Listeners Preferred the AI Song… Rithik Kundu on What a Blind Test Reveals About Taste, Originality & the Future of Music
What happens when listeners don't know which song was made by a human and which was made by AI?
Rithik Kundu is a Brooklyn-based music-tech researcher, producer, and investor working where generative audio, AI, and music rights meet. He founded NYU Generative Audio & AI, is a managing partner at Joker Deck Ventures, and contributes to the World Intellectual Property
Organization's technical network on AI and copyright.
His blind listening study — human-made tracks from Grammy-winning producer Mr. Sonic set against AI-generated parallels made in Suno — has been presented at Sony.
In this conversation, Rithik and I dig into what that study found: across 95 listeners and four genres, the human-made music won in reggaeton and cumbia norteña, while the AI tracks were often preferred in the more Western-leaning genres, even as listeners described them as familiar and generic.
We talk about why familiarity drives taste in a world of more than 100,000 new uploads a day, why training data is the "fruit in the blender" behind every AI song, and what happens to genres that barely exist in that data.
We also get into the fear of the "death of the process," why the major labels may want their own AI more than they want to stop it, and what fair compensation could look like, from attribution technology to royalties for the artists whose work trains the models. Then we zoom out: open-sourcing training data, the Napster parallel, synthetic data, spoken-word poetry, and what young creators lose if they skip the fundamentals and go straight to the prompt.
If you're a musician, designer, or builder trying to figure out where human craft still matters in the age of generative AI, this one's for you.
