open All Channels
seplocked EVE General Discussion
blankseplocked GIVE US NPC AI
 
This thread is older than 90 days and has been locked due to inactivity.


 
Pages: first : previous : ... 5 6 7 8 [9] 10 11 12 13 ... : last (21)

Author Topic

Thebro Nobrunder
Schrodinger's Renegades
Posted - 2008.11.14 18:46:00 - [241]
 

To cut down on the computations you could always organize npc's into squads.
There would be one AI "agent" who commands the actions of all the others in his squad.

Another idea might be to allow NPC's to call for reinforcements. The reinforcements would be tailored to the pc's currently attacking the npc. For instance if the players were flying all amarr lasers ships the reinforcements should have em/ther resists and be fitting tracking disruptors.


Seth Ruin
Minmatar
Ominous Corp
Circle-Of-Two
Posted - 2008.11.14 18:56:00 - [242]
 

Originally by: Thebro Nobrunder
To cut down on the computations you could always organize npc's into squads.
There would be one AI "agent" who commands the actions of all the others in his squad.

Another idea might be to allow NPC's to call for reinforcements. The reinforcements would be tailored to the pc's currently attacking the npc. For instance if the players were flying all amarr lasers ships the reinforcements should have em/ther resists and be fitting tracking disruptors.


Both great ideas. Adding to them...

The AI lead "agent" -- What if this were a pilot with (at least minimal) leadership skills (and therefore bonuses)? It would give more meaning to higher "ranked" NPC ships in missions, and would give the player(s) a good reason to primary them.

As far as the NPC reinforcements being tailored to the PCs, I think it might call for a bit more data mining than is feasible. However, reinforcements should always be a possibility. Especially belt rats within the same system!

Neth'Rae
Gallente
State Protectorate
Posted - 2008.11.14 19:00:00 - [243]
 

Edited by: Neth''Rae on 14/11/2008 19:03:52
Edited by: Neth''Rae on 14/11/2008 19:02:27
Edited by: Neth''Rae on 14/11/2008 19:01:12
Originally by: Thebro Nobrunder
To cut down on the computations you could always organize npc's into squads.
There would be one AI "agent" who commands the actions of all the others in his squad.



NPC FCs :D
Nice idea :P
Edit: Ah, just read it again..
This would mean everyone would act the same in a squad though?
NPC FCs would be a cool idea though.

One thing I thought of was, waves that come and go, warp in and warp out..
Interceptors swooping in to web you down, and then bombers sweep in launch rockets(?) and then get out.
And then the interceptors would scatter and try to get away once you engage them.
But it wouldn't really add anything other than the coolness factor..

I try to think of things that require skill, timing, and target priority..
But there's not much to work with when you consider EVE's combat system and that you're mostly doing missions solo.
If there's target priority, people will just learn the order and rinse/repeat.
If there's timing involved people will just learn when to do what and rinse/repeat.

Originally by: Thebro Nobrunder
Another idea might be to allow NPC's to call for reinforcements. The reinforcements would be tailored to the pc's currently attacking the npc. For instance if the players were flying all amarr lasers ships the reinforcements should have em/ther resists and be fitting tracking disruptors.


Hm, what's the range of a shipscanner btw?
Maybe there could be close range scanning ships that you'd have to shoot before they give away your fittings ^^




Seth Ruin
Minmatar
Ominous Corp
Circle-Of-Two
Posted - 2008.11.14 19:05:00 - [244]
 

Originally by: Neth'Rae
I try to think of things that require skill, timing, and target priority..
But there's not much to work with when you consider EVE's combat system and that you're mostly doing missions solo.


Does it have to be that way? What if the AI were gradually increased from level 1 to level 5 missions in such a way that by level 3 you'd practically need some kind of buddy, and most definitely couldn't do 4s solo? (of course, risk/reward will have to be adjusted, as Incognito mentioned earlier... ignore risk/reward for the sake of argument, though)

Originally by: Neth'Rae
If there's target priority, people will just learn the order and rinse/repeat.
If there's timing involved people will just learn when to do what and rinse/repeat.



Eh, admittedly, that's pretty much how PvP is too. For the most part, if you have good intel and know what's coming at you, you know before they jump in (or before you jump in) what's going to be primary, approximately how far they're going to be from the gate, etc etc, unless they have a very creative FC.

Marleau LeRoi
Caldari
Praetorian Freespace
Posted - 2008.11.14 19:44:00 - [245]
 

From someone who has run his fair share of missions, I would certainly support AI in some form or another. From an EVE universe perspective, I've always found NPCs to be a bit of a buzz kill in the sense that in a given L4 mission, one might destroy what, 40-50 ships? If you figure each had a crew, and the sheer number of missions being done by all players, the human loss would be pretty extreme. So while I'm not really the RP type, in the back of my mind I'm always thinking, why would any sentient creature put themselves into a situation where they weren't doing everything possible to not die? Why would one sign up for a pirate corporation to be loaded into a BS fitted with a single cruise missile launcher and no armor to speak of. Wouldn't that person be worried about the fact that 1.5 million of the guys in front of him died a quick/pointless death in the last month?

