The first set of sprints on the machine learning research project was wrapped up early in April of this year. An instance was spun up and it was capable of learning how to play tic-tac-toe from a completely trial and error basis. It was able to learn to play, to follow the feedback provided by the environment, and even to create memes from experience and behave completely from its meme stack. While it was able to do this and much was learned from that element of success there were several elements of failure from which much more can be gleaned, learned and applied to the next set of sprints. The error state in a form of questions, "Why didn't the instance learn to play tic-tac-toe better? Why is it blind to my strategy while mindful of the rules?"
Wednesday, May 06, 2009
1. Externalized Judgements: The source material for judgement formation was reflecting external environment and did not arise directly from internal reason. This passivity precluded independent thought and any meaningful grounds for speculative interpretation of another actor's behavior. The instance needs to judge independently, relying on its own internalized feedback system.
2. Lack of Motive: The instance had no reason to play, it was forced to play and it had no internal valuation of why feedback was positive. It learned how to play but not why to play. It requires internal conditions that must be maintained to maintain existence, a reason to play. This is related to the error brief above.
3. Cartesian Error: The instance had no type relation to its environment and was positioned as an outside observer that could effect but not be affected, basically I reproduced Descartes silliness in splitting the mind and body, actor and environment. The instance needs to arise from the environment, fundamental sharing type, and learn to identify elements of its environment and not just know them. This is challenging, but provides exciting possibilities.