Several people who read A Plan for Spam
have asked if I'm worried that Microsoft has already been granted a
on some aspects of Bayesian spam filtering.
I'm not. A patent doesn't mean
much until it is tested in court. Especially for something
like software, where the patent office regularly grants
patents for ideas that are not new at all.
Jason Rennie's ifile,
a Bayesian mail classifier, predates Microsoft's patent application by two years.
Pantel and Lin's paper
about using a Bayesian classifier specifically to filter spam also predates
the application by three months. The one novel idea I see in the patent is using
non-word features of the message (e.g. the arrival time) as if they were words
in a Bayesian calculation of spam
probability. But (a) this is an obvious idea to one skilled in the art, and (b)
you don't need to do this to make an effective Bayesian filter.
Even if the patent were valid, I don't think it would be dangerous,
because I think big companies apply for
patents mostly as a defensive measure.
Big companies apply for patents on everything that
comes out of their research departments as a matter of course,
more to protect themselves against patent suits than to
use as a weapon against competitors.
Like many big companies,
Microsoft wins by dominating distribution
channels, not by having better products.
Having a technical edge over competitors is not
critical to their business.
Patents are even less of a worry for free software. Even
Microsoft is constrained by public opinion. Can
you imagine the stink it would raise if Microsoft tried
to shut down an open-source project for patent infringement?
I've never heard of any company,
big or small, trying to shut down an open-source project
over a patent.
However, if you're worried about ideas being taken out
of circulation by being patented, the thing to do is
publish every idea you have as soon as you have it.
No one can patent an idea that has already been published
by someone else.
And if you want to start a startup and are worried about
getting caught in a web of patents, build a
That kind of project is far too messy and hands-on for anyone
to get very far into it in a corporate R&D department.