Creative computing

Margaret Boden believes that computers can teach us a lot about human creativity. Aisling Irwin went to meet her. The story of Henry Ant, composed by a computer program called Tale-Spin, is ammunition for those who say computers cannot be creative: "Henry Ant was thirsty. He walked over to the river bank where his good friend Bill Bird was sitting. Henry slipped and fell in the river. He was unable to call for help. He drowned. The End."

The story illustrates one of the difficulties of modelling literary creativity via computer programs: writing a story depends on layers of background knowledge that we did not even know we had. In the case of Henry Ant the programmer forgot to tell Tale-Spin that people generally notice what is happening to others near to them. Bill Bird should have noticed, and rescued, Henry Ant.

This program was written more than ten years ago and more sophisticated work has been developed since. But much more interesting than the perpetual discussion about whether computing - from ordinary, step-by-step programming to neural networks - can ever really be creative is the question "What insight can computer programs give into human creativity: can computing science explain human originality?" One of the first people to consider this question was Margaret Boden, now professor of psychology and philosophy at Sussex University. Her work involved taking human concepts such as "representation" or "meaning" and considering whether computer programs could help in developing a scientific explanation for them. At the least, Boden managed to convince a generation of non-computer scientists that artificial intelligence was worth a glance.

Boden's living room walls are adorned with pictures created by computers. She sits there, larger than life, among the pictures, a collection of antique glassware and myriad books, and describes the moment when she discovered the possibilities of artificial intelligence.

It was just after she had arrived at Harvard University as a doctoral student in the early 1960s. Standing in a bookshop she read Plans and the Structure of Behaviour by Miller, Galanter and Pribram, which tried to apply computer programming to the whole of psychology. "It seemed to me like a flash of lightening - a promising way of asking and perhaps answering a number of questions that I had been asking for years.

"It hadn't struck me that you could actually use computers to explore theories about psychology in general," she says. "I was an idiot."

Attempts to link the two were just beginning in the United States. "The ideas were in the air but there weren't courses in it." She was in the right place at the right time. She was also at the right time personally: the wanderings of her background education had led her to the point at which she was ready to develop such ideas. Boden studied medical sciences and philosophy at Cambridge with the intention of becoming a psychiatrist. Then, just as she was going to start training at St Thomas's hospital in London, she was offered a post teaching philosophy at Birmingham University. "I soon got bored with it. I wanted to do things that didn't really exist then, like philosophy of biology or psychology."

So she went to Harvard to do her doctorate in social psychology, specialising in reducing purposive explanations of behaviour (explanations couched in terms of our intentions) to mechanistic ones (physical and chemical explanations) This formed the core of her first book, Purposive Explanation in Psychology which considers the ways in which people apply the concept of purpose across a range of psychological theories, using intentions to explain behaviour. Boden studied a psychologist who absolutely denied that there could be mechanistic explanations for intention - she reinterpreted his purposive theories using computer programs.

Artificial Intelligence and Natural Man followed, a guide for perplexed non-computer scientists to the guts and implications of AI's attempt to build robots and computer programs capable of mimicking human behaviour.

But it is a later book, The Creative Mind, which strives for a scientific understanding of creativity, which has proved her most popular. In order to avoid putting critics off at the first page Boden divided her investigation into four questions: can computational ideas help us understand how creativity is possible? Can computers produce work that appears to be creative? Can computers recognise that a work is creative? And can computers ever produce a work that is truly creative?

The book starts by avoiding computer concepts completely. Instead Boden explores how we think in certain areas - for example in music - in ways that may generate creative ideas. Each area is fenced in by rules and mental structures. All types of thinking - music, maths, literature - operate within constraints without which thought would be impossible. There are genres, theories or systems that keep out the unthinkable using fences made of rules and mental structures. Inside the fencing is an area in which a disciplined form of thinking can occur. Boden calls this a "conceptual space".

To explore these conceptual spaces we need mind maps, which tell us where to go, what there is and where the limits are. We spontaneously construct mind maps of the processes going on in our minds. We use these maps to explore our conceptual spaces. As we get to know the subject we develop maps of our maps.

The idea of mind maps has brought insights into the creativity of children and how it becomes more flexible as they grow up. Older children, for example, can on request draw abnormal people - perhaps with two heads or no legs. Younger children, although they can draw normal people as fluently as the older children can, just do not have the flexibility to branch out into two heads or no legs. They have not developed good internal maps of their skills, says Boden. The younger children have a skill which can be switched on but cannot be varied. The older children have explored their skill and redescribed it at higher levels, breaking it down and analysing it as they go.

The conceptual space that we navigate with a mind map is created by a "generative system". It "generates" a list of cans and cannots. The space is explored by cranking the generator. New chemical elements were discovered by exploring the Mendeleyev's periodic table - it was a generative system.

In order to move in a coherent way through a big conceptual space created by a generative system and described by a mind map, we need a set of heuristics. "'Protect your queen' directs you into some chess paths and away from others," says Boden.

