WideEyes and some Future Plans

I wrote a while back about WideEyes, a generalized AI that attempts to play Gameboy Color games with mildly interesting results.  I'm looking to expand the AI in the future and possibly tie it into some Twitch integration.

Some stuff I want to add/try:

• Culling Data: A big problem with the first iteration of WideEyes is that it gets flooded with irrelevant data.  I plan to solve this using a number of different algorithms to mark certain sections of data as important or unimportant.  This will save on processing time, allowing me to implement more processor intensive tasks later, and help WideEyes not to be confused by irrelevant data later.
• Basic Causality: I plan to allow WideEyes to understand and use cause and effect relationships to learn more about the game worlds he finds himself in.  This will be implemented using a modified N-tree, along with some accompanying data structures to model what a "cause/effect" actually is and how it can be observed.
• Bayesian Probability: I plan to model WideEyes model of causality after a simplified version of Bayes Theorem.  This will allow WideEyes to talk about a sequence of events as being somewhat likely, or very likely, or not likely at all, rather than just in absolute certainties.  WideEyes will be able to weigh the likelihood of different reactions occurring and choose his actions more intelligently.