Two engineers who haven’t seen each other in years, eating and drinking at a restaurant in Korea Town, talking about the TikTok algorithm: The question is what we would do if we were hired to write a recommendation engine.
The most obvious solution is to feed more of the same based on what you have already seen and on what other people who have watched the same videos have watched before. It’s like the classic Amazon scheme, “people who bought this also bought.”
This is how the algorithm got started originally and worked pretty well for us engineers, but the same logic turned out to be disastrous for normal people.
You, as a normal person, would watch, say, makeover videos. Just watching one of them would lead to a series of related videos about your appearance. You come across one video about how to make your nose look smaller, and you watch it to the end because you feel your nose is a bit too big. Suddenly, your feed is filled with videos about noses, including dubious medical advice from self-proclaimed experts.
Since you are not an engineer, you don’t know how these algorithms work; so you assume everyone is watching all these videos about noses. Since your nose is not perfect, you become increasingly insecure and unhappy in the world increasingly obsessed with noses.
We engineers design all these things thinking that how we think is normal and wreak havoc on normal people’s lives. Designing a video platform thinking that it’s a learning tool is like thinking that going to the beach is a learning opportunity; only engineers think like that. The last thing normal people want to do outside of work or school is to learn.
TikTok took this to the next level; they deliberately designed the algorithm for normal people with the express goal of turbo-charging the feedback loop between you and the world of your own projection.