an actual AI, instead of the coding approach.

Let's say I want to make a war game. The Primary objective of the enemy AI would be to survive. The secondary objective would be to kill you and your allies. The essence of war, I suppose. Of course there's more, but it's irrelevant to the example.

Once your objectives are stated, you get to the teaching part. Implement the notion of health. Implement the notion that bullets and grenades are deadly threat to your health, and therefor your primary objective. Implement the notion of running, ducking, crouching, kicking, throwing, and so on and so forth.

Then sit back, press play and watch the simulation. What you should get is a very primitive display of AI. From there on you have the basis to build upon.

K^2, I think I put it wrongly in the first place. It's not the If Then that is the problem per say, after all this is how our own intelligence functions to a certain degree. What I mean is, there is no future in AI for If Then CODING, because it means the input of the programmer for EVERY SINGLE ACTION taken by the AI. Nothing will ever come out of it that wasn't initially programmed, hence the lack of intelligence.

As for actual AI today, if my statement is incorrect, please point me in the direction of a machine that has displayed intelligence in the past.

Funny you should mention war, because that's sorta the premise of the book: they have the ability to create an INTELLIGENT AI, but they're not sure if it will be intelligent enough to not harm humans. They build a space station as a closed, controlled environment, load the AI and let it run the station for them. They then systematically attack it, to simulate millions of years of evolution, and see what it does. By attacking, I mean they would shut down power sources randomly, singly, and in groups. They programmed the AI to survive, and of course it figured out that power souces dropping off the grid was A Bad Thing(tm).

I think you can see where this is going, but it's definitely a good read.

FrankCSIS
12th Dec 2008, 02:27
k^2 I'll admit to my own lack of knowledge on the subject, but I think you're placing intentions in the AI where there simply are none. In at least 2 out of 3 cases here, you're basically saying the AI has the capability of foreseeing an outcome that hasn't been previously programmed by putting 2 and 2 together on its own, based on previously implemented knowledge.

If there was such a thing today not only would games be immensely different, but so would technology in general. So far we're still debating on whether or not some animals can actually do it.

Maxx, looks like a good read indeed. I'll have to look it up. Thanks for the heads up.

K^2
12th Dec 2008, 03:04
Frank, there are plenty of games that can do things that are not directly programmed in.

Have you ever played a game called Worms? There are a bunch of worms on a 2D battle field, view from the side. They fire various projectile weapons at each other.

The shooting AI works roughly as follows. First, it is the rule of the simulation that all projectiles travel in parabolic trajectories. Wind effects these parabolic trajectories, but they remain quadratic curves. So given an origin and target, the AI can find a power/angle pair that hits the target dead on. Multiple pairs are possible, and they have to be checked. That part is just aiming algorithm. No real AI involved.

However, the trajectory may, and often does, intersect terrain somewhere before it reaches the target. That is, there is an obstacle in the way, which is where the charge will explode. AI tries a number of different angles simulating the shot, damage to terrain and everything that follows.

Now, imagine that there is a little hill near the enemy, and there is a mine sitting on top of that. One of the trial trajectories comes close, the projectile explodes and throws the mine into the enemy. Mine detonates throwing the enemy to the side where he dies. Worms explode when they die, so maybe he explodes taking out an explosive barrel which releases napalm onto another nearby enemy. That enemy gets washed down by burning napalm and ends up in the water, thereby drowning. Two kills for a single shot.

Now, could the programmers have coded this in? No. AI can't honestly predict such a chain either, but it can stumble upon it while searching for the best shot, evaluate the amount of damage dealt, and chose it as a best shot. AI effectively invents a way to do damage in complex ways using rules of the environment.

And these kinds of shots happen in the game. Not very often, because there are CPU limitations on how many trajectories can be checked, but they do happen, and it looks amazing when they do.

This is just the most rudimentary example of AI exhibiting emergent behavior (behavior not programmed in). There are games that push this kind of a thing further, but they are usually strategy games or such, and the end user rarely sees that happening. In a game like Worms, where the complexity of interaction is seen directly you can actually see it happen.

FrankCSIS
12th Dec 2008, 03:24
I've played Worms before, really fun game, but I admit I didn't pay that much attention to the details. I wasn't as interested back then as I am today for such topics.

I don't want to argue just for the sake of it, and I'm too limited to keep up, but I'd like to think out loud for a second and see where it goes.

What I'm getting out of this is the only limitation to AI right now would be CPU speed, correct?

Let's say I'm in the worm's situation. I see the mine. I see the barrel. I see the water. With my intelligence, assuming I actually think before I take actions, I'd consider attempting such a shot if I were sure of my throwing abilities. In other words, I'd analyze the terrain and take ONE shot, so long as I have the same guaranteed accuracy as the CPU. Or I might actually go for the sure kill, if I didn't think all the objects were lined up well enough for my chain of explosions to occur. Depending on my personality and past experiences I'd either risk or be conservative. That's a conscious decision taken after a thought was generated.

Now what bugs me with the worm example is the simulation part. To a certain extent, you could say our brain works the same way, but I'm not prepared to make that leap. The CPU here tries ALL possible shots in what I assume about a millisecond, and then makes the one that generates the most points according to the system. In other words, it does all possibilities and watches the results, something I don't have the luxury of doing as the human player. The CPU doesn't think which shot would be best, it simply does them all and picks the one that scores the most. It didn't invent the complex way, like you said it stumbled upon it by trying them all. The only reason it "picked" that shot was because of the scoring system most likely implanted.

If you gave the worm the possibility of doing only ONE shot, without simulation and only with the knowledge of the terrain, are we sure that's the one he'd pick? Would he hesitate between this attempt and the sure kill? Would he even notice the possibility? Most likely he'd just take the first simulation and throw. And to me, that's not intelligence, that's just quick processing.

If I didn't understand the process though, then please correct me. Like I said, I've never really paid attention to such things before.

K^2
12th Dec 2008, 05:08
What I'm getting out of this is the only limitation to AI right now would be CPU speed, correct?
Not quite. The fact that we are limited by CPU speed (also memory and access times) limits our understanding of how it all works on a really big scale. Currently, we can do about as much processing as some very simple animals. On that scale, we can build AI that behaves the same as these animals. So we seem to be on the right path. And I'm pretty sure that if we did have the resources, we'd be able to build some sort of conscious AI by simply upscaling what we do now.

Of course, whether that AI will have any intention to perform tasks you want is a tossup. I don't think we'll end up with a robot rebellion like Sci-Fi suggests. What we'll probably end up doing is designing architecture around preference for certain tasks. But that's just speculation. Point is, yes, processing capacity is the main problem, and the rest is doable.

Now what bugs me with the worm example is the simulation part. To a certain extent, you could say our brain works the same way, but I'm not prepared to make that leap.
That's precisely what I'm going to say, but I'll try to clarify to show how I see these things as similar.

When you, as a human, try to figure out the angle and power of the throw, you don't just magically arrive at the right values. You do still "simulate" various paths. You do so according to your own model, of course, and not the code in the program. You also do it slightly differently. Instead of starting with tracing the entire path, you examine certain chunks of the map, trying to picture all possible trajectories passing through it. You then look for the ones you can connect continuously from one such chunk of the map to the next. As such, you analyze far more trajectories than AI does, but you do a rough analysis instead of exact computations.

There is one more difference. When you look at that chunk of the map, you see all the paths in your head "at once". That's because your brain is massively parallel. Instead of analyzing paths each path along one by one, you just look at the whole picture for multiple paths at once. CPU cannot do that. So instead, it goes over paths one by one, step by step along each path. Also, the fact that all these things happen at once, you can trace your own model of what is going on roughly. If a trajectory you are tracing is starting to look wrong, your brain discards it before you even realize you were tracing it. If two traces don't match exactly, it can make small jumps. Things like that. Part of it is because your model of trajectory is experience-based. Part is because, again, you are looking at the whole picture at once.

Overall, what AI does is optimization. Keep in mind that original Worms ran on an Amiga hardware of the early 90's. Looking at the entire map and tracing only from source to target saves it a lot of unnecessary computation. It also makes it seem a little mechanical when you play against. There are types of mistakes that humans make that such AI never will. And, as you said, there is the factor of risk vs certainty. AI is always certain, because its model isn't experience based. You can fix that. You can have the AI watch projectiles and "learn" how they work. You can have it analyze different portions of the map instead of tracing trajectories. You can even have it learn that certain objects do certain funky things that can lead to extra damage, like that mine/barrel thing. And if you make the model approximate, you can make the AI play with chances. You can even have it learn which kinds of risks pay off and which don't. With some number of fortunate/unfortunate coincidences you can even end up with a superstitious AI. But all these things cost CPU clocks and memory. Something that games can't usually spare a whole lot of on AI.

Anyways, Worms was mostly an example of emergent behavior. Usually, you get more interesting results with learning AIs. You can't have real intelligence without learning. That's where knowledge bases come in. With these, games can really come up with new things. Not by just stumbling upon them, but by designing, experimenting, and correcting for errors. Problem is, AI like that takes up a lot of time to put together, eats up most of your CPU power, and overall doesn't end up making enough impact for most players to notice. I'm afraid, really good AI will start making into games only when chat bots with speech synthesis and recognition will become sufficiently good to be part of a game. When players will be able to ask question of NPCs, devs will have no choice but to implement top of the line AIs. Right now, they go for what's CPU-light instead.

GmanPro
12th Dec 2008, 05:27
I love worms lol. Great game. :D

I remember the AI making mistakes in that game though. Quite frequently. But who's to say that they weren't done intentionally based off of the difficulty settings? They probably were...

Here's a great question: Which came first? A reaction or a stimulus? Because that's all that AI really is. Actually, that's all that plain intelligence is, reacting to stimuli.

It seems to me that if a computer was capable of "choosing", then it wouldn't be AI, it would just be intelligence. By definition anyway, Cogito ergo sum.

TrickyVein
12th Dec 2008, 06:17
But surely you can see why If Then will never work out if you push its own logic to the limit.

I wholeheartedly agree - every variable, foreseen or unforeseen, would have to be accounted for by the programmer (rather impossible, no?) - perhaps it is better to engineer nature instead of nurture? To program mechanisms of adaptation and learning which would allow this strict adherence to self preservation no matter what the input?

