On the Notion of Framing in Computational Creativity
John Charnley, Alison Pease and Simon Colton
Computational Creativity Group
Department of Computing, Imperial College
180 Queens Gate, London SW7 2RH, United Kingdom.
ccg.doc.ic.ac.uk
Abstract
In most domains, artefacts and the creativity that went into
their production is judged within a context; where a context
may include background information on how the creator feels
about their work, what they think it expresses, how it fits in
with other work done within their community, their mood before,
during and after creation, and so on. We identify areas
of framing information, such as motivation, intention, or the
processes involved in creating a work, and consider how these
areas might be applicable to the context of Computational
Creativity. We suggest examples of how such framing information
may be derived in existing creative systems and propose
a novel dually-creative approach to framing, whereby
an automated story generation system is employed, in tandem
with the artefact generator, to produce suitable framing
information. We outline how this method might be developed
and some longer term goals.
Introduction
Michael Craig-Martin’s 1973 work, An Oak Tree, comprises
a glass of water on a shelf and an accompanying text, in
which Craig-Martin claims that the object which appears
to be a glass of water is really an oak tree. The text takes
the form of a question and answer session written by CraigMartin
about how he has changed the water into a tree:
A. [.....] I’ve changed the physical substance of the glass of
water into that of an oak tree.
Q. It looks like a glass of water.
A. Of course it does. I didn’t change its appearance. But it’s
not a glass of water, it’s an oak tree.
...
Q. Haven’t you simply called this glass of water an oak tree?
A. Absolutely not.
Craig-Martin is rather mysterious as to how he has accomplished
the change:
Q. Was it difficult to effect the change?
A. No effort at all. But it took me years of work before I
realised I could do it.
Q. When precisely did the glass of water become an oak tree?
A. When I put the water in the glass.
Q. Does this happen every time you fill a glass with water?
A. No, of course not. Only when I intend to change it into an
oak tree.
The status of the piece as a work of art is then raised:
Q. Do you consider that changing the glass of water into an
oak tree constitutes an art work?
A. Yes.
Q. What precisely is the art work? The glass of water?
A. There is no glass of water anymore.
Q. The process of change?
A. There is no process involved in the change.
Q. The oak tree?
A. Yes. The oak tree.
This is an example of human creativity which is taken seriously
in its field. First shown in 1974, it was bought by the
National Gallery of Australia in Canberra in 1977, and has
been exhibited all over the world with the text translated into
at least twenty languages. Like many important works, opinion
is divided: artist Michael Daley referred to it as “selfdeluding”
and “pretentious” (Daley August 31 2002), while
art critic Richard Cork wrote:
“I realise that one of the most challenging moments occurred
in 1974 when the Rowan Gallery mounted an
exhibition of Michael Craig-Martin’s work.” (Cork October
9 2006).
Researchers in Computational Creativity (CC) can learn
from this work. The main point that we consider in this paper
is that the artefact (the glass of water) has no creative
value without the title and accompanying text. The value,
the creativity associated with the piece, lies in the narrative
surrounding the glass of water. This point has clear implications
for CC, which has traditionally focused on artefact
generation, to the extent that the degree of creativity judged
to be in the system is often considered to be entirely dependent
on characteristics of the set of artefacts it produces
(for instance, see (Ritchie 2007)). Very few systems in CC
currently generate their own narrative, or framing information.
Artefacts are judged either in isolation or in conjunction
with a human-produced narrative, such as the name of
the system and any scientific papers which describe how
it works. Researchers in CC will be familiar with what
Bedworth and Norwood call “carbon fascism” (Bedworth
and Norwood 1999) (the bias that only biological creativity
can produce valuable artefacts), and, for the most part,
computer-generated creative artefacts are not taken seriously
International Conference on Computational Creativity 2012 77
by experts in the domain in which the artefacts belong. We
believe that enabling creative software to produce its own
framing information will help to gain acceptance from these
experts. While An Oak Tree may be a rather extreme example
of the importance of framing information, we hold that
such information almost always plays some role in creative
acts, and is a fundamental aspect of human creativity. We
consider here which types of framing information we could
feasibly expect a piece of software to produce, and begin to
propose ways in which we could formalise this. Specifically,
we consider three areas in computational terms: motivation
(why did you do X?), intention (what did you mean when
you did X?), and processes (how did you do X?). We make
the following contributions:
1. We highlight the importance of framing information in
human creativity.
