Wolfgang Heckl’s work is reminiscent of Paul Klee’s expressionism with its harmonious, vibrant colours. Only at second glance does it become clear that Heckl has taken a completely new approach to creative expression using artificial intelligence.

Robert Ketterer, Munich

HECKL began his project of transforming images into music and back again with the help of algorithms 15 years ago. Now he is taking the project into a new era and, as an AI artist, is using neural networks for this task for the first time.
Heckl’s works of art are not the result of chance, but of a veritable transformation. This is demonstrated by the reversal exercise. If the image is fed back into the AI, the sounds of the music come back (largely) original.
The artist came up with the idea during a visit to the Lenbach House in Munich. There he saw “Impression III (concert)” – a painting by Wassily Kandinsky, which was created as the essence of a concert by his friend Arnold Schönberg at the beginning of 1911. Kandinsky was enthusiastic about the twelve-tone music he had heard with Franz Marc and other members of the “New Artists’ Association”. Under the impression of the performance, he created “Impression III (concert)”, one of his later most important works. For Kandinsky, the concert experience was not only of a musical nature, but also led to other sensual, above all visual, impressions. In his book “On the Spiritual in Art”, he described in detail only a short time later which colours correspond to certain instruments or tones.

I realised that Kandinsky heard sounds when he saw a colour. And when he saw colours, music played for him,” explains Heckl. “Kandinsky had created his work using only his natural perceptual intelligence. Now I wanted to know whether I could at least do the same with the support of computers and artificial intelligence.

W.M. Heckl

Heckl took the following approach: a neural network was trained using tens of thousands of images to encode visual information and reconstruct the original image in a functional test after decoding. In a second step, this process was also applied to tens of thousands of pieces of music (midi files). Colours were then assigned to the notes and finally a neural connection network was created that links music and image.
“We also carried out functional tests” explains the computer artist, “by having the AI convert the same piece of music into a painting several times. Each time, the results were extremely similar, but not completely identical.”
However, science has by no means solved all the mysteries. “It remains to be seen whether AI can actually generate added value independently of analogue human artists,” says Heckl. “We already know this in the field of pattern recognition, human language comprehension or suggestion algorithms on various sales portals. There are still some unanswered questions when it comes to art. But the idea that AI itself can be an artist is in the world.”