David Marr (1945-1980) was a British neuroscientist who became one of the founders of computational neuroscience and a central figure in the theory of vision. Working at MIT, he integrated psychology, neurophysiology, and artificial intelligence into a new account of how the brain builds a representation of the visual world from raw light. He died of leukemia at the age of 35, and his ideas were collected in his book “Vision: A Computational Investigation into the Human Representation and Processing of Visual Information,” published posthumously in 1982 by MIT Press.
Marr’s most enduring contribution is a way of thinking, not a single algorithm. He argued that any information-processing system should be understood at three distinct levels: the computational level (what problem is being solved and why), the algorithmic level (what representations and procedures solve it), and the implementation level (how those are physically realized in hardware or neurons). As the MIT Press description notes, this framework “influenced a generation of brain and cognitive scientists.”
Marr’s three levels are still taught and argued over whenever people try to relate what an AI system does to how it does it - for instance, in debates about interpretability, where understanding a network’s “algorithm” is distinct from describing the task it performs or the circuitry that runs it. He is also one of the field’s poignant might-have-beens: a thinker of unusual clarity who shaped computer vision and cognitive science despite dying very young.