🔒 Locked

New Concepts - dynamic game stories

Started by llvllatrix Jun 16, 2002 at 10:58 PM 6 replies 1.8k views

Original Post

Jun 16, 2002 at 10:58 PM

Just so u know, im a programmer, so im a bit wierd. Anyway, i was just about to go to sleep when i came up with an idea. Ive been trifiling for some time now, attempting to create a totally dynamic engine - one where the computer decides the story line. One problem, the computer has to be able to "think", and do it creatively for it to be able to even compare to u guys. Anyway i think i might have come up with a solution, but there are a few catches (excuse me if i get a bit technical here but i just want to explain what the catches are). A while back i read something on ANNs or artificial neural networks (simulate a brain). I thought, great an easier way to do artificial intelligence, but that was then. It just occured to me that i could not only use these networks to simulate the actions of my AI units, but i could also use them to define things like graphics and such. I could even take that concept a little further and use the output of every neural network as an input to another neural network. This would give you group intelligence. The creatures in your world would interact like a family. Heres the catch, their functions would be very limited. They would have no notion of common themes in stories, nor even what would make a good story. It would be like taking a player and sticking it in a heard of furbies that can learn, grow, and perhaps reproduce. There are also several choices involved here. On the one hand you could "hard code" several elements like a house for example can be seen as one structure with one purpose, and building it is instaneously possible. Or you could let this system start from scratch and "figure out" how to build one for itself. The neat thing about doing a story like this is that it offers the player almost infinite possibilities as to what they want to do, and doing stuff should never run out. The disadvantage is that this stuff is not always going to be fun. However, it being a program, you could introduce external elements, like addaing a player and even an external enemy AI (which has not evolved with the rest of the system). Anyway, if you are still with me, what do you think? Would a game like this be sucessful, i mean its "story" (or perhaps more appropriately it's "history")? [edited by - llvllatrix on June 16, 2002 12:41:30 AM]

superpig

Jun 17, 2002 at 2:10 AM

Hmm...

I''d have to see it implemented to believe it. I read up on NN theory, but couldn''t figure out how to actually apply one to a game (i.e. where to use it, what to feed in, what to do with what you get out, etc).

It''d be difficult to apply to any kind of conventional game, too - although, that''s not necessarily a bad thing.

Superpig
- saving pigs from untimely fates
- sleeps in a ham-mock at www.thebinaryrefinery.cjb.net

Jun 17, 2002 at 2:05 PM

I believe neural nets are much better for simulating instinctive type intelligence than this sort of thing - for example I went to a talk recently by a company who are making humanoids walk using neural nets to control the muscles. As far as stories go, the herd of furbies thing is probably all to true - their motivations and intellect would be far to simple to provide an interesting story (furby 1 want food ... furby 1 steal furby 2''s food ... furby 2 kill furby 1 ... so what, why should I care?)

But then, perhaps you''ve got some original ideas about how to apply neural nets to the problem ...

superpig

Jun 18, 2002 at 1:52 AM

quote:
Original post by g
furby 1 want food ... furby 1 steal furby 2''s food ... furby 2 kill furby 1 ...

Now *that''s* entertainment.

Caravaggio

Jun 19, 2002 at 12:58 AM

Sorry, as a graphics oriented person my head exploded halway through the first post. :D

I read an artilce on MSNBC awhile ago that talked about a method of AI that''s been in production for a few decades now and involves simplye giving the computer facts, not opinions or guidelines. Chlorophyl makes plants green, the earth revolves around the sun, it takes it 360 days to get around once, it rotates on it''s on axis once every 24 hrs, humans are animals, heat causes pressure, gravity pulls down, 1+1=2 the year is 2002, scooby doo sucks. Things lke that, then it would take all this and consider abstact questions entered somewhere else, kind of like the brain in I, Robot.

I was wondering if maybe that couldn''t be applied to a game. This creature is this heighth, heighth denotes strength, you are this height. Then the creatures could judge which one they could and couldn''t take on.

It''s a snappy thing there.

llvllatrix

Jun 19, 2002 at 7:24 PM

What youve pointed out is basically what ive said somewhere in the first post (about the houses)

Glad to see your catching on

A game like this is definitely going to be my next project. Imagine a game, which has basic concepts (like physics, and some primiative chemestry system, basic "facts"), and uses them to generate a abstract world from these facts that the player plays a part in.

It also means a new game every time you reset the game.

Springcoil

Jun 22, 2002 at 5:48 PM

llvllatrix you have a very interesting concept here.
I see you are trying to create a completely interesting experience but how are people supposed to discuss the game and reveiw it if it is all random. By removing the story you could effectively remove the social part of games. In theory it works but if you ironed out the flaws you could have a great system there. Just some people play games for a story. However if you took a world like tolkiens and added certain rivarlies and certain worlds and certain characters, then it could work. The same world with some similar characters but different fights and terrain now that coukd work.

"Being mortal is what makes us capable of the impossible"
- Akira Toriyama.
www.xgameproject.com
Live each day like it is your last because that way you''ll eventually be right.

Kylotan

Jun 29, 2002 at 10:54 PM

The idea is not new. The problem is that a neural network is not a silver bullet. All it does is take a set of inputs and map them to outputs. If you want certain outputs, you still have to train it. Alternatively you can use some sort of unsupervised learning but there''s little chance of it producing ''intelligent'' behaviour. If you could detail how you would use a neural network in the way you describe, that would be truly interesting. (But this would almost certainly be the wrong forum for it, too.)

[ MSVC Fixes | STL | SDL | Game AI | Sockets | C++ Faq Lite | Boost | Asking Questions | Organising code files ]

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.