Hmmm.…So what exactly is the Roon recommendation algorithm doing? I didn’t want to bring my personal background into this, but since it’s relevant, I have worked in AI research for 35 years (long before the tech companies got interested in AI). Recommendation algorithms by and large fit very simple statistical models to usage data. In short: they are dumb!Your simple characterization of the Roon Algorithm is simply untrue. I’ve discovered many esoteric music connections in classical, jazz and otherwise that I bet even your favorite FM disc jockey from the hazy nostalgic past wouldn’t have a clue of.
I love and miss the old days of FM radio, but that’s no reason to dump on something that is really wonderful in the here and now.
Geek discussion warning: let’s take Netflix as an example. You have a giant matrIx whose rows are users and whose columns are movies. The matrix entries are ratings. The matrix is extremely sparse, as most entries are zero. There are hundreds of thousands of programs to watch. Each user of Netflix barely watches a tiny percentage of the available choices. In this formulation, the recommendation problem is called matrix completion. Seems a hopeless problem: only a tiny percentage of entries are non-zero. Enter some cool math on what is called low-rank matrix completion. This assumption allows you to fill in the missing entries. I will skip the math — it involves ideas related to compressed sensing that incidentally can be used to create efficient digital codecs far better than FLAC.
But, here’s the main point. The Roon recommendation algorithm, like the Netflix method, has no clue what it is recommending. Books, movies, music, shopping behavior on Amazon — all use the same technology. All it cares about is the user data, which is just a highly sparse matrix. The algorithm cares nothing about the content. It knows nothing of the content.
So, unlike the FM music host, who is a trained professional in classical music or jazz, the Roon algorithm, like the Netflix algorithm is ”artificially stupid”. Next time you hear an article on AI, mentally replace the word “intelligence” with “stupidity” and you’ll get an accurate picture of the current state of the technology (I say this as someone who has spent his entire career in the field). We’d like to create an artificial music or movie critic, a program that actually listens to music or watches movies and forms its opinions. We have no clue how to do this since it involves solving the AI problem, which no matter what the hype you hear on the web, is many decades away from being solved.