The downside of Predictive algorithms

Some time ago Facebook launched social graphing, here it enables you to search and see what your friend like and other stuff, this is what Wikipedia says “the global mapping of everybody and how they're related”. Facebook and many other social media platforms use simple (yes I said simple) algorithms to determine association and thus make certain recommendations. For example give that person A knows B and B knows D and both D and A know C there exists some (naïve) probability that person B knows person C thus suggest C to B as a friend (i.e. if A=B and B=C=D then A=D). Like I said it’s that simple. From this premise you can add any complications you want but the idea remains the same. Social graphing thus says, if A is friends with B,C and D and spends more time reading status updates from D than either B or C then ,ceteris paribus, maybe they are more alike and if there is an item that D likes then chances are A will also like. This also helps the like of Facebook to decide which status updates to show you, and the rest of your friends status you must search for.

The logic is very simple and the execution could be more complex than I have indicated and anyone who has ever had to build a model will agree. But what is the drawback of such algorithms? The idea is that we should push to you what you want to see or what you would most likely respond or react to. The unintended consequence of this is that it limits your imagination/creativity. Say for example you search for a particular make of a car just because you are considering purchasing a new car, then the smart algorithms at Google will start showing you adds about this particular car and maybe from competitors. This is the power of suggestion, they put this car everywhere until either you are sick of it or buy it. But this has deprived you of an important part of the shopping process, thinking. How? Every time you see something you maybe possibly like, it looks better and better. You forget to start critiquing the item and look at the exaggerated good features. It also means that you are likely experience a lot more buyer’s remorse/regret as effectively this purchase was influenced.

The other fact is that these algorithms may re-enforce bad behavior, take as an example a person had a gambling or alcohol problem. If you looked at their searched and likes what would likely show up as top search is their favorite alcoholic beverage to their favorite casino. If they go to rehab and come back, what adverts to you think they will see? Since their search patterns, used by the models, is the same as before all this poor guy will see is an invitation to the life they are trying to leave behind. This for me I think is the worst of the downsides. Since the internet always remembers the history of our bad behavior will follow us.

In the case of Facebook, the idea is to say, if your friends rated or commented on a restaurant or a place or whatever and they liked it, you are likely to like it to. This will thus be pushed to you as suggestion. It maybe spot-on and you like the places that your friends like or dislike the same things. Take a look at your friends on Facebook, some of them are completely different from you and there are very few if any that are 100% like you. The algorithms we have fail to model one thing, EMOTION, or better yet our emotional attachment to people and things. Take your Facebook friends list again and look at your call record, what proportion of those people have you made contact with? The reason for that last step is since you can’t detect emotion from text and I doubt that a meaningful conversation can be had using an IM. Most people who have ever been in sales will tell you that people buy with heart, you seldom act what you think but you do act what you feel.

These algorithms are displaying information to us that they ‘think’ is of better value and importance to us. They are confirming what we are and not inviting us to change, to explore and to take a chance. I work a lot with building models focusing on customer behavior and trying to use this to predict what a customer is more likely to buy and I always emphasis the need to remember that each customer is different. And of course I can experiment with option and test my hypothesis (Data Science) and give to the business a working profitable model.

Please note this is my opinion based purely on my personal experiences and don't claim to be based on any research or study, but it would interesting if some did do a study.