How is social media changing the way we think?

Social media is changing us a great deal, and not always for the better.

It seems that we humans are constructed in such a manner that we are drawn to content that evokes negative emotions like fear and anger. We spend far more time looking at and/or reading about stuff that makes us unhappy than we spend on content that makes us happy and relaxed.

So, it’s no surprise that social media companies who want you to spend more time on their website, (and that is most companies) will by design recommend items, (Items being text to read, products to buy, movies to watch or most anything else.) that are most likely to keep you engaged by making you angry and scared.

Do you think this might be a big part of why our society has become so polarized?

I should point out that these companies are not, (at least not explicitly) pushing “hateful material”. Its just that their algorithms which optimize the “time-on-site” learn, (very quickly) what will keep you enraged and therefore engaged.

It’s the algorithms stupid! Specifically, the recommendation algorithms which are the key part that suggests relevant items to users. The scary part is they are very good, in fact they are better at predicting what will keep you glued to their site than even you are.

Recommenders are critical because they can generate tremendous revenue if they are efficient. A few years ago, Netflix had a challenge, (the “Netflix Prize) where the goal was to produce a recommender system that outperformed its own algorithm. The prize was one million dollars.

Here is another example, in 2016 Google changed the You Tube algorithm to a new one created by their Artificial Intelligence company Deep Mind. It uses deep learning neural networks to optimize the content to maximize the users watch time.

Like most of the high-end recommenders it has a two-step process. Both of which utilize the most advanced multi-layer neural networks.

• The first step is candidate generation. This step selects a few hundred possible videos from You Tube’s entire library. It uses a rather crude personalization criteria based on videos commonly viewed by people of similar demographics who watch similar videos as the user.
• The second step is ranking. In this step the smaller subset of videos is scored for expected watch time. The algorithm uses hundreds of inputs such as videos the user has previously viewed in that channel, length of time since the video was viewed, search queries made by the user before watching the video, likes and other more nebulous criteria.

The two-stage algorithm is necessary due to the large number of candidates and the fact that they are constantly changing. Just running the second step is impractical. But it leads You Tube to recommending addictive videos selected from a small subset of popular videos while a large number of videos are not considered. In the words of Francis Irving, it is like getting TV recommendations from a savant toddler rather than a professional TV critic.

If you need more convincing, (or you just want to scare and anger yourself) you can read Google’s paper about the algorithm titled Deep Neural Networks for YouTube Recommendations.

So, think about this the next time you find yourself spending hours watching Facebook, You Tube, Instagram or any other social media site. You are being manipulated by a machine! Who is the master now?

And remember — always back it up!