I was just discussing with my friends the utility of setting up mechanisms for electronic poll-taking, so that elected officials - or anybody else - would have ways of finding out opinions of various social groups and population as a whole on various issues.
It may be useful to set up a service where one could offer any question for an opinion poll, suggest possible answers, and receive statistics that can be broken down by various social factors, as well as free-form comments. The service could also have mechanisms for identification of respondents and verification of their profiles.
One problem with such a system, as well as existing electronic polls, is that they are "not representative" - that is, while reflecting the opinions of people who are taking the poll, the statistics do not necessarily reflect the average opinion of the population. (One may doubt though that diluting the opinions of active people who understand the Web with those of people who don't care about the issues enough, or can't figure out how to click on buttons, would result in better decision-making).
This is also a problem with physical voting: people who show up at the booths are in many ways different from people who don't show up, and so the votes are always, to some [unknown] degree, non-representative.
Interestingly enough, this problem with representation can be
alleviated with electronic votes.
I would expect that, with a set of several opinion polls conducted both electronically and physically, one could train a neural network to quite accurately predict the opinions of people who would show up at the physical polls from the opinions of people responding to the electronic questionnaire. Of course, the mixing function for the votes will be different from arithmetical average, but it the point is, it will work.
This can be taken further.
Similar statistical methods can easily predict the opinions of people who would _not_ show up at the polls as well, and so they can represent a much more accurate indicators of the actual public opinion on a given subject - and with a lot lower cost, as the price of the software would be offset by not having most people show up at the polls.
One could also predict what kind of people would show at the polls, and who wouldn't.
A real indicator of whether a certain decision would serve the interests of the public, would be the opinion of the public about the issue _after_ the decision has been taken and brought some results. This would require some clairvoyance on the part of the public, though. Currently, people hope that arithmetical averaging of votes before the decision is the best possible way to predict their opinion of its results. This expectation seems ungrounded and most likely untrue. A learned statistical function may be able to predict future opinions of the general population from relatively small sample polls much better than global pre-decision averaging, and thus better represent actual interests of the public.
The question is, is the society really interested in finding better ways to realize people's interests, or it is just perpetuating existing decision-making power structures?
Maybe, somebody could set up such a service, work on the prediction functions, and then the results would speak for themselves?
This polling service may be combined with some idea futures functions as well as reputation brokering and collaborative filtering.