an
international and interdisciplinary journal of postmodern cultural sound, text
and image
Volume 19, Fall/Winter 2022/2023, ISSN 1552-5112
Writers and Writers of Writers: Creativity
and Authorship in the First AI Novel
Introduction
Right
now, you can go onto Google, and write a poem with the aid of artificial
intelligence. ‘Verse by Verse’ is described as an ‘experimental AI-powered
muse’[1]. It allows you to choose the classic
American poet or poets whose style you would like to emulate, your poem
structure, syllable count and rhyme scheme, input just one opening line, and
your ‘muse’ will suggest the rest. Ray Kurzweil’s cybernetic poet invites
readers to guess which poems were written by its poet and which by a human
author, in a type of Turing test for poetry[2]. Google’s Magenta project spans a range
of creative enterprises from writing to music, exploring the potential for
‘machine learning as a tool in the creative process’[3]. Creative algorithms have been used to
come up with ‘what if’ scenarios for children’s stories and West End musicals;
to produce writing in the style of Harry Potter, and even to create a script
based on all the previous episodes of Seinfeld[4].
In
2018, Christie’s claimed to be the first auction house to offer an artwork
created by an AI. It was entitled ‘Portrait of Edmond de Belamy’, and the name
of the artist in the bottom left-hand corner belonged to an algorithm[5]. It sold for $432,000. The work was
created by a Paris-based collective, which fed the AI portraits from the 14th
century to the 20th century. Hugo Caselles-Dupré, one of the members
of the collective, said they felt that portraits, more than other types of
artwork like nudes, or still lives, demonstrate that ‘algorithms are able to
emulate creativity’[6]. Assertions that AIs are creators in
their own right should be considered cautiously. The human input, the
collaboration with artificial intelligence, is sometimes concealed in favour of
the more sensational headline that an AI has produced a work of art. In his
commentary on the ‘Portrait of Edmund Bellamy’, Browne notes that ‘there
remains significant disagreement over who or what deserves to be an AI artist’[7].
This
article sets out to explore issues of authorship in AI art. It begins with a
sketch of some conceptual contexts, including Deleuze’s outlook on creativity,
Du Sautoy’s ‘Lovelace Test’, and Miller’s views from The Artist in the Machine. These writers all deal with the struggle
to adapt existing definitions of creativity, artistry, and artwork to processes
and products that include an artificially intelligent component. To examine
these issues in more depth, the essay then examines the novel 1 the Road as an example of AI art. A
novel is defined by a number of factors, including its authorship, its context
and the text itself, and the analysis considers the novel from these three
perspectives, concluding with a summary of whether or not the novel passes Du
Sautoy’s ‘Lovelace Test’ and what further fields of exploration the novel opens
up.
Contexts
and Concepts
In
‘What is the Creative Act?’, Deleuze he says creativity is about invention. He
as a philosopher invents concepts, and scientists ‘also invent … they create as
much as an artist’.[8] However, for Deleuze, that creativity
is of a different sort and happens in a specific realm. A scientist, he says
knows how to ‘invent and create functions. What is a function? A function
occurs when there is a regulated correspondence between at least two sets’.[9] Such functions are rules that are
highly codified, and seem far removed from the invention of philosophical
concepts, even though Deleuze emphasises the similarity between the two in that
both are carried out in space-time. The difference between the creativity of
the philosopher and the scientist, as outlined by Deleuze, is similar to the
difference commonly perceived between the creativity of the human and the AI.
Human creativity appears to be more spontaneous, more chaotic, whereas AI
creativity usually takes place in the context of the regulated input of data
sets and programming.
Ada
Lovelace implied a similar fundamental difference between a computer’s ability
to create and that of a human when she pondered the implications of Babbage’s
analytical machine in the 1840s. She cautioned against expecting too much from
this primitive calculator: ‘The Analytical Engine has no pretensions whatever
to originate anything. It can do whatever we know how
to order it to perform’.[10] In other words, with a machine, you can
only get out what you put in. In the 21st century, is that still the
case? In his book The Creativity Code,
mathematician Marcus du Sautoy proposes a new kind of test for machine
creativity. He calls it the Lovelace test, after Ada: ‘to pass the Lovelace
test, an algorithm must originate a creative work of art, such that the process
is repeatable (i.e. it isn’t the result of hardware error) and yet the
programmer is unable to explain how the algorithm produced its output’.[11] In this scenario, the machine must
prove itself to be able to create on an ongoing basis (not just a fluke or a one-off
error), and must produce something new, unexpected, and unaccounted for.
