Streaming service Deezer recently conducted an experiment with the aid of research company Ipsos. Discovery – this 97 percent of people I can’t tell the difference between fully AI-generated music and human-made music – it was disturbing. But that’s not the whole story either.
In the survey, 9,000 participants listened to three songs and were asked to guess which, if any, were entirely generated by artificial intelligence. If the participant failed to guess all three correctly, they went into the failed pile. This means that if two out of three are correct, Deezer and Ipsos still claim that it is impossible to distinguish fully AI-generated music from real music.
Deezer sent me the three tracks used in the study, so I decided to run my own (less scientific) experiment. I had 10 people listen to the same songs and gave them the same encouragement. People have had trouble determining which songs are fully AI-powered. Only one person managed all three. But if I hadn’t combined the answers, the results would have been much less dire. In 43% of cases, users managed to recognize whether the song was created by artificial intelligence or by a human.
It’s also worth noting that several people told me that one of the songs was so terrible, so obviously AI, that they thought it must be a trap and guessed it was real.
Not surprisingly, Deezer’s study participants were a bit surprised by how poorly they performed. 71 percent were surprised by the results, and 51 percent said they felt uncomfortable not being able to tell the difference between AI-created art and human-made art.
Opinions on the impact were divided: 51 percent believed AI would lead to “more low-quality, generic-sounding” music. Somewhat shockingly, only 40 percent said they would skip AI music without listening to it if they were consciously exposed to it.
However, one area where most agreed was the need for transparency. Eighty percent want AI-generated music to be clearly labeled. This is Deezer’s current approach. He created a system that can automatically detect and tag 100% AI-generated content from the most popular models such as Suno and Udio. Deezer also excludes music marked as artificial intelligence from its algorithmic recommendations.
Recently on Spotify announced steps to combat AI sloppiness on its platform, but stopped low of saying it would clearly label AI-based content. It announced rules on artificial intelligence impersonation and a novel spam filter that should keep many of the worst actors off its platform. But rather than operate blanket labels, he is working on a standardized credits system, saying: “The industry needs a differentiated approach to AI transparency so that it doesn’t have to classify every song as ‘is AI’ or ‘not AI.’ This system, however, would rely almost entirely on labels and artists honestly disclosing when songs use artificial intelligence, even if it’s simply to aid in the mixing process.
Says Manuel Moussallam, Research Director at Deezer Edge that there is a gray area around hybrid content that can use elements of artificial intelligence. But he says “it’s not a technical issue. It’s a transparency issue and an ethical issue” that will require responsible action from all parties involved, from creators to music distribution services like DistroKid to streaming platforms.
It’s clear that the amount of AI-generated music streaming is staggering and growing. Deezer says it receives more than 50,000 AI-generated songs daily, which accounts for more than 34 percent of the music added to the service.
Moussallam says that while the company receives a staggering amount of AI-powered content, it only accounts for 0.5 percent of streams, and the vast majority of that tiny slice is fraudulent content. While he acknowledges that this creates some challenges for Deezer simply due to the volume, he doesn’t believe it drastically changes the user experience. “People are still making music and will still listen to music made by real artists,” he says.
Holly Herndon, who has extensively used custom artificial intelligence models in her music, confirms this eloquently Edge that “just because anyone can make polished, kitschy stuff doesn’t mean anyone will care about those songs. Artistic practice is much more refined.”
Seventy percent of survey respondents believe that fully AI-powered songs pose a threat to musicians’ livelihoods, and 64 percent believe that AI could lead to a reduction in creativity. But Moussallam is less cynical, saying: “We are not heading towards a future where humans are removed from the creative process and only artificial intelligence is incorporated into the creative processes.”