40 What Drives Creativity?
Programmability, of course, is not the only feature of computers which makes people doubt their relevance to creativity.
—Margaret Boden18
We can apply the criteria of novelty, usefulness, and surprise to the products of creativity. But how do we assess the thought processes that lead to them?
What is it that drives us—and perhaps computers, too—to be creative? This question goes to the very core of the creative process. Finding an answer will take us a long way toward unravelling the mystery of creativity both in us and in machines—though that will in no way lessen this fascinating quality. And in fact computers may well turn out to be our guide in understanding the human brain.
Cognitive science explores the ways in which people and computers move from problem to solution, following the most promising paths, by using effective methods for problem-solving: heuristics. Herbert Simon, inventor of the BACON program, explored creativity some thirty years ago. An essential element in his work was the assumption that, for example, the only difference between Einstein and his fellow physicists in 1905 was that Einstein had better problem-solving abilities—better heuristics. Simon made this assumption when he designed his BACON software, which he based on extensive analysis of how ordinary people solve problems.
But this assumption is hardly accurate in the case of minds on the level of Bach, Einstein, or Picasso. People operating at that level often invent the problem, whereas in Simon’s formulation the problem is already there to be solved.
Margaret Boden and Computer Creativity
In her seminal work The Creative Mind, Margaret Boden breaks the creative process down into three types: combinatorial creativity, combining familiar ideas in new and illuminating ways; exploratory creativity, pushing the boundaries of the familiar; and transformational creativity, the great creative leaps of the imagination. What happens when we apply these processes to the way computers work?
Boden claims that AI uses all three types of creativity, though the results may sometimes be attributed to the human operator. But, she continues, these three types of creativity “aren’t found in the proportions one might expect” and there is also a degree of overlap.19
Apparently, there are few computer systems that exhibit combinational creativity. This is surprising. One would expect computers to be adept at coming up with links between stored facts. Boden suggests joke-generating programs, to which I would add programs that generate metaphors. These use programs that employ pattern-matching rules and draw on huge databases of words and puns. Another example is Herbert Simon’s discovery programs, which manipulate data using equations (rules) to select ratios of quantities, such as time and distance. They also demonstrate exploratory creativity in that they explore new combinations of words and numbers. But none of these programs have come up with anything startlingly novel.
Harold Cohen’s picture-making program AARON also shows combinational and exploratory creativity, combining images—of arms, legs, flowers, trees, colors—that Cohen has programmed into it to create new art.
In Boden’s view, exploratory creativity is the most relevant description of the way computers work. She considers David Cope’s method of composition to be purely exploratory and capable of producing results indistinguishable from human achievements.20 In my view, his work could also be said to employ considerable combinational creativity in the way he uses Markov chains to draw on his database of music as he seeks strings of notes to create pleasing combinations and construct a composition.
Boden also suggests that computer-generated art exemplifies exploratory and transformational creativity.
Certainly we can see these leaps of the imagination in action in, for example, Tony Veale’s inspired algorithm for generating metaphors, in Ross Goodwin’s development of Char-RNN for writing scripts, and in François Pachet’s Flow Machine, which not only creates music but is also designed to enhance human creativity.
Boden remarks that the results of transformational creativity may be so outlandish that they seem jolting or even repellent.21 This is one factor that Ahmed Elgammal had to take into account when he developed his CAN, with the express aim of seeking out artistic styles that had never been seen before.
How do the concepts of little-c creativity and big-C Creativity apply to computers? To recap, little-c creativity is an everyday experience such as thinking up a new route to work, a discovery that has novelty, value, and surprise at least for the discoverer. Big-C Creativity, conversely, is a transformational idea, an idea that no one has ever had before.
The terms little-c creativity and big-C Creativity most commonly appear in books dealing with human creativity, but Boden applies them to computer creativity, too. In her books she uses different terminology. She replaces little-c creativity with P-creativity—psychological creativity, personal everyday creativity—and big-C Creativity with H-creativity—historical creativity, an idea no one has ever had before in the entire history of the world. Her books are much read among computer scientists, and as a result these are the preferred terms in the computer science world.
Computers can certainly be little-c creative, as we have seen in examining the art, literature, and music they have produced. It may be only a matter of time before they become big-C Creative. In The Creative Mind, Boden poses what she calls the four Lovelace questions—after Ada Lovelace, who developed the first algorithm to be used on a computer. These are as follows:
- Can computational concepts help us understand human creativity?
- Could a computer ever appear to be creative?
- Can a computer appear to recognize creativity?
- Could a computer, no matter how impressive its performance, ever really be creative?22
On the basis of examples from around 1990, when the book was published, including art, poetry, and scientific discoveries, she concludes that the answer to the first three is “Yes.” In her book, she focuses primarily on the first Lovelace question: Can computational concepts help us understand human creativity? She goes on to say that questions one to three are empirical and scientific, while the fourth is not scientific but philosophical. In her last chapter, she discusses it from a philosophical perspective and concludes that the answer has to be “No.”
In 1994, Boden was invited to publish a chapter-by-chapter précis of The Creative Mind in a special issue of the Behavioral and Brain Sciences journal, along with replies from creativity researchers—testimony to its importance.23 Robert J. Sternberg, a cognitive scientist and prolific writer on creativity, mentioned his disappointment that Boden discussed what he considered the “fundamental question”—Could a machine really be creative?—only in her final chapter and then only in a philosophical vein. It should, he wrote, have been the “focus of the book.”24
One wonders whether Boden continues to believe that the question of whether computers can really be creative is purely philosophical.