Algorithms are now capable of creating original works of art by taking inspiration from thousands of images and inventing new artistic styles. Here follows an overview of these “artist-algorithms” that are shaking up the art world and questioning our conception of creativity.
One of the first examples of algorithmic art - art generated by algorithms - dates back to 1973, when English painter Harold Cohen wrote a computer program called AARON that could produce original drawings. American artist Jean-Pierre Hébert drew the outlines of this artistic movement twenty years later and invented the term “algorist”. An artist is an algorist when they create a work of art from an algorithm that they have designed themselves. The act of creation is in the writing of the code, which becomes an integral part of the final work.
Advances made in artificial intelligence (AI) are questioning this definition and bringing about a new generation of models. Thanks to machine learning, algorithms no longer simply follow a set of pre-defined rules by the programmer-artist. Fed with a large amount of data, they assimilate the aesthetic characteristics of artistic corpora and become ever more autonomous in the production of content. Since the 2010s, many families of algorithms have been used to explore new practices and keep on pushing back the boundaries of “artificial creativity”.
Composers have taken hold of Markov chains
Musicians have been forerunners where algorithmic art is concerned. Very early on they looked into using computer programs to compose music. As early as 1957, two Americans, composer Lejaren Hiller and mathematician Leonard Isaacson, were programming supercomputer ILLIAC I to generate musical suites using Markov chains .
Today, Markovian models are trained via machine learning, using existing pieces of music. They analyze the musical characteristics (rhythm, tempo, melody, etc.) and the sequences of notes to determine the probability of one note following another. They can then generate new pieces of music in the same style as the original corpus. Markov chains are used in jazz improvisation for example.
Evolutionary algorithms imitate creative thought
Less widely-publicized, evolutionary algorithms are also used to generate credible works of art. Inspired by Charles Darwin’s theory of the evolution of species, they are based on the three fundamental principles of natural selection. According to these principles, there are differences between individuals of a same species (principle of variation). Some traits are more advantageous than others and enable the individuals that have them to survive and reproduce better than their counterparts (principle of adaptation). These traits are passed on from one generation to the next (principle of heredity). The idea behind creative evolutionary algorithms is to reproduce the intellectual approach of the artist, who imagines, tests, and selects new ideas. This means modifying entry data randomly and in a variety of ways, selecting the best-adapted variant or variants, and repeating the process until a satisfactory idea emerges. During this iterative process, the artist intervenes to choose the most aesthetic variations of a generation, but it is also possible to automate this step.
