Slashdot Mirror
McAfee Granted Far-Reaching Spam-Control Patent
Titusdot Groan writes "Infoworld is reporting that Network Associates, makers of McAfee, have been granted a broad anti-spam patent. The patent covers "compound filters, paragraph hashing, and Bayes rules" and was filed in December of 2002. The patent appears to affect Spam Assassin, Spam Bayes and many other anti-spam products and services. As an aside Paul Graham's "A Plan for Spam" was published August 2002."
We were always encouraged to write ideas up as patents; lots of the people there received regular royalties or bonus payments from their personal patent portfolios, sponsored/owned by NAI. With the buy-out of SpamAssassin, I'm not terribly surprised at this news.
One tiny peeve, though: it's pronounced "muh'k AFF-ee ".
I believe the article is available online here, though right now it looks like this specific issue is kind of broken. It's called "An Essay towards solving a Problem in the Doctrine of Chances" anyway.
The Internet Archive of Early Journals is a great resource for 18th century journals and magazines. The Philosophical Transactions in particular are very interesting to history-of-science-minded science geeks everywhere.
Let's not forget that this kind of broad patent can be easily circumvented. You only have to do one single improvement over the patented method, and you no longer fall under the patent. It's not the idea of doing Bayesian filterting that can be patented, it is rather HOW it is done. You patent a specific implementation. Ideas cannot be patented, in Europe (don't know about the US system, but I would say it is similar in intent).
That's why people look at prior ART to dispute a patent.
I don't know about SpamAssassin, but I use SpamBayes and I know that their algorithm involves more than just a Bayesian filter. I doubt they would fall under the patent that was just granted, or it would be really bad luck, doing the exact same things the exact same way as it is documented in the patent.
Is there anywhare a statistic pf granted vs. denied patants by the PTO. Otherwise it seems save to assume the PTO is just rubberstamping ...
Has anyone thought the only way to combat this maybe to have the FSF start patenting things? I'm not sure of the cost, but at least it'll prevent evil corps from doing it first.
Daniel
aha! i just knew this book would come in handy one day! (The author is the late Georges Perec, who in 1969 took up the challenge of producing an entire novel without once using the letter "e." and was translated into english by Gilbert Adair, also without using an "e")
I'm curious... was she ever confronted directly about this ?
I.e. by a journalist or otherwise quoting their earlier statement about being vehemently against software patents, and then voting otherwise ?
I'd love to know what her explanation for this about-face would be.
I know that in practice this is what politics has come down to - pay lipservice to your voters, even if it means saying the opposite of what you said even just a few hours ago to a different group of constituents, but I'd still be interested.
Even if it were to hold up, the patent is useless. Spam filtering is a trivial application of text classification: given a piece of text, you classify it as belong to the "spam" class or the "non-spam" class. People have been doing text classification for decades and there are hundreds of methods for doing it. The kinds of naive Bayesian filters used by current anti-spam software are actually some of the worst text classifiers around (they aren't called "naive" for nothing). The fact that they work so well on spam shows you how easy the text classification problem actually is in this case.
If you want to see lots of other approaches, look on Google for "decision tree spam filtering", "svm spam filtering", "neural network spam filtering", "latent semantic indexing spam filtering", "boosting spam filtering", and "vector space spam filtering", to name just a few approaches. All of those methods are published, and NAI's patent doesn't read on them.
As for NAI's patent, I suspect it is actually fraudulent: the widespread use of naive Bayesian classifiers for spam detection, in place of better text classification methods, was a historical accident, and the fact that they patented this rather than any kind of better method strongly suggests to me that Bryson and Ekle didn't actually "invent" this, but that they applied for the patent after observing that the method was becoming popular.
Thanks for that link. That paper references this paper (PDF) by M. Sahami, S. Dumais, D. Heckerman, and E. Horvitz."
They seem to be the first Bayesian Spam Filterers. So if the patent belongs to anyone, its Microsoft and Stanford University.
Doesn't that make you feel better.
Athletic Scholarships to universities make as much sense as academic scholarships to sports teams.
The filing date of the patent is Decemeber 2002, so this Dec 2001 date you're coming up with is totally and completely irrevelevant.
Speaking as someone who is about to take the patent bar exam (so I acutally know what I am talking about) you're wrong.
Yes, McAfee would have to be the original inventors to get a patent. However, it's going to be hard, if not impossible, to prove they aren't. What can be done is show that the information on this invention was in the public domain prior to filing of the patent.
35 USC 102(b) provides that you can't get a patent on your invention if the invention has been published over a year prior to the filing of the application. That's where December 2001 comes from.
Paul's article may still be relevant. However it does not prove that McAfee didn't invent it first. If you really think Paul's article shows prior invention, why not submit a third party request for reexamination based on that article?
Oh, I forgot, you're just a slashtroll who doesn't actually know what you're talking about.
"I decry the current tendency to seek patents on algorithms.
There are better ways to earn a living than to prevent other people
from making use of one's contributions to computer science."
-- Donald E. Knuth, TAoCP vol 3
And yet this is somehow worse, the only algorithms are "use a combination of other people's algorithms" and "apply some tools to some task". Seems now that NAI owns "combining and applying"...
Perhaps I'll patent a method for "solving problems with obvious solutions" (which isn't done often nowadays in any event).
/* MAGIC THEATRE
ENTRANCE NOT FOR EVERYBODY
MADMEN ONLY */
Where I have a problem with all this story is that someone in McAfee had to have known that there was prior art in this area before the patent was applied for. I think that applying for a patent when prior art is known about should be a criminal offense and punishable by a court of law.
This patent needs to be fought.
Not a meaningful statistic (sample of one), but... My name is on four different software or software architecture patents. Of the four, two were denied on initial filing. Of the two that were denied, one was granted after several claims covered by prior art were removed, and the other was granted after writing a several-page submission that showed how the prior art cited by the patent office did not apply to our situation. Casual conversation with other people listed as inventors on patent applications made by the giant corporation where we worked suggested that my experience was not unusual -- a substantial portion of applications seem to be initially rejected.