The Art in The Machine
This is a review of the Creative Horizons of Computational Systems. Can we build intelligent tools that are creative, help us be more creative or shed some light on the creative process itself? Let’s review the meaning of creativity and the ways through which the machine can be a helpful participant.
I. Creativity, A Human Experience
The Oxford dictionary describes creativity as “the use of imagination or original ideas to create something inventiveness”, but if we ought to find a common agreement across the literature with regards to describing the phenomena of creativity per se (from a human perspective), that would probably be that creativity is, at least, an elusive subject . It seems to trigger a fascinating effect due to the mystery with which it is surrounded. “To produce something out of nothing”, the creative process as performed by humans, can naively be seen as an alchemical process through which the dross of common thought is transformed into golden exceptional ideas. Yet, the most creative people on earth have proved to be unable to break down their own creative process at the peak of its execution. A testimonial tool is needed, that which is able to witness the magic when the magician is in trance, that which is able to whisper back the trick and allow the awe to charm even more. But in order to make a tool understand, can we demystify the concept and bring it closer to achievable and definable terms, reaching the necessary conditions for the artificial modelling something? Soon some definitions we humans have found so far.
Creativity is not marginal. Boden, the godmother of creativity in the grounds of A.I., depicts it as a universal aspect of human intelligence not restricted to a limited few . While there have been many attempts to distillate a solid definition, we may assume the complexity of such a quest and use several sources to draw a descriptive landscape of the unagreed matter:
- Greek philosophers like Plato rejected the concept of creativity, preferring to see art as a form of discovery. When asked in The Republic, “Will we say, of a painter, that he makes something?”, Plato answers “Certainly not, he merely imitates” . This draws from his conviction that, epistemologically, everything already exists. This, at least sets a space for discussion in the matter of originality.
- In 1926, G. Wallas quoted Helmholtz in , describing the creative process in a bottom-up fashion by dissection into four stages: (i) preparation, preparatory work on a problem focusing on exploring its dimensions, (ii) incubation, internalizing it into the unconscious mind while apparently idling the external activity, (iii) insight or illumination, with the arrival of the apparently serendipitous “a-ha” moment that pushes the creative idea from its preconscious processing into conscious awareness, and (iv) verification, where the idea is consciously verified, elaborated and then applied. This segmentation, while appealing to the heuristic reader, also brings an apparent conscious/unconscious duality to the game.
- In 1964, Arthur Koestler attempted to develop a general theory of human creativity by defining the essence of creativity as “the perceiving of a situation or idea […] in two self-consistent but habitually incompatible frames of reference” , thus implying some kind of disruption in a matrix of thought by the non-obvious combination with another. Koestler also suggested that originality, emphasis and economy are universal features of creative thought, differentiating between domains (arts, business, sruvival, …)of creative application by its emotional context.
- Csikszentmihalyi expands on Wallas/Helmholtz by presenting a five-step architecture in his 1996 book , where he divides the verification step into a two-folded deployment, (i) evaluation, introducing an aesthetic/semantic assessment measurement to the product of the internal creative process, and (ii) elaboration, where the process becomes explicitly deliberate and intentional with the clear aim to provide the environment with an artifact. At this point, rather than needing to acquire knowledge, the process must dispense it . Furthermore, and possibly most important, Csikszentmihalyi stresses the importance to consider these steps (preparation, incubation, insight, evaluation and elaboration) as parts of a multiple-iteration, possibly recursive process in which the steps may be revisited multiple times, in varying order as necessary, which adds a considerable amount of freedom degrees to the paradigm.
- Finally, Margaret Boden, who has published multiple work on the field of creativity and its possible modelling through computational toolsets   , describes creativity as the ability to generate novel, and valuable ideas. This author defines valuable as being, at least, interesting, useful, beautiful, simple, and richly complex, attributes that land on the concept of “ideas”, meaning precisely: concepts, theories, interpretations, stories or artefacts that could range from graphical images, sculptures and houses to jet engines. As for novel, Boden tackles on the field-recurring notion of originality by discriminating between two types of outcome ideas:
a) P-creative: psychological novelty, meaning that is new to the person who generated it.
b) H-creative: historical novelty, a p-creative idea that has never occurred in history before.
