The game of the future.

The place to hang out and talk about totally anything general.
User avatar
BrianBlessed
level4
level4
Posts: 867
Joined: Sat Dec 24, 2005 9:33 pm

Postby BrianBlessed » Fri Jan 05, 2007 12:55 am

The problem with 'true AI' is what you base it on, I mean darwinian AI can be as complicated or a simple to program as possible as they are a creation and therefore there is no reference for the AI. However if you wanted to programme an artificial entity to act and respond exactly as a human would you are posed with the problem of what to base it on. Do you analyse every thought, emotion, movement and action of every person on the planet past, present and future and programme in the entirety of what it is to be human aswell as any potential adaption and evolution in the future. Unfortunately that's bordering on the Doctor Who-esque impossibility, while you could create an incredibly complex entity and one that had some level of human behaviour, it's far easier just programming a soldier for a shoot 'em up and boxer for a boxing game and so on.

While you could program the whole plethora of human behaviour for a soldier in a shoot 'em up, as the game stays on the battlefield the uses of said behaviours are going to be fairly limited. While you could upgrade AI so soldiers can move barrels or crates to use as cover or open the door of a car and roll it along for cover. You are incredibly unlikely to need to simulate the dynamics of a mass of human interaction such that if you kill an entire platoon bar one man, that man then finds a vehicle and escapes the war zone to a still relatively peaceful area where he sets up a weapons research lab. After several months/years of development, his new weapons/vehicles/armour appear on the battlefield in the hands of your opponents.

Although the comicbook causality is rather intriguing, so if you are playing GTA and you kill someone's parents infront of them then they vow to fight crime and eventually find and come after you for vengeance. Or if you just shoot a civilian in the leg, they get taken of to hospital and depending on the exact trajectory and entry point it would determine the severity of the injury. Perhaps there are complications and the leg has to be amputated, you then see them later with a horrific peg leg aswell as a far more bitter personality.
User avatar
shinygerbil
level5
level5
Posts: 4667
Joined: Wed Dec 22, 2004 10:14 pm
Location: Out, finding my own food. Also, doing the shinyBonsai Manoeuvre(tm)
Contact:

Postby shinygerbil » Fri Jan 05, 2007 3:00 am

Montyphy wrote:THE GAME!

You lose.


You f--

mumblemumble
User avatar
KingAl
level5
level5
Posts: 4138
Joined: Sun Sep 10, 2006 7:42 am

Postby KingAl » Fri Jan 05, 2007 3:55 am

BrianBlessed wrote:<- snip ->


Essentially, it's the issue of approaching it top-down or bottom-up. Do you design AI via a complex network of digital neurones, or simply create high-level abstractions that perform the required behaviour?
User avatar
xander
level5
level5
Posts: 16869
Joined: Thu Oct 21, 2004 11:41 pm
Location: Highland, CA, USA
Contact:

Postby xander » Fri Jan 05, 2007 7:11 am

KingAl wrote:
BrianBlessed wrote:<- snip ->


Essentially, it's the issue of approaching it top-down or bottom-up. Do you design AI via a complex network of digital neurones, or simply create high-level abstractions that perform the required behaviour?

The problem is that, whichever approach is used, you have to get from one to the other. If you create complex networks of digital neurons, they have to be able to perform tasks which require high level abstractions. If you create high level models, they have to be able to deal with real world complexities thrown at them. We humans can do a very good job of writing both kinds of programs, but we have so far failed to create anything that can do both -- it is like quantum physics and relativity; there must be a way to unify them, but we haven't done it yet.

xander
Cheesecake?
level2
level2
Posts: 75
Joined: Wed Jan 03, 2007 4:40 am

Postby Cheesecake? » Fri Jan 05, 2007 8:21 am

Image
User avatar
ghostDancer
level2
level2
Posts: 90
Joined: Tue Jul 12, 2005 4:50 am
Location: Land of Confusion
Contact:

Postby ghostDancer » Fri Jan 05, 2007 11:46 am

Xocrates wrote:That reminds me of a "game" I once played. basically you were able to spawn soldiers and set their priorities (like "approach the enemy" or "Attack the enemy") and design the scenario so they would learn how to fight and move. The ones with the best results would be kept and the weaker ones eliminated.

After training, you could set several armies against each other and they would fight by themselves as you watched the result.

The game was rather frustrating because as were teaching them one thing, they were forgetting another. But is was interesting.

Unfortunetely I can't remember the name of the thing :(


Maybe that game is Nerogame http://www.nerogame.org/ i haven't tried yet , it looks promising but also that you must put a lot of time into it.

ghostDancer
User avatar
Xocrates
level5
level5
Posts: 5262
Joined: Wed Dec 13, 2006 11:34 pm

Postby Xocrates » Fri Jan 05, 2007 12:33 pm

ghostDancer wrote:
Xocrates wrote:That reminds me of a "game" I once played. basically you were able to spawn soldiers and set their priorities (like "approach the enemy" or "Attack the enemy") and design the scenario so they would learn how to fight and move. The ones with the best results would be kept and the weaker ones eliminated.

After training, you could set several armies against each other and they would fight by themselves as you watched the result.

The game was rather frustrating because as were teaching them one thing, they were forgetting another. But is was interesting.