HMMMMMMMMMMMM

GmanPro
12th Dec 2008, 06:21
Lol, maybe you could program the computer to program itself :eek:

TrickyVein
12th Dec 2008, 06:32
I remember the AI making mistakes in that game though. Quite frequently. But who's to say that they weren't done intentionally based off of the difficulty settings? They probably were...

[/I]

MJ12 still run into their own live grenades on realistic. Never ceases to crack me up. And UNATCO troops still shoot at each other to get to the enemy.

Yes, the end product of Descarte's systematic doubt - Intelligence can't doubt that it itself exists and is doubting itself. Is that right? Or did I just come up with a paradoxical strange loop? Wait -

Here's something to chew over - when a program encounters an undefined variable it will usually crash, not being previously told what to do with it or how to handle it. When we humans encounter a similar convoluted true/false statement, a paradox (try, "I'm lying"), we don't "crash" - we can live with this kind of inconsistency. How do we program this kind of adaptability into AI? Something which would allow the AI to jump out of its own predefined system and define new variables for and by itself? The AI would have to be able to create its own reality as it was introduced to more and more of the real world - or at least the world which we would allow it to "see."

GmanPro
12th Dec 2008, 06:47
^^My CPU is a neural net processor. A learning computer. :cool:
But seriously, if a computer can learn, adapt, and make choices, then it is technically, for all intents and purposes, intelligent.

-Damn! Now I have an urge to play Worms again. :nut:

K^2
12th Dec 2008, 23:13
I remember the AI making mistakes in that game though. Quite frequently. But who's to say that they weren't done intentionally based off of the difficulty settings? They probably were...
Precisely. Most of mistakes are actually done by taking a perfect shot and messing up the power/angle slightly. Sometimes, that resulted in a collision with another surface. This is especially apparent with grenade.


Actually, that's all that plain intelligence is, reacting to stimuli.
Yes. Though consciousness arises only when it can generate its own stimuli and react to these.


Cogito ergo sum.
"Cogito ergo sum," is a logical fallacy.

GmanPro
13th Dec 2008, 00:44
Because if you don't think, that doesn't mean you aren't? :scratch:

Care to elaborate?

K^2
14th Dec 2008, 02:33
Implication is I think => I am. It also means I am not => I don't think. Or in terms of basic elements "I don't think OR I am".

It does not require "I don't think => I am not" implication. Direct statement does not imply converse statement. So that isn't the problem.

The problem is that there is no basis for such a statement. It is arbitrary. You can use it as an axiomatic definition of existence, but as I said, it is rather arbitrary.

For example, consider statement "Nothing exists". That statement is equivalent to "Everything exists", because the difference between existing and non-existing is only a result of contrast. When you separate existing/non-existing into a full set and empty set, the difference is erased. Another way to think of it, if it makes it easier for you, is that if nothing exists than there is no rule saying something can't just spontaneously appear (because if there was, something exists) so everything would then exist. But now, if everything exists, something that thinks also exists. So we arrive at a situation where nothing exists, yet something thinks. Violation of "Cogito ergo Sum".

It is a good utility rule, because it makes the logic of pragmatic thinking a bit easier. It allows you to just assume you exist, and therefore, should act to protect your interests. But it does not work as an absolute rule. Furthermore, you can usually arrive at pragmatic logic following Newton's square. If you exist, and you don't protect your well being, you'll get harmed. If you protect your well being and you don't exist, then there is no loss anyways. And you arrive at all the same conclusions.

TrickyVein
14th Dec 2008, 07:04
if nothing exists than there is no rule saying something can't just spontaneously appear (because if there was, something exists) so everything would then exist. But now, if everything exists, something that thinks also exists. So we arrive at a situation where nothing exists, yet something thinks. Violation of "Cogito ergo Sum".



Aren't you also committing the logical fallacy, as you say, of (1) implying the converse statement from the direct statement (there are no rules forbidding existence, therefore without rules something will exist), and (2) implying that the converse statement must be true because the direct statement must be false in this instance (if you are saying that Descartes has indeed committed a logical fallacy)?

I'm not convinced by what you're saying, that it is a purely arbitrary saying. Descartes arrived at his famous saying because it was the only thing that would not succumb to the snares of Cartesian dualism. It was the only thing that he could not doubt, that it was he who was doing the doubting.

Also, surely something can exist without thinking so itself? You must agree, if you adopt the popular view of self-consciousness, that it is only us humans and perhaps few other animals who are actually imbued with self-aware intelligence - and yet obviously self-aware creatures are not the only things in existence.

TrickyVein
14th Dec 2008, 07:14
Yes. Though consciousness arises only when it can generate its own stimuli and react to these.



Centuries, nay, Millennia of philosophy and science, inquiry into the workings of the human mind, and at long last, here, on the Eidos forums for DX3, our very own "K^2" has answered mankind's deepest of questions.

Look no further! This is what consciousness is. So...expertly, so...effortlessly summed up in these few words. Why continue to study the brain? Who needs this neuroscience anyway? K^2's got all of the answers you need.

I've seen a deer pleasure itself at the zoo. Does this mean that it's actually...self aware? :eek:

K^2
14th Dec 2008, 08:00
Aren't you also committing the logical fallacy, as you say, of (1) implying the converse statement from the direct statement (there are no rules forbidding existence, therefore without rules something will exist), and (2) implying that the converse statement must be true because the direct statement must be false in this instance (if you are saying that Descartes has indeed committed a logical fallacy)?
I'm really taking a short cut. I mean, yes, it isn't necessarily true, but it cannot be proven false either. And as long as it can be true, Descartes' statement still fails. Did you follow or do I need to draw a diagram? (I mean no offense, it is a bit convoluted.)

Centuries, nay, Millennia of philosophy and science, inquiry into the workings of the human mind, and at long last, here, on the Eidos forums for DX3, our very own "K^2" has answered mankind's deepest of questions.

Look no further! This is what consciousness is. So...expertly, so...effortlessly summed up in these few words. Why continue to study the brain? Who needs this neuroscience anyway? K^2's got all of the answers you need.

I've seen a deer pleasure itself at the zoo. Does this mean that it's actually...self aware?
The millennia of thought on consciousness can be put into a pipe and smoked. The philosophers of then had not a single clue on how central nervous system works.

What we know now about consciousness is entirely summed up in that it is a result of various self-responses of the nervous system. Note that consciousness is not necessarily the same as self-awareness. Self-awareness requires subject to construct a complex enough model of the world to include self in it. Consciousness is simply ability to construct some model of the world, and self-interaction of the nervous system is key in that ability. Any number of modern works on neural nets agrees with this, including some studies of simulations.

A mirror-and-paint test is a great check for self-awareness, by the way. In that test an animal has a spot of odorless paint placed somewhere they cannot see (tip of nose, forehead, etc), and is then placed in a room with a mirror. Very few animals can look at a reflection and recognize it as such. Beside humans, only a couple of primates will attempt to wipe the spot of paint off after seeing reflections.

It doesn't necessarily prove that a particular animal is not self-aware, but I think there is little doubt that a chimp looking at a mirror, and starting to wipe off the spot on paint on self, rather than reflection, is sufficient proof that they are self aware, and therefore, without a doubt conscious. This requires the animal to construct a model of mirror's behavior, that it shows a copy of something in front of it. Furthermore, to realize that itself is part of the surrounding, and therefore, will be reflected. Then to conclude that if a reflection has a spot of paint, then it itself does to. Thus completing relation between self awareness and construction of the world model including self.

FrankCSIS
14th Dec 2008, 08:48
Sorry to get this back on somewhat of a track for a minute, but I just thought I'd say I see most of your points and it makes a lot of sense.

I still have somewhat of a difficulty with the definition of AI illustrated by those examples but I'll put that on the account that as a being of very short life span, I never witnessed the millions of years of potential intelligence evolution (assuming intelligence development wasn't almost instantaneous), and so have a hard time accepting those very primitive display as intelligence, rather than scripted coding with no independent thoughts.

I still think this approach has limited development potential for the mid to long term due to its very nature and logic, but I salute everyone working on the evolution of such technology. Hopefully they will prove me wrong. For all I know, although it doesn't appear as such right now, the progress curve might turn out to be exponential, with a spark, a specific point or status reached where the learning process would become similar to ours.

K^2
14th Dec 2008, 10:43
The developments in AI are frighteningly fast, and Moore's law plays a big role in this. About a decade ago, the general statement was that most advanced AI's have smarts of an insect. Now, advanced AIs can out think most reptiles. Tens of millions of years of evolution in less than a decade. Reptile's intelligence might not seem like much of an intelligence to you, but human intellect has the same basics. Just much, much higher scale. If Moore's law keeps on working, we'll have capacity to simulate a neural network as complex as human brain within 15-20 years. If we won't have conscious self-aware human-like AI in 25 years, you can call shenanigans on the whole thing. But you should see qualitative changes much sooner. Within a decade from now you should be able to see artificial systems that behave the way you expect a thinking thing to behave. They'll learn, comprehend, interpret, and adapt far beyond capability of any scripted system.

GmanPro
14th Dec 2008, 22:30
The millennia of thought on consciousness can be put into a pipe and smoked. :lol:

I think therefore I am ... I think. :D

Nine thousand and one internets to whomever knows what game that is from.

Mindmute
14th Dec 2008, 22:36
:lol:

I think therefore I am ... I think. :D

Nine thousand and one internets to whomever knows what game that is from.

That was actually first said By George Carlin, not a game, unless I'm mistaken.

GmanPro
14th Dec 2008, 22:37
This may be so. But the game I'm thinking of was released in 1999. IMO the second best game experience of all time.

K^2
14th Dec 2008, 22:55
I think therefore I am ... I think.
Don't know who originally came up with this one, but it is probably the most precise way of putting it without making it into a book.

NK007
14th Dec 2008, 23:57
I remember they promised that characters could go anywhere and you could kill anyone in STALKER... then the game came out and it wasn't so. So some guys messed with the game's AI folders and let loose with the Xray engine... The most important characters would stroll around The Zone and get killed by an anomaly or a boar. At other times army fellas would go out of their outpost and slaughter the entire first stalker camp, who are armed with 9mm pistols XD. The Barrier, which was a place some faction was constantly fighting to protect and was always succeeding, got completely overrun with mutants. I started the game and the trader told me "go see Wolf", so I went to a location on the map, 3 loading screens from where he's supposed to be, and there he is, lying on the floor cold as ice XD. It was crazy fun to just do it and find out what happens every time you restart the game, even if you couldn't possibly finish it. My god that was when I saw the future of gaming...

