Algorithms play an increasing role in our online lives. In their neverending quest to accurately profile their users in order to maximise ad revenues IT companies employ more and more sophisticated data mining methods incorporating information from your activity history, your friends, your location and even strangers with similar interests to you.
Ever since I’ve watched Gary Kovac’s shocking TED talk “Tracking the Trackers” I’ve become increasingly privacy conscious and have taken several precautions to not be as easily tracked. From activating the ‘Do not track’ option in my browser to installing privacy enhancing browser extension such as Disconnect to opting out of targeted ads within the Google settings.
Thanks to these steps my advertisement profile with Google is now rather unspecific. Without taking such measures however, Google has been able to profile people surprisingly well.
I actively try to avoid the personalised features that sites present me with. In Facebook I never use their EdgeRank algorithm that sorts my feed according to “Top Stories” – I use “Most Recent” instead – simply because most of the Top Stories unsurprisingly is paid content from pages I subscribed to, not posts from my friends. Another reason is that I prefer to keep an open mind and personalised filters tend to create a filter bubble which not only distort people’s view of the outside world according to their own preferences and beliefs, they also do so invisibly.
For this week’s exercise of exploring algorithms I have decided to take a look at YouTube, since I have a long standing history of using that site. My main interaction with the site is wih the “My Subscriptions” tab which is always more relevant (and recent) than the algorithmically populated “What to watch” feature.
Logged into my account, this is what my front page looks like
I can immediately tell why YouTube is recommending these videos to me. All of these videos are closely related to videos I have watched on YouTube within the last 48 hours. I watched one “CinemaSins” video, one Pink Floyd song, one Kygo song, a fail video and a Strokes song. While the songs that I played were actively sought out the other videos showed up on my subscription feed which made me click them.
If I scroll further down, it seems that YouTube still takes the same 3 or 4 videos from earlier and shows more related videos. Additionally, it suggests videos by la belle musique, a channel I am already subscribed to.
While the suggested videos generally meet my taste, they don’t necessarily entice me to watch them now, especially since they don’t lead me to interesting new channels which I might want to subscribe to.
If I log out and visit YouTube in an incognito mode I am greeted with the following suggestions.
None of these videos have any relevance to my search or watched videos history but looking at the channel names (Ad Council, RadioKRONEHIT) one can assume that these videos have been placed on the front page because someone had paid for it.
Let’s take a look at the comments section of YouTube which has long been famous for its disastrous reputation. Apparently Google sorts the comments according to your Google+ profile which I, however, never use. This shows in the comparison between the two comments pages of a random video I clicked on.
As we can see, probably due to the lack of usable profile data Google has from my Google+ account, the comments shown are the same both when logged in and logged out. Again, the combative nature of YouTube comments shines through once more. It seems that no matter how sophisticated the algorithms in place, unless YouTube actively censors comments, it will always fight an uphill battle against the culture that has developed within the YouTube comments universe.
In conclusion, suggestion algorithms like the one used by YouTube can somewhat enhance your user experience to a certain extent, provided that you are okay with sharing enough data about yourself. Given the problems associated with filter bubbles and privacy concerns however, at present I still prefer a carefully manually selected subscriptions list to algorithmically derived suggestions.