2. We propose an approach to automatically generating
framing information, in which a separate creative act of
automated story generation is performed alongside traditional
artefact generation.
Framing in human creativity
Sir John Tusa is a British arts administrator, radio and television
journalist, known for his BBC Radio 3 series The
John Tusa Interview, in which he interviews contemporary
artists. These interviews have been reproduced as two books
in which he explores the processes of creativity (Tusa 2003;
2006). We have analysed all thirteen interviews in his most
recent collection, in order to provide the starting point for a
taxonomy of framing information. In the following discussion,
unless otherwise specified, all page numbers refer to
this collection of interviews (Tusa 2006). We identified two
categories which artists spoke about: INTERNAL, or inward
looking, in which the artist talks about their own Work, Career
and Life, and EXTERNAL, or outward looking, in which
the artist talks about their view of their Audience and Field.
The artist’s work
Discussion about artists’ work is very common in the Tusa
interviews. This might concern a specific piece; such as how
an artist feels about a piece, what they think it expresses, or
how it relates to everyday concepts; or this might concern
details of the generative process, such as how the work is
created, how processes involved in its creation fit together,
or whether a new technique or material changed the way that
something was done. As an example below, Cunningham
(MC) relates his work to scientific and religious concepts:
MC: ...it was the statement of Einstein’s which I read at that
time, where he said, ‘There are no fixed points in space.’ And
it was like a flash of lightning; I felt, Well, that’s marvellous
for the stage. Instead of thinking it’s front and centre, to allow
any point, very Buddhist, any point in the space to be as
important as any other. (p. 66)
The artist’s career
A picture of the structure of an artist’s career, in terms of his
or her past, present and possible future directions, can aid
understanding of current work. Questions about previous
work include asking how two pieces differ; what category
work from certain periods falls into; classification of a career
into different stages, such as early and late, or pre-work X
and post-work X.
Examples from (Tusa 2006) include the questions: “So
you think you are recognisably the same person, creatively
the same person as you would have been if you’d stayed in
New York?” (to Forsythe, p. 93); “What’s the next stage of
your evolution as a maker of ballets?” (also to Forsythe,
p. 105), and “When you look back over the last twenty
years, would you ever have guessed that the work that you
do would have travelled so far ..... I mean this is an extraordinary
journey. How aware have you been of the evolution
as you’ve been through it?” (to McBurney, pp. 181-2).
The artist’s life
Audiences are interested in the personalities and influences
behind society’s “creative heroes”. Topics of interest include
political, intellectual, personal, cultural and religious influences;
value systems; reasons for working in a particular
area; important events in the life of the artist, and so on.
John Tusa asks many questions in this vein. For instance,
he asks: “Are you an optimist or a pessimist as a person?”
(to Rovner); “When did you discover that you had this condition
called Dysgraphia, where I think the brain wants to
write words as pictures?” (to Viola); “What do you feel, as
you’re coming in to work?” (to McBurney); and “What music
do you like?” (to Piano), and has some rather poignant
exchanges, such as one with Viola in which he asks about a
near-death experience (pp. 221-3); and this exchange with
Rovner (MR):
JT: Are you lonely as an artist?
MR: You mean as an artist or as a person?
JT: Well as a person who is an artist.
MR: I’m alone. I don’t know if I’m lonely. I am single, you
know I’m a single person, I’m a single person. (p. 213)
The artist’s view of their audience
The perception that an artist has of his or her audience may
influence their work. Queries in this topic included questions
about effect of a particular field on audiences, and what
the effect of certain pieces of work are on the collective subconscious.
