The Creative Computer as Romantic Hero?
Computational Creativity Systems and Creative Personæ
Colin G. Johnson
School of Computing
University of Kent
Canterbury, UK
C.G.Johnson@kent.ac.uk
Abstract
A popular definition of computational creativity is that
it consists in behaviour that would be regarded as creative
if performed by humans. This raises the question
of which humans, as there are many different styles of
human creative behaviour. This paper unpacks a number
of ways in which human artistic creativity can be
characterised, compares them with the kinds of creative
actions found in computational creativity, and explores
some aspects of human creativity that are underrepresented
in computational creativity systems.
Introduction
Computational creativity (CC) has been pursued for
decades, with intense activity in recent years. One of the
most common definitions of CC is “building software that
exhibits behavior that would be deemed creative in humans”
(Colton and others 2009). This paper explores this
definition.
Creative behaviour is not monolithic; there are many different
ways in which to validly be a creative person. It might
be asked what kind of creative behaviour in humans we want
to compare computer systems with. This paper is concerned
with creative systems in artistic domains such as visual art,
music and literature, not with so-called everyday creativity.
Creative Personæ
There are clear parallels between the idea of CC as exhibiting
human-like behaviour and with Turing test-style definitions
of intelligence. In both, the system is designated as acting
in a creative/intelligent way if it can generate behaviour
or product that would require, creative/intelligent action in
humans to produce it, that a human observer would recognise.
Creative acts in artistic domains are not actions that most
people will carry out regularly. So, care must be taken in selecting
which humans are taken as exemplars. In traditional
Turing style AI tests, the exemplar human is a member of the
general population. In these tasks we will have to assume
that the exemplar has some specific skills and knowledge.
One approach is to choose beginners in the domain as exemplars.
This fits with the approach to CC system development
that sees the development of systems that can do
beginner tasks as the first step towards the development of
more sophisticated systems. Another approach is to build
systems to be compared with mature creative work, where
the exemplars are mature artists. There is a danger with any
of these exemplar-based systems (Pease and Colton 2011)
that they encourage pastiche; but, if evaluators are primed
sufficiently, perhaps this can be avoided.
McCormack (2005) notes that CC algorithms will be valued
when they “produce art recognized by humans for its
artistic contribution (as opposed to any purely technical
fetish or fascination)”. This seems, reasonable; yet, it might
just be a temporary state. Once we have got beyond the
point at which the products/activities of computational artistic
creative systems are acknowledged as valid artworks, we
might become interested in “biographical” aspects of them,
and produce works that reflect on the origins of the work
without this seeming like “technical fetish”. This reflection
on origins might become part of the depth of the works.
An important point is that not all creative people are creative
in the same way. This paper will consider a number
of dimensions of what will be termed creative persona
space: an informally defined space representing broad attitudes/approaches.
This paper considers three dimensions:
the social vs. individualist dimension; the importance (or
not) of ongoing tradition, development and “craft skills”;
ideas of new and old media and the way in which technology
is used in the artistic production itself.
Ritchie (2007) uses the inspiring set in evaluating CC: a
set of human works as exemplars of what a successful CC
system would generate. The intention here is similar, but
with regard to the creator rather than the works: for a particular
CC system, can exemplars of the creator that is being
represented by the system be given?
Dimension 1: Socially Embedded vs.
Individualistic
One dimension of difference between creative artists is between
those that work as individuals and those that create
work in a socially embedded context. No creative artist is
entirely divorced from social context, but we focus on those
that work directly with others in collaborative creation.
This is rare in the literary arts and uncommon in the visual
arts; occasionally small groups will work consistently
together over a long period of time (e.g. artistic duos such
as Thomson & Craighead, the Chapman brothers and Cardiff
& Miller), but literature, visual art and theatre/film writing
are dominated by individual creators (comedy writing—
International Conference on Computational Creativity 2012 57
especially for TV—is a notable exception, as are the works
of groups such as the Dogme 95 filmmaking collective). In
music, collaboration is more common. This reaches its peak
in performance styles such as free improvisation, where a
number of performers work together to collaboratively create
a work without a preformed plan or an idea of leadership
or direction. In some more commercial creative domains,
such as advertising, group creative work is standard.
