Thursday, May 1, 2008

Artificial Intelligence Thoughts

Algorithms cannot create intelligence. Programming algorthims is analogus to building "instinct" but humans have evolved out of instinct into somethimg more, so when we try to create AI algorithms we get in a loop because we are trying to create intelligent instinct which is nothing but an oxymoron.

A more abstract approach is needed, one that starts with emotions/instinct and builds on it.

Intelligence requires "emotion" on inputs, and unlike non living things, living things constantly strive away from equlibrium whlist feeling anxious if taking from equlibrium "against will"??? Controlled change driven by individual is +ive, uncontrolled changed driving my environment is -ive? maybe

Ideas to simulate intelligence

a) Simulate a Cause/Effect system/network. All inputs are scanned constantly and compared and crossreferenced. The network continues to breakdown inputs in fixed time sequences and make guesses as to what Input A -> Input/Output B. (ie: If I observe that B happens many times after I observe A, I can define a relationship between A & B

b) For each input generate a weight (-ive /ngt) based on individual "view" of the input (eg: heuristic ALL CAPS wieigh more -ive for keyboard input, STRONG text could be -ive or +ive/)

c) Create a natural "drive system". Implement an input system that generatives +ive and -ive. Entity will seek (ie: create outputs) in a drive to constantly remove itself from equilibrium. Weither by generating +ive (favorable) or -ive inputs

d) Implement Pablo's dog system. if A -> B and A -> B - > C eventually A -> C. Food causes salavation reflex, ring bell each time food is shown, eventuall bell causes salavation (expectation of food)

An example is to connect battery power levels as example/analogy to food, and "charging" status as being fed

Source Dump..

class event

level: (int +ive or -ive) strength of event

name: (undefined) seek to find what to call events?

group of events (function the same as events)

EVENT (or group of ) -> causes -> EVENT (or group of)

Make guess on EVENT based in prior events

Inputs can't be too exact. (ie: 2 spoken words "hello") need very similar input strings to assume it's the same word. If not, system would have to relearn word for every person who spoke it. Ironically a video input would need to be able to distiguige more detail then an audio input?

No comments: