We’re being flooded with AI news this week, so for the sake of our sanity, we’ve once again rounded it all up for you to digest in one crispy AI nugget:
Earlier this month, Google and YouTube’s parent company Alphabet claimed that it has the right to use songs uploaded to YouTube to train its Lyria 3 AI-music model. Note carefully that Alphabet is not admitting that it has done this, but that its terms of service would hypothetically allow it. The caveat is that this only affected independent musicians and producers who have uploaded their work to YouTube, not the major players who have different contractual agreements.
For indie artists and rightsholders, finding the leverage or budget to negotiate terms with YouTube is much harder, and may feel that their only other choice is to not upload music to the world’s biggest streaming platform. Now Billboard, which originally published Alphabet’s claims, has shared the responses of indie artist advocates like A2IM’s CEO Ian Harrison, the Music Artists Coalition’s executive director Ron Gubitz, and AI-campaigner Ed Newton-Rex. TL;DR: they’re not happy at all. “It’s outrageous,” Newton-Rex says, “And I suspect if people aren’t angry about it, they don’t know about it.”
Another day, another social media platform slipping AI into its creator tools: this time it’s Facebook, which is “reimagining” its Facebook Creator Studio as a standalone “AI-powered companion app to help you grow on Facebook.” That means it’ll be a separate app that shares creator-specific tips from its AI assistant, has an AI-powered comment-reply drafting tool, and highlights creators’ priority comments to reply to. It’s aiming to make rummaging through the data dashboards easier: instead creators can ask natural language questions and get answers and recommendations. The new app is currently being tested with selected creators.
We’re still fascinated by AI-music labels that seek to launch AI artists, and are eager to find out who these artists are enjoyed by and what the commercial use-case is. We’ll find out by watching companies like India’s Eros Innovation, which now has a label powered by what it calls a “Large Cultural Model”: Eros Music Worlds, with seven AI-native artists, who perform music composed by humans. The intended output of these AI-artists reflects the flexibility of AI-generated content: not just singles and videos but microdramas and live experiences. On the latter front, the platform has also partnered with the family of late playback singer Mohammed Rafi to collaborate on new music recordings and a live concert experience.
Unbaking a cake is hard: detecting a fully-AI-generated track might now be easy enough, but things get trickier when AI was only used for part of the composition. Modulate has now launched a detection model that it says “helps determine whether music contains AI-generated vocals, instrumentals, or other musical elements” with a target audience everyone who is dealing with the onslaught of tracks that are fully or partly made by AI. Modulate says its system runs three independent models – for identifying AI-generated vocals, speech, and instrumentals – and then combines these indicators to create what it says is a more reliable judgment. Its target is distribution platforms, music streaming services, rights organisations and so on – companies which, depending on their internal policies are either clamping down on AI-music, or letting it fill up their servers.
