From the archives: Unravelling the mysteries of creativity

This article originally appeared in The Skeptic, Volume 7, Issue 5, from 1993.

Creativity can be scientifically understood with the help of computational concepts. This may seem surprising, not to say absurd. Most people take for granted that there can be no interesting relation, only utter incompatibility, between computers and creativity. Indeed, most people assume that there will never be any scientific theory of creativity – for how could science possibly explain fundamental novelties? The very notion seems to be a contradiction in terms.

In fact, however, computers and creativity make interesting partners with respect to three rather different sorts of project. One is psychological: understanding human creativity. One is technological: trying to produce computer-creativity. And the third is pragmatic, even educational: using computers to aid one’s own creativity.

Human creativity is something of a mystery, not to say a paradox. Artists and scientists rarely know how their original ideas come about. They mention intuition, but cannot say how it works. Nor can they say clearly how creativity can be recognised. One new idea may be creative, while another is merely new: what’s the difference? And how is creativity possible?

Some ideas we call creative are merely novel combinations of familiar ideas. Much poetic imagery, or analogy in science, is of this type. Psychometric measures of creativity usually identify ‘creative’ ideas in terms of their statistical improbability [1], and such tests are best-suited to capturing creativity of this ‘combinational’ type. As well as measuring it, the psychologist needs also to be able to explain this sort of creativity.

Imaginative association (whereby vastly different ideas, as well as similar ones, may be related), and the recognition of analogy, can each be illuminated by computational models. Association takes place by means of mental processes comparable to (though often much richer than) the parallel-processing defined within connectionist AI-models, or neural nets [2]. The origination of the haunting imagery of Coleridge’s poem The Ancient Mariner, for instance, can be largely understood in these terms [3, Chapter 6]. As for analogy, which often effects a permanent change in one’s perception of something, this too has been modelled in AI-terms [4, see also 5].

Combinational creativity, surprising though its results can be, is less arresting than some other cases of originality. A novel combination of two familiar ideas is something which did not happen before. By contrast, some creative ideas surprise, shock, and delight us because it seems that they could not have happened before. Relative to our expectations, they are not just improbable, but impossible.

Combination-theory cannot capture this distinction. An adequate scientific account of creativity needs to be able to do so. Also, it must show clearly how apparently impossible ideas can in fact arise. And it must explain in what sense one impossible idea can be more impossible, more creative, than another.

Non-combinational creativity involves the exploration, and in the most interesting cases the transformation, of conceptual spaces in people’s minds. Conceptual spaces are styles of thought. They include ways of writing prose or poetry; genres of sculpture, painting, or music; theories in chemistry, biology, or mathematics; habits of couture; systems of choreography… in short, any reasonably disciplined way of thinking.

Within a given conceptual space, many thoughts are possible – even if some of them are never thought. Others are impossible. If you are skilfully writing a limerick, iambic pentameters simply cannot drop from your pen. But if you want to write a new sort of limerick, or a non-limerick somehow grounded in that familiar style, then blank verse could perhaps play a role. The deepest cases of creativity involve someone’s thinking something which, with respect to the relevant conceptual space present in their minds, they could not have thought before. (It follows that constraints are essential to creativity, even if creativity largely consists of overcoming them.)

The obvious next question is how this supposedly impossible idea could possibly come about. And the answer is that the creator must change the pre-existing style in some way. It must be tweaked, or even radically transformed, so that thoughts are now possible which (within that space) were not conceivable before. To understand how this can happen, we must understand clearly what conceptual spaces are, and what sorts of mental processes could explore and modify them.

Styles of thinking are studied by literary critics, by musicologists, and by historians of art, fashion, and science. And they are appreciated by us all. But intuitive appreciation, and even life-long scholarship, may not make their structure entirely clear. Indeed, the unclarity may be proclaimed as unavoidable by the most scholarly of critics. (An architectural historian said of Frank Lloyd Wright’s Prairie Houses, for instance, that their ‘principle of unity’ is ‘occult’ (cited in [6]). By this he seemed to mean not merely that he had not managed to identify it, but that it must remain forever hidden to human eyes.) Even when a scholar does claim to have discovered a certain style’s principle of unity, it may not be stated clearly enough for scientific purposes.

This is the second point at which computational methods can help. Conceptual spaces, and ways of transforming them to produce new ones, can be clearly described by using computational concepts. These concepts are drawn from artificial intelligence (including, but not confined to, connectionist AI), and they enable us to do psychology in a new way. A conceptual space can be thought of as a generative system: a more or less complex set of rules which define the relevant dimensions, and specify ways in which a range of structures (ideas) can be generated.

In this way, the structure of tonal harmony [7], the ‘grammar’ of Prairie Houses [6], and spaces of many other sorts can be clearly expressed. Then, the power of the computer may be exploited to study various ways of exploring the relevant domain. And methods (‘heuristics’) for navigating, and even for changing, highly-structured spaces can be examined and compared.