Egoyan, for instance, discusses responsibility to
one’s audience with Tusa (p75).
The artist’s view of the field
Embedding a particular artist’s work into the context of a
body of work is one of the purposes of framing information.
Queries include definitional questions about particular
fields, and their relationship to other fields; how a piece fits
into a field; in which field an artist sees themselves; the in-
fluence of external characteristics such as politics, or how
modern advancements such as new techniques have affected
a field; the history of a field and directions in which it could
go, and so on. For instance, Egoyan discusses how he thinks
video compares to film (p76), and Forsythe talks about how
his work fits into great classical ballets (p. 106).
International Conference on Computational Creativity 2012 78
Framing for Computational Creativity
Analysis of the interview responses suggests a new direction
for CC: enabling creative software to generate some of
its own framing information. As with human artworks, the
appeal of computer creativity will be enhanced by the presence
of framing. However, there are obvious restrictions on
the scope to which the various forms of framing apply in the
computer generated context. Here we consider three areas in
computational terms: motivation (why did you do X?), intention
(what did you mean when you did X?), and processes
(how did you do X?).
Motivation
Many creative systems currently rely upon human intervention
to begin, or guide, a creative session and the extent
to which the systems themselves act autonomously varies
widely. In some sense, the level to which these systems
could be considered self-motivating is inversely proportional
to the amount of guidance they receive. However, it is possible
to foresee situations where this reliance has been removed
to such an extent – and the human input rendered so
remote – that it is considered inconsequential to the creative
process. For instance, the field of Genetic Programming
(Koza 1992) has resulted in software which can, itself, develop
software. In the CC domain, software may eventually
produce its own creative software which, in turn, produces
further creative software, and so forth. In such a scenario,
there could be several generations in an overall geneology of
creative software. As the distance between the original human
creator and the software that directly creates the artefact
increases, the notion of self-motivation becomes blurred.
Beyond this, the scope for a system’s motivation towards
a particular generative act is broad. For example, a suitably
configured system may be able to perform creative acts in
numerous fields and be able to muster its effort in directions
of its own choosing. With this in mind, we can make a distinction
between motivation to perform creative acts in general,
motivation to create in a particular field and motivation
to create specific instances.
In the human context, the motivation towards a specific
field may be variously influenced by the life of the artist,
their career and their attitudes, in particular towards their
field and audience. Several of these are distinctly human in
nature and it currently makes limited sense to speak of the
life or attitudes of software in any real sense. By contrast,
we can speak of the career of a software artist, as in the corpus
of its previous output. This may be used as part of a
process by which a computer system decides which area to
operate within. For example, we can imagine software that
chooses its field of operation based upon how successful it
has previously been in that area. For instance, it could refer
to external assessments of its historic output to rate how
well-received it has been, focusing its future effort accordingly.
The fact that a computer has no life from which to draw
motivation does not preclude its use as part of framing information.
All those aspects missing from a computer could, alternatively,
be simulated. For example, we have seen music
software that aims to exhibit characteristics of well-known
composers in attempts to capture their compositional style
(Cope 2006). The extent to which the simulation of human
motivation enhances the appeal of computer generated artefacts
is, however, still unquantified. The motivation of a software
creator may come from a bespoke process which has
no basis in how humans are motivated. The details of such a
process, and how it is executed for a given instance, would
form valid framing information, specific to that software approach.
Intention
The aims for a particular piece are closely related to motivation,
described above. A human creator will often undertake
an endeavour because of a desire to achieve a particular outcome.
Factors such as attitudes to the field contribute to this
desire. Certainly, by the fact that some output is produced,
every computer generative act displays intent. The aims of
the process exist and they can, therefore, be described as
part of the framing. In the context of a computer generative
act, we might distinguish between a priori intent and intentions
that arise as part of the generative process. That is, the
software may be pre-configured to achieve a particular goal
although with some discretion regarding details of the final
outcome, which will be decided during the generative process.
