Well, it happens that I also think that this could be an interesting approach to a social news website. It could also help us avoid the inevitable problem of large user bases and abuse from more vocal (annoying) minorities - such as Ron Paul supporters.
There is also one more problem that I'd think could be addressed: language. I'd like to see a website where the user could have links in any language that s/he can speak or read. When reddit was still interesting, half of its user base was non-american. I bet that more interesting people would hang around in reddit if they could also send links in other languages.
So, with that in mind, I started working on a side-project, and it is precisely what inovica is describing: a social news website that could use Bayesian filtering to score links for each user. Check it in http://news.communick.com. For lack of a better name, I'm calling it SmartNewz.
I just started, and it is only a side-project. The site, as it is, was done in a weekend. I hope I can also use this as a means to learn how to develop applications for facebook and OpenSocial. But those are only ideas.
The part that I believe is interesting is open sourcing the filter, as well as providing an API for those that want to implement their own filters.
So, if anyone is interested in such a service (or if you indirectly want to help disqus have more users), please, register at the website, check the filter project repository at http://code.google.com/p/smartnewz, or the trac wiki for more information, in http://trac.communick.com/news.