In 1999 I was working for a web search company and one of the things we really wanted to do was to make search personalized. Eventually we started talking abou

In 1999 I was working for a web search company and one of the things we really wanted to do was to make search personalized. Eventually we started talking about what years later became known as "filter bubbles". It would be bad if people were constantly fed stuff from the same information universe, right? In retrospect, it didn't get all that much of our attention in terms of work done. Sure, we got very obvious (and measurable) feedback mechanisms and we did spend considerable amounts of time countering the aspects that more directly affected percieved quality of the product. But in terms of what it means, we figured "we have time to deal with it later". Just figuring out how to do this inside a few milliseconds per query was more than enough of a challenge in the near term. Since then (and probably before then) I'm sure thousands of teams have had roughly the same experience. You know that there are going to be problems ahead, but you build it anyway because if you don't someone else will. Your paycheck depends on it. When you implement software that anticipates the preferred path a human likes to take through content you end up amplifying what tendencies are present in the person. This is "harmless" in the sense that you are not imposing some editorial direction. You can't be criticized for imposing your opinion or notions of good/bad, right/wrong etc. However the results appear to be polarizing in that they allow people to navigate to certain maxima. The really hard problem here is how you counter this effectively without imposing your view. I think there's interesting work to be done in that area.