Perspective on "Creative" AI

by Nick Briz

Oct 5, 2017

At Branger_Briz we've been working on Machine Learning projects as far back as 2014 with one of our long time clients Sportsmanias. But it wasn't until more recently that we've devoted much of our research and development efforts to exploring the latest in the field of Artificial Intelligence. Much of this R&D has already found it's way into a couple of our commercial projects, for example the AI software pipeline we produced for the band Muse which auto-generated daily supercut remixes for the band's latest single Dig Down. We are now working on two interactive installations on for the Miami Children's Museum which feature a set of custom trained Machine Learning "style-transfer" models.

Throughout our time researching technical progress in the field we've also paid close attention to the way the general public has perceived these developments. The way AI is discussed in social and mainstream media at times sells the technology short and other times expects it to perform miracles. Both reactions are understandable, these technologies move at such a fast pace that it causes many to overlook some serious advancements as well as assume anything and everything is possible. But what concerns us more than mis-calibrated expectations is the anthropomorphization of these technologies. We call the field Artificial "Intelligence", we refer to the latest designs of deeply nested linear-algebra functions artificial "neural networks", we call the process of using data to adjust the internal parameters of these functions "learning" and we refer to the final arrangement of values a "trained" model. While these are all useful metaphors, they can have the negative side-effect of leading us to believe that this math is a conscious being. When these algorithms can be leveraged to generate new data (images, music, texts) we mistakenly ask, "is the AI being creative?"

AI is "creative" in the same way an assembly line robot does "labor". These are tools designed for specific tasks, they are not sentient beings exerting mental effort.

In January we published a blog post sharing some of our developments creating artificial neural networks which can produce new pieces of music. Work like this has lead others to ask the question, "if the AI produces music does it deserve the copyright?", "what will happen to musicians when they're replaced by these creative machines?" Though understandable, these questions are misinformed. I recently spoke at Copy Camp, an amazing conference in Warsaw Poland devoted to discussing the role copyright and intellectual property laws play in our society as well as how best to improve them. I shared some of the music AI research we've been doing, and more importantly, the perspective we're approaching this conversation from.

The question, "if the AI creates a work, does the AI deserve the copyright?" was raised in a couple of presentations, our response is: of course not. AI is "creative" in the same way an assembly line robot does "labor". These are tools designed for specific tasks, they are not sentient beings exerting mental effort, no matter how well that mechanical arm welds or how well the algorithm arranges musical notes. how might these new AI tools change the way musicians make music? The notion that these musical algorithms will one day replace human musicians is hyperbolic and uninteresting to us. A better question might be, "how might these new AI tools change the way musicians make music?"

This is the question driving our current research. We think these new tools have the possibility to change the way musicians make music in the same way the sampler did. The sampler opened the door to entirely new musical genres. It was a new meta-instrument which could become any other instrument. The sampler, and the sampling techniques which preceded it, introduced new forms of auditory collage we'd never heard before. And we think a similar door has just been opened. Below is an excerpt from my presentation which introduces our vision.

Have some thoughts to share? Join the public conversation about this post on Twitter, or send an email, we'd love to hear what you think!

machine learningAIalgorithms