If you’ve applied for a job before, you’ve probably fretted over how to answer questions like “are you inclined to rely more on improvisation than on careful planning?” or “do you like to create challenges for yourself when you take on a new project?”. Companies commonly use personality tests filled with questions like these to assess the fit of a potential employees with the company. There’s a whole laundry list of these personality tests here. Some companies, like Google, have even developed their own. All of these tests are sets of written questions meant to poke and prod at a candidates mind to get a real sense of their ability and personality. However, the meaning behind these questions is relatively transparent, motivating candidates to give the answers they think their employer wants to hear. Startup Pairwise is taking a different approach to personality tests. Instead of words, Pairwise will use images to test a candidates mentality using data gleaned from their LikeBetter picture game. LikeBetter is a flash based game that shows you a series of pairs of images uploaded by users. For each pair, you pick which image you prefer. Based on the choices you make, LikeBetter makes a guess about your personal traits, which you then confirm or correct. As more people use the system, LikeBetter discovers the strong correlations between the choices people make and the attributes they express. Based on this data, Pairwise creates a quiz using some of the most highly discriminating pairs, chosen to have the strongest and most confident predictive power across the broadest spectrum of personality traits. They can then track a candidate’s behavior through the test and make an educated guess about their personality based on the correlations they made in LikeBetter. Pairwise does their best to make the test harder to read into by being a completely image based test and using non-obvious pairs (no GI Joe vs. Barbie). Under the hood, LikeBetter is using an iterative application of Bayes rule called Naive Bayesian inference. The method uses a lot of dense statistics involving proposing hypotheses and dependent probabilities. If you really want to learn about it, check out the Wikipedia entry. On the other hand, the employment quiz is not making and testing hypotheses, but comparing the user’s behavior with the statistics they collected through LikeBetter and determining the the applicants tendency toward either extreme of an attribute (i.e → Read More