For
Miller, author of The Artist in the
Machine, the definition of creativity need not create an immediate divide
between human and machine: ‘Creativity is the production of new knowledge from
already existing knowledge and is accomplished by problem solving’.[12] By this definition, creativity is
something that machines and humans could equally possess. Indeed, Miller
imagines a scenario when machine creativity might become commonplace: ‘Perhaps
the next Turner Prize winner might be a machine, or the next Pulitzer Prize
winner… Imagine if the next Keats was a computer, or the next Beethoven.
Suppose the next legendary chef was a computer, or the next stand-up comic?’[13] Machine creativity may emulate human
creativity or it may be entirely dissimilar, according to Miller: ‘The great
question is whether computers too can develop these qualities, or whether they
will develop their own ways of thinking and begin to operate and function
autonomously – not as replica people, but as an altogether different and
independent form of intelligence’.[14] To
explore these questions about AI art and artists, in more detail, a text that
claims to be the first novel fully written by an artificial intelligence, makes
an interesting case study.
An
AI Artwork: 1 the Road
1 The Road is a road trip novel about a journey
between Brooklyn, New York and New Orleans, Louisiana. Ross Goodwin created the artificial intelligence
– he describes himself as an ‘artist, hacker, a code inventor and a gonzo data
scientist’.[15] This
is not the first project of this kind that he has been involved in. He
previously collaborated with artist and designer Es Devlin on the projects Poem Portraits and Please Feed the Lions, which invited the public to contribute words
to poetry created by artificial intelligence. He was also involved in Sunspring, a science fiction short
written by an AI named Benjamin. This AI novel project was sponsored by Google
and he was accompanied by Kenric McDowell, the leader of Google’s Artist +
Machine Intelligence program, his sister Beth, who is a food writer, and
Christina Caro, a photographer for Google. The technology used involves
convolutional neural networks, which ‘classify entities in images, along with
long short term memory (LTSM) recurrent neural networks (RNNs), which write
text letter by letter’.[16]
AI
art, like human art, requires input to create output. For this AI novel, the
input was very specific and limited to four streams: a camera mounted to the
boot, a GPS on the roof, a microphone inside the car and a clock. The manuscript was created as the car was
driving in a process that involved several stages. The camera captured an image
every twenty seconds, which was then ‘textualised’ into a black and white image
made from letters and characters. These black and white images appear alongside
the final text of the novel. Then, the image was described in a sentence, which
in turn was fed into a neural net created from 200 books and using statistical
probability to prompt text generation from the original description.[17] The manuscript that emerged was printed
as the car drove along. With its experimental appearance and content, and its
ambiguous authorship, the novel raises many questions about narrative,
creativity and authorship in general. The following paragraphs frame some of
these questions by looking at the novel from three different perspectives
–context, authorship and text.
Context
All
novels are partly defined by their contexts: their industry context – where
they are published and how they are marketed and reviewed; their readership
context- who reads them and why and how they interpret them; and their
situation in a dense web of literary contexts, both fictional and non-fictional
– texts that have directly or indirectly influenced the novel, texts that the
novel can be compared with, texts that the novel has influenced.
This
novel is linked with road novels of 1960s’ America, and Goodwin cites Kerouac’s
On the Road, Hunter S. Thompson’s Hell’s Angels and Tom Wolfe’s The Electric Kool-Aid Acid Test as
inspirations.[18] McDowell suggests in his introduction
that the novel is a contemporary reimagining of the thematic preoccupations of
those writers: ‘Those authors set out in search of freedom, masculinity,
enlightenment, hedonism, 20th century values currently under
renovation’.[19] Goodwin even links the spread of ‘LSD
and radical ideas’ in 60s’ America with 1
the Road. He notes that ‘Hallucination is the technical term for the
activity of a machine learning system when it generates content and technology
is like a drug to the extent that it helps us to reach beyond ourselves and
attain experiences that might otherwise be out of reach’.[20]
1 the Road is also contextualised by its
relationship to writings about technology. Goodwin references Nietzsche,
Heidegger and Kittler in his introduction to the novel, including a photograph
of one of the first typewriters, the Writing Ball, which Nietzsche used and
which perhaps influenced his statement, quoted by Goodwin that ‘Our writing
tools are also working on our thoughts’.[21] Goodwin
points out that the Writing Ball was originally created as assistive technology
for the blind. Its potential was soon discovered, however, as a tool for
writing that was faster than handwriting and enabled the production of multiple
copies. Just as the typewriter, a tool, shapes the creation itself, so too does
the personal computer, and so too does artificial intelligence, in this
experimental novel.