One implication of this account of creativity is that a creative system, whether mind or machine, needs internal ‘maps’ of its own conceptual spaces [3, Chapter 4]. These maps, which may exist on many different levels, enable the system to move within the relevant spaces, to test their limits and boundaries, to modify them, and even to transform them. Thus one way, probably the most important way, in which Mozart was different from the rest of us is that his mind contained more richly-detailed maps of musical structures, and more ways of negotiating them fruitfully, than other people’s.

Computationally-informed work in developmental psychology suggests that young children are unable to vary their skills in flexible and imaginative ways until they have developed internal maps of the skills concerned [8]. Thus an older child can draw many ‘funny’ animals (two-headed dogs, or centaur-like monsters) at will, whereas a younger child simply cannot – even though the younger child’s skill at drawing normal animals is no less fluent. Variation of size or shape (of parts and/or whole), repetition or deletion of parts, and mixing of categories (to produce centaurs, for instance) appear in a predictable sequence. The theoretical explanation is that a skill which is fluent may be merely ‘compiled’: it can be run, but not varied. To be varied, it must be redescribed at some higher level (or levels), in terms of specific parameters and subroutines. These ‘representational redescriptions’ not only transform line-drawing (and language [9]) into a more creative activity, but underlie the development of conscious self-reflection.

What of the second link between machines and creativity? Can computers be creative? That is, can they produce performance of a kind which we would regard as creative if we saw it in human beings? (This is a scientific question. For present purposes, we may ignore the non-scientific, philosophical, question of whether a computer, no matter how humanlike its performance, could ‘really’ be creative [3, Chapter 11].)

The answer to this question is ‘Yes’. A number of programs already exist which can explore a given space in acceptable ways. For example, a computational grammar has not only shown what the principle of unity of Prairie Houses actually is, but has generated designs for new ones, previously undreamed-of [6]. Some programs can generate thousands of line-drawings in a certain style, pleasing enough (if hung on one’s walls) to be spontaneously remarked upon by unsuspecting visitors [10]. Others improvise unpredictable melodies, and accompaniments, from a modern-jazz chord sequence [11, 12]. Yet others come up with (occasionally, brand-new) scientific hypotheses [13, 14].

A few programs can even transform their conceptual habitat, alter their own rules, so that interesting ideas result. For instance, programs using genetic algorithms (see below) have come up with optimal solutions to a number of difficult problems. One such problem concerns accidental leaks in huge gas-pipelines, running across many different countries. A self-transforming, ‘evolutionary’, program can infer the position of the leak on the basis of hourly measurements taken at various points along the pipeline [15]. These measurements include the gas-inflow, gas-outflow, inlet-pressure, outlet-pressure, rate of pressure-change, season, time of day, time of year, and temperature. As well as finding the leak, the program can recommend emergency-action (specifying which pumps should be turned off, which valves should be closed, etc.)

Many of the ideas generated by current computer-models of creativity were already known to human beings (though not specifically prefigured within the program’s initial conceptual space). But the definition of creativity adopted within this article is a psychological one: the ability to come up with an idea which, relative to the pre-existing domain-space in one’s mind, one could not have had before. Whether any other person (or system) has already come up with it on an earlier occasion is irrelevant. That is a historical question, not a psychological one.

Moreover, at least one entirely unknown mathematical theorem has been suggested by a computer, whose programmer had never even heard of the branch of mathematics concerned [16]. The program, using its initial concepts and its exploratory/transformational rules, found its own way into the new mathematical space, by creating that space for itself.

The third way in which computers and originality are related involves the use of these machines to help our own creativity. A number of graphic artists, film-animators, and industrial designers are using computers as genuine partners in their creative quest. The machines in question execute ‘evolutionary’ programs, continually making random changes in their current rules so that entire new forms, new species, of structure result. The human being continually chooses the ones he or she finds most interesting, and the machine concentrates on them.

For instance, an image-generating program [17] uses self-modifying ‘genetic algorithms’ (modelled on biological mutations) to generate new images, or patterns, from pre-existing ones. At each ‘generation’, the selection of the ‘fittest’ examples is done by the programmer – or by someone fortunate enough to be visiting his office while the program is being run. That is, the human being selects the images which are aesthetically pleasing, and these are used to “breed” the next generation. The programmer’s aim, in this case, is to provide an interactive graphics-environment, in which human and computer can cooperate in generating otherwise unimaginable images. (At present, this program runs only on the Connection Machine, a massively parallel computer costing 15 million dollars.)

