- This topic has 19 replies, 7 voices, and was last updated 19 years, 7 months ago by
.
-
Posts
-
-
Here’s a pot of gold at the end of the rainbow situation (or some other metaphor for things that are hard to find) –
finding a college with an A.I. research department wherein at least one member is even slightly interested in, or aware of, games development and the woeful state of A.I. methodology as practiced in the industry.What I’m wondering is if anybody knows, on a global scale, where they combine these interests, of games development and A.I.?
I’m running out’ve new department websites from which to raid papers – and that ain’t too cool.On that note, thank god for Dr. Charles –
http://www.gamedevelopers.ie/news/index.php?article_number=194
:Dp.s. belated congrats to aphra on the 2nd bday bash…
-
I think Abertay have a sister course in Canada somewhere and they have AI on the course I think. Don’t know what the uni is, but you might possibly be able to google it *shrug*
-
the woeful state of A.I. methodology as practiced in the industry[/quote:3356a52034]
really?
i thought that gamedev ai programmers were pretty advanced as ai programming goes – now thats a vast generalisation, but come on, just cause some games ai (allright, a lot of games) doesn’t perform adequatly don’t tar them all with the same brush.and dont forget what a difficult job it is to get a bot/enemy ai of some sort, behaving in an intelligent or lifelike way.
each game is different and has requirements for ‘intelligent’ play.in most cases, i’d imagine the algorithms used are fairly cutting edge.
[note] craig is not an ai programmer :P
-
most of albertas stuff specialise in classical games(checkers, go) although they have produced some new path finding algorithms and developed a prog script ease which makes the scripting of biowares never winter nights easier.
also alberta have strong links with ea and helped them on several fifa games. improving on corner kicks specifically.
heres the linkhttp://www.cs.ualberta.ca/~games/
as for most game dev ai programming being state of the art, that i dont think is accurate. game dev have a different approach to ai as opposed to academics.
game dev ai = make a character whos interesting and exercises the players ability. they cheat cheat cheat.. use complete knowledge, see through walls, super accuracy in shots etc.
academic ai= build an ai which solves the problem, where you have incomplete data.
michigan and usc have heavy research in this area also.
using SOAR (a rule based system)
http://ai.eecs.umich.edu/people/laird/gamesresearch.htmlalso this girl has some good work done in this area (shes contributed to ai game programming wisdom ?)
http://www.itee.uq.edu.au/~penny/index.htm
actually ill be publishing my thesis on this stuff probably end of next week. so ill post a link, ive got alot of stuff in their about this difference and the various ai and how they can be used in games, and why some cant. if ye’re interested in a read.
-
i stand corrected. :)
okay ok so they cheat, but surely some things do need to work well, for a strategy rts type game you’d need a smart ai, surely they wouldn’t just use a massive state based one based on stats that it knows about the player(through cheating).
and what about pathfinding type stuff? smarter = more efficient = less cpu cycles etc.
-
i stand corrected. :)
okay ok so they cheat, but surely some things do need to work well, for a strategy rts type game you’d need a smart ai, surely they wouldn’t just use a massive state based one based on stats that it knows about the player(through cheating).
and what about pathfinding type stuff? smarter = more efficient = less cpu cycles etc. [/quote:069bec89ce]
well i had this talk with graeme devine at awakenings and yeah pretty much states are good, because easy to code and learning curve easy. pretty powerful.
rts can store look ups which have your last few moves and can predict off that what you might do. also you should see ai programming wisdom. great article about how simple deduction can help enormously in rts.i.e. player last time they attacked had iron weapons, so might they can deduce the players in iron age and also all the steps which the player took to reach this stage. possibly developed workshops, large amount of people chopping wood for the fires. so good action might be to pan out and find these people who probably are unprotected. probably not a very clear idea, but you see what im getting at. smart behaviour with pretty much no traditional ai, just look ups and simple if-then-else.
again on pathfinding, true you always want more cpu cycles freed up, personally i think the opt a* algorithm is pretty tight at the moment. not sure theres much else to optimise there but maybe?
i do think reinforced learning is going to be used alot more in the future in games, because it offers the ability to learn from reward\penalty scenarios.
anyway you should take a look at those sites. very good.
