| Kwaad said: Thank you for replying however let me list off a few AI's that are very... smiple. Blue Gene. (Think a board game AI) Possible Checkers AI. (much smipler than a Blue Gene AI system, yet can still take hours to make a single move) I just wanna say that the AI in GTA3 sucked. Infact, most AI on console games suck. Some games have had people put *real* AI's on them. However if you put more than 1AI bot, it tends to each a computer for dinner. My point on this is. When you have 30 'objects' It can take alot of computing power to do this. 30 'objects' (object defined as an aware entity, car driver, pedistrain) They are watching traffic monitoring where the other AI's are going, and how they are moving. If you cut infront of one of them, the car you cut infront of, would take evasive manuvers depending on what personality the car has. While the car you just cut off runs into the sidewalk, the 10 pedistrains start running away from the 'psycho' driver and they run for cover, get out of the way, or just get squished if their not fast enough. All the while, the other cars nearby are witnessing this, while the other pedistrains are watching the carnage, and acting accordingly. Yes, they are all running a script, however where they go is decided by what their personality is, and where they are. So a gangster might pull his gun out, and start killing everyone, while a cop might run to a injured pedistrain to make sure he is ok. Basically it's simple scripts. However there are ALOT of simple scripts running at the same time, for each person, while there are 30 diffrent 'objects' running the many diffrent scripts. And rember, the physics, and everything else is going on at the same time. This is more than a simple application to figure out what each AI would do. Sure. On a home PC, that might take .5 seconds to compute the AI that is going on. Instant. Right? Well that's not fast enough. The AI has to be done in at least 1/30th second intervals. It cant take 1/2 second to run the AI scripts, as that would cause a 1/2 second freeze on the game. Or it would delay AI reactions by around 2-3 seconds due to the fact that 90% of the processor is NOT doing AI. Either way, that would completly ruin the 'feel' to any game. My point on that is, just becuase the PS3/360 has enough power to do it. Does not mean a home PC can. (read RISC processor for 360, and read CELL processor for PS3) Both are many many times better than a home PC in architure, and both are bigger as well. (more performance for chip size, and bigger chip overall) The Wii does not have near the processor as a home PC. The PS1 is about 1/10th the speed of the Wii (in processor). While the SNES is about 1/10th the speed of the PS1. Basically the SNES is 3000-4000times SLOWER than a PS3. (this is not accurate, dont quote me, but it's some crazy big ammount) If the PS3 could do it in 1 second, the SNES would take 50minutes. If the PS3 could do it in (10% processor use over 1/30th second) the SNES would do it in 30 *SECONDS* Basically 10% of the processor over 1/30th of a second, would take the SNES 30 seconds to do the same ammount of math. Basically the SNES is 00.03% of the total power on the PS3. .03% PS3 = 100% SNES. It's bed-time. If my wife dosent nag my ass off. I'll find accurate numbers on this, and compare the processor in percentage from NES/SNES/N64/PS1/PS2/GameCube/Wii/PS3/x-box/360 The PS3 numbers will be created from what I know about the processor on the PS3. Not what I have read. It runs at about 1/2 of what everyone says it can run. (today at least, I'm sure it will run at least 80% in the future) But I'm kinda curious about this now. How much power does it take to run AI? I will also read up on that. I will take a quote that would say something like... "It takes 25% of a P4 2.8ghz to run the AI in this game." I would then consider the generation, the ammount of power overall, and then scale the numbers with percent to show how much power 'that' game would run on a modern console. And then whatever % was specifyed, I would put it on the 360/PS3 level, and then work the way down showing the % of processor needed to run 'modern' AI on systems. |
Blue Gene, Checkers: i assume "blue gene" was the chess playing machine. This kind of AI (although in reality is far from "intelligent") is basically brute-force, and again, the problem size is exponential--in principle anyway (there are ways to reduce the problem size). here machine power doesn't matter as much as it appears---yes, you need an IBM super cluster to beat Kasparov, but for the remaining 99.99% of chess players, a machine 1000 times slower will still win handily.
getting beat by brute force from a computer is absolutely no fun--but that's a different topic.
What you have in mind is a totally different beast--it appears in a lot of fields these days so let's just call it "agent-based". yes, in agent-based algorithms power matter more noticeably, because, many algorithms scale linearly, or quadratically (meaning going from 1 to 2 the amount of power used up increases by a factor of 4) with problem size, and are generally parallelizable (good for machines with multiple processors).
In those cases, PS3 and 360 would have advantages over the Wii. I assume that is what the spore developer had in mind when he went on his rant at the GDC.
i don't know the estimate just how much slower the Wii is compared to PS3, i have no idea what benchmarks people use. let's just say it's a factor of 20. that means assuming linear scaling, you can handle 20 times as many objects. assuming quadratic, that'll be about 4-5 times... and yes, that'll mean a difference of 0.2s and 0.05s for some situations.
"AI" %--it's completely dependent on games. PS3 and 360 might as well go the agent-based route. however, do keep in mind that that alone does not a good game make. it's the INTERACTION that the gamer observe that matters, and indeed, if you go the agent based route, you better come up with good interaction codes. they go hand in hand. it also provides for (in principle anyway) gameplay that's different every time. however, i don't know if this approach would qualify as "AI" most of the time.
i think agent based games would be very interesting, but i don't know if you'll see good examples any time soon. putting together a good set of interaction logic is key, IMHO. also, in my mind, a factor of 20 in processing power is nothing (graphics aside). truthfully, i think that if you're dependent on that factor of 20 to make your game "fun", you're in trouble anyway.
i have some ideas for some agent-based games involving learning algorithms and evolutionary algorithms (same thing, really) that i'd like to implement at some point in my life, and i dream of being the first to come up with some totally convincing gameplay. ah, dreams.
the Wii is an epidemic.
