The paper "Large-Scale Study of Curiosity-Driven Learning"[1]
talks about AI that likes to take actions that yield results that it didn't predict.
The idea is that curiosity is good intrinsic reward function. The problem though is that it can get stuck doing something that is completely unpredictable, like changing the channel on a TV.
I think this explains a lot about human addiction to endless videos as well. Ironically I learned about this from a YouTube video[2].
The idea is that curiosity is good intrinsic reward function. The problem though is that it can get stuck doing something that is completely unpredictable, like changing the channel on a TV.
I think this explains a lot about human addiction to endless videos as well. Ironically I learned about this from a YouTube video[2].
[1] https://pathak22.github.io/large-scale-curiosity/
[2] https://youtu.be/fzuYEStsQxc