The Non‐Redundant Contributions of Marr's Three Levels of Analysis for Explaining Information‐Processing Mechanisms


A favorite hobby-horse of mine.
I do think that it is very useful to use Marr’s three levels in stating both a problem and solutions.


There is something about that magic number three, same as in canonical “introduction, body, conclusion”.
Probably because majority vote doesn’t work with two.
Otherwise it’s arbitrary, subdivisions in the explanation should reflect those in the subject matter.


The original exposition in Marr’s book “Vision” makes a clear and compelling case for the three levels.
I strongly suggest that this book should be in any experimenters library.
Actually reading it would be a plus.


I skimmed linked article, it doesn’t look clear and compelling to me. Marr’s is a distant memory, but I can think of no reason why there should be exactly 3 levels, regardless of complexity of actual subject.
The subject is king.


Perhaps this will help?

In chapter 1.2 of Vision , David Marr presents his variant on the “three levels” story. His summary of “the three levels at which any machine carrying out an information-processing task must be understood”:

  • Computational theory : What is the goal of the computation, why is it appropriate, and what is the logic of the strategy by which it can be carried out?
  • Representation and algorithm : How can this computational theory be implemented? In particular, what is the representation for the input and output, and what is the algorithm for the transformation?
  • Hardware implementation : How can the representation and algorithm be realized physically? [Marr (1982), p. 25]
    Found in this paper:


Yes, I remember that, and I think this subdivision is arbitrary.


Perhaps it is.
I still find it very useful where reading some paper to try and place what I am reading in context. These groupings seem to be good dividings lines. This provide me with a useful frame to parse and understand what is being communicated.

This leverages the well known principle of the memory palace.

I suppose that you are correct that some other grouping could provide the same function.