Computational Poetry Workshop: Making Sense of Work in Progress
J. Corneli⇤
Goldsmiths College
A. Jordanous
University of Kent
R. Shepperd
Goldsmiths College
M. T. Llano
Goldsmiths College
J. Misztal
Jagiellonian University
S. Colton
Goldsmiths College
C. Guckelsberger
Goldsmiths College
Abstract
Creativity cannot exist in a vacuum; it develops through
feedback, learning, reflection and social interaction
with others. However, this perspective has been relatively
under-investigated in computational creativity research,
which typically examines systems that operate
individually. We develop a thought experiment showing
how structured dialogues can help develop the creative
aspects of computer poetry. Centrally in this approach,
we ask questions of a poem, inviting it to tell us in what
way it may be considered a “creative making.”
Keywords: computer poetry, social creativity, flowcharts,
Writer’s Workshops
‘We can talk,’ said the Tiger-lily: ‘when there’s anybody
worth talking to.’
Through the Looking Glass, Lewis Carroll
Introduction
We are writing in a large part to champion Alan Turing’s
proposal that intelligent machines should “be able to converse
with each other to sharpen their wits” (Turing, 1951).
The formalism that we propose builds on the notion of social
cybernetics that flows from the following propositions
of Heinz von Foerster’s, which he uses to theorise systems
in which participants can responsibly specify their own roles
in relationship to other system participants:
“Anything said is said by an observer.”
“Anything said is said to an observer.”
(Von Foerster, 2003 [1979])
According to Jaako Seikkula and Tom Arnkil, who draw
on the philosophical and literary analysis of Mikhail Bakhtin
(Bakhtin, 2010 [1986], 1984 [1963]) in their approach to
psychosocial work,
“Dialogues could be called ‘the art of crossing boundaries’.
Instead of trying to control others, the parties
reach out towards each other to hear their views better,
to generate shared languages and to join resources.”
(Seikkula and Arnkil, 2014, p. 23)
⇤Corresponding author. Email: j.corneli@gold.ac.uk
This paper outlines a study of social creativity with a dialogical
emphasis, taking computer poetry as our working domain.
It uses the Writer’s Workshop model (Gabriel, 2002)
as the virtual laboratory in which to conduct a thought experiment.
The findings of our study are applied to the FloWr
system (Charnley, Colton, and Llano, 2014). We focus on
the following questions in turn:
– How has the social dimension of creativity been explored
in CC to date?
– How can a created artefact tell us about its making, and
what can this contribute to CC?
– How can computer poetry contribute to developing a
process-based theory of poetics?
– What would have to change about the FloWr system to
implement the computational poetry workshop approach?
– What are the pros and cons of the workshop approach?
– What might be the future role of dialogue in CC?
Background
Social creativity in CC
Minsky noted that computers need to be social if they are to
deal with problems of any great complexity (Minsky, 1967,
1988). We believe that this is particularly true for challenges
in computational creativity, since the essence of creativity
lives in its appreciation by the creative entity itself
and its audience. With creativity in ‘the eye of the beholder’
(Cardoso, Veale, and Wiggins, 2009), the ability to respond
to evaluation during the creative process (Poincaré, 1929
[1908]; Csikszentmihalyi, 1988) becomes pivotal. Social
creativity expands this paradigm by introducing co-creators
to the process, and creating works that rely on dialogue, re-
flection, and multiple perspectives (e.g. the stages suggested
by (Gervás and Leon, 2014)). ‘Results’ may be steeped in
process and are not always based on consensus.
The Four Ps of creativity – the creative Person, Product,
Process and Press (i.e. environment) (Rhodes, 1961) – have
been emphasised in general creativity research. Pluralising
these terms (Persons, Products, Processes) calls further attention
to a social dimension of creativity, and would emphasise
the way the “Press” accommodates multiple multidirectional
perspectives akin to a social network in both the
modern and original senses. The Pluralised Ps remind us
1
Proceedings of the Sixth International Conference on Computational Creativity June 2015 268
that in order to understand creativity it is not sufficient to
model a lone creator or to generate an attractive artwork.
To date, computational creativity research has achieved
many successes in computational generation of creative
products, but the question of how these systems could adapt
and learn from feedback to improve their creativity is
less-explored in computational creativity (Jordanous, 2015).
Evaluation has been advanced as a pivotal contributory part
of the creative process, but researchers often give priority
to generating artefacts that could be seen as creative over
the task of incorporating feedback and evaluation within the
processing of a creative system (Jordanous, 2011).
