{"id":192,"date":"2017-06-15T21:27:30","date_gmt":"2017-06-15T20:27:30","guid":{"rendered":"http:\/\/www.danieledavi.com\/blog\/?p=192"},"modified":"2017-08-05T15:01:36","modified_gmt":"2017-08-05T14:01:36","slug":"okcupid-algorithms-and-social-experiments","status":"publish","type":"post","link":"https:\/\/www.danieledavi.com\/blog\/2017\/06\/okcupid-algorithms-and-social-experiments\/","title":{"rendered":"OKCupid algorithms and social experiments"},"content":{"rendered":"

\"\"<\/a>Matchmaking companies exist on the web, OkCupid is one of them.
\nThe way these things work is customers sign up, enter a profile, and the company tries to find compatibility between user profiles. Based on what you’ve told about yourself, you’ll be recommended people to meet that the company thinks might be compatible with you.
\nDetermining compatibility is, of course, extremely difficult.
\nAnd a company like OkCupid has a proprietary algorithm to try to predict this compatibility. They do so by means of developing a compatibility score that is computed through some proprietary function of all the different things that you have in your profile.<\/p>\n

Of course, they’re trying all the time to improve their algorithm and improve the predictive power of the compatibility score that they create. And for this purpose they’re constantly monitoring how well they do. They’re constantly tinkering with the algorithm, trying to improve it, change some weights, change parameters. And some of these changes may be successful and may be adopted. Others may be rolled back.
\nAnd so in this sense, if you’re a user of a website like OkCupid, you’re constantly being experimented upon in that what you’re experiencing is an algorithm that is perpetually in development.<\/p>\n

Now, OkCupid actually took this thing couple of steps further.
\nThey tried slightly bigger experiments. So for example, a thing that they started to do is something called the Love Is Blind Day.
\nAnd on this day they would suppress photographs in user profiles. And what this meant was that you would not be able to see he profile photographs of potential people to meet.
\nYou would, of course, be able to read what they had written in their profile in terms of their interests and other text attributes.
\nWhat they found was that even though as humans we all, very much, care about photographs and that is likely to be a large part of the input in terms of our deciding if somebody looks interesting and attractive, that not having those profile pictures actually resulted in more conversations, more dates.
\nAnd in that sense, their platform was actually more successful on this day, on the Love Is Blind Day.<\/p>\n

So taking this kind of counter-intuitive success another step further, OkCupid decided to try an experiment where they wanted to understand the impact on the success of a date of simply being told that you were compatible. And so what they did was took people who were actually not very compatible, had a low compatibility score, and falsely told them that they had a high compatibility score, just to bring people together. Conversely, they also took people who had a high compatibility score and told them they had a low compatibility score. And they actually did a three by three matrix of this sort with 30, 60 and 90% compatibility score actual versus 30, 60, and 90% declared.
\nAnd what they found was the success rate was the highest when people were compatible, and were told they were compatible, that is at the 90-90 entry of this matrix.
\nAnd it was lowest where they were not compatible and were told they were not compatible at the 30 30 entry of this matrix.
\nThat if people were not compatible, but were told they were compatible, they actually did pretty well. And in fact, they did just as well as if they were compatible but were told they were not.
\nThis is an interesting insight into human behavior. And this experiment was reported by the CEO of OkCupid in a blog post. When he reported this, he was skewered on the Internet by a whole range of people.
\nAnd fundamentally what this says is, in doing this experiment, interesting as it might be, that OkCupid broke the unwritten expectations of society. The CEO Christian Rudder obviously didn’t agree with this general social consensus, or maybe there wasn’t enough of a social consensus, that he, and other people like him, would think it perfectly acceptable to do what they did in terms of reporting compatibility scores that were intentionally incorrect.
\nReally what we’re doing here is, I think, socially, drawing a line between experiments that companies conduct in the normal course of business because they’re trying to improve their product, or improve their algorithm. And yes, sometimes hoped-for improvements may turn out not to be improvements. Sometimes companies actually tell you something completely wrong, but that’s unintentional. And that kind of thing we accept as just the way the world is. On the other hand, having a company intentionally lie to you, intentionally give you a wrong score, is something that most people would consider socially unacceptable.
\nAnd because of this particular experiment and the discussions that followed, I would say that this is a matter on which there is greater social consensus today then there was at the time that OkCupid actually did conduct this experiment.<\/p>\n

More on this story:
\nhttps:\/\/theblog.okcupid.com\/we-experiment-on-human-beings-5dd9fe280cd5<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Matchmaking companies exist on the web, OkCupid is one of them. The way these things work is customers sign up, enter a profile, and the company tries to find compatibility between user profiles. Based on what you’ve told about yourself, you’ll be recommended people to meet that the company thinks might be compatible with you. […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[65],"tags":[150,152,155,154,153,151],"class_list":["post-192","post","type-post","status-publish","format-standard","hentry","category-courses-notes","tag-algorithms","tag-data","tag-ethics","tag-experiment","tag-science","tag-social"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t