Significant advances in artificial intelligence, including machines that play master level chess, or make medical diagnoses, highlight an intriguing paradox. While systems can compete with highly qualified experts in many fields, there has been much less progress in constructing machines that exhibit simple commonsense, the kind expected of any normally intelligent child. As a result, commonsense has been identified as one of the most difficult and important problems in AI (Doyle, 1984; Waltz, 1982).
The Importance of Commonsense1
It may be useful to begin by listing a number of reasons why Commonsense is so important:
Any general natural language processor must possess the commonsense that is assumed in the text.
In building computerized systems, many assumptions are made about the way in which they will be used and the users’ background knowledge. The more commonsense that can explicitly be built into systems, the less will depend on the implicit concurrence of the designer’s commonsense with that of the user.
Many expert systems have some commonsense knowledge built into them, much of it reformulated time and again for similar systems. It would be advantageous if commonsense knowledge could be standardized for use in different systems.
Commonsense has a large element that is environment and culture specific. A study and formalization of commonsense knowledge may permit people of different cultures to better understand one another’s assumptions.
No attempt will be made here to define commonsense rigorously. Intuitively, however, commonsense is generally meant to include the following capabilities, as defined for any given culture:
knowing the generally known facts about the world,
knowing, and being able to perform, generally performed behaviors, and to predict their outcomes,
being able to interpret or identify commonly occurring situations in terms of the generally known facts – i.e. to understand what happens,
the ability to relate causes and effects,
the ability to recognize inconsistencies in descriptions of common situations and behaviors and between behaviors and their situational contexts,
the ability to solve everyday problems.
In summary, commonsense is the knowledge that any participant in a culture expects any other participant in that culture to possess, as distinct from specialized knowledge that is possessed only by specialists.
The necessary conditions for a formalization to lay claim to representing commonsense are implicit in the above definition; a formalism must exhibit at least one of the attributes listed there. Virtually all work in the field has attempted to satisfy only some subset of the commonsense criteria.
Commonsense Representation Formalisms
In AI research, work on common sense is generally subsumed under the heading of Knowledge Representation. The objective of this article is to survey the various formalisms that have been suggested for representing commonsense knowledge.
Key Terms in this Chapter
Non-Monotonic Logic: A logic that attempts to overcome the restrictions of monotonicity.
Logistic Models: Modified logics that attempt to overcome the problems of representing commonsense knowledge in the classic predicate calculus.
Belief Maintenance Systems: Systems of logic that permit theorems to be updated as new knowledge becomes available.
Commonsense Knowledge: Knowledge of the basic facts and behaviors of the everyday world.
Monotonicity: A characteristic of logic that prevents changes to existing theorems, when new information becomes available.
Representation Formalisms: Theoretical frameworks for representing commonsense knowledge.
Dispositional Models: Representations of things or facts.
Propositional Models: Descriptions of representations of things or concrete facts.