Do Artificial Musical Artists Dream of Electric Cars?

A question about desert-island discs takes a dystopian turn

The idea of desert-island discs seems to have depended upon a couple of aspects of technology that have been rendered obsolete. First, the wind-up gramophone, which doesn’t need to be plugged in, can be replaced by a solar-powered digital-audio player. Second, There is basically no limit to the amount of music that could be brought to the desert island, and no need to limit, to a mere handful, the number of LPs that would be brought along. I own a 1 TB microSD card, which could probably hold all the music that public intellectuals have ever told the world they would bring with them to a desert island. I also own LPs. I would definitely take the microSD card to the desert island, especially when I think of how filthy LPs would soon become, as even at home they require regular cleaning when using them.

I can’t write an article with 1 TB of CD-quality audio, or even hi-res audio, to give account of. And with solar power and satellite internet, even a desert island might have satellite internet capable of supporting high-quality streaming audio, so the storage technology would then be moot. Given technological innovation since lists of desert-island discs were first being made, there seems to be no reason one has to choose at all. The always-on, available-everywhere nature of wireless internet and the accompanying streaming audio allows a person to make as many choices as they like, when they like. There’s no need to limit one’s listening to a list of favorite songs, pieces, or albums.

I can’t speak about anyone else’s use of streaming services, only mine, and I accumulate favorites, rarely removing them but constantly adding them by clicking a heart-shaped box. I own music files, CDs, LPs, 45s, and even a few cassette tapes, and by downloading tracks to my devices, I am able to borrow and take on the “road” streaming audio music. A well-known philosopher, David Chalmers, has written about the “extended mind”, making an argument that the human mind extends beyond the human body, and without invoking electromagnetism or New Age ideas; rather, we can carry smartphones with us, say, that contain things we want to be able to recall at a moment’s notice, so the information we access via such a device can be thought of as being part of our extended mind. My music library is effectively extended, in a similar way, by my being able to access, at a moment’s notice, much of the music that is accessible via the internet.

I have many more “favorite” artists and tracks than I used to. The music I enjoy listening to is much more varied, and contains many more artists, than it once did and was. I have a taste, a favorite-recognizing capability. I find things to be my favorites depending upon what I, given my always-evolving taste, find appealing for whatever reasons. Computationally speaking, something that is able to recognize a member of a type is much less powerful than something that is able to generate all members of that type. It is not surprising, then, that it is easier to recognize one of my favorite songs or artists than it would be to describe what unites them—creating a way to create a potentially limitless set of songs that would be my favorites. This brings me to the question of generative artificial intelligence and the generation of artificially created music.

The recommender systems in streaming audio platforms are favorite-recognizing algorithms, which, though imperfect, attempt to perform the relatively computationally simple task of identifying songs that I would want to listen to again and again. These have been around for at least as long as streaming platforms that suggest music to their users have existed. The development of relatively sophisticated generative AI has led to the ability to generate potentially limitless numbers of music tracks. This is a big topic today and is one I have not educated myself about, so I can only speak tentatively about it. What generative AI allows for is the constant generation of new music tracks, which then get evaluated by a recommender system and suggested to the streaming platform’s user. I have read an article in which the large numbers of artificial music tracks that are being created is regarded as laughable, but, I object, when one has a good filter—a working recommender system—one can pluck out the “favorites” and serve them up to a user.

Maybe I will try to comment on this in the future. For now I set before you, dear reader, the proposition that it has become possible to generate and select artificial music that a given user will love to listen to. The question of the role of musicians in today’s world of generative AI, improving every day, will be answered sooner rather than later. Going so far as to suggest a role for android-like robotics or even just computer images, there is a possibility, however frightful or remote, that generative AI will soon, almost completely, replace living, breathing musicians. The creation of artificial people who serve as the faces of artificially generated music, complete with gossip columns about their love affairs and clothing, hair, and makeup choices—and endorsements of politicians—is plausible on its face. I imagine a Taylor Swift of the year 2050, one that is completely digitally created. I find it horrifying to contemplate.