At the previous year’s International Conference on Computational
Creativity (ICCC 2014) the opening session had
the theme “co-creation.” However in the main proceedings
of the conference, 36 out of 49 papers (approximately 3
in 4 papers) do not appear to mention social interaction or
the ability to respond to feedback. Some notable exceptions
highlight the usefulness of interaction and feedback
for creative systems (McGraw and Hofstadter, 1993; Colton,
Bundy, and Walsh, 2000; Sosa, Gero, and Jennings, 2009;
Pérez y Pérez, Aguilar, and Negrete, 2010; Pease, Guhe,
and Smaill, 2010; Saunders, 2012). Some of this work is in-
fluenced by the DIFI (Domain-Individual-Field-Interaction)
framework (Csikszentmihalyi, 1988). However, social interaction
between creative agents and their audience is often
overlooked or relatively simplified: some examples in
the domain of computer poetry presented below give the
flavour. Increased development of the interactivity of creative
systems, especially where this affects the way these
systems works, has been highlighted as deserving more attention
(Colton and Wiggins, 2012).
FloWr is a framework for implementing creative systems
as scripts over processes that can be manipulated visually
as flowcharts (Charnley et al., 2014). Its general approach
consists of linking the inputs and outputs of code modules,
called ProcessNodes, together to create a linear flow of data.
The resulting Flowcharts can be constructed and executed
visually through a GUI; however, they are ultimately represented
as scripts, which are the main medium of FloWr.
Experiments with automatic process generation in FloWr,
reported in (Charnley et al., 2014), highlight the ability of
the tool to do meta-programming and modify its own flowcharts.
This suggests that FloWr has potential as an environment
for modelling social creativity, where the observers are
nodes and flowcharts, and their languages are, respectively,
programming and meta-programming instructions.
. . . and in computer poetry
In the domain of poetry-generation, there have already been
several attempts to simulate social creativity by incorporating
multi-agent systems. In WASP (Gervás, 2010), social
behavior is simulated by incorporating a cooperative society
of readers/critics/editors/writers consisting of specialized
families of experts that cooperate during the poetrygeneration
process. The McGONAGALL system (Manurung,
Ritchie, and Thompson, 2012) incorporates diverse
modules as operators in evolutionary algorithms that produce
poems fulfilling the constraints on grammar, meaning
A. how to become a writer1
r W i et for 8 hours a day
Readfor8hoursad ya
B. “the Other”
Statement
Speak
Response
Listensandinet pr er st
C. proto-workshop
Poet
Writes Poem
Responds
Contex
S
t
peaks
Reader
Responds
Figure 1: (A) gives a simple recipe for the growth and development
of a writer; (B) response always has dimensions
that goes beyond the utterance that is overheard; (C) adds a
reader who shares the context with the writer and responds.
and poeticity. This approach facilitates the pursuit of several
alternative solution paths in parallel, focusing on more
promising results or coming back to former ideas. However
McGONNAGALL does not provide any communication
between modules. In the MASTER system for computeraided
poetry generation (Kirke and Miranda, 2013) a society
of agents in various emotional states influences each other’s
moods with their pieces of poetry. The poetry-generation
process is based on social learning as the agents interact by
reciting their own pieces of poetry to each other. The generated
poems are based on repeated words and sounds, and
are closer in some ways to music than to typical language.
Montfort, Pérez y Pérez, Harrell, and Campana (2013) and
Misztal and Indurkhya (2014) use blackboard approaches to
poetry-generation, in which independent specialized modules
cooperate via a shared global workspace, à la (Baars,
1997). “Experts” exchange information using the blackboard,
but without direct communication and without feedback
about the reception of their created artifacts.
In connection with our work in the current paper, we did
a limited proof-of-concept reimplementation of some of the
core methods of blackboard poetry system inside of FloWr;
we include one of the generated poems and the corresponding
flowchart.
Methods
“What are the proposed ‘lab rats’?”
The generative side of the cycles in Figure 1 has been studied
more than the reflective side. Our “lab rats” are, accordingly,
not poems per se, but rather, instances of reading and responding
to poetry. Naturally, such responses could be more
or less “canned” (as with Michael Cook’s humorously nonspecific
AppreciationBot2), so the question becomes: what
constitutes an interesting and useful response, and how will
these be developed? The idea of responses is useful at various
levels. We focus here on staging an encounter between
writer and reader.
Writer’s Workshops
Quoting (Gabriel, 2002, pp. 2–3):
The original idea behind the writers’ workshop was to
do a close reading of a work... looking at the words on
1
According to (King, 2000).
2
https://twitter.com/appreciationbot
2
Proceedings of the Sixth International Conference on Computational Creativity June 2015 269
the page rather than the intentions of the author or the
historical and aesthetic context of the work. Under this
philosophy, the workshop doesn’t care much what the
author feels about what he or she wrote, only what’s on
the page.
