A new MIT Media Lab study reveals a surprising disconnect: people often say they prefer AI-generated music, yet human-composed pieces are significantly better at actually making them feel the emotions they're designed to evoke. The research challenges how the music industry measures the success of generative AI tools like Suno, suggesting that preference surveys miss what truly matters in functional music applications.

Why Do Listeners Prefer AI Music When It's Less Emotionally Effective?

Researchers from MIT's Opera of the Future group conducted a controlled experiment with 152 participants who listened to instrumental music designed for either calming or energizing wellness contexts. Half the tracks were composed by human musicians at Myndstream, a global health and wellness music company; the other half were generated by Suno AI. The twist: some participants were told the correct origin of each track, others were deliberately given false information, and some heard the music without any labels at all.

The findings revealed a striking pattern. Participants frequently reported preferring the AI-generated music, yet the human-composed pieces were measurably more effective at eliciting the targeted emotional states. Even more telling, participants were more emotionally moved by tracks when they believed a person had composed them, regardless of whether that belief was accurate. This suggests that authorship perception shapes emotional response in ways that current AI music models don't fully account for.

"These findings point to a more nuanced emotional landscape that current generative music models don't fully capture. We need more human-centric approaches, ones that measure not just what people say they like, but how music actually makes them feel, both psychologically and physiologically," said Kimaya Lecamwasam, researcher in the Opera of the Future group.

Kimaya Lecamwasam, Researcher, MIT Media Lab Opera of the Future

How Should the AI Music Industry Rethink Success Metrics?

The research team plans to extend their work in several directions to deepen understanding of how AI music performs across different contexts and populations. Key areas for future investigation include:

• Genre and Cultural Variation: Testing whether emotional efficacy differences between AI and human-composed music vary across musical genres and cultural backgrounds.
• Physiological Measurement: Incorporating objective measures like heart rate and skin conductance to assess emotional response beyond self-reported preference.
• Expert Perception: Examining how professional musicians evaluate AI-generated music differently from general audiences, which could reveal whether trained ears detect emotional nuances that casual listeners miss.

The implications are significant for companies developing generative music tools. If emotional efficacy, rather than listener preference, is the true measure of success, then the industry may be optimizing for the wrong metric. This is particularly important for functional music applications like wellness, meditation, or therapeutic contexts, where the goal isn't entertainment but measurable emotional or physiological outcomes.

"This research reinforces what we see every day: music's true power lies in its ability to move people and resonate with them on a deep, emotional level. As AI becomes more prominent, emotional connection will be the difference between what's heard and what's felt," said Jordan Galvan, Head of Music at Myndstream.

Jordan Galvan, Head of Music, Myndstream

The study, titled "Exploring Listeners' Perceptions of AI-Generated and Human-Composed Music for Functional Emotional Applications," was authored by Kimaya Lecamwasam and Tishya Ray Chaudhuri of Myndstream. It will be presented at NIME 2026 (New Interfaces for Musical Expression) and SMPC 2026 (Society for Music Perception and Cognition), two major conferences in music technology and perception research.

The research design itself reflects a collaborative approach between academia and industry. Jordan Galvan, drawing on his expertise at Myndstream, crafted the creative briefs that guided both the AI generation and human composition processes, ensuring that both types of music were created with equivalent artistic intent and quality standards. This methodological rigor strengthens the findings and suggests that the emotional gap isn't due to careless AI generation or weak human composition, but rather reflects something more fundamental about how AI models capture emotional nuance.

For listeners and consumers of AI-generated music, the takeaway is nuanced. While AI music tools like Suno have become increasingly sophisticated and commercially viable, this research suggests they may excel at creating music that sounds appealing in the moment but falls short in delivering sustained emotional impact. For wellness applications, therapeutic contexts, or any use case where emotional efficacy matters, the human touch appears to remain irreplaceable, at least for now.