I've always thought that recommendation services would be better of this logic were layered in. E.g. I'm on YouTube and 1 in 10 recommendations are popular/tren

I've always thought that recommendation services would be better of this logic were layered in. E.g. I'm on YouTube and 1 in 10 recommendations are popular/trending videos that the algorithm doesn't think I'll enjoy. Aside from the obvious necessity of bubble-breaking, it has the nice side effect of retraining the recommendation engine without rampant confirmation bias.