But I digress: in the context of a game, none of this matters I suppose. But from an immersion perspective, it would make sense to mission against 4-5 ships fitted in such a way that the people you are fighting against wish to remain not dead vs. 40-50 ships lined up to die. So while this is not AI per se, it is computationally cheap and easy to implement (without taking into account economy effects, player balance issues, etc of course).

Now, tack on any semblance of actual AI (as in actually doing something with the fittings), then hey, I'm all for it. Smile

J'Mkarr Soban
Posted - 2008.11.14 21:24:00 - [246]
 

I'm working on a paper now for my PhD (in AI) where I'm trying to find some relations between games and the 'AI' that is used in those games. Probably everyone can recognise or has experienced this themselves: when you are playing a single player game, you aren't acting through the game as if you were involved in it. Yes we might make some overarching choices about what style of character we play, but at the end of the day the only thing we end up doing is working out what the AI is going to do so we can beat it. That's not a problem - that's how you solve problems, by analysing the problem space and determining the best (or near best) solution. The problem is that that problem space doesn't really vary from encounter to encounter - or if it does, it's because there are different 'types' of encounters. The 'intelligence' put in to these games is entirely deterministic - rule based. If x, then y. (Yes, that's a significant simplification, but just imagine x as a the set of all inputs and y as the output. If the two input sets are the same, you get the same behaviour.)

Now perhaps in a single player game, that isn't really a problem - I find I still get surprised by some actions that NPCs make (like one of the companion characters in Fallout 3 not using the big explody rocket launcher inside a small tunnel - that was a surprise) - however that's because I only have my own, and sometimes my friends', input into what the rules are for the NPCs. It's a different matter entirely when you have the collective intelligence of thousands of players all playing the same game.

This is where we bring in Eve. We have those thousands of players all playing the same game. And over the course of the years, a collective intelligence has emerged based on those experiences. I've been here from day one, and as far as I know I'm the oldest character on the server - so I've noticed the huge shifts that occur after each and ever patch: "ARGH! This has changed!" "Ok, let's try this" etc. (I mean this patch is a perfect example - look at what's happening with the missiles.) This has partly lead to how 'easy' missions are today - they are entirely deterministic, and the players have sussed out exactly how they operate under virtually all circumstances. That's why there's a website dedicated to mission running, how all of the missions operate, what rats are there, even links to the capabilities of the rats. That's why I know without even looking that if it's a Sansha mission I load up my Nightmare, shove on 2 EM, 1 Therm hardner and a tracking enhancer, and go and squash them. If it's a mission I've not looked at before, then I've got some forewarning of what to expect.

And of course, there is the obvious consequence of this: because the mission NPCs (as a subset of all NPCs) are deterministic, it is possible to farm them. I'm a 'career' mission runner, and I will categorically say that missions are boring and too easy. (The fun for me comes from trying to find the absolute optimal setup for my ship for certain situations, before anyone asks 'why')

I've argued that the reason missions are farmed is because they are too easy, and they are too easy because they are deterministic. So what's the solution?

J'Mkarr Soban
Posted - 2008.11.14 21:24:00 - [247]
 

Well, that's surprisingly easy to conclude, if not to implement - make them non-deterministic. Now some might say 'it's possible to give all NPCs rules for using all modules: how, when, in what combination - and then randomise their fit. The players will never know what's being used.' And this is true - for a while. The more observations are made (and given how many missions are run each day, it wouldn't take long time-wise) then we could work out the rules that the rats have for using their modules. Then it's a case of scouting out before hand to work out what the rats are fitting, come back and prepare your ship accordingly. In fact you might find that you need more than one ship to complete the mission. But at the end of the day it's still farmable, even if it takes a little more effort.

No, true non-determinism comes from a completely randomised source of input. The most random generator I know if is 'humans'. The easiest way to harness this randomness is by having the NPCs learn from the humans, and alter their tactics to suit. There are many ways of learning, and one of the next steps of my research is trying to find out how AI in games can improve based on the classification of the game, but at the end of the day it's the only thing that will make PvE, and thus a major part of the gameplay more interesting.

A number of caveats though: to make the learning AI a) beatable and b) viable, it would have to exist in the same environment as the players they are learning from. This means same ships, same availability of weapons, same everything. This has two good points: first, it means the NPCs will no longer cheat. 6 turrets and three defenders? Nine high slots? Gone. Secondly, it would actually make things easier for the devs - there's just one loot table, the rats could be subject to the same loot mechanisms that are in place for PvP.

