This is certainly not essential for Finals. But if you're serious about AI, you should read this section: the papers below represent work that's unlike almost anything done elsewhere, and they are not easy to get hold of (I've put copies in the library). They may be the wave of the future.
Rumelhart and Evans' programs are both inflexible. The properties by which they're to categorise experiences (i.e. the representation) are given in advance, unchangeable by the programs. In most problems, choosing these properties is by far the biggest task. Try to read at least the first reference below to convince yourself that this is so. If you feel keen, go on to the others.
Start with Chapter XIX of Gödel, Escher, Bach by Hofstadter (in the Psychology library). Look for references to the Bongard problems and note the range of different properties that can characterise solutions.
Next, read Chapter 24 of Metamagical Themas, also by Hofstadter. If you can't find this book, go to Scientific American September 1981, from which the chapter is is reprinted.
How might we write a program to solve letter problems fluidly? Now go onto the Hofstadter papers. These are in the Psychology library in the ``AI Box'' in the Photocopies section.
item Copycat: a proposal for computer analogy solving [?] by Mitchell (Centre for Research on Concepts and Cognition). Photocopy number M143.
The final one surveys a number of other attempts at modelling human analogy solving.