A similar method is used by the sculptor William Latham, to generate 3D-forms likely to satisfy specific aesthetic constraints [18]. So as not to jeopardise those constraints, Latham allows self-transformations only at relatively superficial levels in the program (changes in parameters, not functions). In consequence, the varying forms are much less diverse than those produced by Sims’ system.

Sims’ computer-generated images often cause surprise greater than that caused by mere random unpredictability. It is as if, besides our being unable to predict heads or tails when tossing a coin, the coin sometimes showed a wholly unexpected design. If one compares a parent-image with some of its descendants (or even some of its immediate offspring), one may be amazed by the difference. The change(s) effected to the earlier image on the way to the later one sometimes seem like relatively unadventurous exploration, or tweaking. The colour of what is clearly the same image may have been altered, for instance, and/or the lines may obviously have been blurred. Sometimes, however, one cannot say, merely by looking at the chosen image-pair, how they are related – or if they are related at all. The one appears to be a radical transformation of the other, or even something entirely different. (It does not follow that Sims’ program is ‘better’ than Latham’s: if the artist-programmer is trying to achieve a particular type of aesthetic effect, the system’s freedom of transformation must be limited.)

In sum, there are many intriguing relations between creativity and machines. Computers can sometimes do creative things, and some can help us to do so. Not least, a computational approach can clarify many questions about our own creative powers. It gives the psychologist a way of seeing more clearly into the rich subtleties of the human mind.

References

1. Sternberg, R.J. (ed.) [1988] The Nature of Creativity: Contemporary Psychological Perspectives. Cambridge: Cambridge University Press.
2. Rumelhart, D. E., & J. L. McClelland. [1986] Parallel Distributed Processing: Explorations In the Microstructure of Cognition. 2 vols. Cambridge, Mass.: MIT Press.
3. Boden, M. A. [1990/1992] The Creative Mind: Myths and Mechanisms. London: Weidenfeld & Nicolson, 1990; paperback edn. (expanded) New York: Basic Books, 1992.
4. Hofstadter, D. R., M. Mitchell, R. M. French, D. J. Chalmers, & D. Moser. [in press] Fluid Concepts and Creative Analogies. London: Harvester Wheatsheaf.
5. Holyoak, K. J., & P. & Thagard. [1989] ‘Analogical Mapping by Constraint Satisfaction’, Cognitive Science, 13, 295–356.
6. Koning, H., & J. Eizenberg. [1981] ‘The Language of the Prairie: Frank Lloyd Wright’s Prairie Houses’, Environment and Planning B, 8, 295–323.
7. Longuet-Higgins, H. C. [1987] Mental Processes: Studies In Cognitive Science. Cambridge, Mass.: MIT Press.
8. Karmiloff-Smith, A. [1990] ‘Constraints on Representational Change: Evidence from Children’s Drawing.’ Cognition, 34, 57–83.
9. Karmiloff-Smith, A. [1986] ‘From Meta-processes to Conscious Access: Evidence from Children’s Metalinguistic and Repair Data.’ Cognition, 23, 95–147.
10. McCorduck, P. [1991] Aaron’s Code. San Francisco: W. H. Freeman.
11. Johnson-Laird, P. N. [1988] The Computer and the Mind: An Introduction to Cognitive Science. London: Fontana.
12. Johnson-Laird, P. N. [1989] ‘Jazz Improvisation: A Theory at the Computational Level.’ Unpublished working-paper, MRC Applied Psychology Unit, Cambridge.
13. Buchanan, B. G., D. H. Smith, W. C. White, R. Gritter, E. A. Feigenbaum, J. Lederberg, & C. Djerassi. [1976] ‘Applications of Artificial Intelligence for Chemical Inference: XXII Automatic Rule Formation in Mass Spectrometry By Means of the Meta-Dendral Program’, Journal of the American Chemistry Society, 98, 6168–78.
14. Langley, P., H. A. Simon, G. L. Bradshaw, & J. M. Zyktow. [1987] Scientific Discovery: Computational Explorations of the Creative Process. Cambridge, Mass.: MIT Press.
15. Goldberg, D. [1987] ‘Computer-Aided Pipeline Operation Using Genetic Algorithms and Rule Learning. Part I: Genetic Algorithms in Pipeline Optimization’, Engineering with Computers, 3, 35–45.
16. Lenat, D. B. [1983] ‘The Role of Heuristics in Learning by Discovery: Three Case Studies’ In R. S. Michalski, J. G. Carbonell, & T. M. Mitchell (eds.), Machine Learning: An Artificial Intelligence Approach. Palo Alto, Calif.: Tioga.
17. Sims, K. [1991] ‘Artificial Evolution for Computer Graphics’, Computer Graphics, 25 (no. 4), 319–328.
18. Todd, S., & W. Latham. [1992] Evolutionary Art and Computers. London: Academic Press.

This article was first published in the Journal of Creative Behavior, volume 26, (1992) and is reproduced here with permission