AI behaviour when noticing alerted AI's.

[Obsttorte]

Hi everyone,

 

I updated the (what I call it) hitmanstyle approach, allowing to setup AI that is neutral to the player unless he does something forbidden.

 

Anyways, this setup has still a major nitpick. Currently alerted AI (thus meaning AI that has already identified the player as a criminal) does only alert other neutral AI, if the other AI does see the player. You can think of it like the alerted guard is saying something like "The guy over there is a thief!" or so.

 

This setup leads to the fact, that unalerted AI will stay unalerted if they see an alerted AI but not the player. I think his is wrong.

 

The question is if the previously unalerted AI should only become alerted (in terms of searching), alerted in terms of knowing the player or anything else.

 

My personal opinion is, that it should go to the place where th already alerted guard is, without knowing yet that the player is causing the trouble if he can't see him.

 

The question is, if for example an alerted AI causes an unalerted one to come to help and the alerted one gets killed or knocked out before the AI gets to a position where it can see the player. The AI comes around a corner for example and only sees the player. Does he suspect him now for beeing the murder? Does he do this only within a certain timeframe? What if he comes around a corner and another unalerted AI who has not seen the murder comes there, too? Does it depend on who the other one is? Surely, if it is someone of the same faction it is pretty clear who is to blame, but what if it is some sort of burglar (is this the right word?)?

 

I'd be interested in your opinions about this matter, as it seems like a quite complicated situation to me.

[RPGista]

In terms of core gameplay, I dont see a lot of difficulties - an unarmed AI, when scared off by the player, will run and alert guards on his way. I see that as the unarmed AI letting the others know what he saw and where, when they "meet". The same should happen in your setup, AI should be able to pass on the player's data when they see another AI (guards would run towards the place the player was last sighted, neutrals would become alerted and investigate/run too).

 

 

Map wise, it might be a bit chaotic to have every neutral AI running off aimlessly when alerted by other running AI. So my guess is that there should be more than one reaction possible, investigate the area should be preferred I think, even by unarmed (they would go into the "cautions" stance anyway, so it would look appropriate) - they are trying to understand what made the other one flee like that.

 

 

AI that is alerted by witnesses haven’t seen the player, but it is implied that the witness will "describe" the player to everyone else, so it is acceptable that they would attack/flee from us even though they weren’t there when the crime happened.

[Obsttorte]

That's true. But what if you see a murder but didn't see the murderer. You go to the corpse (because you are a guard) and see an unarmed person (the player=a thief, what you don't know)? Would you suspect him to be the murderer? Or would you think the murderer is surely long gone?

 

And if you are the player, which behaviour would you expect? And if you would expect the guard to identify you as the murderer if he comes around the corner shortly after the deed, what amount of time would you interpret as "shortly"?

[grayman]

This all depends on how you define friendship and neutrality.

 

A friend is someone who will come to your defense, recognize your enemies, help you search, listen to your warnings, greet you as you pass each other.

 

A neutral person is someone who will greet you, but do no more. He will not recognize your enemies unless they are already his. He will not help you search, or listen to your warnings. He will not warn you about things that have alerted him. (And up until changes I made a couple weeks ago, he wouldn't even greet you.)

 

So an AI who witnesses the player doing something illegal is only going to expect his friends to respond to his cries for help (because he knows AI neutral to him will ignore him), is only going to ask a friend to help him search (because he knows AI neutral to him won't respond), and is only going to warn his friends as he patrols (again because he knows AI neutral to him won't care).

 

I don't understand this sentence: "Currently alerted AI (thus meaning AI that has already identified the player as a criminal) does only alert other neutral AI, if the other AI does see the player." Are you saying that an alerted AI is alerting an AI that's neutral to him? Given the rules above, this won't happen. Perhaps you meant to say "does only alert other unalerted AI"?

 

The mechanics of the game are pretty simple. Friends warn each other and come to each other's aid. They share the same enemies. Enemies fight. Neutrals couldn't care less. The player is a predefined enemy of practically everyone, which gives him an incentive to stay out of sight, which is the point of the game. He's in a place where he doesn't belong.

 

Allowing relationship changes lets mappers tinker with this mechanic, to create situations that go against the basic mechanic. Up to a point, this works reasonably well.

 

However, the social behavior you're trying to attain is problematic. Having an AI round a corner and spot a corpse and then try to play Sherlock Holmes to determine who among those present might be the murderer is a bit beyond our current capabilities. As I mentioned elsewhere, all our aspiring detective knows is that there's an arrow in the body. Does that mean the only bow-wielding person in the room is the culprit? Does it mean the true culprit vanished and the archer currently in the room is just a guard who responded to a death cry? Does it mean that the archer thinks the detective is the murderer if the detective is carrying a bow? Perhaps the detective shot the arrow, then climbed down from some perch to get through the door the dead body was guarding, and stumbled upon a room full of people he didn't know would be there.

[RPGista]

Thats a tougher one and I remember people discussing this very issue a few weeks ago. This problem only ever arises when the AI involved is neutral to the player, so they have no reason to attack him unless attacked/threatned in some way.

 

A case could be made either way - neutral AI should only become hostile to the player if he is *seen* doing something bad, so he could always blend in the crowd if there were no witnesses to his crimes.

 

However, that might become exploitable, and some kind of inference should be accounted for, if we want to keep things more challenging; as you say, a chance of the player being blamed for the crime should be there, specially if it happened a short time before an AI spots him. He is dressed as a ninja and sneaking around, after all. So yes, certain crimes should be linked to the player more easily, but there are problems - a valuable item that is stolen should only be important to certain AI (the guards and servants inside the mansion, say), not to every AI in the map. So how would you know which crimes an AI would care for/recognize might also be a question.

 

EDIT: Now that I read Grayman's explanation, Im not sure what I wrotte applies.

Edited March 14, 2013 by RPGista

[Obsttorte]

I don't understand this sentence: "Currently alerted AI (thus meaning AI that has already identified the player as a criminal) does only alert other neutral AI, if the other AI does see the player." Are you saying that an alerted AI is alerting an AI that's neutral to him? Given the rules above, this won't happen. Perhaps you meant to say "does only alert other unalerted AI"?

I misspelled this sentence a bit I guess. I'm lacking a good word, but what I've meant was that AI that is hostile to the player will only turn other AI hostile to you if the other AI does see the player. So it can point at the player and say "this is the thief" but he can'T describe him. (This is all not meant wordly ).

The mechanics of the game are pretty simple. Friends warn each other and come to each other's aid. They share the same enemies. Enemies fight. Neutrals couldn't care less. The player is a predefined enemy of practically everyone, which gives him an incentive to stay out of sight, which is the point of the game. He's in a place where he doesn't belong.

I'm not talking about the default AI behaviour. I'm talking about my specific setup. I used stims to carry the hostility form one AI to another. At the current state this only happens, as already said, if the neutral AI can see the player and is within the stim range of the hostile AI. I just think it may be a bit extented.

as you say, a chance of the player being blamed for the crime should be there, specially if it happened a short time before an AI spots him.

That's what I've meant, but I'm not sure what "short time" may be. Maybe it is something to keep for the mapper to adjust. I was also curious if such behaviour would appear expectable to the player.