At 7:49 AM +0200 4/3/08, Beerends, J.G. (John) wrote:
A good theory is never wrong, it only has a limited context in which it
is valid.
But a bad theory can be wrong. For example, if one had a theory
based on noncoherent detection, saying that in certain experiments a
human's time resolution would be limited by the uncertainty principle
in relation to the signal bandwidth or frequency resolution, and that
theory predicted such limitations in situations where human time
resolution is based on fine time structure, rather than on energy
envelopes, that would be bad. That why it's important to ask whether
humans outperform theory: it can give you a good clue as to whether
your theory is not so good.
When you're in such a situation, it's very hard to build a machine
that will outperform a human, because you don't have a theory that
fits how well humans perform. That's one way to look at our
difficulty so far to build machines to do a lot of things that human
do well, like recognize words, talkers, sound sources, etc.
Experiments help you determine whether you're in such a situation or
not, even for very much simpler tasks.