From Above And Below:
Problems With Boden’s Model Of Human Creativity.
Prof. John Vervaeke
For most of psychology’s history as a legitimate and recognised scientific endeavour, creativity has remained a fringe subject. In the past 20 years however, an earnest research project of defining and then explaining creativity has emerged, with some provocative results. Serendipitously, this research has emerged amidst a growing pressure in the psychological sciences to develop rigorous and naturalisable theories, ones that above all, could potentially be mechanized. I call this serendipity because for the same reasons that creativity was so long ignored by psychology – it is a subjective notion, it is difficult to characterize, it is in short, ill-defined – there is the potential for theories of creativity, once they emerge, to be non-rigorous or even vacuous. The emphasis on naturalisable theory however, ensures that those theories that emerge have at least made an attempt at rigour and possibly even mechanizability which in turn allows them to make testable predictions. When critically examining a theory of creativity from a naturalistic perspective then, we can ask of it three central questions: does it account for the experimental findings in the area; does it actually present a rigorous and naturalisable model; and does it answer the question it set out to ask.
One of the more dominant theories of creativity that has emerged in the last decade has been the work of Margaret Boden (e.g. Boden 1990). In her writing she has developed important notions of personal- versus historical-creativity, which have helped to define creativity in a much more formal context. Her work, which contrasts with that of Koestler (1975) and others, defines creativity as more than just novelty-producing thought, but rather of novel exploration of and creation of mental representations (Boden 1990, 1994, 1995). She also formulates her theory within a computational framework thus ostensibly leading to a theory that is well grounded in concepts of mechanizability. I will argue in this essay however that Boden’s theory fails to address important considerations both from a top-down or conceptual perspective, as well as from a bottom-up or mechanical perspective. That is, there are important ways in which it does not account for the experimental evidence, and surprisingly enough, ways in which it fails to be a truly mechanizable model. Finally, I will argue that it does not get to the root of our intuitions about what creativity is, that it mislabels certain cases. First though, it will be helpful to review the model as it currently exists.
To begin with it is important to remember that Boden’s theory provides us with an excellent definition of creativity, possibly the most lucid so far,
There are two broad types of creativity, improbabilist and impossibilist. Improbabilist creativity involves (positively valued) novel combinations of familiar ideas. A deeper type involves METCS: the mapping, exploration and transformation of conceptual spaces. It is impossibilist, in that ideas may be generated which – with respect to the particular conceptual space concerned – could not have been generated before. (They are made possible by some transformation of the space.) The more clearly conceptual spaces can be defined, the better we can identify creative ideas. (Boden, 1994; pp. 519-520)
This definition is an apt one, and notwithstanding some issues of applicability that I will discuss towards the end of the paper, it seems to capture our notions of creativity more precisely that Koestler’s creativity-is-novelty theory (1975). Her description of the processes of creativity is more vague. She moves from her definition to argue that since creativity is the investigation and transformation of conceptual spaces, the mechanism of creation must be a sort of mechanized search through and between conceptual spaces,
A generative system defines a certain range of possibilities… These structures are located in a conceptual space whose limits, contours, and pathways can be mapped, explored, and transformed in various ways… probably the crucial difference between Mozart and the rest of us is that his cognitive maps of musical space were very much richer, deeper, and more detailed than ours. (Boden, 1995; pp. 2-3).
This description seems, prima facie, to provide a mechanism for the creative process, perhaps one that can be naturalized. It also handily accounts with the argument, raised by Harnad (unpub. ms.) and others, that the best way to be creative in a domain is to gather as much knowledge about the domain as possible – in other words, it supports Pasteur’s Dictum. This notion: that creativity is achieved through manipulation and investigation of conceptual spaces, is a brief but I think an accurate summary of Boden’s theory. Much of her writing in the area is not so much concerned with developing her theory as with testing various AI’s against it to see how they fare (Boden 1990, 1994).