Most work in CC focuses on the individualistic concept of
creativity: writing of stories, creation of jokes, composition
of melodies, creation of pictures. There has been some interest
(Cook and Colton 2011) in mapping out the various contributors
to creativity in CC; this is explored futher below in
the discussion of computer as medium. Whilst CC systems
might create interactive works—for example in game level
design (Togelius and others 2010)—it is rare that the system
continues to be creatively active during interaction.
There are a number of potential reasons for this focus on
the individual. These are readily criticisable and there is no
reason to believe that all of them are believed by all practitioners,
but listing them gives us an initial scoping out of the
potential reasons:
• The work in CC is coming out of an AI tradition, which
has focused on the idea of the simulation of the individual
mind interacting with a task (though multi-agent systems
are a counterexample). • From an artistic perspective, there is a tradition of the
“lone genius” in the romanticist tradition in art (Lovejoy
1948), which sees the role of the artist as developing their
own authentic and original voice. This idea of the artist
as romantic hero rejects the idea of collaboration and the
development of an ongoing tradition, instead of which the
great artist is seen as creating an individual body of work
expressing their own personal world-view. • Creativity might be seen as happening because of various
interacting processes within the mind (see e.g.
Koestler (1964)). However, people have a curious
reluctance to admit hierarchical models of interacting
networks, both in intelligence and creativity: creativity/intelligence
might be seen as a product of interactions,
but it is tempting to contain those interactions to one level
in a complex system. It is difficult to conceive of a system
where creativity is a product of interacting systems within
the mind and also a network of interacting minds. • There is a desire to be able to pin down exactly where
the creativity is coming from. If a system is embedded in
a complex social system with both human and computer
agents, it is harder to point to a specific creative act by the
computers. An easy criticism of such systems is that all
of the creativity is coming from the human agents, and the
seeming creativity of the computer agents is replicating,
decorating or making trivial responses to or elaborations
of the creative acts of the human agents. • Individual creativity might be seen as the first stage in
the development of more sophisticated interactive creative
systems; therefore, until CC systems have demonstrated
individual creativity, there is no point in tackling
the “more complicated” task of group creativity. • In practice, most creative work is individual; group creativity,
in the arts, is confined to specialised areas. Therefore,
CC systems are just emulating the world.
There are a number of such collaborative systems that
have been created. Consider Voyager (Lewis 2000), where
heuristics interact within a listening/responding musical system
that improvises alongside human musicians; and Sanfilippo’s
LIES (Sanfilippo 2012), where sound processing systems
are connected, the parameters of these interactions being
adjusted by the user. Is this a CC system? It is easy to say
that the creativity is coming from the user in the form of the
parameter changes; but, those might be provoked by sounds
from the system, what Blackwell and Young call “strong interactivity”,
which “depends on instigation and surprise as
well as response.” (Blackwell and Young 2005).
What might a CC system that was designed to work in a
free, collaborative environment look like? Take musical improvisation
as an example. One source of inspiration might
be the broad guidelines that are given to beginners making a
start in improvisation; whilst improvisation might be “free”,
it is not “anything goes” and there is a strong, often unarticulated,
tradition about what is acceptable behaviour in such
performances. This needs to be learned by participation and
reflection, but guidelines can help to guide beginners so that
they are not floundering totally.
Consider, the three guidelines put forward by Dave Smith:
Listen, Don’t waste sounds, and Develop a sense of social
responsibility. How might a CC system attempt to work with
these? “Listen.” is both trivial (we need to have some means
of getting input from the other performers) but, of course, is
actually a very deep and complex guideline, especially considering
that listening is an active process. Clearly, “listen”
means that improvisation should take account of that listening.
However, account-taking can easily become trivial, and
bejust imitation; learning to develop material, and links between
different heard sounds, is important to produce depth.
How might a CC system “listen” in this deeper sense?
One characteristic of listening is that people frame listening
with regard to what has been heard in the past, making subtle
distinctions between some objectively very similar sounds,
and grouping other sounds together that are objectively different
(e.g. (Goto 1971)). One way to do this would be to run
a system for a long time, and accumulate a set of listenings.
This runs into the problem (Bown and McCormack 2010) of
getting people to interact over a long period of time with a
“naive” system that isn’t providing engaging feedback.
