Subject: Re: Novel anti-software-patent article
From: Lynn Winebarger <owinebar@free-expression.org>
Date: Thu, 6 Jan 2000 20:26:38 -0500 (EST)

On Thu, 6 Jan 2000, Crispin Cowan wrote:

> Speakinf of RSA & co., is it purely coincidental that all the really useful public
key
> algorithms were developed after algorithm patents?  Are we really sure that people
will
> continue to invest in crypto algorithm research motivated by fame, glory, and
> first-to-market?
> 
    We'd have to ask the inventors about that.  My guess is that, yes, it
probably was coincidental.  The emerging prominence of networks at the
time was probably more of a motivating factor, I would think.  
    As long as we're on the subject, I thought the Benson decision was the
flag to _not_ allow patents on software.  The decision I usually see cited
is Diet-something-or-other, where a process for curing rubber that
included the use of a computer program to control the curing was patented,
and the patent upheld.  To me, that kind of patent seems like it could be
within the realm of reason, because in that case the algorithm isn't
patented per se, but only the application of it (and a fairly "hard"
application it seems like to me).  The real question is one of what patent
lawyers call "broadness", which I would prefer to call "abstractness" or
"meta-ness", as it seems to me to better capture the problem.  That is, at
what level of abstraction should a patent not be allowed.  Right now, it
seems that the level is far too high.  But that is not the same as saying
that there is no level at which patents wouldn't be appropriate. 
    Right now, the guidelines for algorithm patents allow for anything
that has any relation to the physical world, or serves "any useful
purpose" I believe.  This means any idea where an abstract algorithm is
thinly veiled in an application (by displaying results on a screen, or
taking in some sort of real-world data) becomes eligible.  I don't claim
to know where the line should be drawn, but this is just ridiculous.  Even
the most abstract mathematical algorithm can be dressed up in an abstract
application (for example, an algorithm for computing the coefficients of a
data set with respect to a given orthogonal basis in a Hilbert space is
very easy to dress up in a variety of fairly abstract applications - say,
the lossy compression of images, sound, or video).  It's disturbing
because, in a real sense, these mathematics (and associated algorithms)
describe the structure of information, in the same way that the
mathematical constructs physicists choose describe the structure of the
physical world.  That is, physicists develop (or borrow) particular
mathematical structures precisely because the correspond well with what
they see in nature (or, really, the lab).  In the same way, computer
scientists (and statisticians, for that matter) develop (or borrow)
mathematical structures precisely because they correspond well with the
information they look at.  For example, graphs for programming language
specialists or harmonic analysis for compression of sampled continuous
data. Hence, to me, patenting an algorithm on processing video information
starts looking suspiciously like patenting a law of nature.
     Ok, I'm rambling.  It's because I'm more confused about the
legitimacy of "software patents" than I used to be (because I'm not sure
where exactly a patent stops claiming only an algorithm, and where it
starts claiming something I would normally consider reasonable that uses a
computer program - just as 7 of Morse's patents on telegraph-related
inventions were considered reasonable because they just _used_ EM's
ability to affect change at a distance).

Blah blah,
Lynn