It depends what you mean by "discoverability". YouTube's algorithmic search may surface related videos, but what if you want to discover other views of a topic, for example, a timeline of all the news, including videos, on a topic[1]? That's not really a decentralization problem, more of a curation problem.
Good question. For discoverability to work well, you'd have to be able to identify a particular user, then track their viewing habits, then develop a viewing profile for them that you can do matching against.
The technical aspect of this is a mostly solved problem (YouTube and other distributors do it) but getting past the privacy aspects will be challenging - the type of person to use a distributed platform will also likely be using it to avoid this sort of tracking. Or they're in a country that has GDPR-like laws.
Why does the algorithm have to show you what content to watch? Just have it a good search and tagging system and the user will figure out what they need.
Surfacing content you don't know you want is something an algorithm can be quite good at. I don't know that I would have ever discovered HISHE, for example, without it.
A search system cannot replace a recommender system and vice versa, they are very different use cases with very different needs.
For any type of media, whether it's songs/bands on a music streaming service, video content on youtube, or even complementary goods in an online store, a good recommender system can and will suggest things that I would never have thought of (because I didn't know it existed) and thus could never have found with any search or tagging system.
How do I search for a song that's similar to those that I listened just now (and genre tags don't really work, they're far too broad, subjective and inaccurate) and that's explicitly not one of the songs which I remembered and just heard already?
How is that solved on the decentralized platforms? How does discoverability work, what assumptions does it make, and can it be "gamed"?