AI: YouTube approaches record labels amid questions over IP rights

According to reports, YouTube is working on an AI tool that would enable users to replicate the voices of famous singers and has approached record labels to work out the logistics for intellectual property in the AI age. 

Why IP matters in AI

Words and images are, thanks to AI, incredibly quick and easy to generate, which is a threat to owners of intellectual property - whether that be in the form of music catalogs or brands. This is by far the biggest current issue to the development and deployment of AI in the corporate world as companies are intensely concerned about their copyrights. With significant trials currently underway, the coming months will be pivotal to the use of the technology. 

For companies, the typical stance has been one of cautious experimentation, as this undeniably impressive technology rubs up against significant legal issues, but early deal-making offers a route to early deployment without the legal risks. 

What’s going on

Bloomberg reports that YouTube has been speaking to record labels to see if there is a way for the video-streaming company to train an AI tool that could mimic artists’ voices. 

It’s worth noting that, with or without labels’ blessings, users are already playing with similar tools but illegitimately, not least an AI-created song that sounded like the rapper Drake, which caught mainstream attention. A tool with licensed music will protect the business, even if the wider questions remain. 

Additionally, there are reports that Google, which shares a parent company with YouTube, was in talks with Universal Music Group about the creation of a similar tool. In April, UMG requested that streaming services Apple Music and Spotify stop AI firms from ingesting the contents of their music libraries for “training” machines.

As with the recent Hollywood strikes, many artists are deeply concerned about the implications of AI, and increasing numbers are calling for protections of their voices. 

The legal context

IP in an AI age is complicated, partly because there is no single AI engine that can be regulated; already some artists/creators have seen how open source systems can make it extremely difficult to enforce opting out of training datasets. Meanwhile, online publishers are working out ways of fighting back. 

In the US, a series of different lawsuits are now moving through the legal system, almost all of them concerning the legitimacy of the training data those models have been built on. Effectively, if the use of large quantities of content on the internet is deemed illegal, then the consequences for generative AI in its current form could be significant. 

Sourced from Bloomberg, WARC, Slate