From OpenCog
Jump to: navigation, search

A detailed design for the ExecutionManager still needs to be written!

However the conceptual idea (due to Jiang Min) is to use game theory, to figure out the optimal procedure(s) (drawn from the ActiveSchemaPool) to execute at each point in time.

Game theory would be a bad choice for deciding what strategy to take ... i.e., what procedures to put into the ActiveSchemaPool in the first place ... but once that decision has been made, we have a choice to be made among a fairly small number of options, and we can assume that probabilities regarding these options can be estimated based on the specific observed context ... so, this is a context in which game theory can potentially be very useful...

Finding a pure strategy is likely not feasible so a mixed strategy approach must be taken, where the probability of success of each strategy is estimated using PLN...