Chronology Current Month Current Thread Current Date
[Year List] [Month List (current year)] [Date Index] [Thread Index] [Thread Prev] [Thread Next] [Date Prev] [Date Next]

[Phys-L] Re: model vs. truth



I am not arguing that I have developed an efficient means of finding truth, only that it exists, proven by the fact that we desire "better" models.

Justin Parke
Oakland Mills High School
Columbia, MD


-----Original Message-----
From: John Denker <jsd@AV8N.COM>
To: PHYS-L@LISTS.NAU.EDU
Sent: Tue, 31 May 2005 14:56:23 -0400
Subject: Re: model vs. truth


Justin Parke wrote:
Some were reluctant -- in the recent thread on religion and science
-- to use the word "truth" to describe the goal of science,
preferring the word "model" instead.

Once Mechanics and Electricity and Magnetism were "nailed down" (I am
told) that physicists believed they had solved the universe, then
quantum mechanics came along. So it is understandable that
physicists would be more cautious now in declaring that they have
found truth since their theories are always subject to change as new
data is revealed.

Being cautious about Truth goes back waaaay before the
advent of quantum mechanics. Off-the-cuff citations
include Lao-Tzu (~500 BC) and Plato (~400 BC).
<<They see only ... shadows ... on the opposite
wall of the cave.>>
<<The Way that can be trodden is not the true Way.>>

But isn't this a limiting process? That is, as theories make more
accurate and more general predictions we consider them better than
older theories which made less accurate and less general predictions.
Isn't the limit as t goes to infinity truth?

Maybe. But the asymptotic result is oftentimes a poor
guide to the everyday result. I run into this time and
time again in the machine-learning business: some guy
announces that he has invented an algorithm that is
"asymptotically optimal". Yeah, whatever. The literature
is full of asymptotically optimal results. Parzen Windows
is asymptotically optimal ... you just have to feed it
an infinite amount of training data. Alas I don't have
an infinite amount of data on hand, and I don't have
infinite computer resources or infinite patience ... and
more-or-less all of these asymptotically optimal algorithms
are horribly inefficient and ineffective in the non-asymptotic
regime.

http://www.google.com/search?q=parzen-windows
_______________________________________________
Phys-L mailing list
Phys-L@electron.physics.buffalo.edu
https://www.physics.buffalo.edu/mailman/listinfo/phys-l