Few things are quite as challenging as using a computer to model something we think of as quintessentially human. And if you鈥檙e Douglas Hofstadter, the cognitive science researcher whose book G枚del, Escher, Bach inspired a generation, trying to model creativity will make you ask yourself tough questions. How much can we learn about human creativity this way? Is 鈥渃reativity鈥� the right word anyway? And will the Cats deliver? Mike Holderness and Liz Else quizzed Hofstadter
How do you define creativity?
I think we would do better to talk about 鈥渄iscoverativity鈥� than creativity. I am convinced creativity is the ability to discover things that are in some sense there for anyone to find: things that the rest of humanity will appreciate because they are beautiful or because they are insightful or because they are truths about nature. Of course, we consider some people great musical composers when nine-tenths of the world has never heard of them, and when most of those who have heard their music didn鈥檛 like it anyway. But they鈥檝e found what I call a 鈥渧ein of receptivity鈥� in the minds of a large number of people.
Advertisement
Are some discoveries more creative than others?
Well, some are easier to make. Take what happened in quantum mechanics in the 1920s. I use the metaphor of people finding a new beach. Physicists rushed to this beach because they knew that it was going to be covered in wonderful shells. It wasn鈥檛 that hard to find fantastic, beautiful shells because it was a completely virgin beach. But after 10 or 20 years it got harder and harder. Then it became time to try to find new beaches.
It is standard to think of science as discovery, but doesn鈥檛 鈥渄iscovering鈥� a novel sound bizarre?
My feeling is that novelists discover patterns or structures of human behaviour, perhaps miniature plot lines. And they also discover idiosyncratic fashions of weaving plot lines together, in multiple layers, and these devices become characteristic of them as an author. They might make a complex analogy: 鈥淭hat marriage is like another marriage that I once saw.鈥� That is how we navigate human relations. We are always asking each other for advice: 鈥淗ow should I treat my daughter?鈥� 鈥淥h, my friends had a daughter who used to do similar things, so鈥︹�� Novelists observe the world and discover events and people in it that they find interesting, and their discoveries pepper their novels. Good novelists are those whose sense of what is interesting coincides with that of many others.
Will we be able to scan brains and say 鈥淟ook, that鈥檚 the creative thing being appreciated鈥�?
There has recently been a very strong push to connect the most complicated cognitive or emotional phenomena to the latest trends in biology and exploring the brain. People yearn to find 鈥渢he鈥� neural correlate of an emotion or even a creative thought. To me, that hope seems about as silly as trying to find the key to the greatness of a novel by closely examining the typographical symbols that compose it. Clearly, the relative frequency of 鈥渒鈥� and 鈥渏鈥� is ridiculously far from what makes a novel great. There are so many levels of description between the level of letters and that of ideas.
How wide is that explanatory gap between neurons and thoughts?
Well, a great poem creates activity in billions of neurons, each changing every few milliseconds over several minutes at least. From my viewpoint, you could assign to each intermediate level of description a factor of 10, and since a neuron firing is 10 or 20 levels away from the level of a poem, you are missing the mark by something like 15 orders of magnitude. That just gives an idea of how senseless it is to try to talk about a neural correlate of poetry.
Would we do better approaching the enigma of analogy to understand creativity?
There鈥檚 no doubt about it. Making analogies is central to being human. Every single word choice we make, for instance, is done by analogy. Consider the word choice you made earlier 鈥� 鈥渆xplanatory gap鈥�. It seemed trivial to you, but for a computer it would be terribly, terribly hard. Every single word choice is analogy-making, because it is connecting prior experiences with a new experience and recognising the commonality. As we grow up, we do that repeatedly at more and more abstract levels until we build up very deep insights about the world.
In the 1980s and 1990s you tried to model analogy-making on a computer. What did you do?
We kept it as simple as we could. Our program Copycat deals with questions like 鈥淚f ABC changes to ABD, what, by analogy, would XYZ change to?鈥� There are many defensible answers, by the way, including XYA, XYD, XYY, XYZ, WYZ, WXZ, DBA and others. Our goal is to make a program that can find all these answers but also that has aesthetic preferences that make it 鈥渓ike鈥� some more than others. Although it finds XYD most frequently, it 鈥渓ikes鈥� WYZ the best. Copycat itself came out of an analogy 鈥� the idea of modelling a computer program on an ant colony. Copycat is composed of many, many small processes, many of which run in parallel. Each small 鈥渁nt鈥� makes its contribution to finding the analogy.
How have you developed it since the early days?
We have a newer project, Metacat. It can say: 鈥淭his analogy is like that one鈥�, and you can ask it: 鈥淚n what way?鈥� Metacat will reply that the pressures that give rise to this analogy are related to the pressures in that one, but there is this difference. Metacat also has a limited sense of its own goals 鈥� and in the future we would like it to have a sense of the goals of the person it is talking to. For instance, if I were to say to it: 鈥淚f ABC changes to ABD, what would ABC change to?鈥�, I would want it to reply indignantly: 鈥淲hy are you asking me what you just told me?鈥�
That sounds as if it is expressing constraints on what is important.
In my book Le Ton Beau de Marot, I discussed the key role of constraints in creativity. Constraints give a framework within which to work. Take, say, the rhythm used in most English sonnets, iambic pentameter. The vast majority of people like this regular periodic beat. As in music, they like a rhythm based on a small integer. You can go out on a limb and try seven or eleven, but the odds are that fewer people will be attracted to your creation. Such preferences are part of our biological nature.
Pandering to inbuilt preferences suggests film producer Samuel Goldwyn鈥檚 demand that his creative people produce 鈥渘ew clich茅s鈥�.
That鈥檚 cute. I can almost subscribe to this motto. I would just add one word: to me, creativity is finding 鈥渘ew future clich茅s鈥�. Everything truly creative eventually becomes a clich茅. The Mona Lisa is the most clich茅d of all the clich茅s you can imagine 鈥� and something similar could be said of Einstein鈥檚 great discovery E = mc2. In each case, someone found a hidden vein of receptivity in the human mind 鈥� something that a lot of people could relate to.
How long do you expect it to be before Metacat delivers a significant insight into human creativity?
Well, I feel this is already happening, and I hope it will increase in the next few years. But I don鈥檛 expect human-level intelligence suddenly to flower out of our work. If that happened, it would mean that our intelligence is a lot simpler than we humans tend to believe. In fact, I personally would be devastated if any of my computer models turned out to be just as smart as a human being. It would be a terrible shock.
