Why representation is needed

A reasoner is situated in the world; in order to survive in their environment, they must be able to make decisions about that world. This does not automatically mean that a reasoner will require the building of internal representations which describe the world.   For example, there is a subfield of AI and robotics known as   reactive reasoning--research there has indicated that reasoners can do quite well at interacting with the world solely by reacting to the environment (see, for example,       [#!Arkin-89!#,#!Brooks-86!#,#!Payton-86!#]). Another approach which has often been utilized is to have the reasoner build internal representations. As with the reactive technique, this explicit representation approach allows the reasoner to interact with the world. Additionally, it is possible to perform long-term planning and higher order cognitive activities such as reflection. This explicit approach is the one the ISAAC model incorporates.

There are certainly aspects of reading which can be described in a reactive fashion; one possibility would be the allocation of system-wide resources based on changing levels of comprehension coupled with other constraints on the reasoning system. In such a position, a reactive reading controller might be the best for the speed it would bring to the re-allocation of sparse resources. However, the vast majority of reactive research has concentrated on a more traditional agent interacting physically with the world. The ISAAC project concentrates on an agent interacting mentally with the textual world. It appears unclear that many of the benefits which a reactive approach brings would, therefore, be realized by a reading agent. If we assume that an internal representation is needed, the first question to address is exactly what to represent. In other terms, what entities exist in the world which would need to be reasoned about in such a fashion as to make an internal representation desirable?