This article is part of Demystifying AI, a series of posts that (try to) disambiguate the jargon and myths surrounding AI.
An AI-made portrait sold for $432,500 at a famous auction last week. This was a story that was widely discussed in tech media in the past week, with some suggesting the development marked a threat for human artists. This is just one of the many stories of progress in deep learning that triggers sensational headlines about AI manifesting artistic creativity that is on par with humans. “AI songwriting has arrived” and “AI will soon write better novels than humans” are just some of the stories that have surfaced on mainstream media in the past few months.
To be true, like any other technological innovation, artificial intelligence—or more specifically, machine learning and deep learning—will affect the way we create music, art and literature. But in the past year, I’ve researched the space and spoke to artists and engineers who were involved in developing and using AI technology that produces works of art, and I can say that nothing could be farther from the truth than fearing AI will be replacing human artists.
And this is true, even if AI creativity sells for hundreds of thousands of dollars.
The difference between human and machine art
To understand the impact of deep learning on human creativity, we must first understand the differences between how AI and humans generate art. Despite its older technology, slower processing power and unstable storage, the human mind undergoes a process that is much more complex than the most advanced AI algorithms when it comes to creating works of art.
When you’re drawing a painting, composing a song, writing a novel (or even this blog post), your life experiences, culture, religion, political and social tendencies all mix into a jumble of emotions and chemical reactions that affect the result of your work. A real analysis of what goes into human creativity is beyond my knowledge or a single post. Suffice it to say that we can’t truly understand the human creativity process and every single work of human art is unique in its own right. Trying to reproduce it would effectively be like trying to step in the same river twice.
On the contrary, we have a pretty good understanding of how AI algorithms generate visual, audio and textual data, even if their inner workings sometimes evade us. At the heart of most recent AI innovations are neural networks, complex structures that are especially good at examining and matching patterns and classifying information. The deep learning techniques used in various art and music generation tools differ, but one specific technology that has become very popular is generative adversarial networks (GAN). GANs involve two neural networks, one that generates new data and a second one that evaluates the first one’s output to see if it passes for a specific class of data.
For instance, if a classifier network has examined enough samples of Irish folkloric music, it will be able to tell you if a new sequence of music notes is of the Irish folkloric class. So an Irish music GAN would have a generator network create music samples and run them through the classifier to see if it passes as Irish music. If the result is not satisfactory, the generator modifies the data, re-runs it through the classifier and repeats the process until the latter rates it as an acceptable Irish music sample.
The same method can be applied to create all sorts of data. Last year, graphic chip giant Nvidia used GANs to create realistic-looking images of people who did not exist. More recently, a team of developers used GANs to create the same painting mentioned at the beginning of the story, the one that sold for more than $400,000. And bad actors used GANs to create fake porn featuring celebrities and politicians.
But let’s be clear, neither the generator network nor the classifier knows anything about the content of the data they’re creating, its artistic value or the potential harm it may cause to other people. There are no emotions involved, no spark of inspiration and imagination. GANs and all other deep learning or AI techniques that manifest creativity use mathematics and statistics to create data and compare them with other samples they’ve previously seen.
But as we’ll see later in this post, a lot of creative power lies in deep learning, if not in the way that some publications and marketing folks like to portray.
Deep learning will automate some creative tasks
While acknowledging the limits of deep learning and neural networks in creating works of art, music and literature, it would be unfair to say there will be no automation of artistic creativity. Like every other industry and domain of human labor that is being affected by AI, human creativity is bound for some disruption and automation. At the very least, we’ll have to rethink our definition of creativity.
My take from talks with different experts is that deep learning will automate some forms of art. For instance, “functional music,” the kind of audio that we play in the background of presentations, ads and some of the simpler video games, can be automated by neural networks that generate new sequences of tunes based on input parameters a user provides. These can be things such as style, tempo and mood. There are already several companies that have developed or are developing such AI applications, and they have a very rich market for their products.
Automated functional music will enable people to access affordable creation of background music without hiring the expensive services of a human composer.
It’s easy to see similar developments happen in the visual arts domain, where AI algorithms can create unique functional visual effects for the background of videos and presentations.
But functional art might not exactly be considered creative content. They’re made to help the user focus on something else, such as the content of a presentation or video. “We think of functional music as music that is valued for its use case and not for the creativity or collaboration that went into making it,” Drew Silverstein, CEO and co-founder of Amper Music, an AI startup based in New York, told me in December. But artistic music, Silverstein explained, “is much more about the process than the use case. Steven Spielberg and John Williams writing the score of Star Wars, that’s about a human collaboration.”
