I skimmed/read the entire article (dated June 2019). It does not not contain anything new or surprising for most of the HN audience (not a critique of the article itself, which is aimed at a broad audience).
There are some very fundamental questions that I don't think we have really good answers to yet.
- Like all social media, Youtube's algo is aimed at maximizing aggregate view time, and people working there are directly rewarded for that so it's difficult to change. You could bring up some obvious criticism, that recommending crap will destroy the value/brand in the long term. But that's extremely hard to measure/proof, especially when all short term KPIs point in the opposite direction. So you need the top leadership (who still has to answer to the Board/shareholders) to take the contrarian position on faith.
- The article/Chaslot claims that Youtube's recommendations are not about "what the viewer wants". The obvious question is then, how do you measure "what the viewer wants"? Youtube / social media's main metrics, engagement minutes does not seem obviously flawed. Even if they were, I have not seen any practical alternative (either in this article, or many other discussions). People often bring up analogies with nicotine and other addictive substances, that engagement is largely involuntary. But the main issue with nicotine is clearly harmful for health. If it weren't, there would probably wouldn't be much legislation around it, even given the adddictiveness. The more apt analog may be soft drugs like cannabis, where the legislative momentum seems to move in the other direction.
- The fundamental issue is not that social media is not giving people "what they want". It's that often people want things can be bad for themselves and/or for society. This is very difficult to resolve without taking a strongly authoritarian/paternalistic standpoint.
Article is just incorrect: few years ago, YT realized that optimizing for watch time decreases engagement, which is obviously most critical for life-time profits, and switched to optimizing for life-time engagement. I.e. watch less today, but return tomorrow. It’s very tricky ML problem, so by now YT ML is probably the most advanced in the world in this area.
PS. Trigger for investigation was the event named “boobacalipse” - when recommender recommended videos with boobs on the first page for couple days.
Nit picky, but nicotine itself isn't really harmful unless you have blood pressure issues to start with. It's tobacco (specifically smoking it) that's the real danger.
There are some very fundamental questions that I don't think we have really good answers to yet.
- Like all social media, Youtube's algo is aimed at maximizing aggregate view time, and people working there are directly rewarded for that so it's difficult to change. You could bring up some obvious criticism, that recommending crap will destroy the value/brand in the long term. But that's extremely hard to measure/proof, especially when all short term KPIs point in the opposite direction. So you need the top leadership (who still has to answer to the Board/shareholders) to take the contrarian position on faith.
- The article/Chaslot claims that Youtube's recommendations are not about "what the viewer wants". The obvious question is then, how do you measure "what the viewer wants"? Youtube / social media's main metrics, engagement minutes does not seem obviously flawed. Even if they were, I have not seen any practical alternative (either in this article, or many other discussions). People often bring up analogies with nicotine and other addictive substances, that engagement is largely involuntary. But the main issue with nicotine is clearly harmful for health. If it weren't, there would probably wouldn't be much legislation around it, even given the adddictiveness. The more apt analog may be soft drugs like cannabis, where the legislative momentum seems to move in the other direction.
- The fundamental issue is not that social media is not giving people "what they want". It's that often people want things can be bad for themselves and/or for society. This is very difficult to resolve without taking a strongly authoritarian/paternalistic standpoint.