Framing and any other contextualisation of the work
as it is intended to be presented is permitted, and receives
critical attention. We define a Workshop closely
following Gabriel’s outline, to be an activity consisting of
these steps: presentation, listening, feedback,
questions, and reflections. The first and most important
feature of feedback is for the listener to say what
they heard; in other words, what, for them, is in the work.
In some settings this is augmented with suggestions.
After any questions from the author, the commentators
may make replies to offer clarification. In related recent
work, we have shown how the Workshop framework can
help foster serendipitous discovery and invention (Corneli,
Pease, Colton, Jordanous, and Guckelsberger, 2015; Corneli
and Jordanous, 2015).
Content as creative process
Giving agency to the poem rather than the poet’s intentions,
the poem illuminates its own creative process. This informs
our approach to Workshop interactions, which are focusing
on the poem observing its own construction. We’re interested
in context not in the literary or historical sense but in
the micro-history of the poem’s creative evolution. The originary
and therefore unpredetermined nature of the creative
process means that the outcome represents a more accurate
and objective evidence of the process than the poet’s attempt
to explain the process. Moreover, to the extent that a creator
knows what is expressed through the creative process,
even he or she learns this only in the course of doing the
work. Observers are only able to consider after the fact how
a creator may have selected and rejected various possibilities.
The content of the poem is no more and no less than
how the poem was made.
“In a poem, objective material becomes the content and
the matter of the emotion and not just its evocative occasion.”
(Dewey, 1958 [1934], p. 69)
P. G. Whitehouse writing on Dewey’s Art as Experience
suggests that Dewey joins Collingwood in separating aesthetic
emotion from any notion of inspiration that could be
considered to be something like raw materials. An emotion
is aesthetic when it “adheres to an object formed by an expressive
act” (Whitehouse, 1978, pp. 149–156). However,
“the art object does not have emotion for its significant content”;
rather, the emotion “belongs to the self that is concerned
in the movement of events toward an issue that is
desired or disliked” (Dewey, 1958 [1934], p. 14).
Aspects of the creative process
Doug Anderson and Carl Hausman take Collingwood’s
study further and map the creative process roughly as follows
(Anderson and Hausman, 1992, pp. 299-305):
Disturbance ! aesthetic emotion ! response !
artist’s decision on components of expression ! feeling
of easement plus a simultaneous emerging of a
unique imaginative expression ! alleviation ! realising
and converting prior psychical emotion ! unique
aesthetic experience including new conscious emotion
The poem is a work of progress before it is a work in progress.
The purpose of a poetry workshop that attends to the
content of the poem as process is to illuminate what the poet
is exploring through his/her creative process and through the
poem. The process of reading a poem is also a process of
poiesis – and in the Workshop, the reader joins the writer in
the process of creation. Asking questions like those listed in
in Table 1 tells us what the constituent parts of the poem are
doing.
Relevance for CC research
From a CC standpoint, asking what the work tells us about
the creative process gives an objective and critical focus on
“creative evolution” (Bergson, 1911 [1907]) and provides
an antidote to the seductions of mere generation. A poetry
workshop gives participants the opportunity to read
the drafts and final versions of poems by other Workshop
participants, a shared culture of critique that can be applied
to previously existing poems, and a structured way to
gather feedback on one’s own work in progress. These analyses,
unbiased by the explanations of the (software) creator,
will allow participants to explore and extend the conceptual
space around poetry, or in practical terms, the toolbox
the agents can access. “Extending” expresses both a refinement
of the tools used and the introduction of entirely new
tools. Moreover, reverse-engineering of the creative process
from artefacts will help to teach agents participating in the
workshop at which stage of their creative process these new
tools or extensions could potentially be used. Dialogue in
the workshop involves “respecting the voices of each of the
participants” (Seikkula and Arnkil, 2014), be they agents,
poems, or individual words – and suggests that we look at
the “art of boundary crossing” that is to be found inside
poems.
Bridges between ‘theory’ and ‘practice’
Our ansatz is that the Workshop could serve as a way to
develop a process-based theory of poetics. There are certain
prerequisites: in particular, an underlying context is required,
shared (with respect to differing points of view) by
the poet and the reader/listener (see Figure 1). Participants
are assumed to have relatively stable, enduring but evolving,
identities – either might be able to ask “Who am I?” and
“Who are you?” (Bakhtin, 1984 [1963], p. 251). Answers
would acknowledge a prepared mind with certain prior questions,
abilities, involvements, and so on. However, within
the Workshop dialogues, the discussion focuses solely on the
work itself. Persistent identities allow participants to learn
from these exchanges.
