3 ms·

Request for product: Ask me ~100 questions, and then tell me how to vote

5 points by arikr 10y ago

I'm expecting backlash against this. Even with the downsides, I still believe this would overall be a force for good. I'd want to know how to vote on the entire ballot

arikr 10y ago
So ask me a series of questions to determine what I care about and in what magnitude, and what my incentives are, and then tell me which candidates and propositions align with my incentives Specifically would like this to exist for all votes, so should get my location so you know my exact ballot If anyone builds a good version of this I have a few decently high traffic websites and I will happily link it on them.
- nerdponx 10y ago
  You would need some kind of labeled data set -- hundreds of people answering these questions whose voting preferences we already know, or who are willing to provide those preferences. The innovation here will be in assembling this dataset. The rest is a collaborative filtering exercise.
  - arikr 10y ago
    I think the v1 could be even simpler -- listing of candidates/choices, things that would indicate preference towards and away from those options, and then questions that reveal those things
    - nerdponx 10y ago
      You still need a dataset linking candidate preferences to other personal preferences. Do major political polls (Gallup, etc) ask non-candidate questions?
bbcbasic 10y ago
Nice idea, but it requires predicting the future in such a way that if you could do it, then you'd make more money on the financial markets :-) You'd need a mathematical model of candidates, predicting what they will actually do. What they promised could be parameters tot he model, but I am not sure how correlated their promises will be to their actions :-) You'd need to model the affect of their actions on the world and then calculate how beneficial that is to the person based on their 100 questions. That's a tough one. Good luck.
- arikr 10y ago
  Yes. I think likely best way around your second point about predicting candidates actions is to base all the data points on how they've acted in the past, and assume that people mostly don't change (I've certainly found this to be true in my own life experiences).
throw_away_777 10y ago
This product already exists, check out https://www.isidewith.com/ https://www.isidewith.com/
- GFischer 10y ago
  Was going to post that. It's fairly consistent with what you'd expect, too. I suspect it'll give a lot of "Gary Johnson" and "Jill Stein" answers (I saw the poll on my mother's Facebook along with a few hundred replies on what it suggested), but people will end up voting Clinton or Trump.
rajacombinator 10y ago
these things are inevitably biased, inaccurate, and little more than propaganda tools.

5 points by arikr 10y ago

I'm expecting backlash against this. Even with the downsides, I still believe this would overall be a force for good. I'd want to know how to vote on the entire ballot

arikr 10y ago

So ask me a series of questions to determine what I care about and in what magnitude, and what my incentives are, and then tell me which candidates and propositions align with my incentives Specifically would like this to exist for all votes, so should get my location so you know my exact ballot If anyone builds a good version of this I have a few decently high traffic websites and I will happily link it on them.

nerdponx 10y ago
You would need some kind of labeled data set -- hundreds of people answering these questions whose voting preferences we already know, or who are willing to provide those preferences. The innovation here will be in assembling this dataset. The rest is a collaborative filtering exercise.
- arikr 10y ago
  I think the v1 could be even simpler -- listing of candidates/choices, things that would indicate preference towards and away from those options, and then questions that reveal those things
  - nerdponx 10y ago
    You still need a dataset linking candidate preferences to other personal preferences. Do major political polls (Gallup, etc) ask non-candidate questions?

bbcbasic 10y ago

Nice idea, but it requires predicting the future in such a way that if you could do it, then you'd make more money on the financial markets :-) You'd need a mathematical model of candidates, predicting what they will actually do. What they promised could be parameters tot he model, but I am not sure how correlated their promises will be to their actions :-) You'd need to model the affect of their actions on the world and then calculate how beneficial that is to the person based on their 100 questions. That's a tough one. Good luck.

arikr 10y ago
Yes. I think likely best way around your second point about predicting candidates actions is to base all the data points on how they've acted in the past, and assume that people mostly don't change (I've certainly found this to be true in my own life experiences).

throw_away_777 10y ago

This product already exists, check out https://www.isidewith.com/ https://www.isidewith.com/

GFischer 10y ago
Was going to post that. It's fairly consistent with what you'd expect, too. I suspect it'll give a lot of "Gary Johnson" and "Jill Stein" answers (I saw the poll on my mother's Facebook along with a few hundred replies on what it suggested), but people will end up voting Clinton or Trump.

rajacombinator 10y ago

these things are inevitably biased, inaccurate, and little more than propaganda tools.