i presume you know about -
-
-
For example, one of the most prominent to have recently started on game AI research is David Aha [/quote:6ace171478]
ya this guy has done some really great work. i once heard from my supervisor that apparently you could ask him about any ai paper and he probably has read it. if you put him into google and look at his website, he’s documented most of them. reads like citeseeker ;)
Also aha is working on a system, which he presented at the aaai 05 workshop. Aha and his collegues were working on a pluggable system design which would make accessing games and extending them easier, so as to change\adapt the ai in them, the system was called tielt (Testbed for integrating and evaluating learning technqiues), i heard at CIG05 last month, that there was alot of people after joining this work. -
actually ill be publishing my thesis on this stuff probably end of next week. so ill post a link, ive got alot of stuff in their about this difference and the various ai and how they can be used in games, and why some cant. if ye’re interested in a read. [/quote:b405f0fe64]
hey I’d be very interested – I’m guessing wildly here, but I was talking/mailing to Barry O’Sullivan of UCC about supervising a proposal of mine, and he mentioned he had another student about to publish in the game AI area: don’t suppose thats you?
IMO, there’s gonna be a major paradigm shift in game production core values toward A.I., simply because of the cost of graphics. I’m sure that the per model/environment costs will drop with clever methods like the Bi-pedal engine at IC CAVE, but even so, all the luminaries are saying (at the GDC) that the trend for giant teams is only going to increase at the next hardware gen.
Maybe theres a niche for games which have 70%+ emphasis on A.I., that can be produced with a reasonable budget and use intelligent game gesign to cover for low quality graphics – Im thinking Kurushi, can’t think of anything more recent. They could co-exist quite peacefully with high-end graphical games like your traditional FPS – similarly to arthouse vs blockbuster cinema?
If so, then there is obviously a giant leap forward necessary in A.I. – its been educating to hear from people actually trying to make that leap, since I’m going to try signing on that payroll next Autumn :D
-
hey I’d be very interested – I’m guessing wildly here, but I was talking/mailing to Barry O’Sullivan of UCC about supervising a proposal of mine, and he mentioned he had another student about to publish in the game AI area: don’t suppose thats you?
[/quote:18c2b58750]
yes indeed that would be me. ;) Final draft went to barry last wednesday, so he’s just to clear it and then it goes to my extern in france i think then in front of the board june 15th and my second reader Dave murphy (one of the course organisers for the games msc starting in 2006 in u.c.c).
ya barry o’ is a good supervisor, pretty much you have full reign over your work. he’ll give you support when you need it, but after all its your thesis. Also i guess he probably has the most experience now in ucc at supervising a games msc. so he’d be the right person to supervise yours.
-
Maybe theres a niche for games which have 70%+ emphasis on A.I., that can be produced with a reasonable budget and use intelligent game gesign to cover for low quality graphics[/quote:47c24ec327]
The problem seems to be that marketers cant sell AI anywhere near as well as a more up to date graphics engine with the latest features. AI will therefore always take a back seat to graphics and the latest buzz feature.
As I’ve reiterated so many times, the concept of a ‘game world’ never seems to filter through to games – every AI character exists only when the player is looking at them. A huge part of developing ‘intelligent’ AI will involve not just acting when the player is interacting with them, but also being able to take logical steps when the player is not, so as to create the illusion of an intelligent, living, breathing (game) world. As an example of this, how many games have people played where you enter a room to have a bad guy facing you with a gun out. And how many games have people played where you go into a room and the bad guy is getting a cup of tea, having left his rifle on the table?
On the issue of player modelling, I’ve read a number of interesting papers on the subject in poker, which seems to be a good starting point as there is a finite (and pre-calculable) set of odds given any set game state. Thus there is a ‘pure’ variable of player actions which can be analysed and a model of different player types drawn up. However, one thing even the most advanced models fail to achieve is an ‘instant’ recognition of style – humans have an ability to read a player’s style within a few short hands, whereas a computer still needs to build a more significant database of hands before it can act in the same way. This problem probably wont be solved until we have a level of AI that can understand variable human appearances, body language and a host of other subtle issues.
To date, I’ve always been far more interested in online/multiplayer games quite simply because the level of AI has been so pathetic in so many games. I really do hope the level of AI improves, but in order for this to happen, I think there does need to be a shift in the perception of where the CPU cycles in a game should be spent, as well as an effort made to change the perception of the consuming public as to what really makes a game enjoyable to play.
Dave
-
Originally posted by Skyclad
I think there does need to be a shift in the perception of where the CPU cycles in a game should be spent, as well as an effort made to change the perception of the consuming public as to what really makes a game enjoyable to play.
