What makes a good recommendation engine

publish Backlink: Musings on philosophy

Most of the content here comes from a conversation I had with Preeti on 2022-04-27.

I argue that TikTok’s engine is not good at figuring out “which things you like”; it’s good at figuring out “which 10 second-long videos you like.” The two are totally separate concepts in my mind.

Proof of this/a related concept: I often add long documentaries to my “Watch later” playlist on YouTube about niche topics (for example: the life of an African warlord). Yet I don’t watch many of these. Am I deluding myself into thinking I enjoy these topics since I don’t watch them? Is this an example of Bad faith? I don’t think so. I think the cost-risk-benefit analysis simply leads me (and others) to consume 10 second long TikTok videos since the time cost is low. When I consider watching a 25 minute documentary about African warlords, the risk is 25 minutes of opportunity cost — pretty significant compared to 10 seconds.

So, perhaps TikTok’s short form videos are a tool for good. They provide a low risk / low cost method for people to break into potential new interests.

But it doesn’t feel that way to me.

Related thought: Neurons die when unused; strengthen and brainwash us when overused. Is there a middle ground?