The details of the underlying intent will depend upon
the creative process applied. For example, as above, software
creators might simulate aspects of human intent.
Intent has been investigated in collage-generation systems
(Krzeczkowska et al. 2010). Here, the software based its
collage upon events from the news of that day with the aim
of inviting the audience to consider the artwork in the context
of the wider world around them. This method was later
generalised to consider wider combinations of creative systems
and more-closely analyse the point in the creative process
at which intentionality arose (Cook and Colton 2011).
Processes
In an act of human creativity, information about the creative
process may be lost due to human fallibility, memory, awareness,
and so on. However, in a computational context there
is an inherent ability to perfectly store and retrieve information.
The majority of creative systems would have the
ability to produce an audit trail, indicating the results of key
decisions in the generative process. For example, an evolutionary
art system might be able to provide details of the ancestry
of a finished piece, showing each of the generations in
between. The extent to which the generative process can be
fully recounted in CC is, nevertheless, limited by the ability
to fully recreate the sources of information that played into
the generative process. Software may, for instance, use information
from a dynamic data source in producing an artefact,
and it may not be possible to recreate the whole of this
source in retrospect.
One system that produces its own framing is an automated
poetry generator currently being developed (Colton, Goodwin,
and Veale 2012). In addition to creating a poem, this
system produces text which describes particular aspects of
International Conference on Computational Creativity 2012 79
its poetry that it found appealing and aspects of how it generated
its output. In order to fully engage with a human audience,
creative systems will need to adopt some or all of the
creative responsibility in generating framing information.
Details of the creative process are valid aspects of framing
information, which are relevant to both computational
and human creative contexts. There is a notion of an appropriate
level of detail: extensive detail may be dull and
the appreciation of artefacts is sometimes enhanced by the
absence of information about the generative process.
Examples of framing for Computational Creativity
There are many ways in which creative systems might generate
their own framing information. For example, an automated
art system, such as AARON (McCorduck 1991),
could store details of all its previous artworks and provide an
assessment of how a new piece differs, in various respects,
from its past output. A poetry system, such as (Colton,
Goodwin, and Veale 2012), might reveal the general mood
of the inspiring source it used as a basis for an affective
poem. Mathematical software, such as HR (Colton 2002),
could be given the ability to compare the conjectures it finds
against on-line mathematical databases and report on how its
output relates to known theorems. By corollary, an art system
could appeal to image databases to suggest similarities
to other artists. A simple enhancement to the collage generation
program of (Krzeczkowska et al. 2010) could see it
provide the text of the news story that formed the inspiration
for the collage. In this mode, the framing information would
become as important an aspect of the overall presentation as
the collage itself. The artwork would be a combination of
both the collage and underlying story, rather than the collage
alone. This list is by no means exhaustive. The varied nature
of framing information that we have been describing shows
that the opportunities for enhancing works with framing are
extensive.
A dually-creative approach to framing
Framing information has the potential to greatly impact an
audience’s assessment of an artefact. In some instances,
framing is arguably as much a part of the overall creative
presentation as the artefact itself: this was seen in CraigMartin’s
An Oak Tree, described above, as well as, for example,
elements of Marcel Duchamp’s readymades series
such as Fountain. The information can be as simple as a title
for the artefact, or might encompass much of the type of
framing indicated in our analysis. Framing can add to the
mystique and mystery surrounding an artefact, as we have
described.