[/quote:64c74a7572]i agree totally although i think nowadays the excuse of no cpu cycles for ai is pretty lame, when you consider the power of processors in the ghz.
ai vs graphics\sound. there the first things which reviews and the consumer see when they first look at the game. to get a fell for the ai you really need to sit down and play for a good length of time.
as regards “And how many games have people played where you go into a room and the bad guy is getting a cup of tea, having left his rifle on the table”. i will say i saw something close to this in NOLF2:Spy in H.A.R.M.S way, but your right the ai should be running when the player hasent seen a character etc, perfect example of this flaw is seen in old fps like doom\rise of the triads\wolfenstein.
player walks into a large room, other character is down the end of the room, and doesnt get activated until the player gets within a certain distance. So the player can rpg its ass as it stands still.
-
The problem seems to be that marketers cant sell AI anywhere near as well as a more up to date graphics engine with the latest features. AI will therefore always take a back seat to graphics and the latest buzz feature.
[/quote:76b2665ca1]This is a poor excuse it seems to me. Anything can be sold if you know how, the question is one of market forces moving resources into areas other than graphics so that there is something more substantial to sell. More as in more than just a slight improvement in enemy squad A.I. in an FPS, so that the box can say “enemy soldiers work together to flush you out…” etc. How many times have we heard that before?
What A.I. should be bringing in the near future is a cohesive, modular, plug and play engine which can be billed as a one stop solution to developers, a human-like opponent to gamers, and be packaged as prettily as Havok – and thats really all you need. What is behind the packaging is the hard part.
Unfortunately, from all reports the emphasis on graphics is only going to grow with the nextgen consoles, with HDR and shaders and all that jazz…some industry insider said that they’re looking at a tenfold art resource increase. They’re gonna be hiring cinematographers out of hollywood to make games.
Its in times of adversity that innovation increases though! One of the bright spots, as I see it, is that in an environment of increased connectivity, player modelling in any given online game could provide a massive global library of training sets for an A.I. learning algorithm that could underpin an NPC paradigm for that game. So with these models as heuristics, you could play this game against NPC’s and it would almost as though you were playing a person – a person who is, in the context of the game anyway, a polymorph of everyone who had ever played it.
Could be cool!
-
What A.I. should be bringing in the near future is a cohesive, modular, plug and play engine which can be billed as a one stop solution to developers, a human-like opponent to gamers, and be packaged as prettily as Havok – and thats really all you need. What is behind the packaging is the hard part.
[/quote:8ec97e86ea]
Unfortunately, i think that a one stop solution for ai is a long way off. Havok works because they fully know how to simulate physics because of hundreds of years of physic research\experiments etc.
ai unfortunately is only around 50-60yrs and we dont know all the rules or enough of them to make a decent ai module which emulates humans(nor do games want them i reckon). Often to create the effects you mentioned (flushing out plans), these dont require real ai. Often diversity in pathfinding can achieve this goal, take a look at some papers on halo, solder of fortune.
I know its possible to make npc characters building slightly more automated(in the interest of self promotion, :) see my thesis when its done). Although the basic behaviours need to adapted for the various game genres.
personally i think the most promising ai ive seen is intelligent environment. like in solder of fortune, nolf2 and this is something which i think could be modularised.
as regards ai and advertising, this is a tricky one for pr because you cant show intelligence(in 2 seconds, unlike graphics). although i guess thats a problem for pr people ;)
-
Havok works because they fully know how to simulate physics because of hundreds of years of physic research\experiments etc.
ai unfortunately is only around 50-60yrs [/quote:1d5fb720e5]This is is a salient point, but then there’s a problem with linear estimations of human endeavour – its not a linear progression. Whatever about all of human endeavour, in technology its fairly clear that the curve is exponential, and in ideal conditions the rate of increase of the rate of increase is also exponential (I’m getting this from Kurzweil, not my own brain, unfortunately. While we’re on the subject of stats, it seems that its as likely for one to have been born today, as at any other time in the history of the human race, given population ratios. Ergo, the human race is today technically capable of as much achievement as has occured in its entire history. I am aware of the flaws in these points, but even when you strip them down with realism, there’s a powerful message behind it – the future will be here before we expect. Maybe I should have started a new thread for that *shrug* ).
Of course, if game A.I. is in its infancy, all the better for us. What better place to work than a field that isn’t over-subscribed?
On a point by point basis, cos Im stuck for time, I just wanted to note:
– flushing out plans are what I was using to describe lazy, unimaginative selling points for AI in games. It would be the least part of a cohesive AI engine, and yet since its a pretty big achievement for most games today, I do realise how far there is to go. Scary far.– send me a link to your thesis when its done. :) by the by, masters or PhD? and how long did it take you? (its personal curiosity, in case I do end up down south)…
– intelligent environments – why did nobody think of that before? ;) I dunno, but Ive always thought of the environment AI as almost a first stop in any openworld game when I try to come up with proto solutions and designs.
