|
John
FYI - you probably know it already. For astronomers in India to work, two invisible planets called RAHU and KETU were hypothesized and then only they could predict eclipses or meridian crossings of stars.
On Sun, Sep 6, 2009 at 1:51 AM, John F. Sowa <sowa@xxxxxxxxxxx> wrote:
Rich,
RC> Although the interesting algorithms aren't that well defined -
> they're evolving like stem cells and continue to this day.
I shouldn't have used the phrase "well-defined algorithm(s)"
because it is irrelevant whether you have a precisely defined
algorithm or a bunch of heuristics to do the data mining. Any
method, formal or informal, that is limited to finding patterns
in a fixed set of data cannot distinguish a law from a coincidence.
Furthermore, it's irrelevant whether the pattern finder is a
computer program, a human being, or an extraterrestrial alien.
In terms of scientific method, any pattern found in any given
set of data is a *hypothesis*. Before it can be considered a
scientific law or theory, it must make testable *predictions*
about new cases that have never previously been observed.
Bode's law about the distances of planets from the sun was
originally formulated by J. D. Titius on the basis of the planets
from Mercury to Saturn. When Uranus was discovered, the formula
had to be adjusted. But then Neptune completely destroyed the
pattern and showed that the so-called "law" had no predictive
power.
That is an example of "data mining" performed by humans based
on a fixed set of data. The pattern they found was shown to
be a coincidence that had no predictive power.
JFS> Those patterns might be the result of fundamental laws,
> or they might be accidental patterns that could be violated
> by the next update to the database.
RC> Or they might be bound to the conceptualizations in the observer's
> cranium, whether fundamental or preaproved or officially not.
The issues have nothing to do with the nature of the observer.
It is irrelevant whether it's human, alien, or intelligent computer.
The question is whether the agent is limited to finding patterns in
a fixed set of data (i.e., data mining) or whether it considers the
patterns to be tentative hypotheses to be tested by their predictions.
Scientific method depends critically on testing hypotheses to
determine whether they can make reliable predictions. And the
nature of the agent is irrelevant.
JFS> Some additional analysis and testing is necessary to
> distinguish principles from coincidences.
RC> Recursion does that very nicely. If you can learn it once,
> you can learn it N times.
No. Recursive methods limited to a fixed set of data can't
do anything to distinguish a law from a coincidence.
The critical issue is whether the hypothesis derived by data
mining from a given set of data can make testable predictions
on new data. And the more varied the circumstances in which the
data is obtained and verified, the more reliable the hypothesis,
and the more likely that it indicates something real, not just
a verbal (i.e., nominalist) formula.
John
--
Thanks.
Ravi
(Dr. Ravi Sharma)
313 204 1740 Mobile
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (01)
|