An entropic view of AI

So what is it to view AI through the expanded view of entropy provided by MEP? Basically, we have a problem, and an agent who starts out with a large amount of "potential energy" which is not related to the solution to the problem. The agent goes through a series of iterations where it the "entropy" of the system is incrementally increased through finding paths of maximum entropy and avoiding restrains. When the entropy in the agent has increased its maximum state, it has grown a solution of some kind to the problem. It settles into the state of least resistence, which is an optimal solution for the agent in the time given.

If the agent is trained, the entropy used for solutions will be the entropy at the end of the training sets. The entropy should be observable, and fallow a curve similar to those you can get by graphing Shannon Entropy with different probability distributions.