Player Features and Ideas Discussion

Bump

As a professional research psychologist, I can assure you that this would be more or less impossible for a company like Eve to achieve, even if they poured all their resources into it.

That sort of thing is cutting edge technology that MIGHT be available in some form to certain military branches, etc. at this point, or dedicated machine intelligence firms that contract to heavy hitting contractors, or university laboratories. Not game companies, at least not at the level that would actually make the rats any sort of match for humans in creative strategy.

Yes, you could certainly add a few SPECIFIC variables that can vary according to learned behavior (like they learn to orbit versus approach or whatever with some percent likelihood that is altered by success rates), but that's not really what you are talking about, I don't think.

At this point in time, having their entire strategy being learned just isn't going to happen. It would be vastly more efficient to just write better algorithms to tell the rats what to do, or at best, to have a handful of predetermined variables that they adjust by learning.



Edit: the other problem with learning AI is that it is only learning one thing (or maybe one thing per region, or something). Whatever CCP can afford computations-wise. Whereas human players are learning 40,000 different things - every person is developing new strategies all the time. I've seen simple attempts at AI learning in other games (like puzzle pirates), and what often happens is that players figure out what the AI is doing in all of about 5 minutes, then it takes days for the AI to learn something new, at best. At which point the players figure it out again in 5 minutes.

Aside from this being impractical for reasons already explained...

...I think you misunderstand the goal of rat-AI.

It's entirely possible for a game-AI to be written to perfectly execute the rules of the game and kill you (in a fair fight) every time. This simply isn't any fun. The goal of the rat-AI is to give you computer opponents which don't mind dying all day long (humans tend to hate playing that sort of role).

Since the dawn of PVE, the challenge of killing NPC mobiles has been more about player optimizing their efficiency to kill such NPCs than about real challenge. This is because if they make NPCs randomly get really pissed and kill you unpredictably, the players who like to kill NPCs kind of hate it. It's okay for NPCs to be really hard, but it has to be in a predictable manner... since, afterall, they are run by the computer and thus have unfair knowledge of and ability to execute the against the game rules.

If you want a really player-like challlenge, go attack players.

No ofc you are right. I don't suggest they have no underling programming. CCP would build onto the preexisting AI adding more subroutines that alter behavior. In order to even approach this in any realistic fashion you have to nail down what is possible to log. Meaning, what information can be pulled from player interaction and made into intelligible information. I don't think we can expect to be able to record the mental state of a player, but humans really aren't nearly as complex as they like to think.

When confronted with a danger you have 2 "Tier One" options, fight or flight. Then if you choose to fight, you choose from a limited array of basic styles (full frontal all out dps, which is how they pretty much fight now. Or you try to stay outside your opponents dps range and kite them, or you... well you get the idea)

Now all these choices are subject to change by incoming information. If I attempt to kite a foe and find that I can't avoid enough dps, I will either require logi or a change of tactic. It is a terrible cop out to have AI's simple stand there firing away until their inevitable death.

Anyway in response to ur comment, I think it is possible to do this in a game, but as most devs know the game engine is one of the hardest parts to build in a game and a better AI might just be harder still. I say hard, but really it is just time consuming. It would no doubt take a gifted programmer working along side sociologist to accomplish this.

Thanks everyone for commenting. This is something I've long felt needed attention in games. I'm not going to go into my back story here, but I have exceptional experience in human behavior and study there of. No I'm not a scientist at least not one that went to school. But in just about any situation I can predict with a startling accuracy what ppl will do.

In a fight it all depends on the type of person whether they are aggressive, passive, egocentric, etc. You can't really know these things without also knowing the person, but you can make a fair estimation by how they look and behave. I'm not talking about making an AI that thinks. I'm talking about an AI that is programmed to read a log and has subroutines that sort that log's information into key area's in the overall code for the AI.

Let's break this down. In any given fight in eve there is an aggressor and the aggressed . This can consist of one ship being attacked by another ship, or a fleet being attacked by 2 or more fleets. The beginning of creating a better AI through my suggested method would begin in quantifying variables; finding what is necessary and/or conceivable to consider in the greater equation. I ofc don't think rats should be able to call in alliance support and wage all out war on every player upon being aggressed.

So we break down situations into categories ( 1on1,1on2,etc)(2on1,3on1,etc) Then once we do that we can order information from such battles to be logged under those categories. How does a pilot react when confronted with 1 other adversary. Yes I know there is much more to it than that. There are ship types and attack styles that will alter the decisions of the player being attacked and so on. But before any AI can be created using this method I purpose someone is going to have to figure out first how to collect the data. I think this is 75% of the work really.

Anyway I could go on about this forever, but I have dogs to walk atm. =)

The incursion rats are super buffed. They aren't more "intelligent".

No, they just are biased to the smallest targets. Kind of hard to even say they have any AI at all really. Honestly I don't think the term AI should even be in use in our language for games. Far as I can tell these rats and badguys in any game just have a set of animated walk/fly/ w/e cycles broken by a crude script of "fire until you die" tactics that are provoked upon tripping a trigger.

There must be more reason for why AI's are so primitive. More than just that no one has figured out how to make them more life like. Maybe it's as simple as Coders not being very understanding of natural behavior. I know very little about programming but what little I know tells me that there are a thousand ways to make an AI better. Maybe it is a problem with data flow. The only thing that limits gfx in a game are the players ability to render them at a passable fps. So maybe this is also the issue with AI's. Maybe to gather the amount of information and then collate it into the proper subroutines is too process heavy.

I'm thinking very srsly about learning some python and using it to effect a better AI in blender game engine. I already am familare with blender so might be the best way to learn it.

Erm, it's not smallest target at all. DPS, ewar and remote reps will all get you targeted by them too. I haven't run the things since before the nerf though. Anyone who runs them nowadays, or any wormhole guys who know about sleepers care to chip in?

Some how it seems that the whole second page of this thread has vanished. So intsted of just letting it die, I'll reiterate that I don't want harder rats. CCP could do this just by buffing them. I want to have to think when I pve. I want a rat that is not predictable enough to afk mission.

Sure at first the rats would seem random and just for the sake of being random, but as ppl play they will begin to see that there is order. They'd be much harder at first, but ppl would learn to fight them as they would a person capable of thinking and changing tactics. You could not kite a human and go watch tv while they die. If you can do this to a rat, the ai if broken. I'm not asking for a full on thinking ai. I don't really even consider what I'm talking about as ai at all. I just think rats should be able to cycle attack protocol depending on what player tactics they face. Scale-ability. If you bring a fleet of a carrier and 2 mach and the rats only have a few bs, they should run making the mission uncompleteable, OR call in reinforcements.

Buffing rats to make them take longer to kill is a total cop out; a way to side step actual work in making them better.