AI theory is mainly about how to go about emulating intelligence.  One of the most well known solutions for this is the ELIZA type program.  The program is based on the ideas of Rogerian therapy, which is basically a patient-centered therapy where the goal is to delve into the person’s head by almost constantly asking why in different ways.  At this stage the illusion of a real person was only supposed to last anywhere from 20 seconds to a couple minutes, as it soon became apparent of how little ELIZA actually understood.  From this stems the Turing Test, one of the only tools AI programmers have to base their experiments on, it is basically a test of whether or not a machine can make a person think that the machine is a person.  So people started to try to figure out what are the best programming strategies for making AI.  Among these ideas are:

Neural Networking:
A system where input and output nodes are connected via nodes that would be equivalent of different neurons,
they of course took this design from the human brain.  As certain pathways are used more often they are weighted.  Also like in biological neural networks tasks are performed parallel when required making a system that is good for adaptation for use in real-world problem solving.

Fuzzy Control System:
Fuzzy Control Systems work under the idea of fuzzy logic that is not to say that the logic used isn’t as good, but rather that instead of just being able to have the black and white true and false arguments fuzzy logic is able to
use the kinds of concepts like partially true.  Thusly the end result is more insightful for a human.

Swarm Intelligence:
This type of structure is based on that of an ant colony or a beehive.  Where there are several computers networked together with input and output receptors, so that there is no central unit governing what all units do but a global pattern emerges.
           

The main split in AI theory is the Neats vs. the Scruffies.

      - Neats prefer to use formal logic and applied statistics so that their solutions can be provably correct.  So neats
        would be drawn more towards Neural Networking kinds of solutions.

      - Scruffies think that intelligence is too complicated to be solved with mathematical equations and proofs and
        use a series of learned or evolved hacks that do not   need to have internal consistency, they prefer to use
        empirical experience to show that their algorithms are working.  As opposed to neats the scruffies would
        probably tend to go towards a Fuzzy Control System kind of solution with bits of other solutions thrown in to
        try to maximize potential.

           
This debate goes much further than just programming styles; it permeates into the philosophy of AI, whether they think that AI should resemble human intelligence, and if human intelligence is inherently scruffy or neat.  Other arguments in AI philosophy are the mind-body problem which is the question of whether or not the mind and body
are one or if the mind is something separate from the body.  This is an important question because if the mind and body are separate how would one go about reproducing a mind via physical means?  Others argue about what kind
of intelligence AI should try to emulate: should we try to make AI like human intelligence or should we try to make
an intelligence of something higher than human understanding?

A simple problem is at the base of all of these arguments, which would be at which point does a machine become intelligent?  So far the only answer to this question has been the Turing test, proposed by Alan Turing that, this question is just a simple question about conversation. If a machine can answer any question put to it, using the
same words an ordinary person would then that machine can be called intelligent.


> Top