Further, this can quite easily be scalable. Obviously although I'm working on this (specific) paper just now, it's not the underlying problem I'm trying to solve in my PhD as a whole, and a number of ideas I've come up with generally would be applicable here. Although I can't go into too much detail (for obvious reasons), imagine if it wasn't just local groups of NPCs that learned. What if the faction as a whole learned. What if it realise that gate camps were the best solution to protecting a strategic system? And that it would need to shift resources towards defending that goal? Whilst missions are a specific set that I don't think should have limits on it, I don't see why the wider pirate factions (or, indeed, any faction) can't have some form of limits on the amount of resources: ships, modules etc. These constraints alone will work in with the learning system, but it could mean that you might randomly face a roaming squadron of Sansha trying to lure an over-zealous squadron of Amarr Navy ships into a trap - all without the input of a player or a dev. There have been alot of calls about increasing the flavour of space, making it feel like there are things going on - this could be part of it. A dynamic universe.

I know that sounds far fetched, but that's where the power of learning could come in to play. And I've got to say, an MMO like Eve that has all the randomness and scenarios and situations for a learning system to learn from the humans is the ideal place to do it. Eve's already got a reputation as forward thinking cutting-edge developer due to their single-sharded, highly optimised cluster - why not add another notch in the belt?

Evelgrivion
Gunpoint Diplomacy
Posted - 2008.11.14 22:37:00 - [248]
 

Originally by: Thebro Nobrunder
To cut down on the computations you could always organize npc's into squads.
There would be one AI "agent" who commands the actions of all the others in his squad.

Another idea might be to allow NPC's to call for reinforcements. The reinforcements would be tailored to the pc's currently attacking the npc. For instance if the players were flying all amarr lasers ships the reinforcements should have em/ther resists and be fitting tracking disruptors.




Didn't I just say that? Laughing

Originally by: Evelgrivion
I was recently toying with an idea on how it could be more feasible for awareness and tactical maneuvers to take place, and came up with a "Spawn AI" (for lack of a better term).

The Spawn AI would be in charge of every ship within a group. It would know the vital stats of every vessel, and would make decisions based upon them, rather than having each vessel have it's own AI and it's own awareness routines.

I imagine this would make it more feasible to direct repairs, retreats that are tactically sound, and other maneuvers than an individual AI system would allow, with equivalent CPU time.

Neth'Rae
Gallente
State Protectorate
Posted - 2008.11.14 23:38:00 - [249]
 

Edited by: Neth''Rae on 15/11/2008 04:53:23
Edited by: Neth''Rae on 14/11/2008 23:39:21
Originally by: Seth Ruin
Originally by: Neth'Rae
If there's target priority, people will just learn the order and rinse/repeat.
If there's timing involved people will just learn when to do what and rinse/repeat.



Eh, admittedly, that's pretty much how PvP is too. For the most part, if you have good intel and know what's coming at you, you know before they jump in (or before you jump in) what's going to be primary, approximately how far they're going to be from the gate, etc etc, unless they have a very creative FC.


Yes, but that's exciting because you're fighting against other players, there's obviously the factor of human error aswell as lots of unpredictability...
Making each npc ship have a different role would make the learning curve for new players go backwards tho :P

Some kind of mechanic that would force you to make tactical decisions and plan a few steps ahead might be interesting though..

Myung Chul
Gallente
Garoun Investment Bank
Posted - 2008.11.15 00:24:00 - [250]
 

nonnononono... then they would get so smart they would lag out the server because they don't like dieing

Havok Pierce
Gallente
Aperture Harmonics
K162
Posted - 2008.11.15 00:45:00 - [251]
 

Originally by: John Holt
You can catch more flies with honey you know? If I worked for ccp I would says "---- you!" and ignore your post.



Obligatory XKCD:
http://xkcd.com/357/

CCP Incognito

Posted - 2008.11.18 14:17:00 - [252]
 

Originally by: Seth Ruin
Originally by: CCP Incognito
...


I'd be extremely pleased if we were given a dev blog on this topic once it's more "solidified" for player feedback. Ideally, something as important as the AI should be given the same amount of player input as the Orca had (which was, to my knowledge, the most successful player/dev coordinated effort I've seen in any MMOG)



I will keep it in mind, I am not a verbose writer, but I will see what I can do, no promises!

On the other hand there is allot of fun of just letting what we come up with lose in wild. then count the form threads as a judge of how good the AI became. Does more NPC griefed players == good AI Twisted Evil?

All I am saying is that is if we detail all the tricks that the AI can do then there is no challenge in figuring out how to counter it.

Already I see a couple of points in the plan that player may not like as they will potential make mission running more difficult, and hence slower to kill. This should make the encounters more interesting, but the people that track X mission per hour of grinding equals new ship. These people won't be able to do as many missions per hour, but the rewards would be balanced against the new time requirements.

<edit, had a couple paragraphs on a a new feature removed>

I had written a bit on a new feature to temp you all. But because everything has a jello feel for it's certainty I will hold off for a while.

Yea I think I will have to do a Dev blog at some point :(

People that know me know I HATE documentation with a passion.

CCP Incognito

Posted - 2008.11.18 14:18:00 - [253]
 

Originally by: Havok Pierce
Originally by: John Holt
You can catch more flies with honey you know? If I worked for ccp I would says "---- you!" and ignore your post.



Obligatory XKCD:
http://xkcd.com/357/


LOL! Laughing

Seth Ruin
Minmatar
Ominous Corp
Circle-Of-Two
Posted - 2008.11.18 14:45:00 - [254]
 

Originally by: CCP Incognito
On the other hand there is allot of fun of just letting what we come up with lose in wild. then count the form threads as a judge of how good the AI became. Does more NPC griefed players == good AI Twisted Evil?

All I am saying is that is if we detail all the tricks that the AI can do then there is no challenge in figuring out how to counter it.


I think some of us here would argue that good AI has no obvious counter Razz

But more realistically, I didn't necessarily mean I'd want a dev blog describing every little detail. Just something along the lines of, "This is the general approach we're taking, this is why we think it's a good idea, and this is when you can expect to begin testing it on Sisi, and this is our plan for future improvements on this approach" Smile

Batolemaeus
Caldari
Free-Space-Ranger
Morsus Mihi
Posted - 2008.11.18 15:10:00 - [255]
 

Originally by: CCP Incognito
Does more NPC griefed players == good AI Twisted Evil?


Yes.
Don't call it griefed..a good ai should not be distinguishable from real players. Or at least close to it.

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.11.18 17:08:00 - [256]
 

I'd almost expected this thread to be dead by now. I'm glad it isn't because it basically shows two sides of the same coin. It also shows that among the players of EVE there are those that really think about the game and what they would like to get out of it.

Allow me, if you will, to throw in some general observations. On the one hand you have several posters commenting on the things they want to see EVE NPCs do, either within missions or beyond. These are very interesting, as they provide use-cases that the AI has to live up to. Things like adaptability, non-deterministic behaviour patterns, ability to cooperate (or at least `show' cooperation), and the difference between NPC/AI and players, are very important measuring sticks to apply to AIs. None are easy to solve (I have no illusions), and to a large extend, none can be done fully with pure scripted behaviour patterns (the state space would quickly outpace available resources). But scripted behaviour can go a long way in make the distinction, lets say, blurry. Artificial stupidity has been mentioned, and this is what I would call it.

And this is where the other side of the discussion comes in, and those are the people who talk about mechanisms, ways to bring about more intelligent behaviour. The one is the result of the other afterall. And this is where most games, and game developers falter. There is an opportunity here to really break down barriers between theoretical, or academic work (of which there is plenty), and practical implementation.

There is, you see, actually quite a lot of work done on AI that can make it adapt to a changing environment, to have AI display non-deterministic yet realistic behaviour patterns, to encourage, stimulate, and have benefits from cooperation between AIs, and certainly take away a lot of the distinction away between players and NPCs. All these ideas are being worked on in fields like social simulations, computational intelligence, non-deterministic algorithms, planning, statistical intelligence, machine learning, artificial intelligence in general, and many others.

What is missing though is the leap of faith by the game developers to actually start using it. Sounds crazy, but it's true. Ohh, some of it seeps through in games like Galactic Civilisations (rule-based system), NERO (rtNEAT, evolutionary learning and neural networks), and others. But none break the barrier of what is possible, what is already out there, all basically walk the same downtrodden paths so many have walked before.

MMOs have a unique status in game development, and also have a unique opportunity. Whereas stand-alone games often freeze their AI upon shipping it, they do lack the feedback mechanism the MMOs provide, in updating and steering the AI mechanisms, towards an AI that is truly adaptable, self-learning, cooperative, and indistinguishable from players.

The point I'm trying to make here is that for an AI to do all that, game developers like CCP have to both embrace and let go at the same time. Embrace the academic world and the seemingly outlandish ideas that float around there. And let go when it comes to complete control over behaviour and to a certain extend content. They have to trust, or acquire the trust, to let the AI mechanisms do their job, and settle for a mechanism to see if it did so after the fact. Only by both embracing and letting go of AI will a standard be set. Anything less will not provide what the players are looking for, not when it comes to adaptability, not when it comes to `realistic' behaviour, and not when it comes to cooperation. The answer in AI to the question `does it work', is neither affirmative or negative, it is `it will' (if done correctly that is). If that's not good enough, forget about AI and hire some script-kiddies. I guess what I'm saying is that what is needed for AI in games, is someone bold and courageous enough to just go with it and then sit back and see where it leads you.

I'll get off my soap box now.

Seth Ruin
Minmatar
Ominous Corp
Circle-Of-Two
Posted - 2008.11.18 17:18:00 - [257]
 

Originally by: Bartholomeus Crane
I guess what I'm saying is that what is needed for AI in games, is someone bold and courageous enough to just go with it and then sit back and see where it leads you.


Absolutely. I think CCP has the capability, and on a good day, the drive. A lot of the old "reckless"/"experimental" attitude has faded in CCP due to their current size, but I have faith it's not gone entirely.

