Can AI actually be creative or is it human beings who are hallucinating? We examine how AI is transforming music, art, and more – while learning some science about the heart of the storyteller.
This episode is about creativity. Specifically, what are the unique attributes of human creativity that are beyond AI for the foreseeable future? But also, what aspects of today’s AI demonstrate outputs or behaviors that we might consider creative? They’re not the same thing.
As humans, we find it easy to imagine an animal or a plant creatively solving a challenge in its environment without human intervention. I mean, the nature shows celebrate the abilities of all kinds of creatures and critters to master their worlds. But if we ascribe one ounce of similar feeling towards a mathematical or a silicon system that’s attempting the same thing, well people freak out. I mean, next stop, Terminator or some other digital Overlord.
So, it’s actually going to be our perceptions and our feelings about the difference between creativity as a specific, unique human endowment and outputs and behaviors from AI that might seem creative. And navigating this narrow passage is going to influence or even determine whether we feel like the master or the servant of AI.
There’s been a steady march of AI powered victories over human intellect and strategy games like chess or Go. However, for a long time, people comforted themselves with the idea that an algorithm could never write a poem, or tell a joke, or create music that moved people or painted visual art that inspired. And yet, if you look around, there are multiple AI researchers who are exploring algorithmic poetry, comedy, music and visual art. Because creativity is intrinsic to intelligence among individuals and in groups. And we’re starting to see some initial examples of actual human AI collaborations to create real art.
In October 2018, Christie’s auction house in Paris sold an AI composed portrait of a man for an eye popping $432,500. The painting is called The Portrait of Edmond Bellamy, and what you see is an image of a gentleman who wears a dark frock coat and a plain white collar. He kind of looks like a pudgy man of the church. There’s an unfinished abstract quality to the work. The facial features are in the right places, but they’re not fully formed, almost as if the artist made several erasure strokes at the end. And at the bottom right corner of the painting is a signature, which is an algorithmic formula. And the painting went for $432,500.
The portrait of Edmond Bellamy is one of a group of portraits by a Parisian arts and artificial intelligence collective called Obvious, which you can find it obvious dash art dot com. And the three founders of Obvious are exploring the territory between art and AI using a method called generative adversarial networks or GANs, G-A-N.
Now, to understand how generative adversarial networks operate, you need to go back to 2014. At the time, researchers were using classical machine learning techniques to teach AI systems how to recognize human faces. They were using photos, and then they were asking those systems to create new faces based on those patterns. It was a great theory. The problem was that the results weren’t very good. And then a researcher named Ian Goodfellow suggested a different approach. He used two AI Networks and two competing AI networks to analyze a data set for patterns and take actions, hence, the idea of adversarial networks.
Picture a tennis match in your mind. On one side of the net is an algorithm called a generator. This algorithm takes in the dataset of faces and creates a new face from those patterns it’s identified like noses are found between eyebrows and lips. Ears can be visible or covered by hair. And based on those patterns, the generator creates a new image and serves it up as an authentic image from the dataset. On the other side of the net is an algorithm called a discriminator that is fed the same database. Its job is to decide whether a newly generated image can be distinguished from the other images in the original dataset. And this algorithmic game of serve and volley keeps going until the generator finally produces an image that beats the discriminator.
To develop the portrait of Edmond Bellamy that sold at the Christie’s auction, the artists fed a dataset of 15,000 digitized portraits that were painted between the 14th century and the 20th century to both the generator and discriminator algorithms. And from that data set, the generator made a new image based on the portraits, while the discriminator tried to spot the difference. And after many iterations, the portrait of Edmond Bellamy emerged once the generator finally fooled the discriminator into thinking the new image is a real life portrait. And so AI created a portrait with minimal human intervention that sold for over $400,000.
Now, the interesting questions start because we’re no longer talking about a science project. Who was the actual artist for the portrait of Edmond Bellamy? Was there an actual artist involved, human or not? Well, according to one of the founders of Obvious, if you think that authorship means you’re talking about the entity that directly creates the object of art, well, then the author is the machine. But if you think that authorship is more about holding a vision and wanting to share a message, then that would be the human founders of Obvious.
And these are just some of the philosophical issues at work. I mean, is it really a creative act to digest thousands of prior examples and mimic them? Because that’s how you train an African gray parrot. It can be trained to sing opera or cuss like a sailor. It’s entertaining. It takes work. It takes skill. But is it art that changes people on the inside? But on the other side of the coin, are the techniques being used by generative adversarial networks actually exposing creative behavior that’s universal?
