Gizmos & Gadgets: AIs Compose Songs Proving Creativity is More than Human

May 28, 2017 – A team of researchers at Kingston University in London asked the artificial intelligence (AI) program they were working with to compose its own music. Music Technology Senior Lecturer, Dr. Oded Ben-Tal, led a team that fed the AI system 23,000 Irish folk songs. They chose Irish folk music as a genre rich in many songs.

States Ben-Tal, “We didn’t expect any of the machine-generated melodies to be very good, but we and several other musicians we worked with were really surprised at the quality of the music the system created.”

Initially, to get the AI going it was fed a few notes. It then took off from there creating its own unique song which you can hear by clicking this link.

One thing this experiment answers is the question whether AI has the ability to be artistically creative. In my opinion, the composition does not approach the musicality of a melody like Danny Boy, but it does have an Irish feel.

States Ben-Tal, “People are reluctant to believe machines can be creative – it’s seen as a very human trait. However, the fact of the matter is, technology and creativity have been interconnected for a long time and this is just another step in that direction.”

Kingston University isn’t alone in experimenting with AI composers. Google Brain‘s Project Magenta has flirted with composition. In this case, the music is a far cry from what Kingston produced. It appears that going into music composition blind, as Google’s AI apparently did, produces disharmony with a beat.

And then there is Professor David Cope, University of California Santa Cruz, who has created a series of computer algorithms to generate complete orchestral compositions. Cope has been at it since 1981 and his AI produces music that emulates Bach and Mozart so well that even music aficionados cannot tell whether the composition is human produced or not. Listen to this prelude produced by an AI on a programmed clavier. It has Bach written all over it but lacks a little of the expression a human player might provide. The advantage to Cope’s approach is similar to that done at Kingston.

The advantage to Cope’s approach, versus Google’s Project Magenta, is similar to that experienced by the Kingston AI. In both cases the algorithms have learned from existing composition and then experimented to emulate the style of the genre or composer. States Cope, it opens “an arena of creativity that could not have been imagined by someone even 50 years ago.”