Thom Lawrence

Steve Eichert talks about Bayesian filtering as more than just spam prevention:

What would be cool is if this could be built into some sort of blogging engine, or content management system. I always forget to set the categories when I’m posting my entries and going back and updating the assigned categories is something that simply doesn’t happen. If my blog engine could do this for me automatically it would be super sweet!

I don’t know why most initiates into the Bayesian Conspiracy concentrate so hard on something as dull as spam, so I like Steve’s idea. But you can go much further - imagine how fun it would be to have a system that combined:

1. A public repository of categories.
2. A system for communities to train the engine to identify a document’s category. Perhaps Attention.XML’s tag attribute.
3. A way to track conversations across the Blogosphere.

If you have the set of all categories to file a document under, a tournament algorithm’s probably not going to help. We’d need a new generation of PubSubs and much smarter newsreaders to help identify the subset of the Blogosphere we want to listen to.

But build all of that and you get some cool stuff out:

• Notifications, not just of posts that match our interests, but whole conversations.
• ‘If you’re interested in that, you might be interested in this’ style recommendations of authors, tags, posts etc.
• Attention gaps! I don’t know why I love this idea, but say your shiny new Bayesian engine is putting documents in all the right categories, and taking the next step to make assertions about what
  subjects authors are interested in. And maybe Attention.XML or plain ol’ PageRank lets us say that person’s an authority about that subject. Then a conversation happens and that person’s nowhere to be seen. If Tablet PCs start setting on fire and exploding, and Robert Scoble doesn’t have a comment, I want a big gray hole somewhere telling me that there’s a gap in the conversation. And I’m sure Robert would want to be told ‘this conversation’s begging for your input’ too.
• Grudge matches! I’d love to be able to teach the engine that X person has positive feelings about Y technology, but Z person absolutely hates Y cos it had sex with her cat. When a conversation happens involving the two, I want a bell to ring in my newsreader and for it to blare out ‘LET’S GET READY TO RUMBLE’ (PJ & Duncan style).

I want all of this and more. RSS is just a better way to read the web, because it’s on your terms. But at the same time, it’s like inviting people into your home. I’ve never thought of my browser as mine but my newsreader’s my turf, and if I’m inviting all these people in I want everyone to feel at home. Using Bayesian reasoning lets you break out of the binary this-feed-that-feed, this-category-that-category dullness and enter a world of fuzziness, implications and guesses. Which, to me, sounds much more fun.