New ideas can come from exploring the conceptual space. But the creative ideas that intrigue us often result from a thinker transforming that space so that the topography or the constraints change and new ideas are possible that could not have been entertained before. In this way, says Boden, Kepler's theory that the planets rotated in elliptical orbits, came from a transformation of the conceptual space that his antecedents had worked within.

A new heuristic could transform the space: "Consider the negative, if applied at a relatively deep level of the generative system, can transform the space so fundamentally that very different sorts of location are created and many previous locations, indeed whole regions, simply cease to exist."

Boden distinguishes between a novelty (for example a nonsense sentence that has previously never been uttered) and a creative idea. "Creative ideas are surprising, yes. They go against our expectations. But something wholly unconnected with the familiar arouses not surprise so much as bewilderment. To be sure the lack of connection with what went before may be apparent rather than real. But someone to whom the connection is not apparent will not be able to recognize the idea as creative (as opposed to new). Nor will they be able to see it as relevant to what they had regarded as the problem domain in question: 'That's not art!', 'Call that poetry?'."

Then Boden slips in the computer concepts. "Fortunately a science already exists in which conceptual spaces can be precisely described: namely, artificial intelligence." The younger children's drawing skills, for example, have been compared with a 'compiled procedure' - a sequence of instructions whose internal structure is not accessible to any higher level routine, so cannot be altered."

Boden says: "You may object, 'Constraints, yes. Computer programs - never!' But since creativity is a question of what thoughts can and cannot result from particular mental structures and processes, anyone seeking to understand it needs to be able to describe those structures and processes clearly, and assess their generative potential rigorously. This is why it is helpful to use AI-terms to describe the creative constraints in human minds."

Boden's insistence that ordinary, step-by-step computer programming can give insight into creativity has been criticised, especially with the rise of parallel processing in computing - or neural networks - which are modelled far more closely on how the brain works.

But Boden says: "It is a mistake to think that sequential computer programs cannot possibly teach us anything about psychology . . . theories may identify some of the specific computational processes which, in human beings, are run in parallel. There is great excitement at present about the recent AI-work on connectionism. This is understandable, as we shall soon see. But it should not obscure the fact that AI-models, despite their 'unnatural' air, can help us investigate the contents, structures and processes of human thought."

Boden agrees that many creative acts involve more flexible mental processes than described in sequential programming - for example seeing an analogy where none was there before. Take the water snakes of Coleridge's Ancient Mariner, "blue, glossy green, and velvet black," which "moved in tracks of shining white" shedding "hoary flakes" of "elfish light".

Luckily Coleridge's extraordinarily wide reading, which included scientific works on sea creatures, was tracked over 60 years ago by John Livingston Lowes. His conclusion was that Coleridge's genius lay in his extraordinary memory and in his inspired powers of association.

To dissect this human achievement, Boden uses ideas from neural networks. These systems "learn" as they go along rather than being programmed. Association of ideas by humans takes place via comparable processes, says Boden.

Connectionist systems can recognise patterns that they have experienced before; they can recognise parts of patterns and distortions of patterns as variations on the original pattern; they can provide "best" matches to patterns (as we would do in judging the aptness of a poetic image) and they can learn to do better.

"People who claim that computational ideas are irrelevant to creativity because 'brains are not programmed' must face the fact that connectionist computation is not the manipulation of formal symbols by programmed rules . . . we must consider the ability of neural networks to learn to associate patterns without being explicitly programmed in respect of those patterns," says Boden.

The use of the old and the new computing has produced some riveting pieces of apparent creativity - from the ludicrous Henry Ant to the artwork on Boden's walls.

Boden herself makes interesting use of analogies. Among the spill of them in her books are comparisons with knitting patterns and threading necklaces. One publisher asked her to remove them on the grounds that they would not be understood. She replied that she was fed up with analogies with radios and motor cars, which she as a woman did not understand - men would just have to go and find out about knitting.

Her tribulations as a woman peek out of her CV as well as her writings. Under "posts", appears: "Invited to Center for Advanced Study in Behavioral Sciences, Palo Alto, in 1973, 1980 and 1990: unable to go, because of children". "This is to remind people that women have it difficult," she says.

There are those who reject the computer program approach entirely: "There are many, many people who think that what we do is fundamentally incorrect. Most people who are not academics, and especially the general public, assume that it's got to be nonsense."

She has categorised objections to the idea that computers could be creative. They are: the brain-stuff argument (computers are made of non-biological material so they cannot be creative); the empty program argument (computers will never grasp meaning); the consciousness argument (consciousness is essential to creativity and AI cannot be conscious) and the non-human argument (creative beings would deserve rights, computers do not have rights, therefore computers cannot be creative). She rejects the first three.

"I believe that in principle computers that we have only dreamed of could produce all of the behaviour which human beings produce including what we call creative behaviour.

"I would prefer to sit on the fence but if I were forced to come off I would say that computers can't be truly creative, for moral and political reasons. If a computer literally has creativity then it literally has desires and interests - it would be part of our ethical world. It would be a question of whether people wished to accord computers this type of respect".