However, there are people who are creating functional music and art for a living. What will happen when their jobs are automated?
Deep learning will augment human creativity
Functional art accounts for a small fraction of the industry. The real development that deep learning and advances in the broader AI industry will fulfill is the augmentation of human capabilities. In fact, the likely outcome is that neural networks and deep learning will make it easier for more people to become creative.
There are already different deep learning tools that are helping enhance the creative skills of both amateurs and professional artists. For instance, neural networks can take a drawing and modify it to give it a Van Gogh or Picasso style. Another example is a tool developed by Google that uses machine learning to examine rough sketches and transform them into crisp drawings.
On a more professional level, deep learning can help artists find new ideas and speed up their creative process. Last year, I wrote about folk-rnn, a deep learning application that created Irish Celtic music. I spoke to the developers, but I also spoke to Irish musician and composer Daren Banarsë, who had the chance to examine the compositions of the neural network.
Banarsë told me that while he was surprised at how good some of the compositions were, it was clear that they still needed a human touch to become complete. And he wasn’t worried about the neural network replacing his job. What was intriguing were some of the new ideas that folk-rnn came up with, things that he wouldn’t have thought of.
Banarsë hoped that tools like folk-rnn could assist him in his job. “I always find it daunting when I have to start a large scale composition. Maybe I could give the computer a few parameters: the number of players, the mood, even the names of some of my favourite composers, and it could generate a basic structure for me. I wouldn’t expect it to work out of the box, but it would be a starting point,” he said.
In the domain of literature, deep learning has also found some interesting use cases in assisting professional writers, even if AI’s understanding of human language is limited. This month, The New York Times ran a story which described how a writer was using machine learning to find ideas for his writing and suggestions for completing his sentences. There weren’t much in terms of the technical details behind the software. But the likely technology used is natural language processing and generation (NLP/NLG), a branch of AI that helps computers analyze and create human text. Again, there’s no creativity involved here, just statistical pattern matching and prediction. But like in the case of folk-rnn, NLP/NLG can sometimes come up with interesting ideas that the writer wouldn’t have thought about.
In some ways, this is reminiscent of the advent application programming interfaces (API) as an independent business model. APIs did not drive programmers out of business, but they made it possible for many more people who did not have any background in programming to develop applications.
Francois Pachet, director of Spotify’s Creator Technology Research Lab, whom I spoke to in December, compared creative AI tools with the digital synthesizers of the 80s, which at the time caused fears that computers would cause musicians to lose their jobs. “What happened was the exact opposite, in a sense that everyone took these new machines and hardware with them and learned how to use them productively. The music industry exploded in some sense,” he said.
Why AI won’t replace human creativity
AI technology will continue to improve and become better at imitating human creativity. At some point, it might even manage to create music and art that is indistinguishable from that of human artists.
But what makes works of art valuable isn’t necessarily the output. Most of the time, the process and labor that goes into a human achievement is just as important and dear as the end result. Take the robot in the following video. It never misses a shot, and in the same period it takes to train a professional basketball player (decades maybe?), you can produce millions of these robots. Furthermore, if you ever develop a new update that teaches the robot a new skill, you can quickly roll it out to all of its kind, something you can’t do with human players. And the robot won’t retire after 10, 15 or 20 years.
But what we appreciate in human athletes isn’t only their stats and performance in the court or field. We learn to appreciate their struggles and efforts to reach and preserve their perfection despite every obstacle that stands in their way.
Likewise, we admire the story of musicians, artists, writers and every creative human because of their personal struggles, how they overcome life’s challenges and find inspiration from everything they’ve been through. That’s the true nature of human art. That’s something that can’t be automated, even if we achieve the always-elusive general artificial intelligence.
What inspired me to write this post was about the story of the GAN-generated painting that had sold at an auction. But what was interesting about it was the humans who had developed it and the controversies that had gone into creating the GAN and naming the painting (an interesting twist on the name of Goodfellow, the inventor of generative adversarial networks). Again, it was the human aspect that made the work creative and interesting, not the neural networks burning through CPU cycles and crunching numbers.
This #AI-generated art sold for over $400k. But what made it significant was the humans involved in the process, not the algorithms that created the painting. AI-generated art will improve, but artistic creativity will remain a human discipline.https://t.co/2wGpmDpWCz
— Ben Dickson (@bendee983) October 26, 2018
If anything, AI will make us more creative, not less.