> I've also not heard a compelling argument that one play / minute streamed should be inherently more valuable than another play / minute streamed.
Here's a compelling irrefutable argument.
Let's start the discussion from a couple of scenarios. These scenarios are not the definitive and complete set of use-cases, but are enough to prove a point.
Scenario 1: User manually searches for a song and clicks play. Again the following day. Again and again for a couple of weeks/months. The user forgets about the song after that period.
Scenario 2: User creates a playlist and clicks play. Then uses that same playlist again, and again and again for years. Alterantive scenario: users searches for a published album/compilation and plays that playlist again and again for years.
Scenario 3: Streaming service generates a new generic infinite playlist every day, based on a biased algorithm that upvotes some artists and downvotes some other artists. E.g. Playlist called - Discover Daily/Weekly.
Scenario 4. Streaming service generates a new generic infinite playlist every day, but personalized to the users taste and interests, again based on a biased algorithm that upvotes some artists and downvotes some other artists. E.g. Playlist called - Recommended for you today.
Scenario 5. User hits random play and a truly random algorithm decides what will be played next, no matter of musical style, language, culture or any other aspect. Truly random, no biases whatsoever.
So I will argue why not all streams are equal or should not be treated equal.
The difference is the specificity of intent. Whether the user wished to listen to some specific song, or if the user was force-fed some song. For the initial discussion I am assuming force-feeding, later I will discuss alternative scenarios (clicking skip).
In scenario 1 and 2 there is intent from the user what should be streamed, and this means that the user is interested in the music to be streamed. It can not be refuted that the authors of that specific song should be paid for the stream.
In scenario 3 and 4, the user just wishes to have something making noise, not anything specific, but not random. This is fine, if the user is satisfied, but the problem is all algorithms have biases. If the algorithm favors the most popular authors, and the user could be satisfied with anything in the same style and tempo without preferences, than the streaming service is not fair to the authors who are less popular and never gives them a standing chance to compete. This creates a boosting effect so that the most popular become force-fed to the audience involuntarily and they become even more popular. This is a well-known side-effect in all platforms that use popularity as ranking or rating metric. So here, not all streams can be considered equal, since some had higher probability to be present in the playlist due to inherent algorithm biases.
Scenario 5 is the only scenario where the user is force-fed music, but in a manner that is irrefutably unbiased and irrefutably fair to all authors and songs. In such a scenario all get a chance and all such streams are equal. Unfortunately, users that would prefer such a scenario do not exist. Nobody would endure the torture of a truly random playlist for more than 3 songs.
So we have use cases where the users know what they want, and use-cases where the users don't know what they want, but know what they don't want and the streaming service tries to guess what it is that they want next, without truly knowing.
How do the streaming services know what a user does not want? By measuring the % of the stream that went through. If the user managed to suffer only <5 secs before clicking skip, then that stream should not be counted at all towards payment to the musician. If it was 60 secs than it's a different case. If the song went through entirely it's even better. If the user streamed the entire song, and was actively using the app during the whole time than it is an indicator that the user appreciates the song even more than if a passive encounter occurred.
So they try to guess what to play next. With a biased recommender algorithm that tries to recommend what would be a good fit to play next, and as I said the algorithm might favor some musicians over others.
This is a hard problem and is always discussed at research conferences, both by researchers from the streaming services, and by independent researchers. The topics of bias, randomness, satisfaction are always present. Some authors are even using personality traits to discover the type of music that should be streamed next, which is a very very deep breach in user privacy. I imagine that some services might consider eaves dropping on their customers to evaluate if there is true interest and appreciation of the served content - whether the customer is eating, dancing, smiling, chatting, ..., having sex, during the stream. To gather as much data as possible on what is the context, to offer a better aligned next song (if you are a paying customer) of better aligned ads (if you are a non-paying customer).
This is why all streams must not be treated equal, and the streaming services must find a way to be as fair as possible towards musicians (having in mind that no-one likes a random playlist).
As this is a hard problem, I suspect that more and more services will divert their research towards automatically generated music, so nobody will care if the algorigths are fair and if the authors are paid. Music is most susceptible to this, as it is easy to generate music that sounds good enough to a fairly large audience without realizing it is all auto-generated.
Advertisers pay for a radio and tv service, not users, right?