The old aphorism that good artists copy while great artists steal actually might be closer to truth than many people care to admit. Human innovators of all stripes have been shameless about borrowing from the past and remixing the results. I’m not yet saying that generative adversarial networks are composing superior aesthetic work. To be fair, the art is interesting, but not yet inspiring. I’ve heard AI composed music that has distinct structure. The music sounds soothing, but ultimately soulless. And the jokes? Well, let’s just say it’s going to be a while before I belly laugh at an AI authored joke.
But the bigger point is that these attempts at endowing machines with the ability to create artistic work have actually taught researchers as much, if not more, about human creativity and how it works. We want to think we’re the only creative species on this planet, because creativity presupposes an awareness of one’s place in the world, plus a willingness and an ability to change that world, and a desire to share those results with other people. We grudgingly ascribe a soft, creative glow to certain animals, like when we see some chimp fashioning a twig to probe for termites. Or there’s a male Australian bowerbird constructing an elaborate bachelor pad on the forest floor.
What we do know from today’s neuroscience is that creativity or creative talent can’t be isolated to a particular section of the physical brain, like breathing or digestion. Creative thought and activity takes place all across the physical brain, as demonstrated by functional magnetic resonance imaging or fMRI. Researchers have placed musicians, painters, comics and many other creative types into fMRI machines while they played a keyboard or sketched on a pad or told a joke. The researchers are looking for visual evidence of some focused activity or area of the brain that lights up differently with creative individuals rather than with the general public. However, after multiple fMRI studies with some interesting results here and there, we still do not have a physical brain architecture we can point to that distinguishes highly creative individual A from another individual B.
Now, if we move up a level from the viewpoints of cognitive science and psychology, our creative circuitry as human beings arises actually from the interaction of three different mental systems, you could call them networks, that are happening inside our brains. And the first system is our actual drive to create. It’s as powerful as the drive for food or sex. This is the engine room of creativity. It’s the mental muscle that enables a visual artist to paint for hours or compels an NBA star to climb out of bed and practice 200 free throws shots before breakfast. This is the instinct to master a discipline like painting or music, dance, martial arts. And it’s called the Executive Attention Network.
And that creative drive will ultimately cause us to try things we haven’t tried before. We need to see the possibilities before they become material things or actions. And you can call this the Imagination Network. This is the ability to see or feel or hear things that do not exist unless and until the artist expresses them. You’ll never see a worker ant or bee or termite try something just because. But human beings do this all the time. We actively project obvious fictions in our minds, just like the Queen and Alice in Wonderland, who bragged of her ability to believe six impossible things each morning before breakfast. The Imagination Network enables creative people to harness their drive into new expressions of things or experiences that do not yet exist in reality.
The third, and probably the oldest, creative network in the brain is called the Salience Network. This is the network of the brain that looks at the environment and says, “Hey, pay attention to this!” The focus of attention might be to avoid a predator, but also to recognize a ripe fruit or a sexy glance that’s cast your way. And this salience network is different from creative drive because that comes from within. While salience is about understanding the environment around you, what’s happening around you, what are the relationships around you, and then constructing models out of that.
So within people engaged in creative thought, there’s this constant toggling and rebalancing of the creative drive, the imagination and the attention of a given artist. And these balances are affected by what is the given creative act. I mean, after all, you could be in the south of France painting solitary in a garden. But it could also be a time out with only two and a half seconds to play, and your team’s got to make it three point play. Both actions are drawing from the same creative well.
And so creativity, at least based on how we look at it from the perspective of today’s neuroscience and cognitive science is not so much a thing or a particular endowment we can study in isolation. So much as creative thought and expression emerges from the interplay of how a person lives their reality, how they express their imagination based on their experience with the outside world. And then what is the ultimate message this person feels compelled to share with an audience?
We did a show two years ago that featured sound clips from a 1999 interview with David Bowie about the Internet and its impact on music and art. Many people don’t know that, along with being an incredible musician and performer, Bowie was a serious fine art collector. And he was a fan of Marcel Duchamp, who, along with Henri Matisse and Pablo Picasso, was instrumental for redefining visual art in the early 20th century with cubism and other conceptual art. Cubism is a style that’s dominated by a fragmented subject matter that’s deconstructed in such a way that it can be viewed from multiple angles.
What drew David Bowie and others to this type of artistic expression was this idea that a piece of art is incomplete until the audience comes to it. The audience experiences it and adds their own interpretation. And what the piece of art is about is the gray space in the middle. Bowie went on to say that the gray space in the middle is what the 21st century will be all about.
