Facebook has decided that I am very interested in the legal and personal drama surrounding the 20th anniversary of the debut album from the Murderdolls... a band I had never heard of twenty minutes before I started writing this post. This is broadly amusing to me. I regularly see posts about metal bands, even though the only metal band I follow at all is Tool--and they're pretty unorthodox in the genre. And I frequently linger on the posts because even if I don't care about the band or the genre, music discussion is somewhat interesting to me. And then you get point-counterpoint posts about who really owns the rights to Beyond the Valley of the Murderdolls, which is an excellent title, I have to admit... but I don't actually care or want to see any more of this content.
Now, it's easy to decry this as "the algorithm" failing, but honestly, this is the algorithm behaving as intended. It noticed that I lingered on a post and decided to feed me more of that. Conceptually, that makes sense, and the algorithm is following that. And yet, it doesn't actually feed me content that I want. So, this is to say, if it's broken, it's broken conceptually.
This is because algorithms are, at heart, dumb. Just like computers are dumb. Computers don't do sophisticated things so much as they do very simple things at such speed and scale that complexity can be built out of it. Similarly, an algorithm is making a set of fairly rote decisions, just on a larger scale. But, unlike computers, which build complexity by modularizing simplicity and layering it in abstractions, "the algorithm" is just picking low-hanging fruit over and over from an immense data set. That's not sophistication, that's just broad-scale laziness.
Some of this comes down to market forces. Making computers do more complex things is driven by competition. Consumers wants computers to do cool things, so computer manufacturers and software developers are incentivized to fill in that gap. But the algorithm is affected differently. Instances of "the algorithm" aren't being sold to consumers but are rather being used to direct eyeballs to advertisers. It's a b-to-b product, not a b-to-c, so the 80/20 rule comes into a starker effect. The algorithm doesn't have to be "good", just "good enough" to provide a meaningful return to advertisers, and competition isn't really a thing because there are only two players in this game: Google and Facebook. And when the inefficiencies are borne by the end-user--who is not a participant in this market exchange at all--then you should expect to see the end product get steadily worse and worse.