The Role of Intensional and Extensional Representations in Simulation
I review three systems which do simulation in different domains. I observe the following commonality in the representations underlying the simulations: • The representations used for individuals tend to be domain-dependent. These representations are highly structured, concentrating in one place all the information concerning any particular individual. I call these representations intensional because two such representations are considered equal if their forms are identical. • With important exceptions, the representations used for classes of individuals tend to be domain-independent. These representations are unstructured sets of predications involving the characteristics of class members. I call these representations extensional because two such representations are considered equal if the classes they specify are identical. I draw out various ramifications of this dichotomy, and speculate as to its cause. In conclusion, I suggest research into the process of debugging extensional class representations and the development of intensional ones.
This paper was prepared as the author's area examination.
MIT Artificial Intelligence Laboratory
MIT Artificial Intelligence Laboratory Working Papers, WP-263