I think from a marketing standpoint, saying, "Yeah. We're the first (and only) MMO to have actual, learning, adapting AI," would be HUGE bragging rights, and certainly attract at least a good group of players who are interested to try it out. And hopefully the rest of the game can keep them interested Wink

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.11.18 17:25:00 - [258]
 

Edited by: Bartholomeus Crane on 18/11/2008 18:04:27
Originally by: CCP Incognito
All I am saying is that is if we detail all the tricks that the AI can do then there is no challenge in figuring out how to counter it.


Wouldn't it be neat though if the AI could learn some tricks that even you don't know about? That the AI would be able to recognise a certain behaviour as advantageous and find a suitable counter for it, without you having to put the trick into it?

Originally by: CCP Incognito
Already I see a couple of points in the plan that player may not like as they will potential make mission running more difficult, and hence slower to kill. This should make the encounters more interesting, but the people that track X mission per hour of grinding equals new ship. These people won't be able to do as many missions per hour, but the rewards would be balanced against the new time requirements.


For those players who equate missions with grinding for isk, any change to NPC behaviour will be bad, because, lets face if, the behaviour can't be worse. And, don't get me worse, balancing risk vs. behaviour will been needed with more suitable behaviour, I don't think CCP should halt NPC development based on those types of players. I expect there to be loads and loads of players out there who'd love more interesting NPCs and more interesting missions, and care less about their ISK making capabilities.

Originally by: CCP Incognito
<edit, had a couple paragraphs on a a new feature removed>

I had written a bit on a new feature to temp you all. But because everything has a jello feel for it's certainty I will hold off for a while.

Yea I think I will have to do a Dev blog at some point :(

People that know me know I HATE documentation with a passion.


I think it's sad that devs would be hesitant to put out some hypothetical idea to the forums to test the waters for fear of being spamwhined, or knucklewrapped by suits. Although I like dev blogs, they are a very stunted way of communicating with the player base. I suspect they go through a lot hands before they end up on the forum, and frankly, it shows. It looks all very controlled and mediated. I suppose it is a safe way of doing things, and perhaps the 'climate' dictates it, but I long for the days back when devs could throw out some rough ideas to the forums and discuss it without so many people trying to read things in it. I suspect some things simply can't be discussed in public, and we've seen some changes of late regarding this, but damn ...

Ohh, and bad coders write documentation, good coders write tests ...

Andrea Griffin
Posted - 2008.11.18 17:51:00 - [259]
 

I haven't read through the whole thread - there's quite a lot of text - doubtless I'm going to be duplicating some of the ideas here.
Originally by: CCP Incognito
What would make a 1v1 PvE encounter challenging and fun?
What would make a 1vN PvE encounter challenging and fun?
What would make a NvN PvE encounter challenging and fun?
Right now, I see predictability as one of the problems with PVE. I had made a short post a while back about a point-based spawn system:
Originally by: Andrea Griffin
One idea would be to set a bounty limit for a mission. Ships of the appropriate type could then be randomly selected until the bounty is reached. For example: Mission X has a total bounty of 100k. This could be a large load of frigates, or a cruiser with some drone support, etc, but the bounties all total around 100k.
Something like that would go a long way towards making PvE much more exciting. The player should have a general idea as to what to expect (which faction they are facing), but the specifics should be decided when the ships spawn. If you always know what is going to happen, the scenario is much less fun and exciting. If you never really know what to expect, the game becomes more interesting.

Also, overall, I find the concept of group aggression to be a bit unbelievable. Many missions have several pockets of rats, but only one of them will attack. The others watch as you slaughter their friends, until you shoot at the next group. "Oh, we're being shot at, I guess we should go do something..."

I'm also surprised at the lack of support that the enemy has. Adding some logistics - shield / armor repair - may make sense.

I would also like to see missions scale with the number of pilots. A pilot could accept a mission as a fleet and, based on the ships in the fleet, the enemy's strength could increase. Perhaps more ships, or they are tougher, or deal more damage. More risk - but also more reward for the fleet.

Also, why don't the rats ever switch aggression? If they're pounding on a ship and not making much progress, well, maybe that other guy in the fleet has a softer belly. This would inhibit the idea of one person tanking an entire spawn while the rest go full-gank in absolute safety, something I have found to be effective but ultimately a bit lame.

I would enjoy a few 1 on 1 missions on occasion. The other ship should be hard to take down, have some form of EWar, and have the ability to warp out and repair. These ships could show up on the directional scanner, giving it some use in PvE. It would be nice if Neuts/Nos worked on them as well, though I understand this would require keeping track of their capacitor server-side.

The OP had mentioned NPCs ransoming the player. Not a bad idea - though, it would require that the player be tackled (otherwise they'd just run away). I don't know about anyone else, but the warp disruption rats are always the first kills on my list!

What about mixed groups? Seems like every mission is one enemy type only. I'm sure that Amarr/Caldari groups, and Gallente/Minmatard mixed groups could add a bit more variety.

There should also be some sort of penalty for warping out. Perhaps the rats get reinforcements, or they run away with the mission item, or they pick through the wreckage of their friends resulting in less loot at the end.

