I would feel more comfortable with such predictions if research were clearly walking the wooded path to AI. Moore's law works because 1. We already have proc

I would feel more comfortable with such predictions if research were clearly walking the wooded path to AI. Moore's law works because 1. We already have processors 2. We have a metric to measure processor speed Then the path we're hiking is one mostly of incremental improvments, with occasional boulder of innovation. Whereas the 'intelligent' agents we currently have do not seem particularly similar in quality to intelligence we already know; the path from contemporary AI to strong AI is hardly a trail at all --- it's all boulders; innovation all the way up. [The claim I make about 'quality' is vague; in part necessarily so. If I could pinpoint my discomfort with current methods, I could propose a new course of action based on a new metric. Nevertheless I feel that the high-mathematical bent of current machine learning techniques (proto-value RL, statistical relational methods, etc) will lead to excellent answers, but does not point towards the flexibility of general intelligence. Yrom the other camp, low-level connectionist methods have not to my mind offered significant results in problem solving.]