Except for the part where the curator (not the artist) explains that the AI is not permitted to copy any elements from either of the two reference works. The AI is tasked to create something new that it deems lies aesthetically between the reference pieces. The way it uses the collection is to catalog it based on similarity and then use that to select the reference pieces. DALL-E does exactly what you are saying. A key point of the MOMA piece is to not do that.
It’s interesting that there are so many different folks with a wide range of viewpoints each convinced that they know how this will play out. When the reality is none of us does.
Frankly I think the jobs of young lawyers are way more at risk than media composers. Imagine a parallel universe where an AI gets the composing job while in ours a human does. Both turn in an initial draft that’s generally on target. But each gets back some notes asking that when the video location is on land the music should sound more ‘green’ while the river scenes should be more ‘blue’. The human is going to ask what they mean by green & blue and try to suss out what the client wants. While the AI is more likely to try and mimic the music it finds in films with greener than average color pallets. So for the land scenes, its next rev has a kind of Christmas vibe with just a hint of films with St. Patrick’s day scenes.
While the end product is a composition or audio recording a bunch of the necessary work is figuring out what the client really wants especially when confronted with conflicting or incomplete descriptions. A human is going to run rings around an AI trying to do that.
UPDATE:
In the parallel universe, even though the movie is a taunt thriller set in the summer, to everyone’s surprise the sleigh bells added just the right creepy feel helping the movie become the 3rd highest grossing movie ever. Meanwhile in our universe HBO bought the movie but has never shown or streamed it.