It would also be fun to have some NPCs on my side once in a while. Perhaps I'm being sent to turn the tide of a battle already in progress. This gives multiple options: I could provide remote repair to friends, shoot the enemies, use ECMs or painters, etc.

Anyways, that's what I have for now. Sorry if I duplicated too many ideas, like I said, I didn't read the whole thread. Cool

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.11.18 18:03:00 - [260]
 

Originally by: Seth Ruin
Originally by: Bartholomeus Crane
I guess what I'm saying is that what is needed for AI in games, is someone bold and courageous enough to just go with it and then sit back and see where it leads you.


Absolutely. I think CCP has the capability, and on a good day, the drive. A lot of the old "reckless"/"experimental" attitude has faded in CCP due to their current size, but I have faith it's not gone entirely.


Lets hope so. I guess size does stifle reckless attitudes, although I've worked with guys from IBM that have and are doing some pretty reckless stuff. So you never know. Personally though, I've send Gabe an email some five days ago, and haven't heard back since. Not quite sure what to make of that.

Originally by: Seth Ruin
I think from a marketing standpoint, saying, "Yeah. We're the first (and only) MMO to have actual, learning, adapting AI," would be HUGE bragging rights, and certainly attract at least a good group of players who are interested to try it out. And hopefully the rest of the game can keep them interested Wink


This is something I keep bringing up with game developers when I talk to them. And although they certainly see the possibilities of that, you'd be surprised to find how utterly conservative game developers are overall, especially the ones with an established game out there. Most of the time it's all bragging about 'cutting edge' and all that, but when it comes to AI, it's mostly: 'don't rock the boat'. It staggers the mind sometimes.

AI is usually something game developers don't know much about. Graphics, scripts, story-telling, all that? Yes. But AI? Not really. And it's easy to go overboard as well. As an academic it's sometimes hard no to go off about all types of rather, well, academic subjects when talking about AI, while the guys sitting in (and usually not allowed to say much) actually developing the game are interested in knowing only one thing: 'does it actually work?', or, more often than not: 'can I actually do this?'. And for the former I have only academic proof to give (almost no one has ever done it within the game context), and for the later the answer is usually either 'no', or 'if you let me show you how' (and this to a guy who's paid to be the expert?). So, it's a tough sell, certainly on technical grounds. Unfortunately, with game developers, champions for AI are rare to find! Not to mention the 'not done here/by me' syndrome! You know, the 'what does this academic know about games' complex.

The marketing guys are even harder to please. Will AI sell more games? Is it cost effective to put person months into AI over, say, graphics? Will people play the game longer if we have better AI? I'd say yes, but I have no marketing research to back all this up.

The good news is that players are waking up to the possibilities. Bad AI nowadays, sometimes at least, means getting a poor(er) review. MMOs depend on subscription, keeping players attention. With poor AI, this maybe a problem when it comes to PvE. Graphics are nice, and there's still a lot to do in that arena certainly, but players have more to choose from as well, and you simply can't get away with poor AI like you used to. Even the most glitsy graphics in a game don't make it a success if the NPCs are ******ed, and players pick up on that. The link poor AI to poor design is laid more often now than it was as well.

So there is hope for AI in games, obviously. But what is needed most is for one game developer, even one game, to lay out a stake that's so far ahead of the competition that anything less simply won't do anymore. EVE has the potential to be that game ... it just needs to want to ...

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.11.18 18:30:00 - [261]
 

I wanted to say a bit more about cooperation, mostly because it's a huge subject in AI at the moment, and also because I kinda waved it away as doable earlier.

It is, doable that is, but the question is how to do it. Until now there has been a lot of discussion about how individual NPCs should behave. But what makes them a group? And how should they behave as a group?

Clearly, the aggro mechanism in use now is not it. It's a mechanism for sure, but not very impressive. It does the job, but barely. You could be slaughtering one group, while just a couple of k further ahead, the NPCs are happily circling a rock.

What you want is for the AI to show behaviour that adapts to the circumstances, and the individual AIs behaving in a rational way within that environment. This is the use-case.

Stripped from all other embellishments, what you need is a classifier mechanism. That mechanism can be NPC based (so taking out the leader who holds the mechanism has an effect). But what it should do is recognise a situation, and order individual NPCs around to react on that situation.

Given a suitable context description, you could script this. Unfortunately, that would lead to cut-and-paste behaviour, and the scripted classification could not cover all state-spaces. Also, it would be rather boring, not only to develop, but to play against as well.

However, you could also use a learning mechanism to use that same context description and develop a 'plan', a reaction or behavioural pattern over time. This learning classifier system would be able to cover all eventualities, although in some circumstances it would make mistakes (just like humans do). Again, the offline-online situation would work, in order to overcome resource constraints, but the system would be a nice way to overcome the high dimensionality of the possible state space.

The neat thing is that this could be applied to various levels. You could have AI gang FCs, an AI fleet FC on top of that, and even a system, constellation, and/or region FC, all working on the same principle (although with different contexts obviously).

