Marr Levels

David Marr, a British neuroscientist and psychologist, explored the complex information processing system of the brain and initially focused on vision. He integrated results from psychology, artificial intelligence and neurophysiology into new models of visual processing. Marr (1982) argued that there are "different levels at which an information-processing device must be understood before one can be said to have understood it completely” (p. 24). Although he was not focusing directly on the same types of behavior, Marr's investigatory outlook is reminiscent of that of Tinbergen. He explains that "the effort is to show that in principle the microscopic and macroscopic descriptions are consistent with one another...[and] one must be prepared to contemplate different kinds of explanation at different levels of description that are linked” (pp.19-20). He broke down his explanation into three distinct, but complementary levels that are now known to many cognitive scientists and other researchers as Marr's levels of analysis, or his tri-level hypothesis (Marr, 1982).

The first level of explanation is one at the computational level. This level asks what the system does, what the goal is and what problem it solves. Once the overarching goal of the system is known, the next computational question is why. Commonly, it is a difficult task to separate the two computational concerns of what the purpose of the system is and why it does what it does. A full computational explanation would answer why the system does these things and the logic of the strategy it carried out.

Although much of Marr's work was on visual processes, many of his general hypothesizes can be applied to other mental phenomena such as information processing and intelligence. One of the "underlying tasks" of the mind "is to reliably derive properties of the world" from sensory information and decipher "constraints that are both powerful enough to allow a process to be defined and generally true of the world” (Marr, p 23). He argued that the performance of our brain, which he often refers to as an "information processor" is characterized by mapping one kind of information to another. This view has been generalized and accepted by many cognitive scientists; the role of the brain is to take in sensory information and create a reliable model of the world.

The second Marr level is at the algorithmic, or representational level. After understanding a system at the computational level, an algorithmic explanation describes how the system does what it does. A proper algorithm would address how the computational theory is implemented; specifically, what processes does it employ to build and manipulate representations. Initially, one must determine a way of defining and representing the input and output of the system. In most cases, there are various choices for the representation of the input and output. Additionally, one must define and formulate the algorithm for the transformation from input to output. Usually, there are several possible algorithms for carrying out the same process; this is especially apparent when various modes of representation of inputs and outputs can be employed (Marr, 1982). Finding an algorithm for which the transformation may actually be accomplished is not a simple task with a complex system such as the brain. Our intuition would tell us that the brain takes in sensory input and outputs behaviors, thoughts and actions. However, this focus on output may not capture the full nature of the brain.

This idea of an intelligent system being defined as exhibiting intelligent behavior has been a cornerstone of Artificial Intelligence research since Alan Turing's work in 1950 on "Computing Machinery and Intelligence." In this work, Turing introduced his Turing Test, which was based upon an imitation game. Basically, if a machine were indistinguishable from a human being solely on the basis of written interactions it would be considered intelligent. Turing believed he could prove that machines can think if they could

Clearly, this is not a sufficient test for intelligence; an intelligent being, such as a human, can be intelligent without any output of behavior. A human with their eyes closed, locked away in a dark, silent room not interacting with anyone or anything is still intelligent. Turing's test of intelligence was based on the end result of an output behavior, which is only one part of intelligence; the test did not address the process or architecture. This interpretation of intelligence being defined strictly by the ability to exhibit complex behavior has stunted our ability to properly explore intelligence and the functioning of the brain. Systems are process-oriented and cannot be defined simply by their output.

Marr's third level of analysis is one of hardware implementation. This analysis would address the physical realization of the computational theory and algorithm associated with the system. A detailed description of the architecture, structure and mechanism would be needed to understand the system at this level. This level is similar to Tinbergen's question of causation. In a biological system, one must discover what neural structures and neuronal activities are implemented by the system. This level lends itself well to the study of neuroanatomy and neuroscience, which explore synaptic mechanisms, action potentials, and inhibitory interactions among other biological occurrences. A reliable conception of the underlying mechanisms is necessary but is not sufficient to fully comprehend the system and its associated phenomena.

Marr warns that a correct explanation of a psychophysical observation must be formulated at the appropriate level. In a simplified metaphor, Marr (1982) explains that "trying to understand perception by studying only neurons is like trying to understand bird flight by studying only feathers: it just cannot be done.” He continues in his description of vision by explaining that "we can understand how these cells and neurons behave as they do by studying their wiring and interactions, but in order to understand why the receptive fields are as they are - why they are circularly symmetrical and why their excitatory and inhibitory regions have characteristic shapes and distributions" we must draw upon the study of differential equations, filters, signals, optics, mathematics and other disciplines that at first may not seem related to anatomy (Marr, p. 27).

Additionally, some phenomena may only be explained at only one or two of the levels. However, he stresses that the different levels of description should be "linked, at least in principle, into a cohesive whole, even if the linking of the levels in complete detail is impractical," they should remain "logically and causally related." Marr advises that in attempts to relate psychophysical problems to physiology, too often there is confusion about the level at which problems should be addressed. However, "each [level] has its place in the eventual understanding of perceptual information processing” (Marr, p. 25) and the brain-mind connection.