Note that one doesn't always need to carry out a complete interpretation
When crossing the road, it's usually sufficient to judge the contact
time of an approaching car from the apparent enlargement of its image. There's
no need to spend time working out its make or year of manufacture. In
other situations, more may be necessary - when cycling, it may be
useful to know which makes are driven by aggressive young men, so you can
- See also my tutorial notes
For examples of ILP, see Predicting secondary structure of proteins
with logic-based machine learning by Muggleton, King and Sternberg (AI
photocopy M166). Unlike many connectionist learning systems, ILP is
backed by principled statistical methods. It also has the great
advantage - as a tool - that the results of its learning are easy to
comprehend. Incidentally, AI will become increasingly important in
pharmacology. Worth bearing in mind if you intend to make a career in
There's also an article by Berliner in the AI Journal volume 1 for
1980, RSL open shelves.
For example, a position might be perceived as ``queen on open QR2-KN8
diagonal; knight can reach KB7 in one move; opponent's king on KR1 is
hemmed in at KN2 and KR2''. Features 1, and 2 or 3, might trigger
several possible countermoves, through associations built up by
This article describes the most famous cognitive model of chess skill.
The main references to it are Perception in Chess by Chase and
Simon, in Cognitive Psychology volume 4, pp 55-81 (1973). A
simulation of memory for chess positions by Simon and Gilmartin, Cognitive Psychology volume 5, pp 29-46 (1973).
Two pieces of research that suggest how this may one day be done. Analogy-making as perception by Mitchell (MIT 1993; PSY KH:M 069). And
Declarative learning by Furse and Nicolson in Proc 14th Cog
Sci Soc (PSY BH:C 065) and Mathematical expertise and the
contextual memory system by Furse (AI photocopy F57). To me, the second
seems particularly promising, as a method of organising what you learn
without making preconceptions about which features to concentrate on.
The sole attempt at this: Building large knowledge-based systems
by Lenat and Guha (Addison-Wesley 1990; PSY KH:L 054).
See An emerging paradigm in robot architecture by Malcolm,
Smithers and Hallam, AI photocopy 149.
Wed Feb 14 23:38:20 GMT 1996