In 2017, AIVA Technologies built an AI called ‘Aiva’, which stands for Artificial Intelligence Virtual Artist. It composes classical music, a form of art classified by its derivation from human qualities. Apart from being used for advertising or soundtracks in films, Aiva was also invited to participate in well-known events including the Artificial Intelligence in Business & Entrepreneurship Summit (London).
How does it learn?
Aiva utilizes reinforcement learning, meaning the AI selects actions in an environment, and those actions change the state of the environment, leading to rewards or losses in the game. Hence, the machine needs to decide which action best maximizes its cumulative reward. This method of machine learning stands in stark contrast to supervised learning, which focuses on classification through inputs and outputs. Regarding the suitability of reinforcement learning with Aiva, this means the AI can fully explore the world of music and embrace the nuances of musical composition. The team has taught the AI by giving it segments of music creations from well-known artists (i.e. Mozart, Beethoven); from there it learnt by itself the concepts of music theory and produced its own sheet music. Then, professional artists make the music come to life with actual instruments, formulating sound recordings.
Why focus on classical music?
The founders chose classical music for two core reasons. For one, it is the most universal type of music used across industries, whether it be in trailer soundtracks or games. Secondly, the training pieces were copyright-free. Even though these partitions are no longer copyrighted, Aiva’s creations are registered under SACEM, so aren’t in the public domain.
Aiva technologies hopes Aiva will be able to learn other styles of music sooner or later. This will be especially difficult with modern music, however, the sound design is often so distinct that it becomes hard to replicate. This is definitely a barrier that the team will seek to overcome.
Furthermore, Turing test results have shown that professional musicians were unable to distinguish between real music and AI music, raising the question of whether technology will fully take over the music realm one day. This is not yet a worrying matter; Aiva works with humans in orchestration of its written pieces. In fact, the team doesn’t aim for Aiva to replace humans, rather wishes for Aiva work alongside them to enrich their creative outputs.
Written by Amanda Y.