AI learning from mistakes

San Jose, CA

May 13, 2026

The music industry has been responding to AI with anxiety and fear just as Napster did years ago. Gen AI companies have invested millions of dollars to build large businesses with songwriters without paying independent music publishers any money. Is it true that gen AI companies and lawyers will use the “fair use” defense when they approve of songs without permission or compensation? There needs to be a deerp understanding of how gen AI works.

There are two income sources at hand: 1) training the models and 2) paying for outputs.

In the case of training the models, royalties will be paid for the human-written songs to generative models that require AI-generated outputs. There is also the issue of whether gen AI companies should be permitted to use synthetic data, artificially-created information that mimics the patterns of real songs. Synthetic data makes it harder to identify which songs are used in an output. Every AI-generated output is derived from human songs by their melody, rhythm, lyrics, and harmonies. If this wasn’t so, Suno and Udio would not have survived using songs without permission. It is necessary that if we can’t trace it back to source songs, it will be difficult to compensate songwriters and publishers.

For the paying for outputs model, there is the compensation of AI outputs meaning the tracks generated by users. Each of these outputs includes the combination of songs baked into the model, but the only way to determine which human songs are present is through an attribution engine: technology that can analyze how the system is generated for each output and what percentage of songs that contribute to it.

This leads to the question of how many “source songs” should be considered for attribution. One way to look at it is to say that 60% of attribution is given to songwriters and publishers. In this case, there is a serious consideration.

As AI becomes more involved with Spotify and Apple Music, the royalty pool paid by each human songwriter will lessen.

Many attribution engines are now available. An example of one that is proving its stay is surreel.ai, which is backed by STIM, a Swedish performing rights society. It has adopted its attribution partner for being a maverick in AI licensing framework. Real world data is identifying these “source” songs and correcitng publishing shares. New issues are expected.

We must learn from our past mistakes such as the transition from Napster-era physical-to-digital and realize that we cannot embrace new technology for too long.

FAQ Section