Table 1 contains a list of questions that a reasonably sophisticated
poetry reader might ask about poems. This is complemented
by a list of questions that could be addressed, in a
3
Proceedings of the Sixth International Conference on Computational Creativity June 2015 270
Question Examples
What are the register(s) of the
poem?
cliched, instructive,
imperative
Who is addressed? reader, poet, friend,
rival, confidante
What position(s) are present in the
poem?
pleading, remonstrating,
ephemeral
What is the poem doing with the
reader?
accuse, bewilder, alienate
Who are the characters in the poem? “the falconer”, “you”,
narrator, “two men”
What is the role of image(s) in the
poem?
“the sea”, “a bicycle”;
multiple meanings
What functions, mechanics, and
paradigms are present for the reader
to engage with?
communication, subverted
cliche
What problems, discomforts, or diseasements
are invoked in the poem?
horror, self-loathing,
rejection, desire
How do these evolve? E.g. an image may
start to take over from
a register
What is the world of the poem,
and how does the poem distinguish
between this and its perception of
this?
“Surely”, “must”;
sacred vs mundane;
perspectival vs surreal;
tale vs telling
What are the overlaps, transitions,
implicit dialogues?
“twinned” lines/ideas,
juxtaposed parts of
the poem
What role does the chronology of
reading play, versus references to
chronology and chronological positions
within the poem?
flagged development,
evolution, movement,
stasis
How are lexical categories used? solid nouns, tortuous
adjectives, indistinct
adverbs
Are there discernible allusive effects?
illustrating the literary
apprenticeship of the
author (or reader)
Where is the poet presented with respect
to the poem?
Confidence, determination,
exploration
Table 1: Questions that we ask when reading a poem
straightforward programmatic manner (Table 2). Each of the
examples listed in the right-hand column of Table 1 (and a
plethora that are not listed) present a way of thinking about
the poem. We can see these as roughly analogous to the
agents in Table 2 (Minsky, 2006).
To illustrate, in response to a computer-generated poem:
Oh dog the mysterious demon
Why do you feel startle of attention?
Oh demon the lonely encounter
ghostly elusive ruler
Oh encounter the horrible glimpse
helpless introspective consciousness
A human critic might offer the following feedback:
1. The use of the word mysterious in the first line has
no resolution, real or attempted, or quest to find one.
Question Agent concerned
Word level
What is are dictionary definitions of
this word?
WordNet expert
What are its etymological roots? Etymology expert
Where did this word come from? Provenance expert
What pronouns are used in the
poem?
Pronoun expert
Phrase level
What are the components? Keywords expert
Do the components have a negative
or positive connotation?
Association expert
What are the modifiers attached to
the components?
Modifer expert
Sentence level
What is the parse tree? Grammar expert
Line level
How long is the line? Counting expert
Where does it break? Breathing expert
Where is there white space? Position expert
Poem level
How are terms that exhibit emotion
distributed within the poem?
Distribution expert
Where is there alliteration (rhyme,
consonance) in the poem?
Phonics expert
Does the poem have a metrical
structure?
Rhythm expert
How repetitive is the poem? Repetition expert
Does the poem cohere? Thematic expert
Does the poem have a progression? Narrative expert
Where are the various elements of
the poem concentrated?
Entropy expert
Table 2: Questions we imagine a computer would currently
be capable of answering when reading a poem
2. The use of the word attention is not being interrogated
or acknowledged for its importance. Its qualifying
word is startle, used here as an adjective; acknowledging
the fact that the attention is noted, but
is not yet part of the transformative of the poem.
3. This is repeated in the next references to the aesthetic
experience as a lonely encounter, exclusive ruler,
horrible glimpse and introspective consciousness.
4. The contact made between the poem and its own
construction is qualified in negative terms attached
to the words demon, encounter and consciousness.
5. This poem does not welcome the intimacy of bringing
anything to aesthetic consciousness so that it
might be expressed. Why do I say that? Because
the words are generalised and horribly imprecise.
6. The poem does not move toward a better understanding
of the ideas it alludes to. The vocabulary seems
to associate exploration with fear and isolation and
this is (paradoxically) quite an interesting acknowledgment
of the poem’s refusal to go anywhere i.e. to
become a thing transformed by a creative process.
4
Proceedings of the Sixth International Conference on Computational Creativity June 2015 271
Each of these six points is dual-voiced in the sense that
the critic is relaying the words of the poem with a new emphasis.
Each such statement is one side of a micro-dialogue
(Bakhtin, 1984 [1963], p. 73). The challenge is, of course,
to bring the observations into the awareness of the computer
poet, across the “analogue divide.” Care should be taken not
just to blythely program the computer with more rules, but
rather to give attention to facilitating the process of learning
new rules contextually. We continue with the example from
this point of view in the following section.
First we will consider a reversal of roles, with the computer
in the position of critic, looking at a passage from an
historical piece of poetry. We have selected a passage from
Robert Burns that might have – but in fact did not – serve as
a model for the poem generated above.
