David Courtnay Marr was a British neuroscientist and computational theorist whose brief career produced one of the most influential methodological frameworks in cognitive science. His 1982 book Vision, published posthumously after his death from leukemia at age 35, articulated three levels at which any information-processing system must be understood and applied this framework to a comprehensive computational theory of visual perception.
Three Levels of Analysis
2. Algorithmic: How is it computed? What representations and processes?
3. Implementational: How is it physically realized in neural hardware?
Marr argued that these levels are largely independent: the same computation can be implemented by different algorithms, and the same algorithm can run on different hardware. A complete scientific understanding requires analysis at all three levels, but explanatory progress at one level does not depend on having answers at the others. This framework liberated cognitive science from the demand that every psychological theory specify neural mechanisms, while simultaneously insisting that theories specify what problem the system is solving.
Marr's most distinctive contribution was the emphasis on the computational level -- understanding what a system computes before asking how. He argued that many failures in neuroscience and AI stemmed from trying to understand algorithms and mechanisms without first understanding the computational problem being solved. This insight directly influenced the development of Bayesian and rational analysis approaches in cognitive science.
Computational Theory of Vision
Marr applied his framework to vision, proposing that the visual system constructs a sequence of representations: the primal sketch (edges and textures), the 2.5-D sketch (viewer-centered depth and orientation), and the 3-D model (object-centered shape description). Each stage was analyzed at the computational level (what information is extracted and why it is useful) before specifying algorithms for computing it.
Legacy and Impact
Marr's three levels became the standard framework for organizing computational theories of cognition. Anderson's rational analysis, Shepard's universal law, and the Bayesian cognition movement all operate primarily at the computational level that Marr championed. His insistence on formulating precise computational theories -- specifying inputs, outputs, and the mapping between them -- established methodological standards for the field that remain in force.