The
novel also has a literary context that relates to the French language and
French publishing. The novel is published by Jean Boîte Éditions, based in
Paris. This publisher specialises in experimental, avant-garde writing. The
introductions to the novel by Goodwin and McDowell are printed alongside a
French translation. These introductions themselves provide context for the
book, explaining the involvement of these two individuals, providing an account
of their experience, and offering their interpretations of the significance of
the novel. These introductions also include images from their trip, including
some of their ‘off-road’ experiences not recounted in the novel itself– Ross’s
sister Beth on a swing in Goldboro, a steak eaten while they stopped at a
friend’s house. The novel is the text that was produced in the car, but it is
contextualised by these documentations of the three-day road trip from its
human participants.
The
journey is also contexualised by a spin-off short film directed by Lewis Rapkin
called Automatic on the Road. The
short film documents the three-day road trip and presents interviews with its
participants. But it also produces its own artistic interpretation on the
making of the novel, having the AI’s writings spoken by a female voice in a
different accent from Goodwin, but speaking with him. This technique suggests
their collaboration – almost as though they are co-authors, even though the
issue of authorship remains ambiguous.
Authorship
A
novel is defined partly by the contexts in which it is produced and the
intertextual contexts of which it is a part, but it is also (perhaps primarily)
defined by its authorship. Traditionally, a novel has an author, a creator, who
imagines a narrative and does the writing; an author who makes something out of
words. But who wrote the manuscript of 1
the Road? Who is the author? Goodwin addresses this question in his
introduction to the novel: “…perhaps the
drivers wrote it, using the car as their pen. Or perhaps the machine wrote it,
using our traversal of the landscape as its means of coherence. Or perhaps
those who wrote the machine’s training corpus have written it together,
reaching out from the past, to influence a new story. Or perhaps I wrote it.
But I am less sure than ever about that.”[22]
On
the cover of 1 the Road, Goodwin
describes himself as the ‘writer of writer’. This may refer to the fact that he
wrote the code for the AI, which by implication is the second ‘writer’ in the
description. Inside the book, on the title page, his name appears as simply
‘Ross Goodwin’, not qualified by the phrase ‘writer of writer’, suggesting that
Goodwin is the author. In the introduction, matters are made even more
ambiguous, as Goodwin states that ‘whether the machine itself is a writer
remains a question that is about as relevant as whether a submarine is a
swimmer.’[23] As imagined by Miller, the AI writer,
if it exists, is an entirely different entity to a human writer. Perhaps this
is why, for an AI-authored or co-authored text, conventional methods of
assigning authorship, even the citing of a name on the title page of a novel,
do not suffice. This AI does not have a name – only the human does – so
claiming authorship in that traditional way is not possible.
In
‘The Death of the Author’, Barthes argues that the author, in the sense of an
individual who holds the key to understanding the literary text, is dead. There
is only a scriptor- a person who writes in the moment. Once a text has been
written, even this relationship with the scriptor is weakened. Any meaning is
put there by the reader, who, like the scriptor, is operating at a specific
moment in time and interprets the text in a singular, temporally specific way,
which is why Barthes says that ‘the birth of the reader must be ransomed by the
death of the Author.’[24] Barthes’ way of thinking about
authorship lends itself to the collaborative creation that went into 1 the Road, as it does not necessitate
any sense of ownership by one individual or entity. Barthes’ ideas also
anticipate the ambiguous status of AI art, by presenting the creative work as a
re-presenting of existing texts, in a similar way to how Goodwin’s neural network
is ‘fed’ selected books, or Obvious’s algorithim is ‘fed’ portraits.
Barthes, again: ‘The scriptor no longer contains within himself humors,
feelings, impressions, but rather this immense dictionary, from which he
derives a writing which can know no halt: life never does more than imitate the
book, and the book itself is only a tissue of signs, an imitation that is lost,
infinitely deferred.’[25]
Rather
than the literary text being an expression of an individual consciousness, for
Barthes it is a web of citations. Barthes avoids the idea that the creative
genius conjures an inspired originality, and instead, like Miller, he sees
creativity as something that emerges out of pre-existing materials.