I’m truly sorry man’s dominion
Has broken Nature’s social union,
An’ justifies that ill opinion
Which makes thee startle
At me, thy poor, earth born companion
An’ fellow mortal!
Naturally, the first problem is for the computer to read the
poem. One of the approaches that is most appealing from
our point of view is the automatic generation of a semantic
network from the input text (Harrington and Clark, 2007).
We could straightforwardly extend the methods of Harrington
and Clark with notions drawn from Table 2.
1. The passage begins with I/me, locating the poor,
earth born poet
2. thee/thy is another person, possibly the reader, who
becomes startled
3. Singular I contrasts with the class man
4. sorry is a sad emotion
5. truly exaggerates sorry
6. dominion is large
7. broken and union are opposites
8. sorry and justifies are opposites
9. union, companion, and fellow are positive words
about relationship and joining
10. broken, ill, poor, startle and mortal are related to
frailty
11. born and mortal are related
12. There are a lot of rhymes in the poem, at the end of
the lines, enjambed.
These comments are very different from the other reading
above, and are differently interesting.
We’ve demonstrated that the computer is capable of asking
objective questions of a poem. A similar semantic network
approach would allow it to listen to feedback and take
it on board, even when it doesn’t understand the ways of
thinking that generate this feedback. Again, this links the
process of reading and writing poetry to a process of dialogue.
Moderator
Flowchart A
(F_A)
Flowchart B
(F_B)
1. Read log
2. Send message
3. Writes message
4. Modify flowchart
Workshop
Log
4. Modify flowchart
3. Writes message
2. Send message
Figure 2: Schematic diagram for a workshop built in the
FloWr system
Seeds for a FloWr Garden
Keeping in mind the current limitations of FloWr – no looping
or conditionals, only running one flowchart at a time and
in one direction – a conversation between ProcessNodes or
flowcharts is not immediately feasible. Figure 2 represents
a hypothetical design in which a Workshop could take place
with a minimally-altered version of FloWr. As shown in
Figure 2, each participant in the Workshop would be represented
by a single node. One of these nodes is a moderator in
charge of dictating the interaction between the participants
of the Workshop, while the rest represent flowcharts that
have the ability to modify their own connections according
to the discussions from the Workshop – this can be achieved
by exploiting the scripting mechanism of FloWr and dynamically
loading the new structure of the flowchart. Moreover,
a shared log would contain the history of the messages exchanged
during the Workshop and a queue of messages waiting
to be delivered. We define four different types of messages
that can be exchanged:
• comments about specific elements of a poem, or more
general statements about how the poem affects this reader.
• questions to facilitate comprehension of this commentary;
for instance, the questions can vary from simple requests
of sources of information (e.g. files, input from
another node, which resources a flowchart uses, etc.) to
process-specific details (e.g. current conditions, purpose,
other outstanding questions, etc.)
• answers would be associated to previous questions and
may contain simple text such as an url for the source of
information, or a piece of script representing a node used
by a flowchart.
• suggestions are changes proposed by one participant to
another. Similar to the answer, this can be as simple as
suggesting the change of an information source, or more
complex, such as suggesting the replacement of a node
for an alternative node.
A Workshop session follows this communication protocol:
5
Proceedings of the Sixth International Conference on Computational Creativity June 2015 272
1. The moderator initialises an empty log and sends a message
to the flowcharts to indicate that the session has started.
2. The flowcharts start writing messages in the log.
3. The moderator checks the current state of the workshop
by reading the log.
4. The moderator selects the next message in the queue and
passes it to the target flowchart.
5. The flowchart reads the message and acts accordingly, by
either (i) modifying its connections or; (ii) sending a message
back, i.e., writing to the log.
6. Step 3 is repeated until no further message are left in the
queue.
Example. Figure 3 shows the poetry generator flowchart
that generates the poem about the “demon dog” presented
above. The flowchart uses two linguistic resources: ConceptNet
(Liu and Singh, 2004), a semantic network of common
knowledge, and Disco (Kolb, 2008), a semantic similarity
words retrieval system. Let us assume the human critic
A has access to the system through a “UI flowchart” like a
Read-Eval-Print Loop (REPL), and the poetry generator B
is mainly concerned with maintaining a generative flowchart
like the one shown in the figure. The following exchange of
messages can occur:
1. Comment from participant A to participant B: The
words “lonely encounter” and “elusive ruler” in lines 2
and 3 are generalised and imprecise.
2. Question from participant B to participant A: I
identify the processes Disco3 and Disco4 as the source
of the problem. Can you suggest an alternative to Disco?
