Lots of notes on the Human Pairing Problem.

HumanPairingProblem.md

@@ -1,3 +1,6 @@
+The Premise
+===========
+
 I've only recently realized what an important factor the relationships I have with other people are on my happiness. It seems obvious, but I vastly underestimated it until recently. Friendship, feeling connected to people, and having peers are central to my happiness, and though I've never been in a romantic relationship, they seem to be even more important.
 
 The relationships people can have with each other vary greatly in quality. Since one only ever meets a tiny fraction of the world's population, and only deeply interact with a tiny fraction of that, the fact of the matter is that most people never meet those they could have the best relationships with. I think that is tragic that these relationships never have a chance to actualize.
@@ -14,8 +17,67 @@ Dating or other human relationship websites have a natural monopoly: the more pe
 
 Consider: The actual website part of the problem is utterly trivial with modern web development tools. The hard part is the pairing algorithms. We want lots of people to be able to experiment on paring algorithms. Further, there may be an argument to be had that for-profit organizations aren't really the group we want to be holding some of people's most intimate data.
 
-Well, here's the idea. Start a non-profit relationship website. "We're on a mission to help people find the best relationships that can possibly exist." It should have a sincere mandate to help people find non-romantic relationships, if for no other reason than reducing awkwardness and stigma.
+Well, here's the idea: start a non-profit relationship website. "We're on a mission to help people find the best relationships that can possibly exist." It should have a sincere mandate to help people find non-romantic relationships, if for no other reason than reducing awkwardness and stigma.
 
 People could opt to make their (anonymized as best possible) data ("for love and science!"). People could access the full data set in a "behind the veil" manner, where they hand over specific queries (corelation between X and Y) and get the response, or paring algorithms for training that then integrate into the website. People could charge for use of their algorithm, so there would still be an incentive for people to work on the problem. Further, this would enable a wide range of people, including academics, to do the sort of research OkCupid did on their blog before being bought.
 
 I don't think this is high enough on my priority list to actually work on this problem, but this seems like a good approach to me.
+
+Why a Non-Profit has Comparitive Advantage
+------------------------------------------
+
+The key premise of all this is that a non-profit has comparative advantage in the area of relationship websites. I think they are able to do it better than a for-profit organization. There are a number of reasons for this.
+
+1. **A non-profit is more trust-worthy.** (And, in fact, further measures can be taken, like incorporating in a region with extremely strict privacy laws, to signal its trust-worthiness.) This is important because users trusting the organization with some of their most intimate information is essential.
+
+2. **A non-profit can reduce stigma.** Online dating is kind of awkward, and it seems like *paying* someone must make the problem worse.
+
+3. **A non-profit is less threatening than a for-profit.** 
+
+4. **A non-profit can externally decenteralize.**
+
+Issues with OkCupid's Approach
+-------------------------------
+
+1. Over-value clusters by repetition
+  * Cluster analysis and unify?
+  * Mutual information
+2. Reflective questions are answered poorly.
+  * Try to avoid reflective questions
+3. No/bad question search
+  * Provide questions search. Or perhaps tagging?
+4. People aren't honest (even when they try to be).
+  * Machine learning can leard patterns in dishonest answers.
+5. Unanswered questions make filtering impossible.
+  * Evaluate unanswered questions with expected utility of a probability distribution over answers.
+
+Importance of Romance
+-----------------------
+
+There is huge positive utility in improving people finding partners.
+
+I intend the following points as many weak arguments style reasoning. I'm not highly confident in any of them individually, but am quite confident that several of them will be true (and they're all compelling reasons).
+
+1. A good romantic partner can increase baseline happiness.
+2. Well-matched partners will provide a better environment for children (eg. less fighting).
+3. Some correlation between scientists having partners with similar interests (which there is a great deal of evidence they very strongly prefer romantically) and productivity.
+4. People in good relationships are less likely to engage in (redacted).
+5. IQ is an extremely strong assortative mating variable. So we expect that improvements in pairing will result in more very high IQ couples. This leads us to expect more very high IQ children. Since utility appears to be a superlinear function of intelligence, this is a positive thing.
+6. Failed relationships and difficulty finding a partner cause a lot of suffering that one might imagine being greatly reduced by powerful matching algorithms.
+
+Persuasive Argument
+--------------------
+
+You will never meet your best possible romantic partner, almost certainly. Nor your top ten. Even if you meet them, you probably wouldn't realize it. There's over 7 billion people on the Earth and you'll only meet a tiny fraction of them.
+
+We think that's a horrible tragedy. And with the power of technology, we're going to end it.
+
+For love and science!
+
+Romance is an important problem: Not only is it extremely important issue to a significant portion of the population, but research seems to indicate that romantic relationships are one of the few things that can permenantly increase one's happiness.
+
+Romance is a tractable problem: We believe that with the right data, artificial intelligence techniques (in particular, neural nets) can discover patterns in high quality relationships and predict them amongst the World's population.
+
+XYZ is a non-profit organization dedicated to improving the quality of human life by empowering people to form the best possible relationships. We are deliberatly constructed to legally obligate us to serve Humanity in this way and protect the privacy of those who trust us with their information.
+
+