Pylyshyn, Z. W. (1978). Imagery and artificial intelligence. from C.W. Savage, ed., Perception and Cognition. Issues in the Foundations of Psychology, Minnesota Studies in the Philosophy of Science, vol. 9, Minneapolis: University of Minnesota Press) pp. 19-55

@InBook{,
  ALTauthor = 	 {Z. W. Pylyshyn},
  ALTeditor = 	 {C.W. Savage},
  title = 	 {Perception and Cognition.  Issues in the Foundations
of Psychology, Minnesota Studies in the Philosophy of Science, vol. 9},
  chapter = 	 {},
  publisher = 	 {University of Minnesota Press},
  year = 	 {1978}
  pages = 	 {19--55}
}

Author of the summary: Jim Davies, 2000, jim@jimdavies.org

Cite this paper for:

• The arguments in favor of analogue representations are weak, the structured-description theory is better.

• are constructed out of a vocabulary of available concepts
• has its semantics defined by an accessing function that is not assumed to be the entire visual apparatus
• possibly cannot be translated into sentences

• The memory of an image is not just degraded (e.g. blurred or pixelated). We do not remember complete images low in detail or precision. (p172)
• Perception is influenced by motivation, expectation, prior knowledge, etc.
• Perceptions include stuff that is not in the sensory data. E.g. causality, "to the left of" and an attack formation in a chess board.
• We can recall scenes by querying on the content of the scene. Therefore the scene is already interpreted in memory an not in need of re-perception.
• When we can't remember something from a scene, it is not like a part of a photo torn out, it is more a kind of perceptual attribute or relation.
• When kids are asked to draw what they saw after seeing a fluid filled beaker tilted slightly, they sometimes make the water level perpendicular to the side of the beaker. This can be simply explained by assuming that the children's internal vocabulary of descriptive concepts is limited or different from adult's.
• Go master's ability to remember Go board configurations depend on whether or not they think it is from a game of Go or Gomoku.
• To posit analogue representations is just to take one side of the data-complexity/processing-complexity tradeoff.

• Argument: (p180) Experiments like Kosslyn's show that determination of whether two images are of the same object take an amount of time linearly related to how much one would need to be rotated to transform it into the other. Though one could come up with a propositional description of how this could be done, it is unnatural. This result suggests an analogue representation. There is actually an image being rotated.
  Rebuttal:
  1. It there were a high degree of correspondence between operations in the world and mental operations one might perhaps be justified in speaking of the mental activities as being analogue. But the correspondence is partial: only certain aspects of some physical operations have correspondences. Too many transformations end up impossible to keep track of.
  2. Discovering even moderately novel readings from a mental image is very difficult.
  3. The reason an analogue representation seems natural is because we are using the whole of human perception to look at both the object and the thing representing it! If the representation is analogue, you need to re-perceive it every time you recalled it. Why push it under that rug? You're only postponing the perception part one level in. If the representation is to similar to the real world, it is uninterpreted, and what good would that do anybody? "The most natural representation (the picture in the head) requires a full-fledged homunculus for its interpretation." p186
  4. People ask "how can you take a data structure through its intermediate positions?" But having a naturalistic medium does not solve the problem-- how do you take an image through its intermediate positions? You can't get it for free-- how does the object retain its rigidity, for example? The laws of physics take care of that in the real world, but there are no such things for the mental entities. To do this we need to rely on operations that are no longer analogue. Then we lose the main attraction of analogue representations in the first place.

Summary author's notes:

• page numbers are from the vol 2 of "Readings in the philosophy of psychology" Edited by Ned Block.