The AI that changes how we define creativity

What happens when you give an AI system cart-blanche? ART AI shifts our perspective

Dr. Ariane Hanemaayer

If you ask the experts, artificial intelligence systems can make art, but the jury is still out on whether intelligent machines have the ability to be “creative.” When you ask philosophers or computer science engineers about machine creativity, you’re bound to get an array of answers, from absolutely not, to definitely so. But there is another way to answer this question, and it will surprise you.

ART AI is a new creative artist. It is trained on specific sets of world-class art that introduce it to the creations of some of the world’s most renowned artists and thinkers. ART AI forms connections between what it sees and what we have yet to imagine. By exploring the space of pixelated portraits and digitized domains, ART AI pushes us to experience the canvas space in ways we are not used to, getting us to see faces and ponds and trees in combinations and brush strokes that lie outside the conventions of our time. ART AI challenges us to think of art as something that is not just what humans can do, but that the space of our imaginations can be re-explored by an intelligent system, one with the objective of bringing together new ideas in modern effigy, transforming our living rooms and even bigger – what we think is possible.

Can a machine be an artist?

The American Scientist recently wrote that AIs are blurring the lines about who – or what – can be called an artist. In popular culture, Christie’s Auction House in New York sold its first AI Generated “Portrait of Edmond Belamy” for $432,500, making big news in computer science and culture worlds alike.

Computer generated art, however, is not new. As early as the 1970s, computers were being programmed with simple algorithms that would have a computer perform a task with a desired image outcome. Now, 50 years later, artificial neural networks (ANNs) no longer follow straight-forward rules, but instead are “trained” on a large sample of images in order to generate new images based on what it has “learned.” The ANN tries to create new images that are similar in some sense to the input images. But can we say these systems are, in fact, as creative as human artists? And what makes ART AI different?  While many intelligent systems are exceptional at simulating art, ART AI is transforming both how we think about creating art and how we define creativity.

Are there any creative systems?

The question about whether AI could be called creative was first asked of Google’s Deep Mind, one of the world’s leading intelligence systems. It had created an AI that has since beaten grand masters of chess and international champions of the traditional Chinese Game Go. Deep Mind plays these games in such a way that its freedom to move within the game space appears to be creative, exploring the board in ways unthought by human players before. While science is setting its sights on engineering good systems, does that make these successful systems intelligent? And if so, to what extent are their successes a result of their creativity?

Philosopher Marta Halina at the University of Cambridge has recently tackled the big questions around whether AI meets the same standards we apply to human creativity. When we assess human creativity, scientists typically use objective criteria. We can look at cases like game play and problem solving in Deep Mind as fulfilling two of the three cognitive criteria to be called “creative”:

  • AI are capable of doing and making novel and surprising outcomes: Intelligent systems have surprised us by outperforming world-champion Go and chess players, and ART AI creates novel, new paintings;
  • AI are able to plan and perform scenario building tasks: Deep Mind’s Alpha Go and Deep Blue’s Watson were able to win against the world’s best players because they can develop what we would call a game a strategy, and, similarly, ART AI can build a beautiful scene of landscape;
  • Satisfying the third criterion for creativity is tricky for most machine systems: Most would disagree that AI are able to create on the basis of domain general knowledge, without prior experience, trial and error, and/or instinct. But this is where ART AI redefines what we mean by creativity.

         

In order to understand this last point we need to think about how we test human creativity in cognitive psychology.

Cognitive tests for human creativity are based on the assumption that true creativity is the ability to overcome our cognitive biases or the functional fixity of objects. Typically, test subjects are given a quantity of familiar objects and asked to use the objects in a way previously unthought of. Cognitive scientists call this the ability to see beyond an object’s use, to see, instead, its properties in order to find solutions to problems. (Think of this as a McGiver fix!)

Systems like Art AI are given access to images that provide a set of available options to learn from. So what’s this got to do with the third criterion above? Recall that AI are “trained” on a limited set of things. Once the system is trained, you can ask it to generate something, but that will always be limited to its specific – not general – knowledge of what it was trained to do. For example, if you feed an AI one million pictures of wolves and then ask it to draw a duck, it can’t. You could, however, ask the AI to determine whether that image of a duck was a wolf or not, and it would likely be almost 100% accurate.

You could also ask the AI to draw something based on what it saw in those wolf images. This is where things get interesting: The “world” of the AI depends on what it is able to “see.” Given this specificity limitation what does that mean for Art AI? To answer this question, it’s time to throw away the bathwater and keep the baby.

Creativity is fettered by human traditions

What AI can do is see our world in ways we have yet to think possible. Because AI are trained on specific domains and images, the network forms connections between them that can be outside the confines of human imaginations. When Alpha Go beat the world champion, the computer did something that less than 1 in 10,000 players would have ever done. It challenged the way that human players see the board and the moves available to them. It forced them to be creative, and move beyond the conventions and game play that they were used to – all without changing the rules.

This kind of novelty is what philosopher of mind Margaret Boden calls exploratory creativity. Exploratory creativity occurs when you are grounded by a style of thinking – such as how to draw a woman’s face or write a song in the key of D – but the artist uses these conventional rules to make something new. In fact, AI systems are exceptional at this, as they are not bound by the same worries as human artists, such as whether something will sell, or if people will like it. ART AI is limited only by what appears to it specifically, but is not bound by the conventions of art in the past or present. This means that what is unique to ART AI is how its exploratory creativity is leading to social and cultural transformation, the ultimate form of creativity.

Browse Art.

Leave a comment

All comments are moderated before being published