• OKCupid Publishes Findings of User Experiments - NYTimes.com
    http://www.nytimes.com/2014/07/29/technology/okcupid-publishes-findings-of-user-experiments.html?emc=edit_th_20140729&nl

    La comparaison avec les expérimentations sur les médicaments est très intéressante. Les pharmaciens expliquent que prendre des individus comme cobayes permet d’améliorer la santé de tous, et qu’il s’agit d’une démarche éthique non pas sur la personne considérée, mais sur l’ensemble de l’humanité... ce qui les conduit d’ailleurs à préférer les expériences dans les pays pauvres mais dotés d’infrastructures comme l’Inde.

    Donc, expérimenter en détournant les algorithmes des réseaux sociaux est un bienfait pour l’humanité... indépendamment des effets sur les personnes concernées. CQFD.

    “If you use the Internet, you’re the subject of hundreds of experiments at any given time, on every site,” Christian Rudder, president of OKCupid, wrote on the company’s blog. “That’s how websites work.”

    The test also illustrates how easy it is for a website to manipulate users without their knowing. The small number of users who received changed compatibility scores, some to 90 percent from 30 percent, were not told about the change before the experiment began. After the test ended, OKCupid sent emails revealing the true compatibility scores.

    “I understand that that experimentation is part of the process,” said Zaz Harris, 37, a user of the site from Redwood City, Calif. “But I do think that experiment is a lot more invasive than the others because it could affect outcomes in a meaningful way.”

    She added: “I would probably never see someone that the site said was a 30 percent match when we were actually 90 percent, so that is not cool, really.”

    He likened them to medical experiments where some participants in a study received a placebo they believed was a drug that might improve their health. “Social science is becoming subject to the same problems,” Dr. Piskorski said in an email interview.

    He recommended that sites use so-called “natural experiments” — that is, observational studies that occur naturally, and from which data can be gleaned.

    “We use natural experiments to overcome ethical problems that arise in randomized experiments,” he said. “I think the websites should consider more of these natural experiments even though they are harder to pull off.”