-
This is is a salient point, but then there’s a problem with linear estimations of human endeavour – its not a linear progression. Whatever about all of human endeavour, in technology its fairly clear that the curve is exponential, and in ideal conditions the rate of increase of the rate of increase is also exponential (I’m getting this from Kurzweil, not my own brain, unfortunately. While we’re on the subject of stats, it seems that its as likely for one to have been born today, as at any other time in the history of the human race, given population ratios. Ergo, the human race is today technically capable of as much achievement as has occured in its entire history. I am aware of the flaws in these points, but even when you strip them down with realism, there’s a powerful message behind it – the future will be here before we expect. Maybe I should have started a new thread for that *shrug* ).
[/quote:5f2c414da0]
sorry, you lost me there.. maybe its end of the day syndrome. ;)what i meant was, physics had centuries of investigation and hypothesis, (re-written over and over again). computer ai probably only really started when the computer arrived (sure people thought in 18th-19th century during industrial revolution, wouldnt it be grand to have a machine which was like us).
Although i agree its a great time to be working in the field, very new, easy to find unsolvable problems.– send me a link to your thesis when its done. :) by the by, masters or PhD? and how long did it take you? (its personal curiosity, in case I do end up down south)…[/quote:5f2c414da0]
its a msc, about 16 months approx.(give or take a month or two for writing and reading, at end and beginning)
intelligent environments – why did nobody think of that before?
i think they probably did, only now its making it into games. it makes extendability a hell of a lot easier.. look at the sims for a perfect example. (all the expansion packs, essential new environment items with new scripts etc, which operate in a plug and play fashion)
-
Yeah sorry about the unintelligible discourse…Ive havent been drinking enough lately and my brain is trying to come back to life :)
The main point was that if you take all the work that’s gone into modern physics (thousands of years worth really), and measure it by man-hours, and then map that onto our immediate future – theres enough man-hours available to replicate the entire growth of the field in a ridiculously short time…100 years is a just a guess, but why not?
So therefore every field of research is speeding up, because there are just more people. sorry, I know its a bit of a tangent.Yeah my final year project was pretty much an intelligent environment. itd be an interesting area to work in again…there’s a lot of interesting fields of study crossing over, to greater or lesser extent.
-
only one flaw, not everyone makes constructive contributions(so probably only 5-10% make the big break throughs, rest make minor but fairly valuable stepping stone contributions).
also i dont think research can be quantified the same as software engs man-years, mythical man like you said above. you cant just set out “today im going to build a human”, put 500 ppl on it and wham, 18months time out pops a living breathing human. sure you can make progress(learn how to make a toe or something ;), but for you to make progress in this small area it also depends on support domains keeping up with you (anatonmy, neuroscience, advanced microscopes etc)
man-year-month i think can only be apply to a task which is fully layed out and completely explored, and the only real task is to grind out the code etc. i.e. word processors, engineering tool, aircraft.
Research on the other hand requires experimenting, brainstorming, coding, scrapping, consulting etc. difficult to put it in terms of time. throwing people on the job wont make it work.
thats why few companies are (especially game companies ea exluded) can afford or bother to invest money in new methods, because they want to be told, in 4 weeks we will have solved x,y,z and you’ve got your dollars worth.
While in reality in 4 weeks we may discover we were approaching the problem totally wrong and have to start again. i.e. money appears to be wasted(from payers perspective), researchers perspective is that they have discovered how not to approach this problem. ;) and this re-iterates.
this is a common problem with VC and research.
-
Hmm, its not really comparable to software development – as I unwittingly implied by using the term man-hours. Heres an argument that’s close without really being the same thing – today, more than ever, we have more universities and researchers than ever before. Why? because south asia has subscribed to the educational methodology and forms of the west – so now we’re all reading from the same page. all sci-fi predictions aside, there are just more people educated in the western style, doing more work in all areas.
As for the old VC vs research problem, perhaps it wouldn’t be so bad if there wasn’t 80 artists on every large project? not that I want to get the artists fired :) but the point is, when are we going to get games that successfully shift paradigm, and allow the focus to go elsewhere than graphics (and where else is there but AI?)
It seems to me that Will Wright is the only one in the industry who is consciously trying to address that issue, with any success…can’t wait to see how Spore turns out.
-
- The forum ‘Programming’ is closed to new topics and replies.