This article is an excerpt from the book “Computational Creativity. The Philosophy and Engineering of Autonomously Creative Systems” edited by Tony Veale and F. Amilcar Cardoso (Springer, 2019)
Computational creativity is an emerging field of research within AI that focuses on the capacity of machines to both generate and evaluate novel outputs that would, if produced by a human, be considered creative. This book is intended to be a canonical text for this new discipline, through which researchers and students can absorb the philosophy of the field and learn its methods. After a comprehensive introduction to the idea of systematizing creativity the contributions address topics such as autonomous intentionality, conceptual blending, literature mining, computational design, models of novelty, evaluating progress in related research, computer-supported human creativity and human-supported computer creativity, common-sense knowledge, and models of social creativity.
Products of this research will have real consequences for the worlds of entertainment, culture, science, education, design, and art, in addition to artificial intelligence, and the book will be of value to practitioners and students in all these domain.
Chapter 1 Systematizing Creativity: A Computational View
Authors: Tony Veale, Amilcar Cardoso, Rafael Pérez y Pérez
Summary. Creativity is a long-cherished and widely studied aspect of human behavior that allows us to reinvent the familiar and to imagine the new. Computational Creativity (CC) is a recent but burgeoning area of creativity research that brings together academics and practitioners from diverse disciplines, genres and modalities, to explore the potential of our machines to be creative in their own right. As a scientific endeavor, CC proposes that computational modeling can yield important insights into the fundamental capabilities of both humans and machines. As an engineering endeavor, CC claims that it is possible to construct autonomous systems that produce novel and useful outputs that are deserving of the label “creative.” The CC field seeks to establish a symbiotic relationship between these scientific and engineering endeavors, wherein the artifacts that are produced also serve as empirical tests of the adequacy of scientific theories of creativity. We argue that, if sufficiently nurtured by volumes such as this, the products of CC research can have a significant impact on many aspects of modern life, with real consequences for the worlds of entertainment, culture, science, education, design, and art.
1. 1. From C to CC
Creativity is a multifaceted phenomenon that manifests itself in different guises in different domains. So creativity in the domain of sports (e.g. as manifest in a team sport like soccer, or an intellectual game like chess or Go is clearly different from creativity in the arts domain (e.g., consider painting or poetry), yet there are enough similarities for exemplary outcomes in each domain to be deserving of the same label, “creative.” This heterogeneity makes creativity a notoriously difficult concept to pin down in formal terms, and definitions that favor one area of human activity (such as art) are unlikely to do justice to other areas (such as science, engineering, or cooking). Our definitions of creativity – and a great many have been considered in the scientific literature – are no more than accepted conventions, and it is in the very nature of creativity to bend and subvert these conventions.
Computational Creativity (CC) is an emerging branch of artificial intelligence (AI) that studies and exploits the potential of computers to be more than feature-rich tools, and to act as autonomous creators and co-creators in their own right. In a CC system, the creative impetus should come from the machine, not the human, though in a hybrid CC system a joint impetus may come from both together. As a discipline, CC draws on research in artificial intelligence, cognitive science, psychology, and social anthropology to explore the following questions:
- What does it mean to be “creative”? Does creativity reside in the producer, in the process, in the product, or in a combination of all three together?
- How does creativity relate to expertise and to what extent does it necessitate specialized domain knowledge?
- How does creativity exploit and subvert norms and expectations?
- How are the outputs of creativity judged and evaluated? How can we meaningfully measure creativity? What knowledge is needed (of the creator or process) before we can label a work “creative”?
- What constitutes creativity in different domains and modalities?
- How does creativity emerge from group behavior and collective action?
- What cognitive paradigms offer the most insightful explanatory theories of creativity (e.g., search in a conceptual space, conceptual blending, or bisociation)?
Each of these questions is just as valid in the study of human creativity as it is to the study of machine creativity. What makes CC different is that it adopts an explicitly algorithmic perspective on creativity, and seeks to tie down the study of creative behavior to specific processes, algorithms and knowledge structures. The goal of CC is not just to theorize about the generative capabilities of humans and their machines, but to build working systems that embody these theoretical insights in engineering reality. So CC is both an engineering discipline and an experimental science. in which progress is made by constantly turning insights into applications that can be experimentally tested and evaluated. The purpose of these applications is to create novel artifacts – stories, poems, metaphors, riddles, jokes, paintings, musical compositions, games, etc. – in which a large measure of the perceived creativity is credited directly to the machine. We believe that the future of intelligent computers lies in transforming our computers from passive tools into active co-creators, and that CC is the field that can make this transformation a reality.
(...)
