learning ai, best route?

Started by
1 comment, last by verdad 18 years, 8 months ago
i plan to start a project that simulates intelligent learning behavior in a small creature. my creature will have the following inputs: *cursor position *mouse state: can be none/left mouse/right mouse *punishment *reward my creature will have the following outputs: (for now) *move left *move right - The creature will want to be rewarded and try to avoid being punished. i want to make it try and learn to follow the cursor and such without being pre-programmed. what advice would you give me as to how i can produce this end result.
-www.freewebs.com/tm1rbrt -> check out my gameboy emulator ( worklog updated regularly )
Advertisement
How do you punish it? how do you reward it? Making it follow the cursor isn't much of an AI challenge.

AI agent follows precept/action pair. Given an input event, it wants to produce an action. Those actions which maximize it's utility, are rewarded. Those which punish it's utility, are to be avoided. You might choose a score, [-10,10] for a precept map. Those actions with better scores are rewarded, those with lower scores should be avoided. If you want it to learn, you need some way for it to reason about this.

y = f(precept)

where y is the score following an action from a precept. If the entity walks into a lava pit and the agent is not immune to lava, than the punishment should be severe, and than the AI should remember this (if it lives) to always avoid any further actions involving walking into lava. Or you can give it hints at birth to know always to avoid lava.
"It's such a useful tool for living in the city!"
Not to mention if it sees someone fall into lava and they are severely wounded or killed or something, it would be able to comprehend, "Okay, lava is not good."

Your post is really vague as to what you're doing, what your goal is, and other various things...please post more if you want more.

This topic is closed to new replies.

Advertisement