Imagine if MJ12 discovered the Silhouette hideout and would murder everybody inside, Sandra Renton going back to the hotel by herself and talking to her father, etc... if the game would then have an answer to that, then videogames would evolve.

spm1138
15th Dec 2008, 00:03
I remember they promised that characters could go anywhere and you could kill anyone in STALKER... then the game came out and it wasn't so. So some guys messed with the game's AI folders and let loose with the Xray engine... The most important characters would stroll around The Zone and get killed by an anomaly or a boar. At other times army fellas would go out of their outpost and slaughter the entire first stalker camp, who are armed with 9mm pistols XD. The Barrier, which was a place some faction was constantly fighting to protect and was always succeeding, got completely overrun with mutants. I started the game and the trader told me "go see Wolf", so I went to a location on the map, 3 loading screens from where he's supposed to be, and there he is, lying on the floor cold as ice XD. It was crazy fun to just do it and find out what happens every time you restart the game, even if you couldn't possibly finish it. My god that was when I saw the future of gaming...

Imagine if MJ12 discovered the Silhouette hideout and would murder everybody inside, Sandra Renton going back to the hotel by herself and talking to her father, etc... if the game would then have an answer to that, then videogames would evolve.

You've just told us why it isn't.

NK007
19th Dec 2008, 00:09
The possibilities of a future, d'oh. What happens when you let loose. How it CAN be.

singularity
19th Dec 2008, 01:38
The phrase "I think, therefore I am" was originally a philosophic quote from René Descartes in 1633.

On topic: In a game that revolves around the merging of biology with technology, it would be pretty ironic if the AI was horrible. However, good AI is, at least in my opinion, not always necessary for a great game experience. Back in the day, Perfect Dark had some pretty horrible AI issues, but is still one of the best shooters on the N64. Today, games that feature mindless hoards, like Left 4 Dead and Resident Evil 4 are some of the best on the market. Awesome AI would be just that -- awesome. But I've seen enough games win my heart over without fancy AI (Silent Hill 2), and had more than one that had some really incredible AI let me down (F.E.A.R, Far Cry 2, etc.).

K^2
19th Dec 2008, 02:38
We know it is by Descartes. We know it is from Meditations. And furthermore, some of us know that he's just using it as a loophole to crawl out of Cartesian Doubt and bring in the Light of Nature. A fallacy built on fallacy as an argument for another fallacy. The most important thing to know about Descartes is that he's full of ****. The only reason he's well known as a philosopher is that he was the first one to write in vernacular, and people who couldn't read Latin at the time were pretty gullible.

Read some Plato, Hume, and Nietzsche. In that order.

Spyhopping
19th Dec 2008, 11:05
I stopped studying philosophy years ago, everything I read was just grasping at straws. I find cognito ergo sum about as infuriatingly flawed as the ontological argument which tries to prove that God exists by the idea that we can perceive the 'greatest' being in our minds. Internal representation just doesn't cause things to spring into existence

K^2
19th Dec 2008, 11:33
I like Ontological argument. If one accepts it as valid, you can use it to further demonstrates that God doesn't only exist, but is, in fact, a particular person/item. E.g. "Lets imagine a most perfect pencil. Surely, a pencil that is a creator of all things is more perfect than an ordinary pencil, so the most perfect pencil must be that. And surely it would be more perfect if it existed, but we are talking about the most perfect pencil. Therefore, it exists and have created the world."

jordan_a
19th Dec 2008, 11:36
and had more than one that had some really incredible AI let me down (F.E.A.R, Far Cry 2, etc.).FC2 incredible AI? :eek: Are you sure you played it?

New article: http://www.gamasutra.com/view/feature/3887/the_technology_of_fear_2_an_.php

Radox Redux
20th Dec 2008, 01:38
To me, this thread seems to be less about AI and more about difficulty. Personally, IMO good AI rewards a player for doing good. Good AI is having a guard stand by a window so that if I were to shoot him his body flying through the window dramatically is my reward. Sure.... if I screw up and start playing Deus Ex like it's a Halo game... I expect the AI to spit-roast me. But if I'm showing mad ninja skills, I expect the guards to be set up to fall one-by-one. A good Ai should feature both risk and reward.

K^2
20th Dec 2008, 01:54
It's not AI's purpose to make the game look theatrical. AI should create a challenge. That is its purpose. Deus Ex is a game of choice, so it should not punish you for playing like it is Halo. What it should do is start plaing like a master Halo player. AI should start strafing, use different weapon combinations, and just generally make your life difficult. If, in contrast, the player is using stealth, AI characters should not be running around like idiots. They should be using search patterns.

In short, AI should adjust to the player's style.

Perhaps a better and more realistic way to achieve his is to have different kinds of guards. There can be sentry guards that either guard strategic locations or walk rounds, and if they notice something out of place, call in a few buddies and start investigating area, checking every hiding spot they know of. And then there could be commando guards, that generally don't do anything useful, but if bullets start flying, they start playing like they are Quake Bots.

Pure stealth player will probably try to sneak by the sentry guards, with commandos not posing much of a challenge. Pure action will go in guns blazing, taking commandos one on one, occasionally picking off sentries as targets of opportunity. An assassin player might use stealth to infiltrate the area, kill commandos silently, and then go guns blazing to clear the place from the sentries.

Note that sentry and commando guards do not necessarily have to look different or have different types of arms. It is just a type of AI behavior.

Radox Redux
20th Dec 2008, 02:00
I kinda disagree with the whole AI = Challenge perspective. To me, AI is simply reacting like a genuine person would... for better or for worse. Don't forget... humans are flawed creatures. To me difficulty should hinge on the set-up, not on an AI, since frankly most of the guards will be out before they get a chance to try any fancy new AI. You don't get much challenge from a corpse.

K^2
20th Dec 2008, 02:04
If you had a real human player for every guard in DX, you wouldn't survive it 30 seconds off the dock on Liberty Island.

Creating a more challenging AI will force them to behave more like humans would. AI already has all the flaws you need to make it believable. It's about getting rid of some of these.

Radox Redux
20th Dec 2008, 02:47
If you had a real human player for every guard in DX, you wouldn't survive it 30 seconds off the dock on Liberty Island.

Creating a more challenging AI will force them to behave more like humans would. AI already has all the flaws you need to make it believable. It's about getting rid of some of these.

Depends on the human. I'm glad you used the Liberty Island example, since it prooves my point nicely. Most of the NSF were just drafted in off the street and given a gun. If I went to the dock, there's a much higher likelyhood of them taking a swim than using professional squad tactics.

K^2
20th Dec 2008, 03:01
Well, sure, I agree. They don't all need to be skillful. But Liberty Island is also an intro level. It's not meant to be difficult. When I go up against highly trained guards, and I can kill a number of them before they can even scratch me, the AI isn't fighting hard enough.

NK007
20th Dec 2008, 04:20
Am I the only one who found DX too easy on realistic? Maybe the GOTY edition is just much easier in general? The game starts getting a tad difficult in the end when running out of resources, but what's with enemies in games always hitting 1\5 shots when shooting at you directly???

GmanPro
20th Dec 2008, 05:10
Am I the only one who found DX too easy on realistic? Maybe the GOTY edition is just much easier in general? The game starts getting a tad difficult in the end when running out of resources, but what's with enemies in games always hitting 1\5 shots when shooting at you directly???


I noticed that realistic mode felt pretty much just like the other difficulties except every once in a while an enemy would headshot u. Which I guess is realistic enough... I never really noticed how much damage enemies were doing to me because I always had regeneration-aug fully upgraded and in use.

AzureForge
25th Mar 2010, 14:05
This thread has not been active for a long time but I have vested interest in its renewal. There was interesting discussions about the philosophy of AI and difficulty but I would like to move it back to the actual Deus Ex AI. I think that the fact that this topic exist in the way it exist is a testament to the fact that Deus Ex require more than a classical FPS AI. As good as Fear and Fear 2 were, they are all about combat and Deus Ex is much more than combat. So here is what I believe to be the core of a good Deus Ex AI:

Combat:

The AI need a good modern combat AI at the very least as good as Fear. It need to be exiting and challenging while showcasing augmentations to at least some degree. The game should be able to stand on its own as a purely fighting game so that if you choose to pursue combat augmentations and fight your way through the game by kicking the butt of every m****r f****r that cross you way Krato style, you enjoy every minute of it.

Stealth:

Deus Ex is all about choice and multipath solutions to a given problem and stealth must also be key here. As far as AI is concerned, we are talking about a good detection and investigation system. The system must be a good balance of challenge and opportunity to repair your mistakes so that the stealth portion is not too harsh.

Non-Combat:

Deus Ex is also about the possibility to interact with the AI in a social, non threatening way. The AI should have many different actions and react to the player actions appropriatly possibly escalating in violence if the player chooses so. The AI must have a good threat analysis system where they detect that you have entered a restricted area, doing an illegal hacking or simply just shooting someone (or them) in the face and react approprietly to what they are (combattant, non-combattant, robot). You must also be able to speak with them.

I think that we can have a Deus Ex game with these 3 core concepts implemented in the AI. One thing that we can notice is that there is nothing exclusively Deus Ex there. Except for the uses of augmentations, these concept could apply as well to an Hitman game. I have therefore two questions for you all.

1) Do you agree with my analysis as to what is core to the AI to have the Deus Ex experience or do you think there is something else that I am missing without which we don't have a Deus Ex game?

2) We can assume that at the point in the development, these core systems are locked in. So what more would you want that would support the Deus Ex experience?

Big Orange
26th Mar 2010, 03:35
The AI should be appropriate to the type of mission and story - at the start of the game you should be worried about security systems and guard routes, since the enemy commando team are looking for you anyway.

Jerion
26th Mar 2010, 04:34
Ok...wow, necroposting.

AzureForge
26th Mar 2010, 13:44
To Big Orange:

The nature of Deus Ex make it so that the player could play it stealth like you suggest or burst through and go at it Rambo. The design and AI should make it both possible.

