This is the central question behind Google Brain's MAGENTA project:
"Can we use machine learning to create compelling art and music?"
The most striking word in that sentence is "compelling" - with it, the bar is set high.
Inspired by Google's DeepDream project, the project looks to also develop novel tools that artists and musicians can use to inform and extend their practice. While MAGENTA is still under development, there are other artificial systems that have been devised to generate art hat one can play around with. For instance, using what is referred to as Deep Neural Network models, an algorithm for artistic style was devised by Gatys, Ecker & Bethge (2015). A very accessible talk by the lead author explaining the science behind algorithm can be found HERE.
An app was developed using this work (DEEPART) that allows for photographs to be uploaded and then rendered into a painting that follows specific artistic styles. I tinkered with it to get a better understanding of what these algorithms were doing. What kind of pictures would be best rendered across all styles? And will these renditions it allow me to learn anything about perception? Will I be "compelled" by the results?
I experimented with one of my favourite subjects and one of his favourite subjects to cover two semantic categories (biological category: human; artifact category: train) with the following results.
"Can we use machine learning to create compelling art and music?"
The most striking word in that sentence is "compelling" - with it, the bar is set high.
Inspired by Google's DeepDream project, the project looks to also develop novel tools that artists and musicians can use to inform and extend their practice. While MAGENTA is still under development, there are other artificial systems that have been devised to generate art hat one can play around with. For instance, using what is referred to as Deep Neural Network models, an algorithm for artistic style was devised by Gatys, Ecker & Bethge (2015). A very accessible talk by the lead author explaining the science behind algorithm can be found HERE.
An app was developed using this work (DEEPART) that allows for photographs to be uploaded and then rendered into a painting that follows specific artistic styles. I tinkered with it to get a better understanding of what these algorithms were doing. What kind of pictures would be best rendered across all styles? And will these renditions it allow me to learn anything about perception? Will I be "compelled" by the results?
I experimented with one of my favourite subjects and one of his favourite subjects to cover two semantic categories (biological category: human; artifact category: train) with the following results.
If MAGENTA and other AI projects reach their goals in the creation of compelling art, will the arts suffer an existential crisis? It's difficult to say, but it is certainly possible. But the spread and popularization of artificial creativity or artificial imagination will also trigger human activity towards novel directions of creation. One needs only to be reminded of the impact that photography had on painters. In the words of E.H. Gombrich in The Story of Art:
"The camera helped to discover the charm of the fortuitous view and of the unexpected angle. Moreover, the development of photography was bound to push artists further on their way of exploration and experiment. There was no need for painting to perform a task which a mechanical device could perform better and more cheaply. We must not forget that in the past the art of painting served a number of utilitarian ends ... Photography in the nineteenth century was about to take over this function of pictorial art. It was a blow to the position of artists ... So it came about that artists were increasingly compelled to explore regions where photography could not follow them. In fact, modern art would hardly have become what it is without the impact of this invention."
RSS Feed