Abstract:
We propose methods of employing autocognitive inference as a realistic, feasible way for an agent to make many inferences about its own mind and about the world that have often been regarded in the past as depending on nonmonotonic reasoning. The keys to realism are (1) to use a computable notion of knowing; and (2) to employ specific, realistic assumptions about the sources of our knowledge, rather than simplified negation-as-failure assumptions. We illustrate our methods with a preliminary implementation of several reasoning examples in the EPILOG system.