3. Suggestions from participant A to participant B: Use
WordNet or the Historical Thesaurus to find more expressive
and specific terms for the core concepts in the poem;
try to link the core concepts together by chaining together
related concepts in ConceptNet or WordNet.
4. Action executed by participant B: Receives suggestions,
creates several alternative versions of the script, executes
them and decides which one is most coherent and
which conveys a sense of narrative.
From this exchange, the computer might learn (without ever
being explicitly told) that expressive terms and narratives are
related, and it might begin to discover a way to produce coherent
poems with a narrative structure.
Since the computer has source code instead of a brain, we
can use it to do control studies with process. However, in
general source code does not uniquely determine process:
contextual effects are what make an experiment an experiment.
As described in (Cook and Colton, 2013), code may
include hints about its expected operating context. This is
related to our theme of embedding process within an artefact.
In this connection, one extension to FloWr that would
help to facilitate dialogue between flowcharts would be to
add machine-readable commentary to ProcessNodes. Commentaries
would label a node’s inputs and outputs, describe
Figure 3: The flowchart that created the “demon dog” poem
its basic purpose, and provide information about procedure,
conditional behaviour, mappings between processes and
elements of a generated poem (like the mapping between
Disco3 and “lonely encounter”).
Altered versions of a flowchart (Charnley et al., 2014)
can be seen as parallel solutions that could be executed and
compared on a population basis with respect to some prespecified
metrics in order to make an informed decision on
which suggestion(s) to follow, as hinted in the last step of
the example. In (Colton, Pease, Corneli, Cook, and Llano,
2014) we explored the related idea of modelling system progress
over time. Learning new rules contextually would offer
one clear measure of progress. Caveat lector: considerable
work would be required to realise the ideas we’ve described
in FloWr or any other platform we’re aware of.
Discussion
Potential applications. The paradigm advanced in this
paper would not remove the “generation” aspects of CC,
but would pair them more closely with reflection. The same
skills that support learning in a writers workshop may support
a form of dialogue with the work itself, leading to richer
creative artefacts that show us more about how creativity
works. Focusing on social creativity does not suggest that
we should devalue works from lone creatives, but it does
suggest that we think about how we knit individuals together
in the social fabric of the CC community. The current model
at the International Conference for Computational Creativity
(ICCC) is similar to many other academic conferences: we
present our work to one another and build our sense of community
in that way. But what about a track for computers
to present their work? The idea of computers interacting in
a workshop-like setting is not unprecendented. As Turing
(1951) foresaw, computational software has become highly
competent at Chess and reasonably competent at Go, partly
through continuous practice pitting programs against each
6
Proceedings of the Sixth International Conference on Computational Creativity June 2015 273
other. Poetry could be approached in a similar way, reviving
the floral games of the troubadours. Other creative arts
may also be amenable to the same sort of approach. Gabriel
mentions “brainstorms, critiques, charrettes, pair programming,
open-source software projects, and even master
classes” (Gabriel, 2002, p. 11). The sort of thinking we have
developed here might be adapted to contexts like these.
Potential criticisms. It can be challenging and timeconsuming
to invite and process feedback, and the Workshop
would often be seen as unnecessary for standardised
production cycles that can already produce artefacts that are
“good enough.” Furthermore, since we often seem to get
the computer to do just what we have in mind when we’re
programming, it might not make sense to treat it as a distinct
other and invite it to participate in a dialogue. (Some
REPL users may disagree, and may already think of programming
as a dialogue.) From our read/write perspective
on computational creativity, the most immediate problem is
that appreciation of works of art is rather hard. Consider
the difference between creating a video game (for example)
and playing a video game. In the first case, the designer
has full control over the rule-set, game mechanics, interaction
devices and so forth. At least one computational video
game designer can play its own games (Cook, Colton, Raad,
and Gow, 2013), and an experiment shows that it is possible
for an artificial game player to learn how to play classic
video games using reinforcement learning, starting from
raw pixels (Mnih et al., 2013) – but both are quite far from
general-purpose game playing. This is itself a topic of contemporary
research, and it serves to illustrate that coping
with feedback is a major challenge for AI research. Finally,
we are not in a position to make strong claims about the
quality of workshopped artefacts, although our experience
with poetry has shown us that high-quality poems are often
exactly the ones which teach us about the creative process.
We hope future research will explore this connection further.
Conclusions
The ideas of social interaction, feedback, and evaluation
have frequently been discussed in CC, but implementation
and theorisation around these topics have been more limited.
In the current paper, we suggest giving artefacts more
agency, designing computer programs with more autonomy,
and focusing research effort on creative evolution. We have
shown that in principle computers can engage in dialogue
about poems, which points to a theory of poetics rooted in
the making of boundary-crossing objects and processes. In
order to move from thought experiment to computational
simulation, FloWr could be helpfully extended with further
programmer facilities including loops, subroutines, and
commentaries, along with the ability to generate-and-test
in a population-based manner, and the ability to learn over
time. Workshops and related approaches are suitable for
autonomous learning and development of the creative process,
but they face technical and also some theoretical limitations.
