The Environment as Theory: An Example Using the ACT-R/SOS Environment

One of the most useful and clarifying ideas in cognitive modeling is Anderson's (1993) schema for understanding cognitive models in terms of frameworks, theories and models. In this paper we propose that this schema is equally valuable for describing the environment that a cognitive model acts within. Physics provides us with the correct models for representing the physical environment. However, the interaction between cognition and the physical environment is mediated by transduction processes, which are not fully understood. Thus you cannot simply slap a physics model onto a cognitive model. One way of dealing with this is to assume a transduction model that delivers what the cognitive model needs. Another approach is to drop physics and transduction completely and model the environment as it appears to the cognitive system after transduction. which stands for Simple Object System, is a modeling system for creating low fidelity environments for ACT-R models to act on. It is meant to compliment ACT-R PM (Byrne & Anderson, 1998), which is a higher fidelity environment system for ACT-R, containing a much more detailed model of how we interact with the environment. The advantage of SOS is that it is easy to learn, quick to use, and very flexible (assuming you understand ACT-R). SOS v 1.0 allows the modeler to build a chunked representation of the environment that functions according to its own rules, built in by the modeler. Changes occurring in the environment are represented the same way they would be in the ACT-R declarative memory system. That is, by making new chunks available or changing the slot values of existing chunks (new chunks are not actually created but toggled from visible to not visible). Both the perceptual and semantic properties of objects are encoded as slot values. SOS objects also contain a special slot for actions. The action slot specifies a target (i.e., another chunk representing some part of the environment) and an action (i.e., how to change the chunk structure of the target). It is also possible to insert lisp functions into the action sequence to create more complex effects. Thus SOS models how the environment works, but represents the environment as how we see it rather than how it is, physically. SOS fits well into Anderson's (1993) schema of frameworks, theories, and models. SOS is based on the framework assumption that the end function of transduction is to create a mental model of …