An alternative would be to take inspiration from the idea
of an adult learner who is new to free improvisation but already
has a body of sonic knowledge from listening, speaking
and playing an instrument. For example, a system might
match listened phrases to a corpus of sonic information—
a set of melodies, or a set of nature sounds like birdsong,
or an artificially generated set such as sound contours derived
from spoken text. Developments in audio information
retrieval might mean that such information could be gained
from web search. The system could then base responses on
these matches, which would bring in a broader, allusive set
of responses than those that just work with varying the input
based on just that input alone.
Consider the second guideline: “Don’t waste sounds.”.
Again, there is a naive interpretation of this: don’t play all
of the time. This is simplistic, but could provide the basis
for an implementation; indeed, this is how many beginning
improvisers might, with some success, interpret it.
International Conference on Computational Creativity 2012 58
However, it has more depth. “Waste” could include aspects
of listening: don’t waste the other sounds that are going
on in the environment, whether by ruining them with
your own sounds or by failing to exploit the sounds that have
potential. There is a suggestion of depth of contribution:
providing a contribution that is constantly striving for more
depth and development and eschewing cliche,´
Again, simple versions of these could be implemented in
a CC system, particularly when combined with the deeper
listening ideas above. For example, the agent could measure
the complexity of the current activity and hold back
in over-complex situations, the agent could hold a mediumterm
memory of what material is being worked on and wait
until a particular piece of material is being “played out” before
introducing new material, and so on.
Turning to the final point, how could a computational
agent “develop a sense of social responsibility”? The “society”
in question clearly includes the fellow players; also
perhaps, the audience. One aspect of being socially responsible
is recognising when behaviour is regarded by others
as gauche or inappropriate. Could a computer agent measure
this by the inference of affective states from fellow
players (Picard 1997)? Clumsily, this could be done by direct
feedback from the players to the agent. Another way
in which humans learn social cues is by observing social
interactions amongst others; could a creative agent build a
model of such interactions within the group of improvisers
and then use this model to develop its own interactions? Or,
more simply, by analyse whether the material it performs is
taken up by fellow players and use that as a proxy for the
appropriateness of certain kinds of interaction?
Finally, how could such a system be evaluated? Evaluation
of “individualist” creative systems is typically done
via the impression that the system makes on the audience.
However, a collaborative system could also be evaluated by
the fellow players, and such players might use different criteria
to that used by (or, indeed, perceptible by) an audience.
For example, a creative improvisation system might
be rated highly by fellow (human) improvisers if it provides
a contextually-sensitive way of provoking the other improvisers
to produce more engaging music.
Dimension 2: Tradition-Classicism vs.
Romantic Individualism
Another aspect to the notion of the “romantic hero” is the
contrast between the artist working in isolation as an individual
genius and the artist working in a tradition. This is
somewhat different to the above discussion: this section considers
the contrast between the artist who pursues their own
“individual voice” against the idea of one who is concerned
with developing out from and contributing to the development
of an ongoing tradition. By contrast, consider a classicist
view which sees originality happening via the gradual
development of a tradition, constantly underpinned by ideals
of balance and proportion.
The idea of romantic originality was a significant shift in
concept of artistic development within the European mainstream
tradition, contrasting with earlier traditions about
artistic creation being about skillful execution within a style,
including reuse and redevelopment of material. The romantic
artist creates work that reflects their own, individual, often
troubled, engagement with the world (Butler 1981).
The relationship of CC to this dimension is complex. The
idea of creativity expressed by Boden’s model of transformational
creativity reflects the classicist notion of a gradually
developing tradition; the space is, after all, transformed,
not rejected. But, perhaps too much stall shouldn’t be placed
on this—after all, all artistic work builds to some extent on
previous work, and the most individualist romantic hero uses
tools developed by previous generations. Indeed, individual
initiative eventually tips over into eccentricity; witness the
reviews collected by Slonimsky (2000), or the reception of
“outsider art” (Rhodes 2000).
In areas where there is no ground truth, radical novelty is
difficult to evaluate. A seemly incomprehensible piece could
be madness or genius. In Boden’s terms, if a transformation
consisted of taking one space and replacing it by another,
how would works in this new space be evaluated? For a more
sober transformation, evaluation can start with our existing
ideas of evaluation and push them a little; in a completely
new space there is no corresponding grounding.