Dialogue may offer a way to creatively push these
limits, empowering both programs and programmers.
Acknowledgements
This research has been supported by EPSRC grants
EP/L00206X and EP/J004049, and the Future and Emerging
Technologies (FET) programme within the Seventh
Framework Programme for Research of the European Commission,
under FET-Open Grant numbers: 611553 (COINVENT)
and 611560 (WHIM).
<references_biblio/>
References
Anderson, D. R., and Hausman, C. R. (1992). The Role of
Aesthetic Emotion in RG Collingwood’s Conception
of Creative Activity. Journal of Aesthetics and Art
Criticism, 299–305.
Baars, B. J. (1997). In the Theater of Consciousness: The
Workspace of the Mind. Journal of Consciousness
Studies, 4.
Bakhtin, M. (1984 [1963]). Problems of Dostoevsky’s poetics.
Translated by Caryl Emerson. University of Minnesota
Press.
Bakhtin, M. (2010 [1986]). Toward a Philosophy of the Act.
University of Texas Press.
Bergson, H. (1911 [1907]). Creative evolution. Henry Holt
and Company.
Cardoso, A., Veale, T., and Wiggins, G. A. (2009). Converging
on the divergent: the history (and future) of the
international joint workshops in computational creativity.
AI Magazine, 30(3), 15–22.
Charnley, J., Colton, S., and Llano, M. T. (2014). The FloWr
framework: automated flowchart construction, optimisation
and alteration for creative systems. In D. Ventura,
S. Colton, N. Lavrac, and M. Cook (Eds.), Procs.
of the 5th International Conference on Computational
Creativity.
Colton, S., Bundy, A., and Walsh, T. (2000). Agent based
cooperative theory formation in pure mathematics. In
Procs. of AISB symposium on creative and cultural
aspects & applications of AI and cognitive science.
Colton, S., Pease, A., Corneli, J., Cook, M., and Llano,
T. (2014). Assessing Progress in Building Autonomously
Creative Systems. In D. Ventura, S. Colton, N.
Lavrac, and M. Cook (Eds.), Procs. of the 5th International
Conference on Computational Creativity.
Colton, S., and Wiggins, G. A. (2012). Computational Creativity:
The Final Frontier? In Procs. of 20th European
Conference on Artificial Intelligence (ECAI) (pp. 21–
26). Montpellier, France.
Cook, M., and Colton, S. (2013). From Mechanics to Meaning
and Back Again: Exploring Techniques for the
Contextualisation of Code. In Procs. of the AIIDE
Workshop on Artificial Intelligence and Game Aesthetics.
Cook, M., Colton, S., Raad, A., and Gow, J. (2013). Mechanic
Miner: Reflection-Driven Game Mechanic Discovery
and Level Design. In Procs. of EvoGames
Workshop.
Corneli, J., and Jordanous, A. (2015). Implementing feedback
in creative systems: a workshop approach. arXiv:
7
Proceedings of the Sixth International Conference on Computational Creativity June 2015 274
1505 . 06850 [cs.AI]. Retrieved from http : / /
arxiv.org/abs/1505.06850
Corneli, J., Pease, A., Colton, S., Jordanous, A., and Guckelsberger,
C. (2015). Modelling serendipity in a computational
context. arXiv: 1411.0440 [cs.AI]. Retrieved
from http://arxiv.org/abs/1411.0440
Csikszentmihalyi, M. (1988). Society, culture, and person: a
systems view of creativity. In R. J. Sternberg (Ed.),
The Nature of Creativity (Chap. 13, pp. 325–339).
Cambridge, UK: Cambridge University Press.
Dewey, J. (1958 [1934]). Art as experience. Capricorn
Books.
Gabriel, R. P. (2002). Writer’s Workshops and the Work of
Making Things: Patterns, Poetry. . . Addison-Wesley
Longman Publishing Co., Inc.
Gervás, P. (2010). Engineering Linguistic Creativity: Bird
Flight and Jet Planes. In NAACL HLT 2010 Second
Workshop on Computational Approaches to Linguistic
Creativity. Los Angeles: Association for Computational
Linguistics.
Gervás, P., and Leon, C. (2014). Reading and writing as a
creative cycle: the need for a computational model. In
D. Ventura, S. Colton, N. Lavrac, and M. Cook (Eds.),
Procs. of the 5th International Conference on Computational
Creativity.
Harrington, B., and Clark, S. (2007). ASKNet: Automated
Semantic Knowledge Network. In A. Howe, and R.