Unfortunetely I can't remember the name of the thing :(


Maybe that game is Nerogame http://www.nerogame.org/ i haven't tried yet , it looks promising but also that you must put a lot of time into it.

ghostDancer


Yes, that's exactly it :D

And yes, you really need to invest a lot of time in it before they do anything half-decent.
Rock
level2
level2
Posts: 206
Joined: Sun Sep 17, 2006 1:50 am
Location: Purcellville, VA

Postby Rock » Sat Jan 06, 2007 6:24 am

Cheesecake? wrote:Image


DAMMIT!

I hate it when I lose...

Going back on topic-

Yeah, that sounds like an interesting idea, but it would need to be expanded upon a wee bit more to make it super awesome.

~Rock
User avatar
wwarnick
level5
level5
Posts: 1863
Joined: Mon Oct 02, 2006 8:44 pm
Location: Rexburg, ID

Postby wwarnick » Sun Jan 14, 2007 12:47 am

In order to create the 'truest' ai, you'd have to create it to be simple and ignorant at the beginning, then let it learn by itself. A game like Darwinia would first require an actual world with actual resources, real physics, etc. Also, the creatures (Darwinians in this case) would be given the ability to evolve. At the beginning, they would have to start out as simple "life forms" if you will, then evolve into a more advanced form as they learn to interact with their surroundings.

wwarnick
K25125
level0
Posts: 2
Joined: Sun Jan 14, 2007 2:49 am

Postby K25125 » Wed Jan 17, 2007 8:05 pm

it could most liklely be done. i am a programmer (sort of) and it could be done with a random number generator, every number having a different code, deciding the darwinia's likes, dislikes, etc.
User avatar
Spacemonkey
level4
level4
Posts: 609
Joined: Thu Oct 05, 2006 1:31 am
Location: Auckland, New Zealand

Postby Spacemonkey » Wed Jan 17, 2007 9:24 pm

I was reading about an AI project where evolution is achieved by randomly swapping bits of the code. If the swap would produce a negative effect, the digital lifeform would die, if it were positive, it would gain an advantage etc.
User avatar
MrBunsy
level5
level5
Posts: 1081
Joined: Mon Apr 24, 2006 4:40 pm
Location: Southampton
Contact:

Postby MrBunsy » Wed Jan 17, 2007 9:40 pm

Won't that take an incredibly long time to achieve anything worth bothering with though?
User avatar
xander
level5
level5
Posts: 16869
Joined: Thu Oct 21, 2004 11:41 pm
Location: Highland, CA, USA
Contact:

Postby xander » Wed Jan 17, 2007 9:43 pm

MrBunsy wrote:Won't that take an incredibly long time to achieve anything worth bothering with though?

Probably, but it took several hundred million years to get from microbes to humans, too.

xander
User avatar
Spacemonkey
level4
level4
Posts: 609
Joined: Thu Oct 05, 2006 1:31 am
Location: Auckland, New Zealand

Postby Spacemonkey » Wed Jan 17, 2007 10:04 pm

MrBunsy wrote:Won't that take an incredibly long time to achieve anything worth bothering with though?


Not really, it takes 30 years between each generation of human, that time could be reduced to only a few minutes in a computer simulation.

And the more computers that are used, the faster the simulation can be run.
User avatar
xander
level5
level5
Posts: 16869
Joined: Thu Oct 21, 2004 11:41 pm
Location: Highland, CA, USA
Contact:

Postby xander » Wed Jan 17, 2007 10:43 pm

Spacemonkey wrote:
MrBunsy wrote:Won't that take an incredibly long time to achieve anything worth bothering with though?


Not really, it takes 30 years between each generation of human, that time could be reduced to only a few minutes in a computer simulation.

And the more computers that are used, the faster the simulation can be run.

Some back of the envelope mathematics:

Yes, but it only takes a matter of minutes between each generation of microbes. Let us be conservative, and say that these theoretical microbes can divide or otherwise reproduce once every hour. Not accounting for the change in day length over the last half billion years (the beginning of the Cambrian, more or less), there are 8760 hours in a 365 day year, or about 4.38*10^12 hours since life first appeared, give or take. Given that we are assuming one hour is one generation, that is about 4.4 trillion generations.

Let us say that we compress every generation of computer simulated microbes to one thousanth of a second. Again, I think this is being generous, as that gives about 3 processor cycles to each simulated organism to reproduce on a modern desktop PC. I would imagine that more simulated organisms would slow things down a bit, though you might be able to use super computers or distributed computer to speed things up a bit. In any case, 1,000 generations per second seems really, really fast to me. We still want to simulate the same number of generations. 4.38*10^12 generations * 1/1,000 seconds per generation is about 4.38*10^9 seconds (about 4.4 billion). 4.4 billion seconds is about 140 years (there are about 31536000 seconds per year, again, not taking into account change in day lengths or leap days).

Also note that we are leaving things out. Evolution occurs, in part, because organisms interact with eachother and the enviroment. As we are evolving our code, what makes "good" code that survives? How will different code organisms compete with eachother? How will they share "genes?" &c. &c.

So, while we can reduce the time it might take to evolve something interesting, we are still dealing with astronomically huge spans of time. And, at the moment, I am just talking about simulating very, very simple microbes, and I am being generous in how much time it might take to compute changes. If you were to try and come up with an AI through an evolutionary process (include things like mutation, "natural" selection, &c.), I would imagine that it would take a very long time.

xander

Return to “Introversion Lounge”

Who is online

Users browsing this forum: No registered users and 3 guests