Can You Spot AI Music? Most People Can’t
▼ Summary
– A Deezer/Ipsos study found 97% of participants couldn’t correctly identify all three AI-generated music tracks, though individual song identification was 43% accurate.
– Most survey respondents (80%) want AI-generated music to be clearly labeled, and Deezer has implemented a system to automatically detect and label such content.
– Deezer receives over 50,000 AI-generated tracks daily, accounting for 34% of new uploads, but these represent only 0.5% of total streams.
– Opinions on AI music’s impact are divided, with 70% seeing it as a threat to musicians’ livelihoods and 51% fearing it will create more low-quality music.
– Experts believe AI will integrate into creative processes rather than replace human artists, emphasizing that artistic value depends on more than just technical generation.
A recent experiment conducted by the streaming platform Deezer, in partnership with research firm Ipsos, reveals a surprising reality about artificial intelligence in the music industry. The study found that 97 percent of listeners could not correctly identify fully AI-generated music when compared to tracks created by human artists. This discovery highlights both the rapid advancement of AI audio tools and the challenges they present for listeners and creators alike.
During the survey, 9,000 participants listened to three different songs and were asked to determine which, if any, were produced entirely by artificial intelligence. Anyone who failed to identify all three tracks correctly was considered unable to distinguish between human and AI compositions, even those who guessed two out of three right still fell into the “fail” category.
To test these findings independently, a smaller, informal experiment was conducted using the same three tracks provided by Deezer. Ten individuals listened and attempted the same identification task. Only one person successfully identified all three AI tracks. However, when responses were analyzed individually rather than as a group, the results appeared less extreme. Listeners accurately identified whether a song was made by AI or a human 43 percent of the time.
Interestingly, several participants noted that one track sounded so artificial and low-quality that they assumed it was a deliberate trick, leading them to incorrectly label it as human-made.
Participants in the official study expressed surprise and discomfort with their own performance. Seventy-one percent were caught off guard by the results, and 51 percent reported feeling uneasy about their inability to differentiate between AI-generated and human-created music.
Opinions on the broader impact of AI music were divided. Just over half of those surveyed, 51 percent, predicted that artificial intelligence would lead to an increase in generic, low-quality music. Perhaps more surprisingly, only 40 percent said they would automatically skip an AI-generated song if they knew it was created by a machine.
One point of widespread agreement was the need for clear labeling. Eighty percent of respondents want AI-generated music to be explicitly identified. Deezer has responded by developing technology that automatically detects and labels content produced entirely by popular AI models such as Suno and Udio. The platform also excludes labeled AI tracks from its algorithmic recommendations.
Meanwhile, Spotify has taken a different approach. While introducing measures to combat AI impersonation and spam, the service has avoided blanket labeling of AI content. Instead, it advocates for a standardized credits system that relies on labels and artists to disclose AI involvement, even if it’s only used during mixing or production.
Manuel Moussallam, Deezer’s Director of Research, acknowledges the gray area surrounding hybrid content that blends AI and human input. He emphasizes that the core issue isn’t technical, it’s one of transparency and ethics. Resolving it will require cooperation across the industry, from creators to distributors to streaming platforms.
The volume of AI-generated music being uploaded is undeniably growing. Deezer reports receiving more than 50,000 AI-generated tracks every day, accounting for over a third of all new uploads. Despite this flood of content, Moussallam notes that AI music represents only about 0.5 percent of total streams, with much of that activity linked to fraudulent or artificial engagement. He remains confident that human-created music will continue to dominate listener attention.
Artist Holly Herndon, who frequently uses custom AI models in her own work, shares this perspective. She points out that accessibility to AI tools doesn’t guarantee audience connection, creating meaningful art involves far more than technical polish.
Seventy percent of survey respondents believe fully AI songs threaten musicians’ livelihoods, and 64 percent worry that AI could reduce overall creativity. Moussallam, however, remains optimistic. He envisions a future where AI is integrated into creative workflows rather than replacing human artists entirely, suggesting that technology will augment, not eliminate, the role of the musician.
(Source: The Verge)
