New Study Of AI Music Production Pits Human Against Machine

A new study into generative AI in music production shows that machine learning tools, while promising, lag human composers in creating music with emotional impact. Conducted by SoundOut, a provider of music testing, and Stephen Arnold Music (SAM), a sonic branding specialist, the study compared music produced by Gen AI with human composed music and found that humans hold the edge for emotional accuracy and appeal especially in producing music for brands.

The study also examined GenAI’s usefulness as a collaborative tool, testing the effectiveness of music produced by human composers working in tandem with GenAI. It concluded that GenAI can be useful in generating ideas and initial compositions, but refinement and emotional content is best left to professional composers.

“While humans outperform AI on the emotional front, this study reveals that AI ‘composing by numbers’ is not far behind,” says SoundOut CEO David Courtier Dutton. “AI is not bad at creating emotionally appealing music, but humans are better.”

“Compelling music is more than its composition, it also involves the careful consideration of instrumentation, performance, mixing and editing,” says Chad Cook, SAM president and creative director. “While AI does not currently meet the quality standards for today’s marketing and branding needs, there are benefits to human/AI music collaboration.”

The study involved two tests. The first focused on the emotional accuracy of AI-produced music. A GenAI music production platform responded to four briefs, such as those used by brands seeking music for ads and other media, spanning emotions such as Sentimental/Compassionate, Inspiring and Funny/Quirky. The platform produced five tracks for each prompt, which were then analyzed using SoundOut’s OnBrand tool. OnBrand analyzed each track against 212 emotional attributes and showed that GenAI produced emotionally accurate results 20% of the time but did not meet professional standards in most instances.

The second challenge focused on appeal and emotional impact of AI versus human-produced music. Four tracks were produced in response to a prompt that specified the criteria “somber and emotive,” “inspirational resolve” and “core instrumentation: piano, strings, guitar and vocals.” They included:

• A GenAI track
• A professional human composer track
• The Gen AI track “improved” by the composer
• The composer track “improved” by Gen AI

SoundOut then asked a random group of 200 U.S. pop consumers to rate each track for appeal and emotional attributes. The results showed that human composed and modified tracks demonstrated a higher appeal than AI composed and modified tracks. Human composers improved tracks produced by GenAI, but GenAI made human-composed tracks less effective.

“Humans are better than AI at articulating specific emotions through music,” says SAM Director of Brand Strategy Russell Boiarsky. “We can write emotional tracks that beautifully encapsulate a creative brief. On the other hand, while AI can create appealing music, it cannot accurately compose music for specified emotions … at least not yet. AI alone is not a solution for world-class sonic branding, but humans plus AI could prove to be a powerful combination.”

Findings from the study reveal that for now, AI can be a valuable tool to speed up the ideation phase of music creation. Human composers can then spend more time fine-tuning concepts with expressive performances, emotional timing, quality production, mixing and mastering.

Read the white paper and hear the tracks here.