From a conceptual standpoint, Boden’s theory is a valuable attempt, but a flawed one. There is an important body of research in creativity of which her model seems to take no account: the role of attention and motivation. One can imagine why Boden might wish to avoid questions of attention and motivation – they are exceptionally difficult concepts to mechanize. It isn’t at all clear how we should create computational systems with any notion of attention or motivation, or even whether such systems can be created. Boden does address this potential argument with her theory, but only briefly and only, it seems, to dismiss it,
The role of motivation and emotion is briefly mentioned but it is not a prime theme. This is not because motivation and emotion are in principle outside the reach of computational psychology. Some attempts have been made to bring these matters within a computational account of the mind… but such attempts provide outline sketches rather than functioning models. Still less is it because motivation is irrelevant to creativity. But the main topic of the book is how (not why) novel ideas arise in human minds. (Boden, 1994: p. 560)
So Boden offers two explanations for not having addressed motivation: the existing computational models are insufficiently well defined, and motivation is concerned with the ‘why’, not the ‘how’ of creativity. To the first point one can only agree with her, we do not seem to have a good computational model of motivation; and she should know since she has advanced her own theory in the area (Boden, 1972). To the second however, there can be strong disagreement. Motivation has been demonstrated to have a much more profound impact on creativity than just providing a reason (a ‘why’) to perform creative acts. Amabile’s work on intrinsic and extrinsic motivation has shown the powerful effect motivation can have on the quality and quantity of creative work produced (1984). Motivation does far more than just provide the impetus to produce creative work – Amabile has demonstrated that it actively shapes the work produced, and can, if the motivation is of the wrong sort, destroy it. If Boden were simply trying to create a computer program that produced interesting things, she could afford to ignore these findings; but to try to produce a model of human creativity without accounting for the profound impact motivation can have seems somewhat myopic.
Worse, where Boden’s theory dismisses motivation, it does not even address attention. The body of work on creative flow attests to the fact that attentional control and selectivity are among the most significant predictors of creative performance (Csikszentmihalyi, 1997). It is unclear how a computational model might address this issue, but it’s clear that to succeed, any model of creativity must explain attention’s role. Again Boden’s theory seems caught between attempting to produce a creative computer and the much harder task of explaining human creativity. It is not a sufficient defence for Boden to argue that concepts like motivation and attention are difficult to mechanize since, in attempting to produce a computational model of creativity, that is precisely the task she has taken up.
Putting aside for a moment objections from a conceptual level, Boden’s theory also faces problems from a computational level. Her theory, though expressed in computational terms, is actually relatively vague when it comes to actual implementation. We are told that creativity results wholly from search through and transformation of conceptual spaces (Boden 1990, 1995). Conceptual spaces in Boden’s theory are at best an ambiguous notion. They seem to encompass broad areas of expertise like music, writing or chess playing, but it’s never clear what constitutes a conceptual space specifically: how big can they be? are sprinting and marathon running different conceptual spaces or the same? what about jazz composition and classical composition, or abstract art and impressionist art? And how does our mind delineate these spaces to begin with? Boden has argued (1990) that these conceptual spaces are developed and solidified as we acquire knowledge in a given domain, but without a discussion of how ‘domains’ of knowledge are segregated, this definition is question-begging. There is also a question of combinatorial explosion that emerges when one postulates a model of essentially unfettered search through one or many broadly defined spaces – especially when there is often no specific goal in mind. A model that invokes this type of open-ended search must propose a clear and formal method for restricting the search space to something manageable by the human minds that are postulated to implement it. Boden’s only clear attempt to restrict the search in this way is to argue that increased exposure to a domain will produce richer, deeper cognitive maps which may guide search (Boden, 1995). This again presupposes a mechanism within the mind for organizing domains of knowledge and for suggesting interesting or fruitful lines of inquiry. I submit that such a module or mechanism would be the very essence of creativity, and thus no theory of creativity can realistically presuppose its existence without becoming circular.
Finally, there are intuitions we hold about creativity that do not seem to be borne out by this model. Most notably, the model places highest creative value on those ideas that could not have occurred before, that essentially mark the creation of a new way of thinking about something (Boden 1990, 1994, 1995). However, if one examines the success of artists, especially in music and visual art, it is often not the creators of an artistic genre that get the highest accolades but rather their successors who fine tune, or perfect the art form. Boden’s theory cannot account for our ascriptions of creativity here, since the successors are merely exploring a space, not performing a radical transforming, as the founders of an artistic school might have. In truth, Boden’s theory seems more apt here for describing scientific creativity, where we accord the first person to bring forth a concept the highest creative insight. Perhaps this is grounds for an argument that scientific and artistic creativity are fundamentally different entities, but to make the argument would be well beyond the scope of this paper, and moreover it does not seem to be an argument Boden would support – she would certainly claim that her model describes both artistic and scientific creativity.
