The computational level is the first level of Marr's (1982) tri-level hypothesis.
The computational level of analysis describes both what problem is being solved by the information processing system, and why the problem is being solved. It is a statement of the system's competence and defines the types of functions the system can compute. In other words, the computational level is simply a description of the input-output behaviour of a particular system. Traditionally, analysis at this level has been the domain of philosophers, behavioural psychologists, linguists, and computer scientists.
By considering the goals of an information processing system--specifically how to capture the properties and events of the world that are significant to the information processing system at hand--we develop theories that show how a reliable and accurate representation of the world can be computed.
The computational level of analysis is necessary but not sufficient when describing an information processing system. If two information processing systems have the same computational level description, then they are considered weakly equivalent systems (Dawson, 1998; Pylyshyn, 1984). Cogntive science, however, requires more than weak equivalence. A strongly equivalent model is one that matches a human subject at the computational and algorithmic levels of analysis, and instantiates its algorithms using the same functional architecture.
- Dawson, M. R. W. (1998). Understanding Cognitive Science. Oxford, UK: Blackwell.
- Marr, D.(1982). Vision. San Francisco: W. H. Freeman.
- Pylyshyn, Z. W. (1984). Computation And Cognition. Cambridge, MA.: MIT Press.