PRODIGY

• Uniform represenation

  All the knowledge used or aquired in any module in Prodigy uses the same type of logic-based representation. This includes all factual and control knowledge. It is this uniformity of representation that allows Prodigy to implment a declarative knowledge system.

• Declarative knowledge: Glass box hypothesis

  This property allows all the modules and their data (control rules, problem-solving traces, etc.) to be seen by the other system modules. This ensures that all the system's knowledge, included newly aquired or learned knowledge, is available to any of the system's reasoning processes.

• Deliberative learning

  Deliberative learning is the ability to decide what to learn, when to learn, how to learn, and whether to remember what it learned. This selective learning is accomplished with a utility evaluator, which compares a new piece on knowledge's benefit to its cost, and based on the result, decides if, when, and how it should learn. This property attempts to keep the learning of knowledge from slowing down the system's performance.

• Multiple learning methods

  PRODIGY's uniform knowledge representation and modular architecture allow it to support multiple learning methods. This allows PRODIGY to be a testbed for learning mechanisms. From a software engineering standpoint, it also allows for easier construction and debugging of the PRODIGY system.

• Casual commitment

  During problem solving, PRODIGY uses a number of control rules to direct its search. If at a given point in the search, PRODIGY has no control rules to direct its next choice, it arbitrarily chooses the next action. This is known as casual commitment. Instead of resorting to clever methods to choose the next action in the absence of appropriate control knowledge, PRODIGY instead arbitrarily chooses one action, and based on its success or failure, constructs a new control rule to deal with that situation in the future.