Dataset ≠ Songs.
While investigating the claim that AI developers trained AI models on how to create music, based on using a data set of 12-million-songs, I discovered something unexpected. I was surprised, since many artists are angry, claiming AI had used their music without their permission. They had assumed that since their song was reported to be in a dataset, that AI had stolen their music. What did I discover?
The dataset does not contain 12 million songs.
It contains about 12 million links to songs, along with metadata:
Song titles, artist names, album name, YouTube URLs, duration, views.
It doesn't contain music.
Think of it as a giant list of songs, rather than a dataset of the songs themselves.
That changed my question to, “What was the pipeline?” How did AI get the actual music?
How were the actual recordings obtained and moved into AI training systems? What, if anything, happened between YouTube and the AI models to get music from YouTube into the AI? You can not train an AI model on links and videos - the music itself must be downloaded or installed (via an MP3 for example) into the training system.
That part of the story seems to be less documented - less investigated - than the talking about datasets that were claimed to have used by AI to steal music.
I have theories. One is the AI companies may have bought the MP3s or albums. Another is what we're hearing is not an artist's actual music, rather what AI created from scratch, based on collecting biometric like meta data on an artist. I don't know how AI got the music. I haven't yet talked to anyone who does know.
One thing I am sure of: the assumption that “my music is in a data set” is not true. The assumption, “My music is in a data set, therefore AI stole my music,” is not true.
What is true? The meta data of some artists songs are in some datasets.
Watch the video here.
Read the article.
Reader Responses
After publishing this short on platforms such as LinkedIn, YouTube, and Instagram, people shared questions, critiques, and alternative perspectives. I've selected a few excerpts and included my responses below. Names have been removed to keep the focus on the ideas rather than the individuals.
Reader: That's a great question! How did these AI companies get a A COPY (aka a mechanical) of these recordings!???
Your Thoughts?
I'm interested in hearing your perspective. Share your thoughts, questions, or alternative viewpoints in the comments below.