Project GAYDAR identifies sexual preference from friends lists

Two MIT students who were taking a course in Ethics and Law undertook a project to write a program that predicts sexual preference from facebook friends and interests, the Boston Globe reports.

They took a sample of approximately 2500 men from facebook of profiles that joined the MIT network and were in their cohort. First they trained their program on profiles of people who had filled in their sexual preference and then they had their program predict sexual preference for a sample of almost 1000 who had not filled it in. They took their personal knowledge of 10 people in the network that are gay but hadn’t filled it in on the profile as a simple check for the predictions and those were all predicted to be gay.  It doesn’t seem to work so well with predicting the sexual preference of women.

They apparently had approval for this project from the ethical review board at MIT, whom I think might want to review their criteria.

5 thoughts on “Project GAYDAR identifies sexual preference from friends lists

  1. The article says the research provides us with a warning, and that seems absolutely true. It’s another argument for anonymity.

    I’m not sure it was bad of the university to let them see if it could be done. I’m wondering why you think it is, if indeed you do.

  2. What I think is wrong with it is that the sample apparently came from their peers and even though the data set is encrypted and only in the hands of the professor, the researchers still were digging into a fact deliberately (one assumes) withheld from the profile of those men. I think that it would have been far better to have taken a random sample of people unlikely to be known to them.
    Which makes validation harder, but… The validation of the predictions doesn’t seem very thorough or voluntary either, although theoretically, they could have asked those 10 gay guys first whether they thought it was ok they were used as validators for this particular personal detail. It would be nice to know that.
    I am not at all surprised that such predictions CAN be made, it is not unlike the AOL search data mistake case an indivudual was identified by anonymised search results.
    I guess it would have been more prudent for these students to have focused on some other characteristic.

  3. I’m not sure I’m quite getting it. The original large sample was of men who had made it public that they are gay. So it was the 10 who hadn’t said on Facebook that they were gay, but still provided some validation that you are concerned about?

    I agree that if they’re drawing up a list of these ten people and it’s in the research report, they certainly should have had permission. I would have thought that with a small number like that human subjects might have a number of issues.

    Anyway, I’m glad I asked, since I hadn’t really thought this through!

  4. I wonder if a ‘gaydar’ facial-processing instrument will be ever be devised which exceeds, according to this study, the high accuracy rate of split-second exposure to faces. It would be interesting to see the results of both studies when applied to women.

Comments are closed.