Sounds crazy? Not really, these types of hierarchical systems are already applied in other fields, like biogenetics and cybernetics. Really very interesting what's possible with them.

Doctor Penguin
Amarr
Sacred Templars
Black Star Alliance
Posted - 2008.11.18 18:42:00 - [262]
 

Combat should be cooler. NPC AI would do that in part, but cool special effects and cool surroundings, and cool mission objectives, and cool music to listen to...

You get the idea. It's gotta be interesting.

Hyperforce99
Gallente
The Scope
Posted - 2008.11.18 22:30:00 - [263]
 

apart from the allready mentioned AI improvements I think some form of "calling in the big guns tactics would be nice"

why should rats only use basic tactics. I mean, I'd like to see rats use some more interesting tactics, potentially make them modular so they won't play the same senario all the time, but take this for example:

NPC in deep 0.0 warps into a belt were lots of people are mining. This NPC could be a scout.

Depending on the defencive mussle in the mining group this scout can then decides to turn tail and get backup.

Potentially add staged combat were rats try to warp away if they are taking to much damage.

If the mining ops is big enough and/or the defence is strong enough this might trigger the rats to bring in the big guns.

I'd love to see rats drop a capital on you on rare occasions. Or a strike force team were the rats are using special EW ships in combination with other ship types that make sense.

Basicly give us some rats that do things players do, still slightly predictable on the combat side but making combat more interesting by adding in a game mechanic that will cause spawns to do unpredictable and potentially risky things but that can be very rewarding.

Add to that the system wide belt expansion were deeper into these belts one might find more interesting threats... I'd say this could be very interesting.

RedSplat
Posted - 2008.11.18 22:42:00 - [264]
 

Originally by: Hyperforce99
apart from the allready mentioned AI improvements I think some form of "calling in the big guns tactics would be nice"

why should rats only use basic tactics. I mean, I'd like to see rats use some more interesting tactics, potentially make them modular so they won't play the same senario all the time, but take this for example:

NPC in deep 0.0 warps into a belt were lots of people are mining. This NPC could be a scout.

Depending on the defencive mussle in the mining group this scout can then decides to turn tail and get backup.

Potentially add staged combat were rats try to warp away if they are taking to much damage.

If the mining ops is big enough and/or the defence is strong enough this might trigger the rats to bring in the big guns.

I'd love to see rats drop a capital on you on rare occasions. Or a strike force team were the rats are using special EW ships in combination with other ship types that make sense.

Basicly give us some rats that do things players do, still slightly predictable on the combat side but making combat more interesting by adding in a game mechanic that will cause spawns to do unpredictable and potentially risky things but that can be very rewarding.

Add to that the system wide belt expansion were deeper into these belts one might find more interesting threats... I'd say this could be very interesting.


I've heard this before, i still love the sound of it.

Imagine NPC pirate factions busting a lowsec gatecamp Laughing

J'Mkarr Soban
Posted - 2008.11.18 23:17:00 - [265]
 

One other thing I'd like to add, actually: you can sell truly learning adaptive AI to the economists, too.

Given they have no intentions of cutting Eve off any time soon, and they intend on keeping adding things to it as they go along, then surely it would be beneficial from a financial and manpower viewpoint to make a learning AI? First, it means you do it now, you never have to change it again, even though it's a big undertaking. Secondly, it doesn't need to be tweaked with each and every new module or ship that comes on the market - the system will learn to cope with anything, or at least should do if it's built correctly.

Yes, granted, it would require a fairly large undertaking both in terms of manpower and finances, but the cost-benefit gain over the period of time they intend Eve to go on for may just be worth it - but the longer they spend thinking about it, the less gain there will be until there is a net loss, and then we'll never see it.

Tarron Sarek
Gallente
Biotronics Inc.
Initiative Mercenaries
Posted - 2008.11.19 01:49:00 - [266]
 

Edited by: Tarron Sarek on 19/11/2008 01:50:58

1. Unpredictability

2. Risk.


Especially when it comes to ratting, yes.

Things like random and more flexible spawn composition (maybe slowly adjusting to the current ratting player's ship type), sudden reinforcements (with some time to react, though) and NPC's that run off, more than atm (to come back with reinforcements?) come to mind as some ways to make ratting more exciting.
Maybe introduce some scrambling cruisers?

With regard to missions it's hard to get the balance right.
If you balance it with faction fitted CNR's in mind, missions might get too hard for many players, or too limited with regard to the amount of viable choices of ship type and fitting.

Unpredictability is probably also the key here. And brains.
L1-L3 is not the big issue, although L3 should already be a bit challenging.
L4 Missions should be manageable if done right and with some thinking, but they should be deadly for a macro.
For example 'aggro everything and then just wait or go do something else' shouldn't work in most cases.
A mission should be done in a ceretain way so that you don't get webbed, scrambled and neuted/EW'ed all at the same time, but instead separately or not at all.
But that right way shouldn't be predictable, i.e. random.
So how about some distinct clues/hints the mission runner gets while in the mission?