Holte (Eds.), Procs. of the Twenty-Second AAAI Conference
on Artificial Intelligence (Vol. 2, pp. 889–
895). AAAI Press.
Jordanous, A. (2011). Evaluating Evaluation: Assessing
Progress in Computational Creativity Research. In
Procs. of the Second International Conference on
Computational Creativity (ICCC-11). Mexico.
Jordanous, A. (2015). Four PPPPerspectives on computational
creativity. In Procs. of the AISB Symposium on
Computational Creativity.
King, S. (2000). On writing. Simon and Schuster.
Kirke, A., and Miranda, E. (2013). Emotional and Multiagent
Systems in Computer-aided Writing and Poetry.
In Procs. of the Artificial Intelligence and Poetry Symposium
(AISB’13) (pp. 17–22). Exeter University, Exeter,
UK.
Kolb, P. (2008). DISCO: A Multilingual Database of Distributionally
Similar Words. In Procs. of the 9th Conference
on Natural Language Processing.
Liu, H., and Singh, P. (2004). Commonsense Reasoning in
and Over Natural Language. In Procs. of the 8th International
Conference on Knowledge-Based Intelligent
Information and Engineering Systems (Vol. 3215).
LNCS. Springer.
Manurung, R., Ritchie, G., and Thompson, H. (2012). Using
Genetic Algorithms to Create Meaningful Poetic
Text. Journal of Experimental and Theoretical Artifi-
cial Intelligence, 24, 43–64.
McGraw, G., and Hofstadter, D. (1993). Perception and creation
of diverse alphabetic styles. AISB Quarterly, 85,
42–49.
Minsky, M. (1967). Why programming is a good medium for
expressing poorly understood and sloppily formulated
ideas. Design and Planning II-Computers in Design
and Communication, 120–125.
Minsky, M. (1988). The Society of Mind. New York: Simon
and Schuster.
Minsky, M. (2006). The Emotion Machine: Commonsense
Thinking, Artificial Intelligence, and the Future of the
Human Mind. New York: Pantheon.
Misztal, J., and Indurkhya, B. (2014). Poetry generation system
with an emotional personality. In Procs. of 5th International
Conference on Computational Creativity.
Mnih, V., Kavukcuoglu, K., Silver, D., Graves, A., Antonoglou,
I., Wierstra, D., and Riedmiller, M. (2013). Playing
Atari with Deep Reinforcement Learning. arXiv:
1312 . 5602 [cs.LG]. Retrieved from http : / /
arxiv.org/abs/1312.5602
Montfort, N., Pérez y Pérez, R., Harrell, D. F., and Campana,
A. (2013). Slant: A Blackboard System to Generate
Plot, Figuration, and Narrative Discourse Aspects
of Stories. In Proceedings of the fourth international
conference on computational creativity.
Pease, A., Guhe, M., and Smaill, A. (2010). Some Aspects
of Analogical Reasoning in Mathematical Creativity.
In Procs. of the International Conference on Computational
Creativity (pp. 60–64). Lisbon, Portugal.
Pérez y Pérez, R., Aguilar, A., and Negrete, S. (2010). The
ERI-Designer: A Computer Model for the Arrangement
of Furniture. Minds and Machines, 20(4), 533–
564.
Poincaré, H. (1929 [1908]). Mathematical Creation. In The
Foundations of Science: Science and Hypothesis, The
Value of Science, Science and Method (Chap. III of
Book I. Science and the Scientist, Vol. Science and
Method, pp. 383–394). New York: The Science Press.
Rhodes, M. (1961). An analysis of creativity. Phi Delta
Kappan, 42(7), 305–310.
Saunders, R. (2012). Towards Autonomous Creative Systems:
A Computational Approach. Cognitive Computation,
4(3), 216–225.
Seikkula, J., and Arnkil, T. E. (2014). Open Dialogues and
Anticipations – Respecting Otherness in the Present
Moment. Helsinki: National Institute for Health and
Welfare.
Sosa, R., Gero, J., and Jennings, K. (2009). Growing and
destroying the worth of ideas. In Procs. of the 7th
ACM Creativity and Cognition conference (pp. 295–
304). Berkeley, California.
Turing, A. (1951). Intelligent machinery, a heretical theory.
A lecture given to ‘51 Society’ at Manchester.
Von Foerster, H. (2003 [1979]). Cybernetics of cybernetics.
In Understanding understanding (pp. 283–286).
Springer.
Whitehouse, P. (1978). The Meaning of ‘Emotion’ in
Dewey’s Art as Experience. Journal of Aesthetics and
Art Criticism, 149–156.
8
Proceedings of the Sixth International Conference on Computational Creativity June 2015 275