Monday, November 21, 2011

When Algorithms Go Wrong

Earlier this year, PC World published an interesting article about the key algorithms that rule the World Wide Web. These algorithms include everything from the Google search and Facebook friends stories, to Amazon’s recommendations and even the eHarmony’s matchmaking algorithm. Very interesting stuff, particularly when you consider the economic impact of such Internet services today.

One of the algorithms is the algorithm that drives ad presentment - the idea is to present you with the most relevant ad based on your profile date. Or, actually, with the ad that you are most likely to click on. But a few weeks ago, I had an interesting experience with Facebook. First, Facebook decided, for no particular reason, to present me with ads all in German. My first reaction was actually positive. Among the few bits of information that I have volunteered to Facebook is the fact that I studied at an university in Germany and occasionally, I even respond to a friend’s post in German. And so I thought that Facebook is so smart that it is trying to appeal to the ‘international-man-of-mystery’ side of me.

But last week, all the ads turned into French. Well, I do happen to get by in French but I am pretty sure that I have not volunteered any info about my French connection to Facebook. Sure, I have friends all over the world, including France, but that’s not enough for even the smartest algorithm to label me as a target for French ads. Besides, I could hardly be expected to act on an ad offering me a skydiving experience in France next weekend. Clearly, something has been going wrong with the Facebook algorithm.

When an algorithm serving ads goes wrong, it is perhaps a laughable matter. After all, nobody gets hurt, right? Well, nobody, except for the companies that paid a ton of money for their ads to hit the right audiences. There have been plenty stories in the past about the innocent looking algorithm changes in Google that end up having a devastating effect on many businesses. If you build your online business that depends on the organic Google search driving your traffic, you can find yourself very quickly out of business when that stops working.

With their tremendous reach, it is perhaps time for the Web's Major League players to start realizing the economic power they have. With thousands and often millions of companies depending on them, Google, Facebook, Amazon etc. have to take their responsibility seriously.

This responsibility starts with the profile data integrity, customer privacy, information security and also the algorithm dependability. Algorithm changes can be very controversial as we've seen when Klout changed its algorithm a couple of weeks ago. It’ one thing to gamble with your own fortune, quite another thing to gamble with the fortunes of those who depend on you. Too many livelihoods are at stake. Abusing this responsibility may be perhaps the greatest risk Internet businesses face today.

No comments:

Post a Comment