To Mr. K:

I just looked at the definition of Necroposting and if I did actually do a necroposting, it was not my intention. This thread is flagged as the AI game discussion threat http://forums.eidosgames.com/showthread.php?t=86931. I wanted to post about AI and I tought that using this thread would be the better way to do it. I don't fell that the discussion about AI is closed or my comment irrelevent.

Jerion
26th Mar 2010, 13:57
To Mr. K:

I just looked at the definition of Necroposting and if I did actually do a necroposting, it was not my intention. This thread is flagged as the AI game discussion threat http://forums.eidosgames.com/showthread.php?t=86931. I wanted to post about AI and I tought that using this thread would be the better way to do it. I don't fell that the discussion about AI is closed or my comment irrelevent.

Hey, you're right. This is the flagged thread, and good call for using it. It's my fault for not checking on the date of the last post for removing it from the Discussion Threads List. :)

Unstoppable
27th Mar 2010, 01:06
A.I. Eidos Montreal. If it's crappy so will be your game. I don't care if it has the best graphics in the world and I'm doing backflips like Trinity. (From the Matrix)

Weak A.I. means a weak game.

Big Orange
27th Mar 2010, 05:29
To Big Orange:

The nature of Deus Ex make it so that the player could play it stealth like you suggest or burst through and go at it Rambo. The design and AI should make it both possible.

I'm meant that in the first stage of the game, the semi-tutorial set up, the enemy grunts are breaking into the labs you're working at and you're a guard yourself.

However in the first proper infiltration mission there should be a learning curve to encourage stealth - in the first facility you break into, most of the guards could be talking shop in the cafeteria or sleeping at their desks and weapons worse than a pistol are locked up, but from a third of the way into the story, the enemy facilities are much more rigorously guarded from that point onwards (with key doorways or gates under security lockdown and less sloppy patrol guards packing automatic weapons, roving around in teams).

MaxxQ1
27th Mar 2010, 15:06
A.I. Eidos Montreal. If it's crappy so will be your game. I don't care if it has the best graphics in the world and I'm doing backflips like Trinity. (From the Matrix)

Weak A.I. means a weak game.

Really? I seem to recall many people (and critics) saying that the AI in DX was crap (and it was - you cannot deny that - even for the time it was released, the AI was crap), and yet, for the most part, they overall enjoyed the game, and think highly enough of it to give it several game of the year awards, one of the top five games of the decade (Maximum PC, just a couple months ago), number one of the top 100 games of all time (PC Gamer, again, just a couple months ago), etc., etc.

Now, don't take this as meaning that I think the AI in DX:HR should be crap. I just wanted to point out that that absolutism you stated was patently false. Granted, for the most part, you are correct, though.

Big Orange
27th Mar 2010, 18:00
The guard AI in Deus Ex was kinda weak in comparison to contemporaries like Half-Life (those legendary Marine black-ops) and even a console FPS like Perfect Dark (that used digitally captured human movements).

AzureForge
29th Mar 2010, 19:35
I totally agree with you that the level design (that also include how the AI is used) should be done to increase the difficulty as the game unfold. In the first level we can have one NPC patrolling a zone with a lot of hiding place with not overlapping patrol by another NPC. As the game progress, it can be made a lot tougher and the player has a lot less time to act.

You can have a good AI but if the level is crap and don't showcase it properly, it doesn't look good. Clever level design is key here.

Argent
30th Mar 2010, 01:09
and even a console FPS like Perfect Dark (that used digitally captured human movements).

What does this have to do with AI?

Big Orange
30th Mar 2010, 01:49
What does this have to do with AI?

The NSA/dataDyne goons in Perfect Dark had scripted behaviour that was more complicated than what was shown in Deus Ex - they had a better line of sight, reacted to distant gunfire, could run for back up, and also surrender to the player if they lost their weapon or were wounded.

Blade_hunter
30th Mar 2010, 02:08
Those black ops haven't a great, great AI they have a very agile and fast movement, I always kept revolver ammo to kill them and it's in one shot.
But in therms of AI, the DX AI wasn't a crap even at the time, it was lower compared to an Half life marine but it wasn't that crap.
A crap AI is what you can see in some games that make the game completely non enjoyable to play.
A too aware AI or a too unaware AI will ruin the stealth but completely.
too aware makes the stealth unplayable because the AI will detect you, and too unaware will make the stealth not interesting to play.
And when I watch the DX players the majority played stealth, so the AI wasn't that crap.
To me the bad aspect of it is the fact that the AI never took care about corpses making the fact we could hide corpses useless.
And the slow reaction time.
for the rest the AI can detect your sounds, avoid grenades if you throw a grenade nearby them, see you correctly, react when you shoot at them, sound the alarm when they spot you, you can even make diversions.

Don't forget that the DX AI have several tweaks and even comportments that a few AI had at the same time.
How much games have AIs that can react at so much things at the same time ?
I think people even doesn't take this in consideration.

To me despite the AI have some "stupid" comportments and ignore some parameters, it doesn't killed the game because it was descent and because you can play with.
But in therms of fightings the AI was calibrated a bit strangely, the AI can only dodge or crouch to shoot you.
That's where the AI was pretty weak, but it's a challenge to take down a moving AI, even if it did really fitted with the style of the game.

Deus Ex is rich in interactions and that makes a correct AI difficult to program.

The AI programming have a lot to do with mapping, unfortunately if mappers never place waypoints, map spots and such things the AI wouldn't react that well.
I mean when a mapper is doing his map, he have to think about the cover spots, the ways that the AI can take to be safe or even catch you from behind.
But and there is a but the AI itself have to be well programmed as well.

Argent
30th Mar 2010, 06:13
The NSA/dataDyne goons in Perfect Dark had scripted behaviour that was more complicated than what was shown in Deus Ex - they had a better line of sight, reacted to distant gunfire, could run for back up, and also surrender to the player if they lost their weapon or were wounded.

Thanks for the clarification. I was a bit confused since motion capturing movements for model/graphic purposes has nothing to do with how an AI actually behaves in a given situation, just how they look ;)

AzureForge
30th Mar 2010, 19:03
Those black ops haven't a great, great AI they have a very agile and fast movement, I always kept revolver ammo to kill them and it's in one shot.
But in therms of AI, the DX AI wasn't a crap even at the time, it was lower compared to an Half life marine but it wasn't that crap.
A crap AI is what you can see in some games that make the game completely non enjoyable to play.
A too aware AI or a too unaware AI will ruin the stealth but completely.
too aware makes the stealth unplayable because the AI will detect you, and too unaware will make the stealth not interesting to play.
And when I watch the DX players the majority played stealth, so the AI wasn't that crap.
To me the bad aspect of it is the fact that the AI never took care about corpses making the fact we could hide corpses useless.
And the slow reaction time.
for the rest the AI can detect your sounds, avoid grenades if you throw a grenade nearby them, see you correctly, react when you shoot at them, sound the alarm when they spot you, you can even make diversions.

Don't forget that the DX AI have several tweaks and even comportments that a few AI had at the same time.
How much games have AIs that can react at so much things at the same time ?
I think people even doesn't take this in consideration.

To me despite the AI have some "stupid" comportments and ignore some parameters, it doesn't killed the game because it was descent and because you can play with.
But in therms of fightings the AI was calibrated a bit strangely, the AI can only dodge or crouch to shoot you.
That's where the AI was pretty weak, but it's a challenge to take down a moving AI, even if it did really fitted with the style of the game.

Deus Ex is rich in interactions and that makes a correct AI difficult to program.

The AI programming have a lot to do with mapping, unfortunately if mappers never place waypoints, map spots and such things the AI wouldn't react that well.
I mean when a mapper is doing his map, he have to think about the cover spots, the ways that the AI can take to be safe or even catch you from behind.
But and there is a but the AI itself have to be well programmed as well.

Great resume of the Deus Ex AI Blade Hunter. The difference between bad and good AI can be pretty slim. Although the AI of Deus Ex did a lot of great things and was a huge part of the game, the way you could exploit it made an impression on a lot of people that it was crap. The AI must really be flawless to be considered great.

As mentioned, there is a lot of ways to interact with the AI and the AI designer/programmer must be very carefull in the way he handle those stimulus. While the NPC should react to the sound of a breaking glass while in a more relax state, he should not in combat unless he lost the position of the player (he's not seeing him currently) and the reaction would be less visual, maybe a sound cue pertaining to the fact that he's investigating a location where he think the player could be now. Obviously, the player could go hide and throw something in the hope to evade the AI and go back in stealth. A clever use of the mecanic that should be rewarded.

On the subject, I would recommand this talk about behavior boundaries. You need to be a premium member but I recommand the site in general if you are interested about AI design or programming:

http://aigamedev.com/premium/masterclass/decision-boundaries/

Blade_hunter
30th Mar 2010, 20:59
The link seems to be interesting, too bad that I have to pay that much to read the rest :D
And thanks even if my resume is a bit incomplete, but I tried to mention a lot of the attitudes that the DX AI had here.
Yeah programming a great AI for a Deus Ex game is pretty hard job due to the variety of interactions that the game tends to offer.
But normally the progress that was made in that matter is good even if there is some imperfections that can be seen.
And even if there is modern games with very buggy AIs sometimes

Unstoppable
31st Mar 2010, 03:35
What is A.I. truly though? It's not just the artificial intelligence in a battle. It is, was, and will be for Deus Ex everything that interacted with the player. The bank terminals, the cryptic messages when you did something in the game. The total package came together very well, even if you could snipe enemies before they knew you where there it was because you are somewhat of a demi-god.

My case is that this game needs great A.I., they need to hire someone who knows how to do that.

Argent
31st Mar 2010, 04:56
huh. I always thought A.I. stood for Artificial Intelligence.

AssassinAgent47
31st Mar 2010, 05:36
If this game turns into the disaster Conviction ,my heart will be broken.

As long as AI and stealth are very strong I will be pleased.

K^2
31st Mar 2010, 08:06
What is A.I. truly though? It's not just the artificial intelligence in a battle. It is, was, and will be for Deus Ex everything that interacted with the player. The bank terminals, the cryptic messages when you did something in the game. The total package came together very well, even if you could snipe enemies before they knew you where there it was because you are somewhat of a demi-god.

My case is that this game needs great A.I., they need to hire someone who knows how to do that.
No. AI is something that governs an artificial intelligent agent. An intelligent agent is something that optimizes output based on input in order to achieve a goal. Bank terminals and scripted messages don't optimize anything. They are just scripts. They are not agents, and no AI is involved.

