Generate and test is too inefficient because of its trial and error nature. Minsky moves to improve on this with his Progress Principle: "Any process of exhaustive search can be greatly reduced if we possess some what to detect when `progress' has been made. Then we can trace a path toward a solution, just as a person can climb an unfamiliar hill in the dark -- by feeling aroudn, at every step, to find the direction of steepest ascent."
But, progress may be hard to recognize for difficult problems. Solution? Decompose the problem! "The most powerful way we know for discovering how so solve a hard problem is to find a method that splits it into several simpler ones, each of which can be solved separately."
Another approach is to embody knowledge in machines, because "the most efficient way to solve a problem is to already know how to solve it. Then one can avoid search entirely." (NB: This point becomes crucial for the discussion of K-lines in Chapter 8.)
But, embodying knowledge is problematic too. "We must discover how to acquire the knowledge we need, we must learn how to represent it, and, finally, we must develop processes that can exploit our knowledge effectively. To accomplish all that, our memories must represent, in preference to vast amounts of small details, only those relationships that may help us reach our goals."
The irony of all of this is the paradox of expert systems, i.e., it is easier to program an expert system than it is to program common sense. THat is why Minsky has focussed on "easy" problems to this point in his book.
(NB: Key note here -- decomposition is a key strategy to solving a very tricky problem: how the mind works. What would Braitenberg say about this strategy?)
Pearl Street | Society of Mind Home Page | Dawson Home Page |