I see this becoming more of an issue than the image generation AI. Mostly because the music industry is much more lawsuit happy. But the same legal arguments apply. Training is not stealing.

One of the biggest challenges is determining who owns the rights to AI-generated music. There are questions around whether the rights belong to the creators of the AI algorithms, the creators of the input data used to generate the music, or the entities that commissioned the creation of the AI-generated music. This lack of clarity makes it difficult for artists and businesses to commercially exploit AI-generated music and can also lead to disputes over ownership and royalties.

I haven't read the MusicLM paper yet, it sounds super interesting! I think it's inevitably going to happen with music as well. The biggest challenge will be getting a big enough data set of music to train the AI with. Stable Diffusion was able to train its model on copyrighted images legally I believe, however I think there are much stricter laws when it comes to music copyright. That said, I do think there are many musicians out there who would be willing to train an AI model with their own music. I think it could be a great tool for inspiration personally, imagine playing an instrument duet with an AI that could match your playing style!

Who we can call owner of the AI generated content? Owner of the tool or user of the tool?

The issue of copyrights has been a serious one for centuries. Currently, AI copyrights are uncertain based on current technology, including AI-generated images' legal documents. All of that should remain the platforms' property, despite the fact that creators and prompt crafters work on those things day after day. And also the music copyrights is quite complicated since copyrights are involved in different kinds of usage, period, and region. Most of the platforms did not provide clear copyright documents for creators and musicians. So that musicians will always be underestimated or underpaid. Therefore, we are the platform that licenses music directly from musicians. Included are all the SPs, OPs, composers, lyricists, singers, musicians, music instrument players, producers, etc. they are all involved! Solving the problems and license from the musicians to prove all the technology, platforms that creators are able to use the music with license. And our AI can recommend music from video and images in seconds. Referencing the contents, reference music, story, etc. And edit audio noise reduction at the same time. And... We just launched on Product Hunt yesterday! Welcome to check it out! https://www.producthunt.com/post...

I believe this tool will change the game based on music is created. A lot of musicians will be able to test for themselves how their lyrics sounds before actually going into the studio. My fears is how readily available and accessible will it be to the general public?

Hey Cristina, great topic and thanks for the heads up on MusicLM! Our company, AudioCipher, is a text-to-MIDI generator and we hold over a dozen top ranking AI music articles on Google. I have been working closely with independent developers to help publicize their Google Colabs and websites, as a grassroots effort to bring real AI music generators into the world. If you or anyone else on product hunt would like to collaborate and get some exposure, please contact me at support @ audiocipher dot com.

The hardest challenge is determining authorship and ownership of the resulting work since it's not yet fully resolved.