But you are right in that the AI is more than just combat. How NPCs react to your actions and presence is also a part of it. If an NPC gives you different dialog depending on what weapons you are carrying, that's already a simple AI. A complex system that keeps track of your notoriety and recent actions could go a long way to making this game great.

Blade_hunter
31st Mar 2010, 15:08
Actually AIs needs innovative comportments in therms of video game AI
AIs in most of the cases is just a simulation of an human comportment.
And that's the difficulty.
Making super aware Ais that's easier than making an human like AI with descent defects, well human defects than making a computer AI that will know where you are ...

If the AI is too aware, the stealth is already dead. It will be impossible to play that way.

To me great AI doesn't mean amusing AI, they can make an AI that can play well the combat, now if you haven't enough interactions with, the fun will be seriously limited.
Also it depends what people think about "great AI" you can have a great AI but with only a few interactions with, and a not that great AI but with a vast panel of interactions with.

Being able to play stealth and combat in a same game needs more means to make a correct AI and much more if the AI have wide panel of interactions.

AzureForge
5th Apr 2010, 18:02
Actually AIs needs innovative comportments in therms of video game AI
AIs in most of the cases is just a simulation of an human comportment.
And that's the difficulty.
Making super aware Ais that's easier than making an human like AI with descent defects, well human defects than making a computer AI that will know where you are ...

If the AI is too aware, the stealth is already dead. It will be impossible to play that way.

To me great AI doesn't mean amusing AI, they can make an AI that can play well the combat, now if you haven't enough interactions with, the fun will be seriously limited.
Also it depends what people think about "great AI" you can have a great AI but with only a few interactions with, and a not that great AI but with a vast panel of interactions with.

Being able to play stealth and combat in a same game needs more means to make a correct AI and much more if the AI have wide panel of interactions.

The definition of AI is up in the air and even in the academic field they don't agree. As far as our discussion is concerned, I understand that AI is anything that requires decision making by an agent of the game. This include all the decisions of the NPC but also static objects like turrets. The information gathering (senses) and part of the execution of those actions are included. There is some caveat to include in the execution because while the pathfinding can be included, the line can be more blurry with animation. It depends a lot on the artistic vision of the game and tends to be its own domain.

An AI tends to be great if the decision it takes is great but depending on the type of game we are playing, those decision entails completely different evaluations. Nevetheless, it almost always entails to how an AI take into account the stimulus it receives to make the decision of what to do. While I would agree that a great AI need to be fun but what is fun depends of the type of game we are playing.

For exemple, in the stealth part, what is hard is not as much what to do (mostly investigate or attack the source of the disturbance) but more on perception of the AI and how to execute the investigation. While in combat the senses are much less important (last known position of threat) and much more in what we do and how we do it (type of attack, where I position myself, how I use cover). Outside of stealth and combat, I could suppose that the greatness of the AI would depends on its hability to interact with what you are doing. For exemple, if you point a gun at them, they should take notice. The devils will be in the details for this part as they could react to a huge variety of inputs from the part of the player and cost/benefits decisions must be taken by the developer. At the very least, they should master how the situation can move into chaos if the player do some crazy stuff and how it can revert back to a more relax state.

Obviously, for the AI of Deus Ex to be truly great, it must excel in all those area, not an easy feat. Any of you have some toughts about what kind of behavior you would like to see in the game. For exemple,an AI behavior could be that when you get close to a guarded area, the guard could come to you and say to go away. If you do not comply, you could expect some consequences that would not be compatible with a stealth approach ;).

Icarcus
5th Apr 2010, 20:22
When an enemy is shot at, and they dont have body armour - most body armour stops only pistol shots - they usually cry for mama istead of run at you guns blazing. Taking their bodyparts - shotgun on the leg, and they cant walk properly, heir leg is hanging on by the poants only. If its on the street, paramedics or a bregade of first aid bots comes in and takes the injured person away. please, if the enemy is too hurt, dont give them 40 medkits and in 5 seconds they are 100% regenerated..... shattering glass, explosions should impait hearing, and vision of npc's. they will walk around blinded and deaf no matter to what is going around them.. when you get shot (something combined like in stalker cop and far cry 2) you start bleeding out. if you fall from a great height, your legs are broken, so it is hard for you to move. use bandages to help stop bleeding; the far cry 2 part of this is that PUT IN AN ANIMATION of your character doing this; putting bandages on, crawling with only a pistol out on you stomach and occasionaly look back at your broken legs. Being on fire leaves a different texture on your skin. Going to a plastic surgeoun or a hospital should help fix that.

i agree about the ai having survival instinct. they should freak and run away as fast as they can, hide in a dumpster, climb in a closet. not to be sexist, but some women characters in ofice buildings and such, when you scare them with a gun they fall unconcious. taking people hostage, will either make them do what ever you tell them to say or if you take the wrong person, they luer you in a trap. Police should use tasers and tackle you, ahnd cuff you and put you in their car, take you to the station. either you brake out of the car, out of the station, if you cant - you spend some time in there (sleeping) and are let out. Having contacts bail you out if you commint something realy bad.

have realistic lives for the people. they go to work, you see them leave their home and go there.
you see them eat, drink at a bar, read, and then leave. you should be able to follow some woman all the way home and when she disables her alarm, say POLICE i have a warrant to search your house. workers occasionaly go to the bathroom, where you can scare them and pump out valuable information.

hahaha, when you breake into an apartement, witness a couple in the act, and when they notice you they freak out and call the cops unless you point a weapon.

a man with a crowbar shouldnt run at a machinegun in your hands.

have FAR CRY 2 aiming system, npc puts their hands up when you point a weapon at them.
others gather, some call police and flee.

KILLING people should have a GIGANTIC TOLL on your character. he would be followed to places by undercover police agents. Bars would have assigned max security guards to you or not even let you in. if you are wanted you walk around , some people that are watching public screens or tele-papers see you and you notice the expression on their faces, the whispering. By standards may try to help others by taking yo down.

IMPORTANT......................................................

please dont make the character just carry an RPG, FLAMETHROWER and a sniper rifle into a bar or something. weapons shouldnt be in his pocket, they should be on his back or in his hands, and its illegal to have them.

having a violin case for your dissassembled rifle when traveling would be nice, storing them in garbage bins or in your hotel room.

dont make MAANY guns avaivable. too unrealistic sometimes, make the nice super puper assault rifles appear with few characters, and LIMIT their ammo. make a weight limit , you can carry backacks of ammo and 5 rpg with the best augs and mechanical upgrades, but your hands will be busy. not on the streets.
its just that you FIND that nice cannon or you request one , hand made from germany or something, you will really apretiate it!!!!!!!!

guns jam, break, can be used in melle.
.............................................

when you kill a few people on the street you will be suspected, it will be extremly hard to get out of that.
the consequences should be such that you will WISH that you hadnt killed anyone. you will be followed, sneaking in someplace will be harder, police will be aware of your position. it will be fun to loose your tail though. the officers undercover wont come too near you, you can confront them.

they will leave, others will replace them. watch out for those earphones and old ladies talking in their sleeve.


make enviorements destructible. set a house on fire. watch it burn down. but you will regret it.
law enforcement, owners of the house that are bandits will hunt you. watch your back.

some things that you do, will aquire you such a wanted level that you wont be able to escape it.
THE GAME GOES BACK A COUPLE OF MINUTES JUST BEFORE YOU COMMITED THE CRIME AND DOES AN AUTOSAVE. YOU WILL NOT BE ABLE TO SAVE AFTER THAT UNTILL YOU ARE DEAD, NOR DELETE THE AUTOSAVE UNLESS YOU RELOADED IT AND DO NOT HAVE THE UNESCAPABLE WANTED LEVEL

the unescpable wanted level will have everything chase you, helicopters arrive, like in GTA4.

it is like fighting zombies. you cannot win, its it only a matter of time.

blowing up houses, cars, heli's, everything, but you will be contained. 40 spiderbots will ceize your hiding spot.
poison gas cans will be thrown at you. eventually the couple of locks will be quaranteened and bots sent in. little aero bots like in FRONTLINES fuel of war. they will see you. destroy you. you will run out of ammo.
they dont carry ammo you can scavage. you will die. you will wish you never started that rampage.

BUT MAKE IT REALLLY REALLY FUNNNNNNNNNN while on the rampage. just it wouldnt go with the story - like you blow up the laboratory yo are supposed to steal something from, in the p[rocess. you cant continure the game. thats why you cannot save. you will ave to start from the auto save. but make the fun.

Blade_hunter
5th Apr 2010, 20:35
Yes often AIs are tailored for one type of gameplay, I agree.

AIs that tells you to go away from a guarded zone have much more to do with scripts than proper AI often the guard is guarding a zone and we often can't do it stealthily, except if we use an alternate path.
That's not always the case but some attitudes of that kind are just made with scripts and some AI tweaks so the parameter that makes the AI being hostile.

But actually AIs are a bit more made for both but oriented in a certain way.
Actually some basic FPS have a component of stealth, and the goal is to change your position and being able to neutralize your own prey but without being detected too much.
And when it come for a stealth game the AI es even able to take cover instead of being super resistant or having artificial handicaps to force you to play stealth.
I mean actually a lot of games tries to combine stealth and combat.
And that's much more your gear, weaponry that can make you able to play stealth or not.
Playing stealth is a lot about being hidden and not spotted but also being able to make a diversion.
So we have movement involved and some gear.
When we make assassinations and knockdowns this is much more about combat but in stealth games where we are able to fight only that way we can do that only against isolated enemies, or very low numbered.
Playing combat is facing 1 up to 3 - 6 enemies by making a gunfight and being able to aim wall and take cover.
The tactics in combat have too much to do with our weapons, the fact hey are fast firing or slow, have a spread, have splash damage/area of effect.
and even the AI have to react in regards to our weapons that we are using.
They will be more prudent if you have a powerful weapon than a weak.
I think the AI should have a "morale" factor.
The kind of factor that affect the combat and even the fact that NPCs can surrender.
But this is AI demanding ....

I forget some things but it's a bit hard to list most of the AIs comportment we can see and that are enjoyable in games

SquidPirate
5th Apr 2010, 23:35
The guard AI in Deus Ex was kinda weak in comparison to contemporaries like Half-Life (those legendary Marine black-ops) and even a console FPS like Perfect Dark (that used digitally captured human movements).