While humans and computers can be creative in the absence of a formal definition of how they are being creative, both still need a level of self-understanding and critical awareness to justify the use of the label “creative.” Computers which generate outputs for an external user to evaluate are merely generative in their behavior, and mere generation does not rise to the level of human creativity. Rather, the generation of outputs must be coupled with an awareness of the value of the output in terms of its novelty and its utility. A creative computer must embody a particular view of creativity that the computer itself understands, so that the computer can justify its outputs much as a human creator would do. Such a computer cannot be a dumb savant that naively flings outputs at an audience. Crucially, it must exhibit an ability to filter its outputs for quality, so that any outputs presented to a user show intentionality and discernment, and just as importantly, it must exhibit an ability to articulate why its outputs may have interesting and unexpected value for its audience Thus, in the words of Sternberg and Lubart’s (1995) investment theory of creativity, a creative computer must be able to articulate its sense of how a particular product or idea can be “bought low and sold high”
Though developments in the field of AI have become fixtures in the technological landscape (e.g., machine translation, natural language question answering, driverless cars, grandmaster-level chess and Go), humans still instinctively cling to the idea that creativity is a uniquely human (or uniquely biological) preserve. In this view, when computers apparently exhibit some measure of creativity, this mere appearance of creativity is due to some specifiable slice of the programmer’s own creativity having been imprinted onto the algorithmic workings of the system. In CC research this idea is known as pastiche, since such computers unknowingly resort to the same kind of stylistic mimicry that is knowingly exploited by uncreative human artists. For instance, careful musicological analysis of the structure of Bach cantatas can allow a programmer to write software that generates its own novel cantatas in the style of Bach. Though these outputs may fool the human listener, and even delight the unsophisticated ear, they are the product of a system that mimics rather than creates. Such a system has no awareness of its inherent limitations, nor does it have any conceptual input into the hardwired (albeit pseudorandom) processes that it follows. Such systems are more like skilled forgers than creative artists; while they can expertly mimic and recycle, they cannot innovate, nor can they surprise. Moreover, because they explore a predefined sweet-spot in the space of possible outputs, pastiche systems take no risks, always produce well-formed outputs, and have no need to self-critique or to ever learn from their failures.
Of course, pastiche has its place, both in human and in machine creativity. One can learn from pastiche, and even good creators occasionally lapse into pastiche (recognizing this tendency in himself, Picasso once noted of his own paintings “Sometimes I paint fakes”). Pastiche thus serves as a useful boundary case for computational creativity. Indeed, there are cases where pastiche is precisely what the human co-creator desires (e.g., “let’s explore more variations on this theme”). Pastiche-based systems are a useful starting point for the computational exploration of creativity, but the goal of CC as a field is to actively transcend pastiche, to demonstrate that computers are capable of true, human-level creativity.
CC is an interdisciplinary research field that sits at the intersection of the fields of AI, psychology, cognitive science, linguistics, anthropology and other human-centered sciences. Given its focus on system building, the field has most in common with AI, and builds on many of the same foundations, such as intelligent search in a conceptual/problem/state space. Nonetheless, the field has a distinctive character of its own, which shapes its use of ideas and techniques from other fields. For instance, CC views creativity as arising from more than merely a systematic search of a conceptual space of possibilities. Rather, it recognizes that these spaces are deeply-rutted with conventional pathways, and that creativity arises from how an intelligent agent knowingly exploits or subverts these conventions. Thus, Margaret Boden (1990) suggests ways in which creativity might arise from the exploration of such a space, while Wiggins (2006) has formalized the CC components of this perspective.
A visual representation of a search in a conceptual space is presented in Fig. 1.1. Here, flowers depict acceptable solutions – goal states at which a search can profitably terminate –while footprints illustrate the paths taken by a cognitive agent as it explores the space. Since this model projects a physical search into mental spaces, we can understand “mental agility” as the cognitive equivalent of those qualities that are desirable for an agile physical search.
For instance, one often needs to backtrack gracefully when at a dead end, and shift smoothly to an alternate avenue of search. Note that the search metaphor is just that, a metaphor, though it is one that some CC researchers nonetheless resent as overly reductive. However, alternate metaphors for creative choice-making may yet be reducible to the nondeterministic exploration of an abstract space.
Adaptability, in particular, seems to be a salient aspect of creative behavior that can be formalized in terms of search spaces. Boden (1990) offers an intriguing view of adaptive creativity, of a kind that not only delivers surprising solutions to a problem, but that also changes the way we view the problem itself. Boden argues that one should distinguish exploratory creativity – of the kind visualized in Figs. 1 and 2 – from transformational creativity. While the former explores the space as it is defined by the problem, looking for previously undiscovered or unappreciated states of unexpectedly high value, the latter actively transforms the space. As illustrated metaphorically in Fig. 3, this transformation redefines the criteria of value that gave shape to the space and which drive the search for value in that space.
