If a machine creates an image that looks like art, can we call that creativity? What if that work has earned a gallery exhibition? Who is writing the poetry if we’ve trained the machine to write?
Artists and Machine Intelligence (AMI) is a program at Google that explores the relationships between art, humans and AI. Blaise Agüera y Arcas, who heads up Google’s Machine Intelligence Lab in Seattle, started the program when his research group had unexpectedly discovered that the “flip side of perception is creativity.”
“A lot of the things that we’re working on have to do with perception, like face or object recognition,” he says of his day job. “A cool finding was that these networks, which are designed to categorize or to classify scenes and objects, can also be run in reverse and synthesize scenes and objects.”
The most well-known project from the group is DeepDream. Alex Mordvintsev, a Google software engineer, invented this generative AI while he was trying to visualize what an image recognition network responded to. Mike Tyka, another software engineer at Google, saw potential in it for art and helped further develop the techniques to create images that were shown in an exhibition in San Francisco last year.
“A network of ‘neurons’ can be trained to distinguish cats' eyes, ears, and whiskers when shown thousands of pictures of cats,” Tyka explains. “Once trained, the network can now create a new image based on the rules and associations that it learned during the training period. Using neurons deep within the network often leads to images with interesting combinations of features the network has learned.”
Ross Goodwin, another artist working with AMI, focuses on written creations. His inventions are word generators, algorithms taught to write by feeding massive collections (in the millions of words) of books or poetry from various genres, depending on how he wants the machine to write. After starting with some “word salad”-type results, he’s gotten to some eerily thoughtful productions, including a sci-fi movie script.
Another project, word.camera, writes poetry based on what it sees in an image that it takes or is uploaded.
Goodwin has also experimented with training an AI on the Oxford English dictionary. He’s turned it into a bot on Twitter called “lexiconjure”, which makes up words and definitions every couple hours.
Clearly perception and creativity are intimately connected; but can we call it creativity? As the head of AMI, Blaise has been thinking a lot of about this. We caught up with him to hear some of his thoughts on the creativity question.
What is creativity if it is artificial?
Blaise: There is no hard and fast line between what creativity is and what it isn’t. Asking, “Is X, Y or Z creativity?” or “Is X, Y or Z art?” are problematic questions to ask.
I feel we try to taxonomize things and make up categories in a lot of different ways, but these categories always break down the more closely you look at them. Personally, one of my themes over the last year has been recognizing that there are not very many hard and fast lines left—if there ever were.
Is the job of creation becoming more like curation?
Blaise: In a way. As the tools become more powerful, what we have thought of as creation becomes the end of a giant lever. Rather than worrying about every work or every brushstroke, you can produce a lot by picking and curating.
It’s similar to the photography business: if you’re a photographer, you’re using tools, your instruments, to essentially curate views on the world. You shape these views using higher-level abstractions and capabilities, rather than worrying about every individual stroke of the brush or pen.
I also think it’s over-simplifying to say, “Well, therefore, everything else wasn’t creativity” or insist, “These are just tools,” which belittles the camera in a way. You can look through a telescope equally through two ends and decide, the camera is just a tool and the photographer is the artist, or the whole camera-photographer assembly is the artist.
Or, you can concede that the artist doesn’t have a well-defined boundary—these questions about creative agency are just too complicated; and I don’t think that it’s easily reducible to a simple statement like, “this is a tool.” Many think that on one side, there are people and on the other side, there are tools—but it’s just not that way.
What’s wrong with wanting to differentiate between human creativity and artificial creativity?
Blaise: I think it is actually quite dangerous to try to draw binaries because that’s exactly the sort of process that makes us want to seek out the things that make us uniquely human and defend them. This defensiveness causes backlash and gnashing of teeth and is a very old story; look at the steam shovel and 19th century industrialization. The judgments and contexts of “that’s uniquely human” seem to me like foolish efforts. Really, let’s fast-forward twenty years and see if our binaries hold up.
I don’t think there’s anything uniquely human about intuition or context or judgment. At an even more fundamental level, I just don’t think the whole question of trying to seek out what is uniquely human is all that interesting.
We don’t have fur because we have been making clothing for a long time. We have a short gut because we figured out how to use fire and harness this kind of chemical form outside of our bodies. These are examples of profound, technologically mediated modifications to humans.
So, isn’t it reasonable to think about fire, clothing and other technologies as part of the human? Or is it the case that the whole idea of thinking about these as categories at all is just a little bit limiting?
I’m not saying that there is no such thing as categories—it is a useful abstraction—but where I think people go wrong is moving from the abstraction, that pattern that seems evident, to assuming that everything is logically deducible from that binary. You can get yourself into real trouble
Where do you think this type of creative work is going?
Blaise: It feels to me that this is a qualitatively different cycle than any that proceeded; and when you have these big leaps, people are prone to hyping. I think we are a long way from having to worry about Ex Machina kind of stuff or some of the questions that I suspect Westworld is bringing up.
These questions are premature in a lot of ways because the actual breakthroughs that we are making now are really about taking the functions that our cerebral cortex does and figuring out how we can make computers, based on silicon, do things that look similar to those kinds of tasks.
The purpose of Artists and Machine Intelligence is to venture into territory where we really don’t know what the outcomes are going to be. We began to see these interesting generative things come out of parts of the human community that are concerned with understanding the generation of media and the generation of ideas.
It has been a really productive dialogue and some interesting things have come out of it. Even some beautiful things have come out of it. It has certainly affected our thinking and enriched the cultural discussion—and that’s good enough for me.
Portrait photos by Sarah Ouellet