Agreed. Half-Life had some of the best AI I'd seen at the time... and frankly, it was better than that from its own sequel.

Halo AI was pretty decent. And Thief, as I recall, did a good job.

Blade_hunter
6th Apr 2010, 00:00
Perhaps but half life AI doesn't managed stealth and alarms, think about it
Thief AI was good for stealth purpose, but when it comes to combat that's much more their super resistance that make them defeating you rather than their own ability to fight properly

Energetic Sentry Bot
7th Apr 2010, 20:55
I agree, challenging AI and thought puzzles, Half-Life 2 esque.

Unstoppable
8th Apr 2010, 02:15
I forgot what game I played recently but the best games are those where the A.I. will react on everything you do. It was probably Grand Theft Auto: San Andreas. That game has decent A.I. Also the F.E.A.R games and Mass Effect 2 in higher difficulties.

Alpha-9
8th Apr 2010, 15:59
Adaptive AI plz, ArmA/OFP style.
I want to be flanked and attacked, not just stood there shooting each other in the face...

Add multiplayer DM mod (always nice)

Alpha-9
8th Apr 2010, 16:00
I forgot what game I played recently but the best games are those where the A.I. will react on everything you do. It was probably Grand Theft Auto: San Andreas. That game has decent A.I. Also the F.E.A.R games and Mass Effect 2 in higher difficulties.

R u srs?
GTA had crap AI, obviously because of the thousands of people walking around the streets, if by good you mean running away screaming because you punched a car, then yeah, amazing :mad2:

Blade_hunter
8th Apr 2010, 21:02
Yeah but for some people AIs that owns you by killing you in one shot, by killing you by being multiplicated, having perfect aims, having and awful resistance and/or doing a ton of damage is great AI.
Just because it finishes to kick our asses even in a stupid process.

Alpha-9
8th Apr 2010, 22:01
Well considering the majority of people are idiots, they're idiots.
A great AI is an AI that responds to its enviroments and makes logical choices based on it. Not "Oh going down this corridor, one will come out on the right here with a shotgun" once you've played it once you know it, but unpredictable, realistic AI, the kind of AI that FIGHTS for its life like its human

Blade_hunter
8th Apr 2010, 22:14
That's the thing, but you know more the AI have choices at its disposable more difficult is to program it.
Also managing a ton of NPCs is quite hard.
But Actually I think that some games even with a high number of enemies can make some intelligent choices.
But it's pretty rare also the greatest progress in video games is much more in the graphics area rather than the AI.

Irate_Iguana
9th Apr 2010, 07:43
the kind of AI that FIGHTS for its life like its human

The same AI that jumps INFRONT of the car instead of away from it?

maikaal
9th Apr 2010, 09:42
I don't know, AI was hardly one of the things that made Deus Ex so incredible..

ⓣⓐⓕⓕⓔⓡ
9th Apr 2010, 18:44
I haven't seen much discussion on this issue at all. The A.I. in Invisible War was pitiful. The Deus Ex A.I. was alright because on realistic you could die from a head shot.

What I'm getting at though is that we as the players need A.I. that constantly challenges us. Without that this game is good as dead in the water. I want special agents that come after me during a mission and use powerful nano abilities that kick my butt.

This happened in Crysis but I was able to beat about 4-5 nano suit wielding A.I.'s a bit too easily. (On the hardest setting)

In conclusion what I'm asking for is that the game punishes me for playing on the hardest difficulty. I want to spend a few hours on a level getting my butt kicked and having to figure out how to out smart the A.I. That way there is something for those of us seeking a true challenge.

Lately though I'm beating FPS games because I can figure out the A.I. in a matter of hours. I guess I just have a small niche for figuring out their weaknesses. I hope the QA team on this one shows no mercy and makes the A.I. tough as nails on the hardest setting.

Because the AI in Deus Ex was THAT GOOD. As good as FEAR! :rolleyes: :rolleyes:

People go on about Deus Ex as if they were the glory days. They were not. Deus Ex was not the pinnacle of gaming.

Ashpolt
10th Apr 2010, 00:11
Deus Ex was not the pinnacle of gaming.

I know! It didn't have free automatic health regen or third person cover or anything! I also didn't feel that my moves were cool enough, I'm an android or something, I should be able to just flip out and kill everyone at the push of a button. In slow-mo.

And what was with all the talking? BORING!

Spyhopping
10th Apr 2010, 00:30
And what was with all the talking? BORING!

Firm hand on the space bar next time they whining about terrorists. Then hit F2 and you got yourself a nice little summary of who you gotta kill next! Neat!

Senka
10th Apr 2010, 08:21
The same AI that jumps INFRONT of the car instead of away from it?

Probably just wanted to sue somebody.

gamer0004
10th Apr 2010, 13:15
Firm hand on the space bar next time they whining about terrorists. Then hit F2 and you got yourself a nice little summary of who you gotta kill next! Neat!

What? I WANT ARROWS TO DIRECT ME. Readin summaries is for ******* dawg.

Alpha-9
11th Apr 2010, 23:51
I know! It didn't have free automatic health regen or third person cover or anything! I also didn't feel that my moves were cool enough, I'm an android or something, I should be able to just flip out and kill everyone at the push of a button. In slow-mo.

And what was with all the talking? BORING!

Sarcasm i hope :nut:
I hate superhuman style games where you can't do anything but win, its all up to you! For some reason.. Who the hell am i anyway!?

Things i liked about Deus Ex: Invisible War the most was the multitude of endings and paths to follow, in fact as soon as I completed it, i started again just to see the different endings, it was just the right amount of hours you could play without being agonising.

Omar ftw

Blade_hunter
12th Apr 2010, 12:03
I don't know if people never thought about that question, because I feel often that AIs have some problems to use some sort of weapons not only those that require a lot of strategies and thinking but even those with particular effects.
So does people sometimes felt that AIs weren't able to use correctly some sort of weapons ?

Jerion
12th Apr 2010, 17:50
I don't know if people never thought about that question, because I feel often that AIs have some problems to use some sort of weapons not only those that require a lot of strategies and thinking but even those with particular effects.
So does people sometimes felt that AIs weren't able to use correctly some sort of weapons ?


I have come accross this phenomenon, where an unfriendly NPC's tactics don't reflect the weapon they carry. If there's one thing online tactical shooters teach you it's that even the smallest tweak to a weapon's characteristics can drastically alter how you play with it. In Bad Company 2 for example, you can take two guns which appear similar in stats, but neccessitate different tactics for use. The M416 is more suited for semi-precise CQC and doesn't care if you spray a bit. By contrast, the AKABAN swaps the M416's RoF and damage stats. However it requires considerably more precise shooting and thus neccessitates somewhat more of distanced, take-your-time approach. Often you don't experience this kind of tactics differentiation in AI. In untrained soldiers you can expect this, but when fighting trained troops it would be nice to see some such acknowledgement of varied weapon handling.

Cronstintein
13th Aug 2010, 02:10
Well I anticipate some corridor style fighting. What I would really like to see in combat AI is this:

-Flanking tactics in open environments.
-Pin down and call for backup in closed in environments.
-Intelligent usage of grenades, flash bangs, smoke, suppressive fire and proper usage of their chosen weapon.
-If you break line of sight you should be able to flank BUT if the AI hasn't seen you for 10-20 seconds they should get suspicious!
-Increased AI accuracy if you continually use the same cover. I would hope the reason for this is obvious but in case it's not...

When in a cover based firefight it's about getting a bead on where they're going to pop out and being ready for them so when they start to come around, BOOM. That's how you likely kill the basic AIs used in these games: Let them take a few shots, then when they go behind cover you come out, set your aim, and wait. I'd like them to do that to YOU, forcing you to go a different way or take a few bullets in the face for being dumb.

In a corridor situation without the ability to flank I'd like to see a combination of this (forcing use of a flash, frag or aug on your part). Perhaps combined with an assault, where if you take too long THEY throw the frag or flash or smoke and bum rush you (with the shot-gunners if available).

Anyway I just hope it's a bit more tactical then we're used to seeing in sticky cover games. I don't feel like I'm asking for a lot but many MANY games fail to deliver on AI and it cheapens combat. The warehouse scene in the demo didn't particularly wow me, but it was pretty short. I'd like to see what happens if he comes in a door, making noise, setting off the hornet's nest.


Longer post than I anticipated, hopefully not too boring.

Jerion
13th Aug 2010, 02:31
You know what I'd like to see every now and then? When an enemy notices that it's allies are getting shot to pieces, and instead of freaking out, gets sneaky. Instead of making itself obvious as so many NPCs do, it should try sneaking and flanking around to get you by surprise.

Cronstintein
13th Aug 2010, 02:41
You know what I'd like to see every now and then? When an enemy notices that it's allies are getting shot to pieces, and instead of freaking out, gets sneaky. Instead of making itself obvious as so many NPCs do, it should try sneaking and flanking around to get you by surprise.

Yes or if a baddie fails a morale check and takes off, instead of running around wildly he should barricade him self in a room with only one door and keep a weapon trained on it. (pushing a desk or something over for cover would be a nice touch but likely requires scripting) That's certainly one of the things I'd consider in that situation.

Walking into an ambush due to alarms would be pretty cool too.

Or a sneaky one-man ambush, waiting just to the side of the door with a pistol ready at head level like protagonists often do in movies.

Pinky_Powers
13th Aug 2010, 03:14
AI is one of the things I've always been most excited about in the advancement of Gaming. It's an ugly, slow progression, but some companies are making headway. Monolith, with their FEAR franchise, is doing better than most. But We're still a long way off from where I think we should be.

Here's hoping EM is not just blowing smoke up our asses with their "intelligent enemies".

AzureForge
13th Aug 2010, 14:54
A lot of good ideas, I will try to tackle them.


-Flanking tactics in open environments.

One of the main problem when making AI is communication with the player. Not only must the AI be bright, the player must understand that he's bright. A problem with flanking is that a lot of players (we could call them "casual" :P) find it frustrating. If he could see the clever way the AI flanked him, maybe his opinion could change, but the experience he got was to be shot in the back. The solution is twofold. 1) Communicate to the player the flanking. It should not be obvious but there should be sound cues and visual hint to help him out. What is cool about a game like Deus Ex is that augmentation can be used to acheive that and give empowerment to the choices the player make. 2) Scale the flanking behavior with the difficulty level. The harder the game his, the more flanking should be used by the AI.


