Various people have been discussing the "attention problem" and attention.xml. The basic idea is that the advent of RSS means that we have too much information competing for our attention, and that we need a way to record "attention" data so which could then be used to perform triage of information to be presented to a human user.

The notion of automated triage of information has been around for a while. To an old Usenet junkie like me, the information overload problem and the need for tools to help triage the flow seems like a no brainer. I remember when s(coring)trn newsreader came out (as a set of patches to trn), and when Gnus, the mother of all news readers made its debut. Gnus was one of the first platforms for collaborative filtering, GroupLens, which then went on to become NetPerceptions (which now appears to be defunct). A lot of what is being discussed feels familiar, in concept if not in actual implementation. So what's new here?

Well, strn and Gnus were not instrumented to record attention data. Although I have some doubts about the accuracy of some of the data (like how long did a user read a post), Steve Gillmor seems quite excited about instrumenting clients (like Firefox) in order to obtain this data. Assuming that you could gather meaningful (or mostly meaningful) data, this seems like a good source of data for triage.

Next, you have the notion of distributing / combining / syndicating / bartering / selling the attention data. If you want to do this, having a standardized format for encoding that data seems good, and since adding XML to data always makes things better, you get attention.xml. This part is easy, and most people agree on that. There's a question about where attention data lives and who gets access to it. This is an important question, at least to me. If people are unscrupulous about the (comparatively) small amount of information that I give out about myself, what will happen when they could get their hands on a detailed model of my attention? The thought of storing my attention data in somebody's VC backed server farm doesn't give me the warm fuzzies.

After that, we get to actually using the attention data to perform triage, which is where there is room for experimentation, variation, and market based competition (at least if you believe in exposing your attention stream). Here's where you get into scoring, bayesian filtering, collaborative filtering, reputational filtering and so forth. It's also where you have to deal with issues of granularity, i.e. single posts versus conversations. It's also where you get into potentially innovative presentations as well.

From where I sit, my attention data, my reputation assignments, my triage preferences and so forth are all part of my personal information, and would be something that I would like a personal information manager to manage. I think we have one of those lying around here somewhere...