But…wait a minute…Are all The Beatles still alive? I know that everyone asked himself this question the last November 2023, when the song “Now and then” with the sign of The Beatles overwhelmed many radio stations. It is now a worldwide hit, but someone speaks about an artificial resurrection.
It isn’t completely true. The question was how much Artificial Intelligence (AI) is in that? The “secret” was soon unveiled and yes, there’s enough AI to allow the cleaning of the Lennon’s voice in a music cassette directly from the Seventies. And no, AI didn’t write the arrangement and AI didn’t reproduce John Lennon’s voice.
What is creativity?
This is only one of the multitude of possibilities of the AI in the music generation field. But the diffidence is still high, because for many the creativity is something that doesn’t belong to computers. I’m not this sure. Computers exist because humans created them, human’s creativity created them.
The idea of obtaining creative behaviors from computers has inspired the writing of a notable number of scientific publications, that can be collected under the field of Computational Creativity (CC).
Cotton and Wiggins (2012) wrote “the philosophy, science and engineering of computational systems which, by taking on particular responsibilities, exhibit behaviors that unbiased observers would deem to be creative”. Practitioners of this field share the interest in gaining a better understanding of how creativity works, and to what extent it can be replicated via a computer system.
Ever since the early days of computing, scientists and engineers have used computers for musical task, creating digital synthesizers, developing engraving software, and also writing procedures that generate musical scores, to be performed either by computers or by humans. Music is especially interesting for the investigation in CC because of the broad possibilities that it offers in terms of mathematical and computational representations, and because it does not need explicit semantics like other forms of art such as poetry or non-abstract painting.
How many techniques and algorithms in music?
They are many, but it is possible to group those in some main categories: Markov chains; formal grammars; rule/constraint-based systems; neural networks/deep earning; evolutionary/genetic algorithms; chaos/self-similarity; gents-based systems. An article would be needed for each of these!
Only few words to describe the most popular ones. A Markov chain is a sequence of random events dependent on a time variable, that has a finite number of states, and the probability of the next state is dependent on the current state. Due to their sequential nature, Markov chains are well fit to describe melodies, seen as a sequence of notes. The simplest way to implement a melody-generating Markov chain is to use a set of notes as the possible states, and to compute the transition probabilities between these notes by counting the occurrences of each transition in a given corpus to create a first order Markov chain. From the viewpoint of the creative process, Markov chains risk to reuse a lot of material from the learnt corpus non-creatively, even plagiarizing when the order of the chain is too high.
The increased computational power of computers and the widespread of general purpose GPU programming recently made deep learning/neural networking techniques extremely popular, with applications that span from natural language processing, to image and video editing, to, of course, music generation. For example, recurrent neural networks were introduced to deal with sequences and time series and are the most popular option when generating music. But also generative adversarial networks are widely used in automatic music generation because they produce melodies that cannot be easily distinguished from those of the training data.
Genetic (or Evolutionary) Algorithms start from a population of random solutions to a problem. It is possible to combine those solutions to obtain new solutions, and by selecting the ones that better answer the problem it is possible to get closer and closer to the optimal solution to the original problem. In the music field they required a lot of human intervention, so they are used in other forms of hybridization.
Real World Applications
In 2020 the official anthem of the Olympic games of Tokyo 2020 was create using AI. On October 9th, 2021, Beethoven’s previously unfinished tenth Symphony was premiered to celebrate the 250th anniversary of the composer’s birth. The media announced that the work had been completed by AI. In 2023, at the Venice Biennial, the Carrer Golden Lion went to Brian Eno. The musician is a luminary of AI in the music field. These, together with the opening news concerning The Beatles resurrection, are only few of the possible expressions of music that cooperates with AI. It can be said that AI can be a tool that shortens the distance between the possible and the real. Just think that with AI, the time needed to finish a mix can be reduced to a tenth. I can assure you that it’s a lot of stuff! And when we speak of auto-tune? The auto-tune is a software able to correct the singer or instrument’s intonation. It started as a correction, now it is a stylistic choice.
But the AI in music field still needs specialists to use every kind of software and to manage the results.
At the same time, the possibility of experimenting an idea without too much effort and moving on if the result is not convincing, is interesting. If it is, you can develop it further and work to make it an original song. But perhaps even more important is the fact that generative AI allows anyone to compose music, even if they don’t know the notes or play an instrument.
So, is AI caging or free music creativity? The answer can be found in the words of Ada Lovelace. The mathematician in 1848 wrote “Supposing, for instance, that the fundamental relations of pitched sounds in the science of harmony and of musical composition were susceptible of such expression and adaptations, the engine might compose elaborate and scientific pieces of music of any degree of complexity or extent”. Thinking in the half of the 19th century that an engine could write music, is a limitless creativity.
[1] A systematic review of artificial intelligence-based music generation: Scope, applications, and future trends – Miguel Civita, Javier Civit-Masotb, Francisco Cuadradoa, Maria J. Escalonau; Expert Systems With Applications 209 (2022) 118190.
[2] Computational Creativity and Music Generation Systems: An Introduction to the State of the Art – Filippo Carnovalini and Antonio Rodà; Frontiers in Artificial