Fig. 1.3. Transformational thinkers alter the space that they are exploring, to identify high-value targets that lie outside the original space, and which would not have been considered in the original formulation of the problem. Of course, a transformation may also put states that were previously accessible out of bounds to the creative agent (From Veale, 2012)
Boden cites the development of atonal music as a dramatic example of transformational creativity, and one can also point to key developments in science, such as the transformational shift from a Newtonian (absolute) to an Einsteinian (relativistic) world view, or from a classical (determinate) to a quantum-mechanical (indeterminate) conception of reality. When searching through a space, whether that space is physical or abstract, a searcher can either contort itself to fit the constraints of the space, or contort the space to fit the needs and values of the searcher.
Transformations of the kind analyzed by Boden are the exception rather than the rule in creativity, in either its small-C (everyday creativity on a mundane scale) or big-C (exemplary creativity on a historical scale) guise. One finds a more commonplace form of agile exploration of a state space in the narrative jokes that are the common currency of social interaction. Jokes exploit the fact that we all navigate through shared state spaces in our everyday lives, to explain the events in the world around us and to understand the behaviors of our friends and colleagues. These shared spaces have well-trodden pathways that correspond to the commonsense norms of conventional thought processes, but these rutted paths do not always offer the quickest or surest routes to a solution. In cases when the best path to a solution is circuitous and nonobvious, mental agility is not a matter of speed but of sure-footedness. The shortest path can sometimes lead to incongruity and failure.
Jokes employ state spaces that have been deliberately warped, so as to fool the unsuspecting explorer into believing that the quickest and most conventional route is also the most intelligent route. In other words, jokes subvert the logic of intelligent search in a state space, and thereby demonstrate the limits of conventionalized thought processes (Minsky, 1980). The mathematician John Allen Paulos uses the framework of catastrophe theory to characterize the kinds of warped spaces that are most used in narrative jokes: as shown in Fig. 4, these typically contain an unexpected “kink” or discontinuity that corresponds to a surprising gap in the logic of the narrative (Paulos, 1982; Veale, 2012). Explorers who jump to conclusions by pursuing the path of the discontinuity can be humbled and surprised by their unthinking use of conventional logic.
It is in the computational treatment of discontinuity, incongruity, and contradiction that CC most distinguishes itself from AI as a discipline. In a conventional state space search, contradictions are viewed as dead ends from which a computational agent must backtrack.
AI makes an assumption that search is important but the avoidance of search is more important still, so contradictions serve as useful boundaries to limit an otherwise costly search. CC, however, views incongruity and contradiction as opportunities for further search, to explore whether anomalies can be resolved on another level of representation to yield results that are surprisingly meaningful. Resolvable contradictions of this kind underpin not just the incongruity of jokes, but the absurdity of surrealist paintings, the semantic tension of metaphors, the pragmatic insincerity of ironic statements, the plot twists of mystery stories, and even the unexpected discoveries of mathematics and science. In his wide-ranging theory of “Bisociation”, Koestler (1964) argued that the creativity in these diverse phenomena emerges from the collision of two seemingly incompatible frames of reference (see also Lavrac et al. (2019) in this volume). Koestler’s ideas form the basis of Fauconnier and Turner’s influential theory of Conceptual Blending (see Fauconnier (1994) and Fauconnier and Turner (2002)), though it falls mainly to researchers in CC to anchor these ideas in the algorithmic specificity that can only come from computational model building (see e.g., Martins, Pereira, and Cardoso (2019), Pereira (2007), Veale and Li (2011), Veale and O’Donoghue (2000) and Veale (2019)).
You can buy the book “Computational Creativity. The Philosophy and Engineering of Autonomously Creative Systems” edited by Tony Veale and F. Amilcar Cardoso (Springer, 2019) from these additional sellers:
Amazon, eBooks, Walmart and Barnes&Noble.
The Association for Computational Creativity
A key development in the history of the field was the establishment of an international Association for Computational Creativity to promote further research in CC and to foster public engagement with the societal issues surrounding CC technologies.
In 2008 the steering committee took the decision to transform the workshop into a conference, thus establishing the International Conference on Computational Creativity (ICCC).
The ICCC conference series organized by the Association for Computational Creativity since 2010 is the only scientific conference that focuses on computational creativity alone and also covers all aspects of it.
The Eleventh International Conference on Computational Creativity, ICCC’20 will take place June 29 – July 3, 2020 in Coimbra, Portugal.
The conference’s main goals are to provide a space where researchers from across the world can meet to debate ideas, hear about novel approaches to the study of creativity, build partnerships, and start collaborations that explore interdisciplinary opportunities. (...)
The proceedings of its past conferences can be downloaded from the Association’s web page, www.computationalcreativity.net.
And you can find The Association for Computational Creativity on Facebook or Twitter
