In our everyday discussion of human knowledge we often distinguish between "common sense" and "formal" or "technical" knowledge. For example, we all know a great deal about how forces affect the objects to which they are applied (a "common sense physics"), how substances combine and how this combination is affected by heat, (a "common sense" chemistry), what motivates individuals to act in various ways under differing circumstances (a "common sense psychology") and even intuitive ideas about mathematics.

   One way in which to approach the investigation of intelligence is to attempt to capture these "common sense" theories within a suitable formal language (e.g. first order logic). If the knowledge is appropriately captured, then questions can be put to this system of common sense knowledge and the answers will correspond to those that would be provided by human intelligence. Note that this emphasizes the importance of knowledge to human intelligence rather than the possession of particular processing capacities.

   A representation language is said to be epistemologically adequate with respect to some domain of knowledge (e.g., naive physics), if

• the language includes the basic terms needed to refer to the entities and actions of the domain (the ontological question);
• the syntax of the language provides the basis for expressing the laws and facts of the domain (the expressiveness question); and
• the laws, facts, and rules of inference preserve "correctness" in the domain.

   Heuristic adequacy is a computational property. The common sense knowledge that is captured in some representation language and a proof (or question-answering) procedure for this language is heuristically adequate if the procedure can be carried out within the computational resources that are available to the device. Thus, the "exercise" of intelligence will be limited by such properties as the memory capacity and speed of the computational device.