[MG] Voting advice application

[Michael Allan]

Alex Keskitalo wrote:
> If you told a finnish person that in other countries they don't have
> the voting advice app ("vaalikone") ecosystem, they would be
> surprised and wonder how you could vote blindly like that...  "what
> are the issues?" and "what are the solutions?" ...  This generates
> data, and lots of it...  What kind of data do you guys have to work
> with?

We are not very data centric, so I have trouble answering.  However
you appear to offer something more than data.  You had Finns wonder
"how [we] could vote blindly like that", which indicates that you
offer a new kind of "voting eye".  Is it correct to understand your
question on those terms?

> > I think so.  The larger the sample, the tighter the confidence
> > interval on the prediction.
> 
> Users implicitly (explicitly?) sort the candidates by rating their
> answers, so they end up with a "best candidates for me" -list
> anyway, with or without the predictive algorithm.

I see.  There is another kind of prediction involved:

> But the predictive algorithm helps to make the process easier,
> faster, etc for the user, by giving the user the most relevant
> things they need to rate, on a personalized level.
>
> so as you have rated stuff in a certain way according to your
> personal politics, the algorithm compares your political fingerprint
> to everyone else's, then it comes up with your closest political
> allies, searches ratings they have made to find a candidate's answer
> you should like, shows it to you, you rate it, algorithm runs again
> with new data, etc.

Based on the behaviour of other users it somehow predicts the
questions that I (as a user) ought to be rating - the ones that would
offer the most traction - in order to improve the prediction of my own
choice of candidate.  I think I understand, roughly.

> of course that's just the generalized concept. under the hood the
> complexity is much higher, multiple parallel mechanisms are at work,
> with a self-regulating control layer on top, and that has
> hand-tunable parameters for stuff like how strongly should the
> individuals opinion be reinforced or challenged, higher order stuff
> like that, not perfect but helpful when you have to adapt to
> changing group dynamics as the numbers grow..

Yes, I see there is some positive feedback to be tuned.  The
prediction should not become a self-generated artifact of the
mechanism.  OK.

> > What will the voting feature be like?
>
> The idea is that the user narrows down their candidate choices until
> they have a winner, then we have them confirm the result.

What happens if they do not?  Suppose they decide to vote for a
different candidate.

> In theory this should guarantee a machine-measurable level of
> confidence in the macro-result...

Could you elaborate?  Any assurance of correctness would seem to
require an independent, external check on the mechanism.

> when an election is over, the generated user population is left high
> and dry, such a waste :D

True!  Although the atmosphere is very different after an election
than before.  This may have some effect on the users.  Will they still
be asking the same question, "Who should I vote for?"

> > Thank you, and likewise!  Tim Bonnemann recently observed that
> > e-democracy is "niche and fragmented", and I couldn't agree more.
> > Anything we can do to remedy it would be step forward.
> 
> Agreed, convergence as a macro-strategy.

I guess there are different kinds of convergence.  Other strategies
might be viable too, depending on the situation.  Do you have
something concrete in mind?  (I do, but am more interested in hearing
your own ideas.)

-- 
Michael Allan

Toronto, +1 416-699-9528
http://zelea.com/



Originally posted to the mailing list of the Metagovernment Project:
http://metagovernment.org/mailman/listinfo/start_metagovernment.org