Hacker News new | ask | show | jobs
by hamsterbooster 2192 days ago
Many AI systems are used in the backend to increase revenue. Netflix has a very complex recommendation algorithm based on deep learning/statistics. Amazon uses a lot of machine learning to optimize transportation (NP-hard problem!) , sales, etc..
4 comments

A while ago I watched a 4-part documentary about the every-day lives of ancient Egyptians. Pretty interesting.[1]

For weeks after that my recommendations were filled with bullshit such as "PROOF ALIENS BUILT THE PYRAMIDS!" and such.

So yeah, maybe those "very complex recommendation algorithm based on deep learning/statistics" is perhaps not always such a great idea. In this particular case, it's just a mere annoyance for me, but imagine a 13-year old watching a few genuine documentary videos on Egypt and then seeing this bullshit; they don't have the capacity I have to see it's bullshit.

And imagine if it was on a more serious topic than who built the pyramids...

If I were to ask a YouTube engineer "why did I get this recommendation specifically?" then the answer would probably be "dunno".

An additional issue is that the YouTube of yesteryear was much better in browsing random videos. Now everything is based on what I've watched before, instead of just "give me a list of science videos" or whatnot. This is also an issue I have with Netflix (or rather, had, since I no longer have an account).

It seems to me that inscrutable mindless AI learning has a part in the spread of misinformation and bullshit. I'm not sure how large that part is, but I suspect it's significant. I'm hesitant of the total value in these cases, regardless of what it may do for the bottom line in terms of revenue.

[1]: I'll just drop the link to it here in case anyone's interested: https://www.youtube.com/watch?v=hnsNwwwHm2I

My biggest issue with YouTube is that it shows me adverts for the same product for a about 2 months at a time and I therefore end up hating those products and would never buy them.
My tastes may be a bit hard to model but I don't get any value from Netflix's recommendation algorithm at all. I find new things on Netflix by reading human curated reviews online and then searching for titles by name.
>Netflix has a very complex recommendation algorithm based on deep learning/statistics

And yet, it's harder to find anything to watch than it was when they'd just let you browse by genre.

More importantly, Netflix is using AI and data to determine where to invest next in acquiring content or producing original content.