Unpacking the algorithms behind online experiences

Relinquishing human free will to the machines generally gets a bad rap. In the media, all kinds of scientists are nefarious in their data-driven ways — even well-intentioned science yields Frankenstein’s monster far more often than the reasonable paleontologist in Jurassic Park. The Terminator franchise in the 1980s was among the first to introduce the concept of the Singularity to mass media — what happens to humans when their systems become self aware? — with predictable dire results. (According to futurist Ray Kurzweil the Singularity is now only 17 years away — so look busy.)

And yet in our online lives, we cede more and more to algorithms and machine learning. Eli Pariser highlighted the risks of our collective inattention to the use of algorithms in his 2011 The Filter Bubble. He pointed out that applications like Facebook show us only a curated subset of our friends, just as Google in 2009 began to filter our searches for us — even when we’re logged out. Apart from intermittent outrage in front page stories (“Target knew my teen daughter was pregnant before I did!”), we don’t often question the convenience that all these behaviorally-fueled algorithms offer us. We may cringe at ads that re-target us around the web, but as long as we use the conveniently-timed coupons offered to us, we are encouraging rather than curtailing that behavior.

Recently I’ve been reading more about Echonest, the technology that sits behind music discovery applications like Spotify and Vevo. In many ways this feels like a highest and best use of an algorithm. Who has time to manually curate playlists for everything from a workout to a night out? The algorithms go further to infer musical compatibility among friends, and have even taken a swing at how musical tastes align with political leanings – handy for those throwing national conventions or debate watch parties. The risks seem low compared to algorithmically-led access to one’s friends, but of course there’s the opportunity to skew music discovery to industry-backed new artists à la Clear Channel. While some analysts are bullish on Pandora’s music genome approach (similar to Art.sy’s attempt to map the art genome – h/t Cesar Brea), other investors are betting heavily on Echonest. A few weeks ago Echonest made public their Discovery playlist, which updates each week with emerging music from around the globe. Algorithms find out what people are listening to and taking about, and offer a great way to find new music around the world in a ways most humans can’t scale — well before it’s piped to you in the mall.

In the end, it’s all the same acknowledgement of the impossibility of the big data in our lives. It requires admitting the information overload is too great, and a degree of outsourcing selection to the machines. So maybe algorithms are less creepy when we know they’re out there working for us, and when they are finding needles from the broadest possible haystack — music we might like, seeing friends through the lens of musical compatibility — rather than surreptitiously defining the status updates we get to see.

Was this useful? Please share.