The author raises some good points about a few gatekeepers controlling the majority of the audience, but then condemns recommendation algorithms and believes no one has attempted to solve the discovery problem. How does one miss the obvious when it's right in front of their face?
Discovery is fundamentally hard because its success is judged by two parties whose interests align on paper -- for publishers to find a receptive audience, and for a consumer to find enjoyable content -- but the market is so vast that a missed match isn't just that, but often a match for a different pair. And consumer attention is vast, but ultimately finite, in that it suffers from oversaturation and requires novel and exciting experiences to stand out from a sea of similar offerings.
So platforms use recommendation algorithms to try and guess the sort of stuff every person likes, and try to surface content they're more likely to appreciate, instead of surfacing something completely random. Content that's brand new has no data on its audience, so a chicken-or-egg problem occurs where popular content can bubble toward its most fitting audience, often growing until it can't, while never-before-seen content languishes in a popularity vacuum, desperate for out-of-band help.
The classic out-of-band help is word of mouth. Sponsored, promoted content is a kind of in-band help, allowing platforms to make money and producers to jumpstart their exposure, but has the same structural properties as popularity: those most equipped have the potential to rise the most, while those least-equipped are disadvantaged once more.
These are good points. Somewhat tangentially, I think my gripe with recommendation algorithms is simply my lack of control. If I had some knobs to fiddle with, I'm sure I could tune it to surface way more diversified content that I'm interested in.
Particularly with YouTube, I've recently gotten the feeling that their algorithm has pigeon-holed me into a couple different interests, and I rarely see videos outside those recommended to me. Even if it were just a slider from "our best guess <------> throw in some risky recs", it would feel a lot better.
Discovery is fundamentally hard because its success is judged by two parties whose interests align on paper -- for publishers to find a receptive audience, and for a consumer to find enjoyable content -- but the market is so vast that a missed match isn't just that, but often a match for a different pair. And consumer attention is vast, but ultimately finite, in that it suffers from oversaturation and requires novel and exciting experiences to stand out from a sea of similar offerings.
So platforms use recommendation algorithms to try and guess the sort of stuff every person likes, and try to surface content they're more likely to appreciate, instead of surfacing something completely random. Content that's brand new has no data on its audience, so a chicken-or-egg problem occurs where popular content can bubble toward its most fitting audience, often growing until it can't, while never-before-seen content languishes in a popularity vacuum, desperate for out-of-band help.
The classic out-of-band help is word of mouth. Sponsored, promoted content is a kind of in-band help, allowing platforms to make money and producers to jumpstart their exposure, but has the same structural properties as popularity: those most equipped have the potential to rise the most, while those least-equipped are disadvantaged once more.