[IceMetalPunk]

I agree with you that clicks and view time do not necessarily equate to desired content, but I disagree with you that Google uses those parameters because they're "evil". They make money when people continue using their products, regardless of whether that's because they're being sucked into a rabbit hole of engaging things they don't want, or because they're being sucked into a rabbit hole of things they actually do want.

I think the reason Google (and Amazon, and Netflix, and every other major tech company) uses clicks/view time as recommendation engine inputs is because... well, what else can they use? What quantifiable metric could possibly be used for large-scale, automated recommendations that more accurately indicates what someone actually wants to see more of? (This isn't rhetorical: if you have any ideas, I'd love to hear them.)

I don't think clicks/view time are the best metrics at all, and I don't think they're extremely accurate. But I also think they're the most accurate measure we've got, with the only other options being either (a) remove recommendations entirely, or (b) have humans manually monitor everything you do on the website, occasionally ask you why you clicked or watched things, and then make personal recommendations to you based on your answers. The latter of which is slow, more expensive, less scalable, more invasive to the user, and more tedious for the employees that would have to sit there monitoring you.

Maybe one day we'll have an AI method of using your comments and search terms as a better indication of desire (I dare say some of the recent LLMs are close to being ready for that task), but we're not there yet.

[saurik]

FWIW, my position is today (and has been for many years now) that recommendation systems are inherently problematic, so you aren't exactly trapping me in some kind of contradiction or paradox here by asking me "but how else could it be done?!"... I'd argue that some things simply shouldn't be done.

In the case of social networks, I think they served a positive function to both society and the people who used them back when they didn't have the recommendation algorithms, and your feeds were curated by you choosing to follow people explicitly; this, however, was not profitable, so we are now here.

If you are to do it, then yes: I think you probably need to do what TikTok is doing, and have humans heavily involved in the recommendation system in a way that attempts to put a hand on the wheel rather than the Google way to approach problems with algorithms on top of algorithms and no humans anywhere.

[danaris]

"They don't recommend people anti-democratic indoctrination videos because they actually want you to become anti-democratic; they recommend them because they don't care, and just want to make money" is not a solid argument against them being evil.