And painting isn’t the only fine art that’s been redefined. In music, Arnold Schoenberg developed a new system based on 12 tones, which was a radical departure from the diatonic scale that dominated Western music from the Middle Ages to the late 19th century. In other words, Schoenberg changed people’s ideas about what is music, and what sounds might be musical.
You can fast forward to Brooklyn in the early to mid 1970s, when Grandmaster Theodore and Grandmaster Flash scratched records back and forth. Did needle drops. Rapped their lyrics instead of singing them. And those sounds found an audience in the gray space in the middle. People who didn’t yet know they really wanted that kind of music until they heard it. But after they heard it, they wanted more. And the more they wanted, the more the artist created with the audience and the more music and music culture innovated off each other.
And so it’s virtually impossible to untangle human creativity from the social context of living as a human being. Creative and aesthetic experience is as much a part of being human as big brains, walking on two limbs, long childhoods and tribal aggression. Therefore, the most powerful creative expressions are those that change our ideas about what it is to be human, and how we fit in the world. So, it’s no wonder that creative artists are suspect by those in power. Because when you change the way a person looks at the world and their place in it — watch out!
According to Sean Dorrance Kelly, who’s a professor of philosophy at Harvard, our human creative achievement will always be socially embedded. And because of that, human creativity will not succumb or be determined by artificial intelligence unless we let it happen to us. If we change the norms of what governs our culture and technology’s role in developing or operating within those norms, yeah technology is going to surpass us in creative achievement. But according to Professor Kelly, that’s going to happen not because of some native advantage of AI and machines, but because we as humans have brought ourselves down to the level of machines.
And so if AI actually enslaves us, it’s going to be because we willingly put the handcuffs on our own damn selves. Once we decide that the data tells us what we are and who we are, we’ve left the gray space in the middle and we have thrown our lot with the machines, and the people who own them.
No one knows why we have this vivid drive to create. Or the ability to imagine things that don’t exist, or the salient understanding of how that new vision might fit in the environment around us. We do know those primal forces are core to how we make the models of the world, and our place in the world. And AI will need similar capabilities if it’s to become intelligent as opposed to just skilled. But let’s not forget, as human beings, we’ve played this creative game a hell of a lot longer than digital algorithms. Life itself is a creative response to a very cold and dead universe.
Perhaps a better way to look at AI is as the next profound creative medium — like cameras. We have little problem separating the artistic vision of a photographer or screenwriter or director from the technology they use to realize that vision. Additionally, we give awards for technical excellence in editing and cinematography, soundtracks and many other subdisciplines that must collaborate to apply intelligence to imagination and make that expression come alive. So, I see the Academy Awards in 2030 having its own category for the best use of data science in a major motion picture.
But at the end of the day, that’s actually the easy work. Far more difficult and far more important for our future on a warming planet is to change how we cultivate creative thought and skills within ourselves. Picasso said, “Every child is an artist. The problem is how to remain an artist once the child grows up.” If you ask a group of four year olds who can draw a doggy, you’re going to see most of the hands raised up. Ask the same question five years later, and you’ll get far fewer hands. And if you ask a corporate audience, you’re going to get silence or nervous giggles at best.
All along the line, our societies, our schools and our organizations make it a point to beat creative thought and expression out of the children so that they can become productive adults, that is, predictable adults. In the United States, a first grader can look forward to taking around 118 standardized tests by the time they graduate high school. You can wordsmith that however you want, but it defies logic to think that our industrial and Internet age education system in this country wants to produce anything other than a talented, loyal worker consumer who will totally accept the environment around them as normal and inevitable.
I’m not saying everyone can be an artist who changes our ideas of what it is to be human. Those extraordinary people have always and will always exist. And historically, it’s been difficult, if not impossible, to eradicate them. Ask all the Kings, party officials, and potentates who tried to do just that.
Mark Twain said, you can’t trust your judgment if your imagination is out of focus. And as humans and as human creators, we can still imagine a lot faster and a lot better than the machines now and for the foreseeable future. And yes, machines can fake us out through mimicry, and that is a serious problem requiring a truly creative solution, which we do not yet have. But the only way I believe we as human beings will be truly enslaved by AI is if we give up on our own imagination.
So, learn about data, learn about AI, apply it, and be open to the idea that machines can create interesting and useful things with minimal human intervention. But keep your freak flag close at hand. Fly it loud. Fly it proud because the world really needs it.