Framing information need not be factually accurate. Information
surrounding human creativity can be lost, deliberately
falsified or made vague for artistic impact. Thus,
the generation of framing information can itself be seen as
a creative act. The overall impact of the package – namely
the artefact and the associated framing information – will
depend on both the assessed quality of the artefact, together
with the impression given by the framing information. We
propose one approach to artefact-with-framing generation,
where the two are produced simultaneously, by a duallycreative
process. Under this approach, the most appropriate
creative paradigm for the framing information would be a
form of automated storytelling. One part of a combined system
would create the artefact itself and a storytelling aspect
would generate a framing story. The framing story could be
as simple or as complex as those which accompany human
creations. Tools which were able to perform tasks such as
metaphor and analogy (see (Gentner, Holyoak, and Kokinov
2001; Gibbs Jr. 2008)) might be integrated into the storytelling
aspect.
In the previous section, we discussed aspects of framing
which might be relevant to the CC setting. This information
could form much of the input to the story generation system,
becoming part of the basis of the story. For example,
purely factual information about how the software arrived at
the final product could be retained. Given that there is no requirement
for the framing story to be factually correct, some
or all of the story might be fictional and there is no prescription
for the extent to which the framing story should directly
correspond with the artefact. Consider, for example, a framing
story which describes all aspects of the creative process
in full detail compared with a framing story consisting of a
random seemingly-unrelated word. Both have artistic value,
but in entirely different ways.
An initial approach to the fact that no configuration of a
particular automated story generation system would be able
to generate the variety of framing stories that we have witnessed
in human creativity, might be to develop a small number
of story-telling paradigms, each based upon a particular
story template. One challenge might then be to achieve an
appropriate balance between fact and fiction in the generated
stories. In future, we might hand such decisions over
the software. For example, a sufficiently-able software suite
might decide which story-telling paradigm is most appropriate
for a particular effect, the balance between fact and
fiction and how extensive the framing should be. In a more
complex manifestation, the story might form an interactive
dialogue, providing answers to audience queries in a manner
akin to an interview. As with human creativity, the answers
to those questions may be entirely at the whim of the generating
system. Going further, software might employ story
generation approaches to simulate aspects of the framing information
which might otherwise be absent, such as a religious
belief or other motivation. This could, in turn, feed
back into the generation of the creative artefact itself. Storytelling
for framing information represents an interesting
challenge for our existing and future automated story generation
systems.
Related work
In (Colton, Pease, and Charnley 2011; Pease and Colton
2011), two generalisations were introduced with the aim of
enabling more precise discussion of the kinds of behaviour
exhibited by software when undertaking creative tasks. The
first generalisation places the notion of a generative act,
wherein an artefact such as a theorem, melody, artwork or
poem is produced, into the broader notion of a creative act.
During a creative act, multiple types of generative acts are
International Conference on Computational Creativity 2012 80
undertaken which might produce framing information, F,
aesthetic considerations, A, concepts, C, and exemplars, E;
in addition to generative acts which lead to the invention of
novel generative processes for the invention of information
of types F, A, C and/or E.
The second generalisation places the notion of assessment
of the aesthetic and/or utilitarian value of a generated artefact
into the broader notion of the impact of a creative act, X.
In particular, an assumption was introduced that in assessing
the artefacts resulting from a creative act, we actually celebrate
the entire creative act, which naturally includes information
about the methods underlying the generation of the
new material, and the framing information, which may put
X into various contexts or explain motivations, etc., generally
adding value to the generated artefacts over and above
their intrinsic value.
The introduction of these two generalisations enabled the
FACE and IDEA descriptive models to be introduced as the
first in the fledgling formalisation known as Computational
Creativity Theory. In this paper we have extended this model
by exploring the notion of framing.
Future work and conclusions
Creativity is not performed in a vacuum and the human context
gives an artefact meaning and value. Implicit in the
Computational Creativity Theory models so far developed
is the notion that the FACE information/artefacts resulting
from creative acts can be seen as invitations to a dialogue.
For instance, when a person appreciates a painting, they are
encouraged to ask questions of it, and look for answers, either
explicitly from the artist or some perceived notion of
how artists work, via visual interrogation of the piece itself,
or through certain cultural contexts; for example, by understanding
the culture in the time and place when the painting
was produced.
