Human wins in competition with machine over commercial music composition - just

08 April 2024 - Press release

Generative AI is a great co-pilot to commercial music artists, but not something which can replace people in accurately reaching the right emotional attributes – yet. That’s the realisation of a new white paper, named ‘Music composition – Human versus Machine’.

Gen AI demonstrates a 20% accuracy rate when used for composing music for specific emotional briefs. It can compete with humans, but not with the precision required for commercial-ready music. 

This objective analysis of the current state of Gen AI music composition, emotional accuracy, and commercial implications for brands was formed by a collaboration between Stephen Arnold Music (SAM), the music branding company, and SoundOut, the sonic branding and marketing testing business. The collaboration sought to measure the current capabilities of Gen AI music composition based on emotional accuracy, appeal, and commercial-readiness for brands. 

Humans demonstrate the ability to compose more emotionally accurate and appealing music than Gen AI, while Gen AI weakens emotional accuracy and appeal when modifying a human-created piece of music. 

The results signify that for commercial-ready music, especially for brands, Gen AI is most effective as a creative tool to assist with ideation and composition, enabling human composers to focus on emotion. This collaborative process coupled with SoundOut’s testing capabilities can lead to efficient and emotive brand music.

 

David Courtier Dutton, CEO, SoundOut, said: “While humans still outperform AI on the emotional front, this study has revealed that AI ‘composing by numbers’ is already not far behind. The AI was not bad, it was just not as good.  

“With some emotional fine tuning we expect that AI will at some point in the not too distant future match the majority of human composers. The AI does not need to understand emotions, just how to invoke them in humans. The 2023 argument that AI cannot be creative has all but been settled, and this study demonstrates that AI can also compose music to move us emotionally, it now just needs a little more technical empathy to be able to do this with sufficient precision for commercial use.”   

Chad Cook, President of Creative & Marketing, SAM, said: “AI has demonstrated that it can be a valuable asset with respect to creative inspiration and ideation. When developing commercial-ready music for leading brands, however, there are additional considerations for evoking the proper emotion at the proper time. Performance, emotional timing, production quality, mixing and mastering are all elements in which the human touch makes a distinct impact. Combining the capabilities of humans and AI has real potential for sonic branding in terms of efficiency and quality.”

 

Test 1: Emotional accuracy of AI composed tracks

In the first test, four music briefs were designed by SAM and SoundOut spanning a variety of human emotions. Each brief was programmed by SAM into Stable Audio, an emerging Gen AI platform. Five tracks were produced for each brief, yielding a total of 20 tracks for the test.

The resultant tracks were then analysed on SoundOut’s OnBrand platform, yielding percentile scores for each track against 212 emotional attributes enabling an analysis of how well each brief had been executed.

Overall the conclusion is that AI has done reasonably well.  Directionally, the performance of the AI is largely successful for most compositions. However, in most cases this falls short of what would be acceptable for commercial usage (80%). That said, around 20% of the time the AI does create something that could genuinely be considered as appropriate.

What is clear is that AI is not yet at a level where it can consistently be relied on to replace human composers on a single shot basis and it should therefore be used as a creative tool — a compositional ‘co-pilot’ — taking the heavy lifting off the initial composition which can then be worked on with more emotional focus by human composers (that we explore in more detail in test 2 below).

Given its 20% accuracy rate, it is unlikely that AI will be the trusted composer for today’s sonic brands, however, when used in collaboration with humans, it provides a very strong ideation platform for music creators. This coupled with SoundOut’s testing capabilities can lead to efficient and effective sonic branding.

 

Test 2: AI versus human

Test 2 built on the first test by examining the comparative appeal and emotional response of hundreds of consumers to music created by AI versus human composers, both from the same music brief. A complex music brief was designed by SAM and SoundOut and given to SAM’s composer, Mack Price, as well as programmed into the popular Gen AI platform, AIVA. The resultant tracks were then tested by SoundOut with its leading consumer music testing platform, Slicethepie.

For both appeal, and for the magnitude of emotional response indicated by BrandMatch, the human composed/modified tracks rated more highly than AI composed/modified tracks. Humans were better at composing or improving AI tracks that deliver the desired emotional impact. Furthermore, humans improved the AI track, while AI worsened the human track.

 

To download the report, click here: https://www.soundout.com/human-vs-machine-ai-music-whitepaper.