The view along this axis therefore gives a contrast to the
previous one. Typical CC systems represent individual creativity
rather than social creativity; but, they represent individual
creativity within a tradition rather than that of the
radical outsider.
Dimension 3: Old vs. New Media
Another distinction that can be made in artistic creativity is
between so-called old media and new media. Defining new
media is complex and contestable. Manovich presents an
initial, naive understanding of new media as those cultural
objects that essentially involve the use “of a computer for
distribution and exhibition.” (Manovich 2001).
He goes on to describe ways in which digital technology
has influenced the process of creating cultural works.
One example is where speedup faciliates a difference in kind
rather than just a difference in degree. E.g., real-time rendering
of 3D scenes makes interactive games possible as well
as improving the production process of traditional animation.
Furthermore, computer technologies provoke artists
into exploring new creative areas: for example, the notion
of transcoding (Manovich 2001), i.e. the ready exchange of
data between different media formats, makes us think about
creating works in which different media streams are created
from the same source material, whether in a supportive or
disruptive manner.
Computational Creativity: Old or New Media?
Are the products of CC systems old media or new media?
Perhaps counterintuitively, the vast majority of CC systems
are in an old media tradition. Whilst the computer is essential
in CC, the role it plays is as creator; in new media, the
computer is essential in the work as it is presented.
The work on computational creation of stories (Gervas´
2009), poetry (Manurung and others 2000) and
jokes (Ritchie 2009) is clearly in this vein: the aim of
the vast majority of such systems is to create a work that is
presented as words-on-the-page; whether these words are
presented on paper or on the computer screen (or read out
International Conference on Computational Creativity 2012 59
loud) is not part of their essence. There are a few examples
of literary creativity systems that are clearly within the new
media tradition: for example, nn by Montfort (2007) is an
example of the generation of interactive fiction.
In the case of CC systems for music, the landscape is
more mixed. Systems such as Voyager produce sequences
of notes to be performed by a synthesis system (sounding
like a traditional instrument) or by a mechanical instrument.
However, there are a number of examples that demonstrate
how a creative music computing system could work in a new
media fashion. For example, Magnus’s evolutionary music
concrete ` experiments (Magnus 2006) and Sanfilippo’s
LIES system discussed above show how creative computer
systems can create electronic music that is concerned with
sound manipulation rather than the production of notes.
CC works in a real new media tradition are rare. There is
little CC work producing internet art (Greene 2004) or multimedia
works. There are few works that use computationallybased
means of organising material such as transcoding,
or dynamic creation of work from a database (Manovich
2001), or whose aesthetic is a computational one such as
the database aesthetics discussed by Vesna (2007).
One example is Dance Evolution by Dubbin & Stanley
(2010), which uses a computer game engine as the basis
for a system whereby characters learn to dance in time to
music. The characters are stock video-game images of soldiers;
it is not clear whether this was simply a use of the
resources that were readily available within the engine, or
whether the choice was deliberate. Regardless, this unusual
choice of avatar provides a provocative image, reminiscent
of various artists attempts to subvert the video-game culture
by, for example, the performance of street theatre within
MMORPG environments (Greene 2004).
The Creative Networked Computer
Despite CC arising during the “Internet age”, the typical
CC system produces creative output in a fixed medium (e.g.
words, pixels on a screen, MIDI notes) that ignores the networked
context of the computer. CC systems that draw on
allusions to the world beyond are rare. Most storytelling
systems (Gervas 2009) produce stories about a fixed set of ´
ideas and characters. Most of the CC systems for visual arts
work in an abstract medium.
Where they do provide external reference, this has typically
been provided by the system designer directly. One
way is that the designer builds into the system some understanding
of the external world: for example, whilst the
details of how people are drawn in Cohen’s AARON system
(Cohen 1995; Boden 1990) are created by the system,
the basic idea of a person-shape is part of the system design.
The second way this that the designer might place processes
within the system that generate allusions to the world beyond,
for example in ecologically inspired systems such as
those discussed by Bown & McCormack (2010) where the
interactions between components of the system are inspired
by the kinds of interactions found in natural ecological systems.
