Re: The A.I. Project
- Posted by tone.skoda at gmx.net Nov 06, 2002
- 510 views
----- Original Message ----- From: "C. K. Lester" <cklester at yahoo.com> > If your Pacman is going to have "enemies," won't you necessarily need a > survival instinct? You say that your pacman will never die, but it will > get > weaker. What's to motivate it to get stronger? I would replace survival with "good-bad". All in its world could be divided between good and bad things: good and bad actions (or sequence or combination of actions) it takes. Sensory information will, in some way, sculpture its neural net. In what way needs to be found out yet. We need to find if there is something in common how different sensory information sculptures neural net. That thing that is in common to learning is present in all intelligent creatures. > A Pacman who learns to survive can't be that dumb. ;) > > Plus, will you be measuring it on a scale?! I mean, what's the difference > between a Pacman with IQ of 20 and one with IQ of 120? And won't either > one > of them get bored eventually? If finding food and avoiding poison is most important thing in their life then they wont get bored. I'm not sure if feelings are really needed in AI and if AI creature will have them. > Remember, we're not talking about smart robots here, we're talking about > artificial intelligence, maybe even artificial life! I can't just make AI creature and world and enemies and leave it running all night and then next morning I will have real AI :) I have to be able to expolore what types of algorithms, neural nets etc are good. > Now, since your "enemies" are really just poison pills, the "survival > instinct" of your Pacman will simply be a "maintain a high number" > instinct. What's wrong with that? I want to achieve intelligence, I'm not interested in what will AI creature feel, what will be its purpose in life, survival or pills. Those "pills" are very important to it. > Not really. Besides, you're severely limiting your Pacman to see only one > square ahead! For simplicity. It can still do a lot with this vision :) I want to be able to actually see how neural net is structured and that's why I need simple input. Later I will move on to different enviroment ofcourse, like 3d game. You have to start small. :) There is a lot of complexity already here if you know that it will have to remember all the sequence of actions it takes in its neural net and if they were good or bad. >All I see for your Pacman is it remembering the pattern of the maze > and maybe locations of bite-sized healthy bits. Remembering pattern is quite important.