-Pin down and call for backup in closed in environments.

Using closed environments defensively should be rewarding to the player but the AI should not go all to their doom. Staying back in defense and waiting for the player could be good. Clever player could then sneak back behind them using augmentation or air vents ;).


-Intelligent usage of grenades, flash bangs, smoke, suppressive fire and proper usage of their chosen weapon.

Gotta love when an AI flush you out of your cover with good use of a grenade. Shot-gunner should rush you while machine gunner suppress and sniper cover them.


-If you break line of sight you should be able to flank BUT if the AI hasn't seen you for 10-20 seconds they should get suspicious!

When they lose sight of you, one of the AI could go investigate while the other cover him in case he find something ... like his death.


-Increased AI accuracy if you continually use the same cover. I would hope the reason for this is obvious but in case it's not...

Good grenade use and flanking should have solved this problem, no? This does not prevent the player from camping endlessly.


You know what I'd like to see every now and then? When an enemy notices that it's allies are getting shot to pieces, and instead of freaking out, gets sneaky. Instead of making itself obvious as so many NPCs do, it should try sneaking and flanking around to get you by surprise.

Civilians should freak out, special ops with augmentation is another matter ;).


Yes or if a baddie fails a morale check and takes off, instead of running around wildly he should barricade him self in a room with only one door and keep a weapon trained on it. (pushing a desk or something over for cover would be a nice touch but likely requires scripting) That's certainly one of the things I'd consider in that situation.

I like the idea of AI defending a location. I hope it's in the game ;).


AI is one of the things I've always been most excited about in the advancement of Gaming. It's an ugly, slow progression, but some companies are making headway. Monolith, with their FEAR franchise, is doing better than most. But We're still a long way off from where I think we should be.

Here's hoping EM is not just blowing smoke up our asses with their "intelligent enemies".

The whole history of AI (independent of video game) has been one of high hope and slow progress. The fact is, it's hard to make good AI. Still, I think conversation like this help move it up in the right direction and maybe EM is reading it.

pringlepower
13th Aug 2010, 14:58
A lot of good ideas, I will try to tackle them.



One of the main problem when making AI is communication with the player. Not only must the AI be bright, the player must understand that he's bright. A problem with flanking is that a lot of players (we could call them "casual" :P) find it frustrating. If he could see the clever way the AI flanked him, maybe his opinion could change, but the experience he got was to be shot in the back. The solution is twofold. 1) Communicate to the player the flanking. It should not be obvious but there should be sound cues and visual hint to help him out. What is cool about a game like Deus Ex is that augmentation can be used to acheive that and give empowerment to the choices the player make. 2) Scale the flanking behavior with the difficulty level. The harder the game his, the more flanking should be used by the AI.



Well Mass Effect sort of had flanking... in that while most enemies were perfectly content to stick to cover as if they were magnetically attached, the strongest, heaviest enemies would actually walk towards you. Which, on Insanity, just freaks you out, since you're used to taking your sweet time in whittling down enemy HP while they don't advance. Goddamn YMIR mechs.

I know this is off-topic, since it's not really about AI, but that sort of enemy behaviour - which completely throws you off your game, can be really effective, especially for "boss" enemies. Or the AI using a a heavily-armoured enemy to charge you while his minions follow, spraying a hail of bullets.

Cronstintein
15th Aug 2010, 00:02
Yeah ME2 had a couple fights with enemies doing that. The lizard juggernaut guys are what I'm remembering. But it was far from the norm. Most of the time it was shooter whack-a-mole which is a really sad direction gaming has been heading.

EDIT: As much as many purists ragged on the most recent splinter cell, the 'last known location' feature was pretty cool. I noticed the original deus ex had a form of this system as well, lets hope they do a good job this time.

sharpie
17th Aug 2010, 10:31
I really don't remember FEAR having great AI, sure it was new and strange (moved around a lot) but it wasn't exactly 'challenging'.

IMO there should be roughly 3 different tiers of A.I.

Civilian: This is basically non-hostile A.I. Could be a guard, bystander or a friendly who has no reason to hurt you. Think, guard from HL2 or pedestrians in GTAIV.

Defensive: Guards inside of a building, or any unit who has been fired upon. Think any guard from SC: conviction (shoot on sight, patrol) or FEAR/Arma.

Offensive: Military units who are on the hunt for you. Think MGS (preferably snake eater)

Then there would be tiers within the tiers.

Offensive-in-tier: For instance maybe one group will prefer to split up, one may choose to stick together. One man may rush you, one may instead try to nade and shoot you. May choose to flank/surround you, or simply attack at the same time as a team.

Defensive-in-tier: One guy may seek the nearest cover, one may choose to fire sporadically at you while seeking better cover. One guy may blind fire at you turning it into a stand where he can accurately shoot you (exact amount depending on accuracy) another may lob a nade and seek better cover.

Civilian-in-tier: May scream and run when you pull a gun or may try to disarm you. May pull own (possibly hidden) gun.

None of these tiers specifies any particular person or unit and preferably all persons can switch between the three at will. I.E. a guard may ask to see papers and when you instead pull your gun he may run for a defensive posture and attempt to call for backup, or may switch to an offensive posture and try to take you down. Or a man on the defense may become offensive. Or a man on the offensive may become civilian (if you throw down your arms)

If you're looking to build A.I. the best bet would play any FPS online. There is an absolutely unending range of different tactics and people. The most realistic to build from imo would probably be AA2 (haven't played/liked arma) because death carries weight (a wait of about 5 minutes :D) and you see people being incredibly cautious and very rarely taking the offensive. I don't know how legible all this is (5:31am here) but I'll be more than willing to clarify tomorrow if it's needed.

*also* Typically the most realistic AI lets you make the first move and acts accordingly. I.E. Most people take the defensive posture as they have no interest in dieing (thus why you see cops enter standoffs with police cars).

*also2* "(if you throw down your arms)" This part gives me a great idea. What if you did exactly this and say a cop moved into arrest you, and you slice him up soon as he's close. that would be pretty great, especially if the AI was good enough where you couldn't safely take on a man who was entrenched behind cover.

*edit*

EDIT: As much as many purists ragged on the most recent splinter cell, the 'last known location' feature was pretty cool.

Problem being that they just moved into the slaughter, no sane person would do that. Find cover and let the enemy come to you, especially if he's Sam freakin Fischer. But more importantly you surround the last known location before you even remotely get a man on that position. Sometimes the AI attempted this, but most times they just funneled in (if it wasn't just a lone guy coming to my last known position and died like dogs. This could of been a product of the maps in SC:Conviction but all the same, even without the insta-kill ability it was too easy to rape on that game.

Pinky_Powers
17th Aug 2010, 13:07
I really don't remember FEAR having great AI, sure it was new and strange (moved around a lot) but it wasn't exactly 'challenging'.

That's because you were a super soldier who could slow down time. Try playing the game without SloMo and you'll see how good their AI was. They were aggressive animals with realistic aim who flanked and pressed and knew how to flush you out with grenades.

Crysis also had quite decent AI. But with your Nanosuit and all the foliage, it was rarely a challenge.

sharpie
17th Aug 2010, 21:42
That's because you were a super soldier who could slow down time. Try playing the game without SloMo and you'll see how good their AI was. They were aggressive animals with realistic aim who flanked and pressed and knew how to flush you out with grenades.

Crysis also had quite decent AI. But with your Nanosuit and all the foliage, it was rarely a challenge.

Fair point on FEAR, I did like my slowmo.

Not a fair point on crysis, at least not all of the time. I quit the game when I tried sniping a small encampment that was down the hill from me. They rushed up to my position (despite using a silenced weapon) and despite me being invisible one of the guards was standing right in front of me and staring at me.

I hate psychic AI, besides there's plenty of instances of the crysis AI being daft.

Pinky_Powers
18th Aug 2010, 04:06
Fair point on FEAR, I did like my slowmo.

Not a fair point on crysis, at least not all of the time. I quit the game when I tried sniping a small encampment that was down the hill from me. They rushed up to my position (despite using a silenced weapon) and despite me being invisible one of the guards was standing right in front of me and staring at me.

I hate psychic AI, besides there's plenty of instances of the crysis AI being daft.

There's a huge difference between AI being perfect and just being better than most. Crysis was much better than most, but it certainly had it's issues.

Also, you could not turn truly invisible. It was a cloaking technology, not magic. I liked that the enemy could get a lock on me or become suspicious even when I was cloaked. It was an aid to my stealth, not a cheat code.

Jerion
18th Aug 2010, 04:19
There's a huge difference between AI being perfect and just being better than most. Crysis was much better than most, but it certainly had it's issues.

Also, you could not turn truly invisible. It was a cloaking technology, not magic. I liked that the enemy could get a lock on me or become suspicious even when I was cloaked. It was an aid to my stealth, not a cheat code.

Couldn't have said it better. It was a stealth aid and a darned good one but not an invisibility blanket. If you wanted to effectively hide, you had to, well, hide. Cloaking made that a little bit more effective and let you dash to and from cover mostly undetected. At distance it was helpful for dodging snipers and stuff but there's nothing like true concealment.

sharpie
18th Aug 2010, 05:30
Couldn't have said it better. It was a stealth aid and a darned good one but not an invisibility blanket. If you wanted to effectively hide, you had to, well, hide. Cloaking made that a little bit more effective and let you dash to and from cover mostly undetected. At distance it was helpful for dodging snipers and stuff but there's nothing like true concealment.

I was crouched in a bush...

Jerion
18th Aug 2010, 06:04
I was crouched in a bush...

Next time, try a tree. They tend to be harder to see through. ;)

"Hey Jake, what's up with that bush up there? It looks all shimmery and stuff."

The AI did have it's moments of "What? How do they see me?" but they were pretty few and far between IME.

sharpie
18th Aug 2010, 11:20
Next time, try a tree. They tend to be harder to see through. ;)

"Hey Jake, what's up with that bush up there? It looks all shimmery and stuff."

The AI did have it's moments of "What? How do they see me?" but they were pretty few and far between IME.

