We’re not saying that there’s an AI-generated Kerbdog album on the way (albeit a new studio album would be grand), but it’s possible the robots have been listening and may have sussed out Cormac Battle nails his guitar tone and lyrics of old.
Kilkenny duo 49th & Main, Robert Grace, The Kilkenny’s and Kerbdog appear among search results from two giant datasets made accessible by The Atlantic in recent weeks.
An article this week via long-serving Irish music blog nialler9.com points to several datasets made accessible by The Atlantic – two of them alone containing over 20m tracks and years worth of music that may be used by tech companies to train their generative AI models or AI music tools like Unio and Suno. A quick search shows the inclusion of Graignamanagh’s Robert Grace, The Kilkenny’s, duo 49th & Main and Kerbdog, with Secure, Sally and Mexican Wave included.
Related Post
It’s not just local music in the datasets, mind you, with some of the biggest names in Irish music – Fontaines DC, CMAT, Lankum, Kneepcap, Sinead O’Connor, U2 and more all making an appearance in results.
AI Watchdog
The Atlantic, who’ve made four datasets in total available to search through, are keen to stress that while artists and songs appear in search results, it’s not an indicator that their material has been used to train up AI models. Their AI Watchdog series is an ongoing investigation by the publication to reveal the inner workings of generative AI, driven by Alex Reisner.
“AI companies may omit certain works when training, so the presence of a work in a dataset is not definitive proof that it was used. Companies often use multiple datasets in training, so the absence of a given work is also not proof that it hasn’t been used. Note that some datasets contain multiple copies of certain works.”
The above named Kilkenny acts – and there may well be more – appear in results between two different datasets, one named SLEEPING-DISCO-9M and the second as LAION-DISCO-12M.
Again, per The Atlantic, the Sleeping-Disco-9m dataset is “A collection of 9,713,413 music tracks from YouTube along with lyrics from Genius.com. The dataset was assembled by Sleeping AI, a group of AI researchers who build training datasets and publicly share their research on multiple aspects of AI development.”
Meanwhile, the LAION-DISCO-12M dataset is “A collection of 12,320,916 music tracks from YouTube, totalling 91 years of music. The dataset was assembled by LAION, a nonprofit based in Germany that builds large datasets and has received funding from Hugging Face and Stability AI’s co-founder and former CEO Emad Mostaque.”
Videos released by Kilkenny Arts Festival in recent years also appear in results from 2025 that looked at millions of YouTube videos being used to train generative AI.
49th & Main
One half of the Kilkenny duo 49th & Main, Ben O’Sullivan, took to Instagram on Wednesday for his take on the news.
“I woke up to the news yesterday morning that 45 of our songs have been used to train AI models. I didn’t even realise we had 45 songs, but there you go.”
“We didn’t give permission to any of these companies to do this, they didn’t pay us, there’s no deal signed; they just stole it to train their models.”
“And they have done this to tens of thousands of artists”.
“It’s scary times. We’re at a crossroads, it seems. AI music is out there. It’s in the charts. You’re listening to it and probably don’t even realise.”
If you can’t see the above video, you can watch it here via Instagram.
What say you?
Is this the kind of thing that should be expected? Would you be annoyed to find your music being used to train systems ultimately designed to create new music? Are we reaching a tipping point in all things music technology?
