Here are some patterns we have identified so far:
- The objective problem spaces of many tasks of scientific model
building involve searching within matrix spaces, typically subject to
conservation conditions or laws, and other constraints; simpler models
correspond to matrices of smaller dimensions.
R.E. Valdes-Perez, J.M. Zytkow, and H.A. Simon,
``Scientific Model-Building as Search in Matrix Spaces,''
Proceedings of National Conference on Artificial Intelligence, 1993.
- Our recent human/computer discoveries in biology, chemistry, and
physics have involved developing new representations of the scientific
task, not building expert systems models. This same pattern is
conjectured to hold elsewhere.
``Some Recent Human/Computer Discoveries in Science and What
Accounts for Them,''
AI Magazine, in press.
- Many discovery tasks in science are quite specific in the sense
that they involve a rather narrow type of inference, but are generic
in the sense that they arise in various scientific fields and similar
representations and computational approaches serve to automate them in
each of the fields. This proposition is grouped under the concept of
generic task of scientific discovery.