Boden’s model places emphasis on the restructuring of conceptual spaces as being essential for an act to be truly creative (1990, 1995). However several artificial intelligence programs exist which, though incapable of performing this restructuring, produce work that is judged creative. An excellent example of this is the painting program AARON. Developed by Cohen (1995) to examine the definition of an ‘image’, it has become a well worn example of computer creativity. AARON works by having some basic notions about closure of shapes, figure and ground, as well as a repertoire of hand-programmed shapes like that of a human, or a plant leaf (Cohen, 1995). It draws by means of a robotic turtle which moves a pen – and more recently several pens, to add colour – over a sheet of mural paper. What is so noteworthy about this system is that observers regularly praise the work for its creativity before, and sometimes after, being told that a computer program produced it. Yet the program doing the creating has no ability to restructure as Boden describes it. AARON cannot change its descriptions of shapes, or experiment with figure and ground (Cohen, 1995). It would seem that the standards of creativity to which human observers hold artistic works is not as rigid as Boden describes. Similarly, Johnson-Laird’s jazz improvisation program has no ‘cognitive’ access – it cannot alter its programming, and yet experienced jazz pianists rate its improvisations as being at the level of a talented amateur (1991).
Boden’s theory, while promising at first, fails to meet our criteria. It does not satisfy our intuitive notions of creativity, since it seems to construe the field too narrowly. It rejects or overlooks important theoretical findings about attention and motivation that must be central to any model that attempts to explain the mechanism for human creativity. And finally, Boden’s theory fails to be suitably formalised. Though Boden is always careful to phrase her theory in computational terminology, the model she presents is subject to combinatorial explosion as it stands, and is confounded by an over general definition of conceptual space upon which the rest of her theory is based. It has, in the current form, serious problems.
In fairness though, Boden’s theory should not be dismissed out of hand. There are implementation issues to be sure, and arguably some important conceptual issues, but there is also a deep sense in which her theory seems sound. There is an important distinction between creativity within a domain – improbabilist in her usage – and the creativity that opens up a new domain completely. Her definition of creativity, if not her explanation, is provocative and well defined. If her theory fails at the grander task of explaining human creativity, it has at least demonstrated that a computational model of creativity is not an impossible notion, and that creativity can be subjected to the same rigorous analysis as other cognitive traits with which psychology has concerned itself for decades. It shall prove important for future theories of creativity to incorporate many of Boden’s ideas; but on its own, her theory does not stand.
Amabile, Theresa M. (1985). Motivation and creativity: Effects of motivational orientation on creative writers. Journal of Personality and Social Psychology. 48(2): 393-397.
Boden, Margaret A. (1972). Purposive explanation in psychology. Cambridge, Mass.: Harvard University Press.
Boden, Margaret A. (1990). The Creative Mind: Myths and Mechanisms. London: Weidenfeld & Nicholson.
Boden, Margaret A. (1994). Précis of The Creative Mind: Myths and Mechanisms. 0 Behavioural and Brain Sciences. 17(3): 519-570.
Boden, Margaret A. (1995). Creativity and Unpredictability. Stanford Education and Humanities Review. 4(2).
Cohen, Harold. (1995). The further exploits of AARON, Painter. Stanford Education and Humanities Review. 4(2).
Csikszentmihalyi, Mihaly. (1997). Creativity: Flow and the psychology of discovery and invention. New York, NY: Harper Collins.
Harnad, Stevan. (unpub. ms.) Creativity: Method or Magic? Available Online: http://www.cogsci.soton.ac.uk/~harnad/Papers/Harnad/harnad.creativity.html
Johnson-Laird, P. N. (1991). Jazz improvisation: A theory at the computational level. In: Representing musical structure, eds. P. Howell, R. West & I. Cross. (pp. 291-326). London: Academic Press.
Koestler, A. (1975). The act of creation. London: Picador.
 A notable but essentially sole exception to this seems to be the Gestalt theorists, who were concerned from their beginnings with questions of creativity and insight, but who failed to provide a strong testable theory.
This footnote was just a note to the marker of the original essay about how my thesis had diverged from the one I had previously discussed with them. It's been deleted since it was a bit long-winded, and is certainly no longer relevant.
 ‘Chance favours the prepared mind.’ Taken from Harnad.
 This observation is credited, though without citation since we were not given one, to Prof. John Vervaeke who made the point in lecture.