It cannot really be that hard to teach a deep learning engine how to surf the web while tailoring itself to a given user's tastes and filtering out all the egre

It cannot really be that hard to teach a deep learning engine how to surf the web while tailoring itself to a given user's tastes and filtering out all the egregious crap. It should be substantially easier in fact, than teaching a DL engine how to play a video game, because all the HTML and JS that generated the page is available to the engine to reason about in addition to the pixels. As an added bonus, it should be easy to teach the engine to "see" all the monkeyboxes and eliminate them while at the same time giving no clue to the upstream server that AI was involved in improving the human user's experience.