Preamble

Before you start on Evans' program, I want you to read a bit about vision: specifically, some work by David Marr. Here then are the first two references, and my comments about them. You will find both books in the Psychology library.

1. Vision by David Marr (CH:M 34). Please read the Introduction and Chapter 1.

2. arr's article on Visual Information Processing in The Psychology of Vision (BC:R 81). This paper is reprinted from Phil Trans Royal Society London, volume B 290, pages 199-218. But the Psychology of Vision is easier to find, being in our library (oversize books section).

To give a feel for the psychological importance of representation, I could have started with one of a number of topics: language understanding, game-playing, planning, ... But I have begun with vision. Vision is obviously a psychological process: if you can get a feel for why one of our best vision people decided AI and representations are important, it may convince you of their relevance to other parts of psychology. (And you can also decide whether you want to take the vision option of the AI paper.) It is also very likely that Marr's view of AI and representations will come up in Finals.

Don't, at this stage of the course, try too hard to understand the technicalities. Read both references fairly quickly, and look especially at the historical context --- treat it as a bit of scientific biography. The references both discuss what Marr considers the important pre-AI work, and its good and bad points. What were people doing before AI? What did Gestalt psychology and neurophysiology contribute? Why did Marr eventually decide the neurophysiology experimental findings were not, on their own, enough of an explanation? Why did he take up AI?

Having done so, how did he try to explain vision? What were the stages he hypothesised? Now you can look over his model of vision. This is summarised in a table on page 37 of his book, and described over 4 or 5 pages in his paper.

What was the output of each stage? What's the difference between the primal sketch and the 2-and-a-half-dimensional sketch? Why are both necessary?

Now, becoming more specific. What does Marr mean by a representation? That's an important notion in AI, and one which, in the rest of this tutorial, I use Evans' Analogy program to illustrate. But that program alone might seem too divorced from real psychology: hence my starting with vision.

What representations were there in Marr's model? According to Marr, a representation makes some information explicit at the expense of other. What information is explicit in the 2-and-a-half-dimensional sketch that isn't in the primal sketch?

Now you can go back to Chapter 1 of his textbook. In general, what does Marr think representations are? Why are they important?