Hence, laying out a grid for originality evaluation. In addition, Boden anticipates the need to categorize creativity in order to better understand its inner design and lend psychologists — and AI scientists — a more fertile ground for study. In this sense, she includes all instantiations of human creativity into one of these three categories:
a) Combinatorial creativity, arising from combining unfamiliar ideas (reminiscing Koestler’s approach) and working by making associations between ideas that were previously only indirectly linked. This kind includes analogy, it is probably the most mentioned, and has proven itself able to achieve h-creativity (e.g. in the metaphorical and associative poetry of world-class authors like Shakespeare or Kurt Schwitters).
b) Exploratory creativity, which rests on some conceptual space constrained by a set of generative rules, which define the nature of the space (i.e. the style), where creators navigate in order to create ideas based on what is to be found within the bounds of the space and, eventually (in the most interesting cases), discovering the limits and potential of the space in question. This kind embraces the majority of h-creative artists and scientist, such as Mozart, who spent most of his time exploring his handful of self created musical spaces (styles).
c) Transformational creativity, achieved by altering one or more of the defining dimensions of a conceptual space or style, hence providing surprising ideas, which are fundamentally different to any previous one, which are still bounded to pre-existing style norms. This kind clearly raises the tide for achievement and poses the question of what it takes for some entity to deviate from a clear pattern and fundamentally develop new tools for re-conceiving a certain working domain.
II. The Creative Machine
The Association for Computational Creativity defines Computational Creativity (CC) as follows: a multidisciplinary endeavor that is located at the intersection of the fields of artificial intelligence, cognitive psychology, philosophy, and the arts.
The goal of CC is to model, simulate or replicate creativity using a computer to achieve one of the several ends:
1) To construct a program or computer capable of human-level creativity
2) To better understand human creativity and to formulate an algorithmic perspective on creative behaviors in humans
3) To design programs that can enhance human creativity without necessarily being creative themselves
Being widely ambitious, these statements arise some issues that prevent the common researcher from having a crystalized framework to work upon, and as Jordanous highlights , should be taken into discussion. What is meant by “human level creativity”? The creativity of an h-creative genius or a p-creative child? It does not clarify the a clear threshold of inclusion, e.g. if an “algorithmic perspective on creative behaviour” produces results, we might feel inclined to describe it as creative, but if this perspective does not really shed any light on human creativity, does this necessarily exclude it from being computationally creative? Furthermore, if computational creativity can be computational work that enhances human creativity, then aren’t we really dealing with computer-enhanced human creativity, not computational creativity per se? The ambiguous nature of creativity itself brings computational creativity to a difficult stand point.
We follow the discussion with further approaches to CC’s meaning. Colton and Wiggins  define CC research as
“the philosophy, science and engineering of computational systems which, by taking on particular responsibilities, exhibit behaviors that unbiased observers would deem to be creative”.
Ventura  affirms that “it seems natural to interpret the creative process, particularly in a computational context, as one of search”. In 1999, Pérez y Pérez  already crafted the Engagement-Reflection model to implement systems to simulate computational creativity by taking a cyclic process of engaging with creative production then critically reflecting on what it has produced, to inform the next stage of creative production.
Surely the scope of CC, while still undefined, orbits around the main concerns of the present review, it presents challenging dead-ends and perspectives, but also hints on practical results that unequivocally affect the way a creator can relate to her/his work. There seems to be an inherent drive to understand creativity by performing it, as if both actions were part of an undetachable bundle, which may be pointing at how we ultimately experience the phenomena of creativity.
Three (unevenly developed) streams seem to have emerged from the studies in computational creativity:
First, and with generous success, the development of systems for artistic production or performance, able to output interesting artifacts (i.e. that potentially could contribute with novelty and value to a given — artistic — domain). Past research  argues that these systems have already seen great prosperity as opposed to the ones addressing creative problem-solving or models of creative cognitive capacities, which still remain in their infancy. The emphasis and attention in these kind of systems is majorly given to the output, which is a measure for evaluation and subsequent design guidance. One could imagine that the utility of this first family of systems lays in producing artefacts that can be directly inspiring to human creators and thus, feeding a creative feedback loop that could empower human creativity. Moreover, reflecting on the construction and inner-design of the systems themselves could be an inspiring source to tweak and enhance the creative process of a given subject without the need to explicitly model or understand the undergoing creative emergence.
Second, the systems that aim at giving an explicit characterization of creativity and achieving an archetypical “algorithm” for creativity. Apart from being far more ambitious, it is true that they have not been as addressed as the former. Most of the literature that tries to define creativity assumes that this milestone has not been yet achieved due to the elusive nature of the matter at study itself. The intent (or possibility) of printing truly autonomous creative features into an algorithm may cross the scientific frontier and step into the philosophical domain . While creativity is definitely a computational (AI) goal, its potential achievement may remain an open question. In an audacious article, Herbert Simon  defends the capability of explaining “ineffable” topics, such as intuition, insight and inspiration through AI, if staying in grounded empirical psychological lands by demystifying such topics, providing objective definitions. In this direction, the focus on modelling creativity itself has partly been put on (i) concept/space blending formalization  and (ii) the search for creative algorithms that explore and try to validate creativity definitions through computational implementations. In the latter approach, Ventura  researches the feasibility and challenges to cover all of five Csikszentmihalyi creative stages within a unique algorithm, pointing out many fugue points where external aid or consulting (the so-called meta-creative processes) would be compulsory to achieve genuine creative performance.
