Maybe everyone should stop mass-stealing media to "train their model"
Of course - we can control ourselves to not copy/plagiarize something, to some extent. As should deep learning models / generators. But we're all influences by someone else.
It will be much more profitable to train AI on what music is popular and sells well, then produce new music which is not close enough to be infringing while still fitting the success formula. This can be done so cheaply (and at some point will be fully automated... requiring virtually no oversight or review) that it can be sold at a fraction of the price of something which took human time to make.
Eventually the only way you'll ever be able to hear human music is to see a human play live or listen/buy directly from the artist. This seems doable, but finding the artists becomes the challenge. Of course we have the internet, so the obvious answer is to make a website to help listeners find artists (exists already, of course). But then, go back to the top of this comment and see how some "artists" (not real musicians, but people who see $ opportunity) will start producing non-human-generated music on that platform. They will sell it a bit cheaper. Algorithms will rank it higher in search or suggestion systems.
And eventually, there's no human music except live (which can be faked pretty believably depending on the audience).
Isn't this the whole point of technological progress? I could buy everything handmade by artisans, but usually I buy mass-produced things because of cost.
> Eventually the only way you'll ever be able to hear human music is to see a human play live or listen/buy directly from the artist.
I would argue that for music this is already true. What you hear on Spotify is so mixed, leveled, autotuned, etc. that it varies massively from a live performance.
I'm not sure there is an agreed upon definition of the goal of technological progress. And if there is, it must include some words that are subjective descriptions.
To use a rather absurd exaple, take John Cage's composition Silence. I'm no John Cage expert, but I expect that the composition (4m33s of complete silence) has some artistic merit in the context of the composer's life and other works. Obviously such a composition could be algorithmically generated virtually instantly, and thereby at virtually no cost - far cheaper than what John Cage's time was worth.
Part of what makes human creations so valuable is that they are a form of communication from one or more people to other people - specifically or generally. Or sometimes they are a kind of self-communication, where the artist is creating as a form of self discovery. Those who experience the creation may find their own value, or they may find value in vicariously experiencing the artist's journey.
To eliminate the human from the creative side and replace them with largely unguided software is kind of eliminating the point of a lot of art... to me at least.
The goal of technological progress as I see it is to aid us in our efforts to do things. It might help us never run out of toilet paper. It might help our car burn less fuel and require less maintenance. It might identify patterns of biometrics and movement which indicate a likely impending medical emergency. I can think of a ton of things which represent significant value from technological progress. Replacing artists and creators is not on that list, or at least not anywhere near the top.
> What you hear on Spotify is so mixed, leveled, autotuned, etc.
Mixing is good; nearly all music needs mixing to balance the inputs and adjust for acoustics of the environment. That's also mostly a human-driven job still (although the software aids are pretty useful in some cases).
Autotune is also a human choice (one which I dislike, but it's not me making the music... so not my choice).
Studio albums made available on Spotify are pretty much just what the artist or record company released. That's not some kind of AI thing.
Training models is not the problem. Mimicing or reproducing through algorithms of human creative work and then passing that off as human created work (which implies time cost, as most humans take significant time to develop skills which allow them to create things) is the real problem.
Already we have "content" on Youtube which is entirely or almost entirely AI generated. Like processed foods, it tends to be hollow and non-nourising to the consumer, but it competes for shelf space with potentially better content. As we capitalistically race to the bottom, the cheap AI-created stuff will be so pervasive that real offerings which cost real money (time) to make will have no space to compete. Consumers will be left with the cheapest garbage which resembles something real. This is what you find in many stores for physical products, and the TV series-ication of video content. Formulas which maximize profit squeezing out real things.
This is how the AI models will increasingly be used, and there's basically no stopping it. I would say that using "indie" services where things are hand/human-made would be an alternative, but Etsy has proven that that approach eventually devolves to the profit-garbage approach.
Shelve space is limited and TV production is quite expensive, thus, risk averse. It's more like a music store with everything, up to you to find what you like.
I understand it is economically better overall (less chance of huge reward, but also much less chance of failure and huge loss) to churn out series using one of the formulas for various genres. That's kind of my point, too. It's not about artistry or telling a story as much as it is about reliable profit. And that's why our content becomes more and more similar and more and more "reliable".
Maybe in a few years there will be another Blair Witch wildcard that will draw a stark contrast from the monotamy that we have. And then the production companies will be hard at work trying to figure out how to recreate whatever the new cool formula is, resulting in a race down a different path to the same bottom.
Build a platform where to view content on the platform you have to agree not to use it for training data and to take action against companies where their work turns up as model output.