They were at the bottom of a hill, I fired and killed one guard. And another guard (from the bottom of the hill) ran up the hill to stand literally in front of me waiting for my suits power to run out. And I'll admit that it may have been a rarity as I didn't even remotely give the game a chance but I still refuse to say crysis had good A.I.

pringlepower
18th Aug 2010, 14:39
They were at the bottom of a hill, I fired and killed one guard. And another guard (from the bottom of the hill) ran up the hill to stand literally in front of me waiting for my suits power to run out. And I'll admit that it may have been a rarity as I didn't even remotely give the game a chance but I still refuse to say crysis had good A.I.

Yeah Crytek games tend to have guards with 50/20 vision.

AzureForge
18th Aug 2010, 19:05
Let me have fun with some speculations. I think that the problem with Crisis and cloaking is the way they implemented their "look at" system. Its a common problem in programmation to have architectural dissonance. AI are usually omnipotent and one of the job of the programmer is to make them know only what they really knows. The AI know that the player fired the shot and therefore it's the target of the AI system. On the other hand, he lost sight of him and uses it's last known position (probably where the shot was fired) to go get to him. This position is probably close to you since you may not have move that much. The lookat system may be a completly different system that uses the current target and when he get close enough, it activates and the AI look at you even if you are cloaked. The cloaking system and the look at system want different behavior and you think correctly that the cloacking should be winning. It's pure speculation here, but I'm not really surprise that it happened. Let's hope the Deus Ex guys are aware of these kind of problems and it will not happen ;).

Cronstintein
28th Aug 2010, 09:00
WanderingKid posted this in an unofficial DX forum (those suckers don't get all the Dev posts like we do) but I thought it was an excellent take on AI:

Squad Leader A:

Aggressiveness: 4/10 (determines how hard and fast he come at you and how much damage he does)
Tactical Awareness: 7/10 (%age certainty of using cover, flank maneuvers, grenades, zig zagging, spreading out within visual range to make multiple kills harder etc)
Situational Awareness: 8/10 (determines threat detection radius, %age certainty of dertermining distance and direction of Player sounds like footsteps and gun shots)
Experience: 7/10 (%age likelihood of retreating to cover if taking damage, likelihood of panicking and running away, likelihood of surrender if about to die with no where to run)

Squad Leader bonus: +2 to all stats for all members in squad (by virtue of being team leader)

Now imagine different stats for the lieutenant and the rookie that roughly gives a sense of what they are about (rookie would have low tactical and situational awareness, would be more prone to friendly fire, more like to run away under fire or straight towards you in a straight line etc).

What can follow is a whole variety of ways to dispatch this squad through emergent possibilities. If on their patrol route you catch the rookie and take him down, the remaining squad strength is still pretty high but you may be able to turn the rookie and use him as a meat shield.

Take down the squad leader first and the remaining team members lose their squad bonus, become disorganized and easy to dispatch one by one. Catch all of them together at a bad time and they find cover quickly, sling grenades and flashbangs and spread out so you are vulnerable in your blindspots giving you a hell of a fight.

Now add another variable outside the AI but which the AI reacts to. Kill the rookie with your gun and drag his corpse down an alley. Baddies killed in this way are flagged to leave 'bloodtrail'. It doesn't need to be visible, its just something that the AI can react to when either it detects its presence and so determines the distance and direction and goes to investigate. Lets say the lieutenant patrols the area and detects the blood trail. His lack of situational awareness causes him to just follow the path to the corpse where you can set up an ambush.

Lets say the squad leader discovers the blood trail first. His high situational and tactical awareness make him move from cover to cover en route to the body and then lay down smoke grenades to foil a potential trap. Or the player could exploit the liklihood of the enemy AI reacting in a certain way and ambush them en route.

Add another variable. etc. What this is, is a move away from scripted events and preprogrammed routines in favour of setting a number of basic conditions or rules that determine the behaviour of an object or enemy AI and then simply letting things happen when you put them all into a box. With physics those base conditions are a crude simulation of gravity which determines how bodies fall (ragdoll) rather than animating a falling body by hand. It can result in some hilariously unique and unanticipated 'death animations' where a corpse flies, spinning through the air like a catherine wheel and bounces off a building on the way down but I've seen some ways where the system breaks down like baddies getting stuck in objects when they collide with them.

The upside is that this procedural approach to game design can create unpredictability, unique events which never happen in the same place twice. The downside is that if the rules system isn't well designed and thoroughly tested, the capacity to 'break' the game can be quite high. Scripted stuff can be more controlled so you can make set pieces that you want the player to see consistently like in Half Life 2 which is essentially a 10 hour sequence of preprogrammed events. It is extremely well done and the illusion is very well maintained but on your second playthrough you start to see where the cracks are.


**********************
END QUOTE

Anyway, I just liked this idea for AI and thought I'd bring it here.

AzureForge
31st Aug 2010, 15:13
Gotta love dev posts ;). I like the intentions of WanderingKid but I will highlight some of the problems and difficulty with his proposed implementation and how we could do it.

First thing first a couple of comments about the chosen categories:

Aggressiveness: For me, this attribute can come in conflict with the tactical awarness. The decision to go hard on a player, I presume rushing him, is not a decision that should be left to chance (like most of the decision that follow, wich is one of the main weaknesses of such a number approach). It should be based on things like am I under attack making a rush stupid, can I easily reach the player, do I have good cover where I am going, do I have visibility on the player were am I going and what are the strenght and weaknesses of my weapon to do such a tactic (A shotgun rock, a pistol or sniper rifle not so much)? The speed is also problematic. Unless you have a good blending system or some great procedural animation system, having lots of different speed is visually problematic. You don't want slipping feets so you are usually constrained by the animation set to walking, jogging and running. This limit the posibilities unless you invest heavily in animation. Damage ruled by number is fine.

Tactical awarness: Again you cannot really use % to decide what strategy to use, you can regulate how frequently they can be used but the hard part is how and when to use it correctly. An NPC that don't use cover will usually look bad. The player don't know the poor guy failed is %, all he see is a fool that stay in the open. Flanking maneuvers are frequently percieved as a cheesy back attack unless the player actually followed the flanking move. They are also guided by more factor than just a will to do it. Do I have a path that don't expose me too much (not a trivial question to answer using programmation), do I have a good flanking position, will it be compatible with my weapon properties and again am I under attack at the moment (maybe I should stay in cover)? Again, the decision to throw grenade is complex. Will I hurt one of my friend, must I expose myself in a dangerous way to throw it, where must I throw it (if the player is in cover, his position won't do the job)? Spreading out and zig zagging is useful only if I do not actually weakens myself in the process so it should be done to actually improve mytactical position like moving toward a better cover. If you are spreading just for the sake of it, it will look bad.

Sitational Awareness: This can be handled fine with numbers. Just be aware that will probably want to change them depending on the situation (relax, alerted, in combat, looking through the scope of rifle can all change the detections property of an NPC).

Experience: NPCs in FPS tends to die fast ;). Panicking and runing away makes sense for civilians but generaly create confusions in the player mind when the combattants do so making him an easy target. I like the concept but it's quite hard to make it look good in a game. The way I know it can be done is that he runs to actually get some help (activate an alarm, go fetch reinforcement) and then come back to fight. Surrender can be cool if there is only one guy remaining but it's the kind of feature that usually fall in the nice to have category. Not a good category to be in in game development :P.

I really like the idea that the way you take on an encounter have concequences. It always does in the way you use cover and if you kill the snipers first or the shotgunner first for exemple. Still, I like to explore new ways of doing it. The hard part is to communicate the intelligence to the player so he can enjoy it. When he kills the leader, he must see the confusions among the troops as a consequence of the lost of leadership, not as bad AI programmation which can easily be the case. Instead of downright confusions and running around, I think it's better to have a change in the quality of the strategy.

For exemple, take a bunch of NPC defending a room. If you killed all the good leaders, they may decide to hide under the bed and in closets. If they have a good leader, they may lock windows and doors and move furniture in front of them. If the leader is great, they may put traps in front of doors and windows and then hide themselves luring you in as weak and jumping on you when you get trapped. In all cases they do logical, understandable and well executed actions but they have different level of usefulness against the player based on how good the player was.

The bloodtrail idea is neat and empower non leathal kills. It should be visible has to not confuse the player.


Lets say the squad leader discovers the blood trail first. His high situational and tactical awareness make him move from cover to cover en route to the body and then lay down smoke grenades to foil a potential trap. Or the player could exploit the liklihood of the enemy AI reacting in a certain way and ambush them en route.

Great exemple, notice that this has nothing to do with chance. Investigate using cover instead of staying in the open is a great way for the AI to look bright and make him more fun. Smoke grenade can be used to create a temporary cover if the space between two physical cover is far. Again, it's a decision that should be based on the situation and if it's a good idea. Notice the quick increase in complexity when you add all the decisions requirements that I mentioned since the beginning of my post. Handling this complexity is one of the main challenge of AI.

Even if procedural systems are great, they have the weaknesses of usually not taking into account the big picture. It's espacially hard to make them do coordinated behavior. Therefore, I think it's important to add another layer above individual decision making that could recognize some tactical situations and influence the NPCs decision based on global data and coordination. Another problem is that designers love to have control. Procedural is great of chaotic combat but outside of that, you may want to make it step aside and use more controled methods.

Anyway, great ideas and this is definitively a good discussion to make AI move forward.

tartarus_sauce
2nd Sep 2010, 03:53
One problem I have in the way of AI is how rarely the enemy as a whole seems to coordinate.

Remember that first assault on Liberty Island? You popped the guys onn the outside perimeter, and the team as a whole was never the wiser. Every little encounter was the same, no matter how violent you were. I think we won't see games that could properly be called "stealth" until the enemy applies proper "punishment." As in, if you take out a guard and he misses the next check-in, the rest of the unit either does an aggressive sweep or fortifies their position- or even makes a violent break-out attempt. Given the advanced communications gear of the near-future, there ought to be some very swift and sophisticated responses to the player's actions.

Personally, I'd like to see a game with a really punishing AI and health system. Something like the original Rainbow Six games, except with enemies that actually communciate and work as a team, and have different response scenarios that they can execute, and with reinforcements availible. I don't think any game has created a truly challenging squad AI yet, let alone a unit AI that responds at the platoon/company level. I don't think it's impossible, though. I think maybe the structure of shooters today just doesn't have room to accomodate such a thing, with the emphasis being on spectacle rather than realism.