In our own work, we found a curious phenomenon: When we suggest
hypotheses to subjects they are much less likely to believe them, and much
more willing to reject them, than when they generated the very same
hypotheses themselves. The motivation to prove the other person wrong
and to prove oneself right is very strong.

Much of the best work in machine discovery is based on the historical
record of the great scientists making the great discoveries. But we have been
very selective in extracting information from those historical accounts. Such
accounts are often filled with statements about excitement, astonishment,
disappointment, envy, doubt, despair. Are these descriptions of emotional
and motivation states irrelevant to understanding science? I doubt it.

Why does it take so long to train a scientist? Is it all due to the slow learning
rate of humans and the huge amount of content knowledge and specific
techniques necessary to work in a field? Why don't we start earlier then? Is
it because we can't? That is, because the kind of domain-general search
constraints are simply not available to young children? That is what the
results reported here today imply, but we have a lot more work to do before
we really understand the nature of these cognitive limitations.

All of the discoveries I have been talking about so far are discoveries about
things "out there": discoveries about devices, or about the planets, or about
the kidney. What about discovery "in here"? When my colleague Bob Siegler
( Siegler & Shipley, 1995) talks about discovery, he is talking about how
children discover new strategies in doing arithmetic, or solving problems, or
playing games. To what extent is what we have learned about discovery
processes in the first sense relevant to discovery processes in the second
sense? Is self-awareness of one's own discovery processes a useful skill for the
scientist who is attempting to discover something about the world? Do the
same heuristics and search constraints apply? It is clear that search in a large
space faces those who would discover more about discovery, and our
challenge is to see whether we can effectively constrain that search as we seek
to discover discovery systems.

Sections of this chapter are based on research done in collaboration with
Kevin Dunbar and Anne L. Fay, and supported by grants from the National
Institute of Child Health and Human Development (R01-HD25211) and
the A. W. Mellon Foundation.

-352-