This could well suggest aspects of the natural world
to viewers of the system, even if the presentation is very abstracted;
but, the decision to make this allusion is that of the
system designer.
There are exceptions. Krzeczkowska et al. (2010) present
a system where the initial source material is drawn from current
news stories, and keywords extracted are used in web
searches for pictures, which are used as the source material
for the creation of a collage using Colton’s Painting Fool
system (Colton 2012).
An important part of much art (particularly visual and
conceptual art) is connotation: the depth of the work comes
from ideas that are suggested, triggering connections that
remain under conscious awareness, or revealed via “ah-ha”
moments where the link or allusion is suddenly revealed (the
idea of CC systems framing information as part of their creativity
(Colton and others 2011) captures some of this).
Computer as Creator; Computer as Medium
CC researchers might eschew working in new media because
of the potential for confusion between two different
roles for the computer. For a CC system in a computer-based
medium, the computer is playing two roles; that of creator,
and that of the medium in which the work is realised. In
theory, there is no reason why such a system should not be
successful. However, in terms of evaluation, the creative
computer working in new media is harder to evaluate. There
is a question of the role that the “creative” computer played,
versus the role played by the broader computational context.
McCormack (2005) has argued that to be taken seriously,
CC systems need to create works that are not just examples
of “technical fetish”, but that are accepted in their own
right. This may be why CC systems have avoided new media
works: many new media works play with the idea of
their technological groundedness in a self-referential and essential
way, and CC system creators avoid building systems
that work in such media to avoid accusations of “mere” technical
obsession. But, awareness of origins need not reflect
an over-obsession with a trivial part of the work; indeed, the
lack of any sense of biographical depth is one of the shallowing
aspects of many CC-produced works.
One new direction would be to build CC systems that explore
and celebrate what computers are good at. One inspiration
for this comes from John Cage’s account of his development:
After I had been studying with him for two years,
Schoenberg said, “In order to write music, you must have a
feeling for harmony.” I explained to him that I had no feeling
for harmony. He then said that I would always encounter an
obstacle, that it would be as though I came to a wall through
which I could not pass. I said, “In that case I will devote
my life to beating my head against that wall.” (John Cage,
Lecture on Indeterminacy (Cage 1973))
A CC system could adopt a similar attitude, to acknowledge
that computers might not be capable of “passing off” certain
aspects of human creativity (e.g. creating fluent natural
language text) but that certain new forms of creativity are facilitated
that draw on the specific capabilities of computers
and the computational infrastructure.
Let us consider two recent works. Kessel’s 2011 Photography
In Abundance where a day’s uploads to Flickr were
printed out and placed in a gallery—hundreds of thousands
of photos. Such a work could not have been achieved without
computational infrastructure; yet, is this computational
creativity? Presumably not, as the creative decision about
International Conference on Computational Creativity 2012 60
what to do with the computational capability of Flickr was
made by the human artist. Yet, the computational infrastructure
seems to have played a stronger role here than that of
medium—the contribution of the computer to this work is
greater than the contribution of paint to a painting. Perhaps
we need to get used to the idea of artists working in a richer
medium, which “pushes back” in terms of creative contribution
much more than a conventional medium.
The second example is my own 2011 piece Blank: nine
printed panels containing image search hits for “blank”.
Most of the images are of empty objects: map outlines,
blank signs, empty music paper. One image in the seventh
panel consists of a photo of a number of gun cartridges—
referencing “blanks” as a cartridge without a bullet.
Had this work been produced unaided by a human artist,
people might ascribe it creative depth. Having set up the expectation
of empty, neutral images, there is this single flash
of violence in the middle of one panel, causing us to reinterpret
the meaning of the emptyness. Of course, there is no
intention to create this as such; it is “mere” serendipity. Yet,
it is deeper than the readerly (Barthes 1970) interpretation
of a purely aleatoric work; the images were “chosen” by a
process that has a huge amount of infrastructure behind it. Is
this any shallower than some purely human-produced work
which has “come to mind” as the end result of the memory
structures produced by the artist’s life experiences?
<references_biblio/>
References
Barthes, R. 1970. S/Z. Editions du Seuil.
Blackwell, T., and Young, M. 2005. Live algorithms. AISB
Quarterly 122:7–9.
Boden, M. A. 1990. The Creative Mind, Myths and Mechanisms.