Despite the importance of framing information as part of
the overall artistic endeavour, we are only aware of a very
small number of systems that generate framing information
to accompany their creative output. We have proposed
one approach to this, whereby automated story generation
is used to generate framing information. There are no real
bounds to what information such framing can contain, its
basis in fact versus fiction, or the format in which it is presented.
Consequently, we suggest that initial attempts be
restricted to a small number of simplified paradigms, taking
their basis from a more complete investigation into how
humanly-produced framing information relates to CC. Expanding
upon this starting point, we imagine software taking
over some of the creative responsibility for the framing
information, such as determining the story-telling paradigm
and the story’s emphasis or level of detail.
Craig-Martin, via his narrative of An Oak Tree, opens up
a dialogue with the viewer on the nature of essence, proof,
faith, matter, reality, art, and so on. The viewer engages
with this narrative, which includes the manner of presentation
of the piece, Craig-Martin’s background as an artist and
a person, critics’ and artists’ responses to the piece, stories
surrounding the work and effects that it has on everyday life
(for instance, there is a myth that Australian customs offi-
cials barred it from entering the country since it was classified
as “vegetation”, and in February 2012 the first three
hits from google images on the search term “an oak tree”
are images of Craig-Martin’s work). We anticipate that the
direction outlined in this paper will form an important axis
of development for CC systems. Our long-term goal is to
help to develop CC to such an extent that one day a piece
of creative software will appear in the table of contents of a
collection of Tusa-style interviews, to discuss its work and
itself, alongside other contemporary artists.
Acknowledgements
This work has been funded by EPSRC grant EP/J004049.
We are grateful to the three reviewers, who raised interesting
points.
<references_biblio/>
References
Bedworth, J., and Norwood, J. 1999. The Turing test is dead. In
Proceedings of the 3rd conference on creativity and cognition.
Colton, S.; Goodwin, J.; and Veale, T. 2012. Full face poetry
generation. In Proceedings of the Third International Conference
on Computational Creativity.
Colton, S.; Pease, A.; and Charnley, J. 2011. Computational creativity
theory: The FACE and IDEA descriptive models. In Proceedings
of the 2nd International Conference on Computational
Creativity.
Colton, S. 2002. Automated Theory Formation in Pure Mathematics.
Springer-Verlag.
Cook, M., and Colton, S. 2011. Automated collage generation
– with more intent. In Proceedings of the Second International
Conference on Computational Creativity.
Cope, D. 2006. Computer Models of Musical Creativity. Cambridge,
MA: MIT Press.
Cork, R. October 9, 2006. Losing our vision. New Statesman.
Daley, M. August 31, 2002. Tracey left on the shelf. The Guardian.
Gentner, D.; Holyoak, K.; and Kokinov, B. 2001. The Analogical
Mind: Perspectives from Cognitive Science. Cambridge, MA: MIT
Press.
Gibbs Jr., R. W., ed. 2008. Cambridge, UK: Cambridge University
Press.
Koza, J. R. 1992. Genetic Programming: On the Programming of
Computers by Means of Natural Selection. Cambridge, MA, USA:
MIT Press.
Krzeczkowska, A.; El-Hage, J.; Colton, S.; and Clark, S. 2010.
Automated collage generation – with intent. In Proceedings of the
1st International Conference on Computational Creativity.
McCorduck, P. 1991. AARON’s Code: Meta-Art, Artificial Intelligence,
and the Work of Harold Cohen. W.H. Freeman and Company.
Pease, A., and Colton, S. 2011. Computational creativity theory:
Inspirations behind the FACE and the IDEA models. In Proceedings
of the 2nd International Conference on Computational Creativity.
Ritchie, G. 2007. Some empirical criteria for attributing creativity
to a computer program. Minds and Machines 17:67–99.
Tusa, J. 2003. On Creativity. London: Methuen.
Tusa, J. 2006. The Janus aspect: artists in the twenty-first century.
London: Methuen.
International Conference on Computational Creativity 2012 81