Short list of ideas:
- Slight alterations of the mission area each time a mission is generated.
- Slight alterations of enemy groups and composition each time a mission is generated.
- Clues about what to attack first, what not to attack, etc.
- Generally more balanced missions with regard to rewards, to reduce cherry-picking
- Generally slightly more powerful but fewer enemies (especially destroyers and battlecruisers!)
- In L3 the need to warp out or turn down missions if doing them the stupid way
- In L4 the high possibility to lose the ship (even officer fitted CNR's) if doing missions the stupid way (ignoring hints)

Sylper Illysten
Caldari
Ex Coelis
The Bantam Menace
Posted - 2008.11.19 02:33:00 - [267]
 

Edited by: Sylper Illysten on 19/11/2008 02:34:42
How about NPC's that use the same equipment as the players of EVE? NPC's seem uneffected by the missile nerfs, torpedo range nerfs, speed nerfs. Their ECM is far better than anything available to players, they are immune to ECM. If you want to make PVE fun and challenging firstly you have to make the playing field level. Fun does not equal I just spent an hour killing one NPC and have 50 hull left and it cost me 10 times as much in ammo and repairs as the mission reward + bounty, some of us are casual players if CCP doesn't want our money, just tell us so we can quit and you can be happy. If CCP does want casual players, missions need to be able to be completed in a reasonable time frame and you have to be able to make some profit (and don't mention grouping when EVE is promoted as a cold harsh place and most random group memebrs are merely out to kill/rob you).

Adeena Torcfist
Caldari
Dark Underground Forces
Posted - 2008.11.19 04:23:00 - [268]
 

Originally by: Sylper Illysten
they are immune to ECM......


wrong, NPC's Can be jammed, so they are not immune, clearlyRolling Eyes

id actually prefer to see much fewer NPC's, however, fielding 5 drones, & having a normal RoF as an average player would have would suffice. Increase the bounties and much better loot drops.

Lesser rats, with greater bounties = ( your so called ) "Need for Speed Initiative"

Problems ive noticed whilst missioning.

Guristas - Tone down the jamming. Having nullifiers & eliminators jamming u thro cycles 1 after the other gets tedious at times....
Blood Raiders - LOL? the damage is that poor, theyve been dead by the time theyve even gotten within there ranges...
Mercs - Great hitting ships, i couldnt see me sustaining the damage they deal out for any length of time (damsel in distress, right hand fo zazzmatazz etc ) so spawns are good, but since when do projectile ships drop lasers? Laughing drops need looking at
Sansha - the only real Sansha mission i enjoy it pirate Invasion. why is that? please look at the other sansha missions, & u will see why....

Sansha Extravaganza would be a nice start ;)

but seriously, good AI is always welcomed. id think of it as a low sec ganking improvement, gank squad warps in, ships can & should target the gank squad too & shoot them. Hell, if The Assualt was anything to go by, the Pirates would be instatly jammed when they warp in, so u could warp out Laughing

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2008.11.19 11:06:00 - [269]
 

Originally by: J'Mkarr Soban
One other thing I'd like to add, actually: you can sell truly learning adaptive AI to the economists, too.

Given they have no intentions of cutting Eve off any time soon, and they intend on keeping adding things to it as they go along, then surely it would be beneficial from a financial and manpower viewpoint to make a learning AI? First, it means you do it now, you never have to change it again, even though it's a big undertaking. Secondly, it doesn't need to be tweaked with each and every new module or ship that comes on the market - the system will learn to cope with anything, or at least should do if it's built correctly.

Yes, granted, it would require a fairly large undertaking both in terms of manpower and finances, but the cost-benefit gain over the period of time they intend Eve to go on for may just be worth it - but the longer they spend thinking about it, the less gain there will be until there is a net loss, and then we'll never see it.



Although this is certainly true, you're still asking them to make an investment now with return that's at best uncertain. There's always interest to pay for that kind of investment and more often than not, internally at least, there's always the question: "can we do it for less". And if you ask developers if they can do things more often than not, the answer is yes.

Also, it appears that CCP is in the grip of agile programming now. SCRUM, sprints, and all that. Well, agile programming teaches not to invest in advance in nicely worked out setups but instead rely on refactoring only when the need arises. Never mind the fact that this is not always appropriate or profitable and has been shown not to work in certain cases.

All in all, I find that it still remains a difficult sell, more so when you're not even given the chance to make a pitch :).

Huurtney Gurdsen
Posted - 2008.11.19 11:29:00 - [270]
 

I was shooting rats in Pirate invasion yesterday and I got down to one last pathetic Thorax. I hadn't scarmbled him or anything and he just orbited me whilst I blew him to pieces. He could have warped off and it struck me that this is rather unrealistic.

I vote for better AI please.


Pages: first : previous : ... 5 6 7 8 [9] 10 11 12 13 ... : last (21)

This thread is older than 90 days and has been locked due to inactivity.


 


The new forums are live

Please adjust your bookmarks to https://forums.eveonline.com

These forums are archived and read-only