I don't know much about AI algorithms, but I have it in mind to write a simulation where AI is represented as a directed graph of nodes representing basic "concepts" (as an abstraction of concept a human might understand). any two nodes would be connected by a "relationship," which are formed dynamically through-out the life of the simulation. The goal is to see if the simulated agents can draw any conclusion based on following trails formed through the graph. A relation ship formed during the simulation would represent a single step, but what if one follows two or three steps through the graph? The question is, can anyone point me to some well known algorithms in AI that follow this model or something similar. I'd like to read what is well already well established in this area.

Quote: The question is, can anyone point me to some well known algorithms in AI that follow this model or something similar. I'd like to read what is well already well established in this area.

This should get you started:
A Brief Introduction to Graphical Models and Bayesian Networks

Or better yet, look into NAG, since it does exactly what you want to do (it joins Bayesian Networks with Semantic Networks) to provide a system for AI argument development based on collected evidence. Check out the work of Ingrid Zuckerman and Kevin Korb, both in Comp Sci. at Monash University.

Cheers,

Timkin

Thanks guys. Food for thought. I've heard these terms mentioned before but knew not what they meant.

I had implemented something similar to what you are describing, but the application was for pathfinding, and I used connectivity matrices which determine how certain nodes are connected in various 'moves' (transitions from one node to another). Therefore, you could use lookup tables to determine the various results 'in the future,' based on the move you take 'now.' This is most similar to something called 'Floyd's all pairs' algorithm, but I optimized it further by representing the matrices as sparse matrices (0 entries are ignored).

I suppose this concept could be used to represent anything, whether it be waypoint nodes in a map of certain types of behaviors in a finite state machine (just depends on how you interpret the data).