One of the continuing problems in mining and making bets off of experimental data, particularly in the fields of finance, quantum mechanics, and to a certain degree inductive reasoning in genetics and proteomics, is evaluating probability based not just on a probability mass function that is event based, but also takes into account the belief of the certainty of the events.
So, when you are statistically describing a situation, you have to take into to account the validity of your instrumentation, context, and history....
In other words, the answer to the question: "What is the most important part of the message" is "The name of the messenger".
I've been evaluating belief and context based probabilities for the purposes of building a trust/evaluation network for my client - who of course promptly rejected it. More specifically, 2 statistical models have been interesting me:
Dempster Shafer belief networks
and
Contextual Probability Theory (CPT)
Both of these basically take bayesian propositions one step further, although bayesian is still deeply rooted in classical probability.
Getting back to trust networks, I put forward this slightly academic suggestion to my client for the purposes of building a more useful trust/FOAF network on their site, akin to tribe.net or friendster but actually possibly useful. In our revised friendster, you could follow chains of propositions about other people, and making evaluations not based on beleif of your friend, but also beleif ABOUT your friend :).
Posted by Da Mystik Homeboy at October 23, 2003 10:47 AM