Weidenfeld.
Bown, O., and McCormack, J. 2010. Taming nature: Tapping
the creative potential of ecosystem models in the arts.
Digital Creativity 21(4):215–231.
Butler, M. 1981. Romantics, Rebels and Reactionaries:
English Literature and its Background 1760-1830. Oxford
University Press.
Cage, J. 1973. Silence. Marion Boyars.
Cohen, H. 1995. The further exploits of AARON, painter.
Stanford Humanities Review 4(2).
Colton, S., et al. 2009. Computational creativity: Coming
of age. AI Magazine 30(3):11–14.
Colton, S., et al. 2011. Computational creativity theory: The
face and idea models. In Proceedings of the 2011 International
Conference on Computational Creativity.
Colton, S. 2012. The painting fool.
http://www.thepaintingfool.com/ (visited April 2012).
Cook, M., and Colton, S. 2011. Automated collage
generation—with more intent. In Proceedings of the 2011
International Conference on Computational Creativity.
Dubbin, G., and Stanley, K. O. 2010. Learning to dance
through interactive evolution. In Di Chio, C., et al., eds., Applications
of Evolutionary Computation, 331–340. Springer.
Gervas, P. 2009. Computational approaches to storytelling ´
and creativity. AI Magazine 30(3):49–62.
Goto, H. 1971. Auditory perception by normal japanese
adults of the sounds “l” and “r”. Neuropsychologia
9(3):317–323.
Greene, R. 2004. Intenet Art. Thames and Hudon.
Koestler, A. 1964. The Act of Creation. Macmillan.
Krzeczkowska, A., et al. 2010. Automated collage
generation—with intent. In Ventura, D., et al., eds., Proceedings
of the International Conference on Computational
Creativity, 36–40.
Lewis, G. E. 2000. Too many notes: Computers, complexity
and culture in Voyager. Leonardo Music Journal 10:33–39.
Lovejoy, A. 1948. On the discrimination of romanticisms.
In Essays in the History of Ideas. Johns Hopkins University
Press.
Magnus, C. 2006. Evolutionary musique concrete. In Roth- `
lauf, F., et al., eds., Applications of Evolutionary Computing,
volume 3907 of Lecture Notes in Computer Science, 688–
695. Springer Berlin / Heidelberg.
Manovich, L. 2001. The Language of New Media. MIT
Press.
Manurung, H. M., et al. 2000. Towards a computational
model of poetry generation. In Proceedings of AISB Symposium
on Creative and Cultural Aspects and Applications of
AI and Cognitive Science, 79–86.
McCormack, J. 2005. Open problems in evolutionary music
and art. In Rothlauf, F., et al., eds., Applications of Evolutionary
Computing, volume 3449 of Lecture Notes in Computer
Science, 428–436. Springer Berlin / Heidelberg.
Montfort, N. 2007. Ordering events in interactive fiction
narratives. In Magerko, B. S., and Reidl, M. O., eds., Intelligent
Narrative Technologies: Papers from the 2007 AAAI
Fall Symposium, 87–94. AAAI Press.
Pease, A., and Colton, S. 2011. On impact and evaluation in
computational creativity: A discussion of the turing test and
an alternative proposal. In Proceedings of the AISB symposium
on AI and Philosophy.
Picard, R. 1997. Affective Computing. MIT Press.
Rhodes, C. 2000. Outsider Art: Spontaneous Alternatives.
Thames and Hudson.
Ritchie, G. 2007. Some empirical criteria for attributing
creativity to a computer program. Minds and Machines
17(1):67–99.
Ritchie, G. 2009. Can computers create humor? AI Magazine
30(3):71–81.
Sanfilippo, D. 2012. Dario Sanfilippo.
http://dariosanfilippo.tumblr.com/ (visited April 2012).
Slonimsky, N. 2000. Lexicon of Musical Invective. W.W.
Norton and Co.
Togelius, J., et al. 2010. Search-based procedural content
generation. In Di Chio, C., et al., eds., Applications of Evolutionary
Computation. Springer. 141–150.
Vesna, V., ed. 2007. Database Aesthetics: Art in the Age of
Information Overflow. University of Minnesota Press.
International Conference on Computational Creativity 2012 61