Third and last, the conceiving of systems or processes that, without necessarily being creative themselves, can potentially enhance human creativity. Let’s flag two convenient approaches to the matter. First, the possibility of considering machines as co-creative partners, and second, to underline the computational horsepower and pattern recognition capabilities of algorithmic approaches (subordinated to the need for interpretability). In  and , four categories of human-computer interaction to promote creativity are discussed: (a) management of creative work (nanny), (b) communication between individuals collaborating on creative projects (pen-pal), (c) the use of creativity enhancement techniques (coach), and (d) the creative act through integrated human-computer cooperation during idea production (colleague). All of them may bring justification to the possibility of conceiving computers as real partners in the creative process intervening at different points in order to generate, evaluate, or refine ideas and bring them to full-fledged products. Finally, one could certainly consider the usefulness of a tool that would be able to unveil hidden patterns of behavior in one own’s work . This is the approach I personally support to consider computation as a powerful tool to model (e.g. using machine/deep learning) a creative process and infer causal mappings that may help the artist/creator in reinforcing or reconsidering her own production routes. This assumes a not so obvious (nor achieved) interpretability of the modelling system in use .
Many are the currently available examples of creative work designed or aided by computational systems. An illustrating example would be the rise of dense layered machine learning algorithms which have given birth to generative algorithms that not only deliver digital aesthetics (mainly sponsored by the ubiquitous new media art of the last two decades), but are capable of creating artworks by mimicking a learn-by-seeing human fashion. The horizons are unclear, but the more we use technology to allow human creators to deepen into their own alchemic process, the more we will be able to seize the benefits of creativity as a human condition.
 Zbikowski, Lawrence M. “Conceptual Blending, Creativity, and Music.” Musicae Scientiae, vol. 22, no. 1, Mar. 2018, pp. 6–23, doi:10.1177/1029864917712783.
 Boden, Margaret A.. “Computer Models of Creativity.” AI Magazine 30 (2009): 23–34.
 Tatarkiewicz, Władysław (1980). “A History of Six Ideas: An Essay in Aesthetics”. Distribution for the U.S. And Canada, Kluwer Boston.
 Wallas, Graham. “The art of thought”. New York: Harcourt, Brace and Company, 1926. Print
 Koestler, Arthur. “The act of creation”. Oxford, England: Macmillan
 Csikszentmihalyi, Mihaly. Creativity: Flow and the Psychology of Discovery and Invention. New York: Harper Collins Publishers, 1996. Print.
 Ventura, Dan. “The computational Creativity Complex”, 2014
 Boden, Margaret A. “The Creative Mind: Myths & Mechanisms”.London: Weidenfeld and Nicolson, 1990. Print.
 Pearce, Marcus. “Boden and beyond: the creative mind and its reception in the academic community
 Boden, Margaret A. “Computer models of creativity”. AI Magazine 30 (2009): 23–34
 Jordanous, Anna. “Four PPPPerspectives on Computational Creativity”, unpublished, 2015
 Colton, Simon, Geraint A. Wiggins. “Computational creativity: the final frontier?” ECSI, 2012
 Ventura, Dan. “No free lunch in the search for creativity”, Proceedings of the 2nd International Conference on Computational Creativity, 2011, p-108–110
 Pérez y Pérez, Rafael. “MEXICA: a computer model of creativity in writing”, PhD Thesis, The University of Sussex, 1999
 Besold, Tarek, R, Veale, Tony. “Editorial: computational Creativity, Concept invention, and general intelligence, Journal of Artificial General Intelligence, 2015, 6(1)
 Simon, Herbert A. “Explaining the ineffable:AI on the topics of intuition, insight and inspiration.” IJCAI, 1995
 Goguen, Joseph. “Mathematical models of space and time”, Reasoning and cognition: proce. Of the interdisciplinary conference on reasoning and cognition
 Guzdial, Mathew, Riedl, Marc O. “ Combinets: Learning new models via recombination” arXiv preprint arXiv: 1802.03605, 2018
 Fauconnier, Gilles, and Mark Turner. The Way We Think: Conceptual Blending and the Mind’s Hidden Complexities. New York: Basic Books, 2002. Print.
 Jordanous, Anna. “Co-creativity and perceptions of computational agents in co-creativity”, 8th International Conference on Computational Creativity, ACC, 2017, p. 159–166
 Lubart, Todd. “How can computers be partners in the creative process: classification and commentary on the Special issue.”, International Journal of Man-Machine Studies 63, 2005, p. 365–369
 Doshi-Velez, Finale, Kim, Been. “Towards a rigorous science of interpretable machine learning.” Eprint arXiv:1702.08608, 2017
 Cohen, H. 1995. “The Further Exploits of AARON Painter. In Constructions of the Mind: Artificial Intelligence and the Humanities”, ed. S. Franchi and G. Guzeldere. Special edition of Stanford Humanities Review, 4(2): 141–160.