Foucault
returns to the subject of authorship in ‘What is an Author’, presenting another
useful way of conceptualising the authorship of AI art. Although accepting of
Barthes’ idea that in terms of interpreting texts, the author is dead, he
points out that culturally, the idea of the author still has a meaning and a
currency. Foucault proposes the term
‘author-function’ to describe the position of the author in relation to texts,
industry and culture, while separating the concept from its traditional
attachment to one individual: ‘the 'author-function' is tied to the legal and
institutional systems that circumscribe, determine, and articulate the realm of
discourses.’[26] The concept of an author becomes an
abstract idea – a blank space where the author used to be. The author function
is a useful way of thinking about the concept of authorship in relation to 1 the Road. There is no single author –
it is not Ross Goodwin and it is not the neural network that he designed. But
there is an author-function that can be filled by both human and AI and
potentially others involved in the creation of the novel. This concept of an
author-function is only necessary because the way in which we think about the
novel and its relationship to the author has not changed. Even if we know that
the author must go in inverted commas, that it is only a notional
author-function, we still need it to be there – this is why in the Works Cited
list attached to this article, every piece of writing has an author attached to
it. However, Foucault was open to the idea that the author-function might not
always be necessary: ‘the form, the complexity, and even the existence of this
function are far from immutable. We can easily imagine a culture where
discourse would circulate without any need for an author’, and would instead
emerge in ‘pervasive anonymity.’[27] Perhaps the concept of authorship will
become transformed in the future for those works created by artificial
intelligences, and for the humans who collaborate with them.
TEXT
For
Umberto Eco, returning to the question of authorship in the 1990s, and as a
creative writer himself as well as a philosopher, a balance is required between
traditional ideas of authorship that attach the label to one individual, and
modern ideas of authorship, like Foucault’s - which see the author as a
placeholder that could potentially be filled in a number of ways. On one hand, there is the
fact that authors sometimes have interesting things to say about their own
texts that both casual readers and literary critics find interesting and even
insightful. This does not mean that the author is the master of the text or
becomes the Empirical Author, as Eco says, or the Author with a capital A that
is critiqued by Barthes. On the other hand, there is the philosophical deconstruction
of the author and the presentation of the reader at the site of meaning
production, Barthes’ ‘the birth of the reader’, a viewpoint which Eco also sees
as valuable. For Eco, the key seems to be the idea of intentionality. As
readers, we need intentionality in order to interpret. This intentionality,
like Foucault’s author-function, can be abstract – it need not necessarily be
that of the literal, physical author. In fact, Eco argues that it may be useful
to see the intentionality as situated in the text itself. He states, ‘Between
the mysterious history of a textual production and the uncontrollable drift of
its future readings, the text qua text still represents a comfortable presence,
the point to which we can stick.’[28] If we follow Eco’s guidance, and ‘stick’
to the text of 1 the Road, assuming
that it has an intentionality that we can interpret, what do we find? What does
it have to say about its creation, its authorship and its status as a novel?
At
first glance, the novel looks more like poetry than a conventional novel. In
one margin there are times - times at which the sentences emerged. In the other
margin are dates – dates of the three days over which the road trip and the
writing took place. The text is made up of short paragraphs, usually of between
one and three sentences. By virtue of the way it is created, from descriptions
of images that are then narrativized, the narrative is quite disjointed, but
setting it out on the page in this way makes a virtue of that rather than
trying to disguise it. Interspersed throughout the text are samples of the
grayscale character images that the neural network was working with. The novel
includes frequent references to another input stream also – GPS location.
Although
there are few characters mentioned, and none that are developed over the course
of the novel, one character is referred to several times– the painter. They are
mentioned in the first lines of the novel and appear again later: ‘The painter
laughed and then said, I like that and I don’t want to see it.’[29]
The novel acquires coherence from the input streams of images, locations
and times, which are highlighted in the novel’s visual layout, rather than
coherence from plot or character development. Its narrative journey is based
around progression in time and space rather than progression in character.
Perhaps for this reason, it lacks articulation of emotion, even emotional
responses to the images that are described.
The
novel offers sentences that are intriguingly ambiguous, and that are open to
diverse interpretations. For example: “The time was eleven minutes after three
o’clock in the afternoon, and soon after we were along together, a few people
were standing in the windows of the train station. It was a relief.”[30] This ‘relief’ may be in the sense of
the visual- the people standing in the window of the train station are standing
in relief. Or, it may be relief in the sense of the emotion.
Indoor
and outdoor landscapes, public and domestic spaces, frequently combine in
unexpected ways, that echo modernist paintings like the melting clocks against
a backdrop of rugged rocks in Dali’s The Persistence of Memory: ‘A small hill
in the distance was immediately opened up and settled on the table.’[31] The narrative creates abstract
descriptions that defamiliarise common objects and settings: ‘A black and white
photograph of his face changed. The leaves are green, the shadows hardened and
the sky was rosy and the hand streamed across the floor’.[32] The
man’s face is a black and white photograph perhaps because he is standing in
darkness, and his hand appears to stream because of the play of sunlight and
shadow seen from the passing car. At one point, the landscape is described like
a theatre: ‘A tree in the background was the size of a stage, and a great crowd
of distant stones had been thrown out of the water. Trees with green leaves and
waves tinged in the sun’.[33] The
scenery is imagined anthropomorphically, and given as much personality and
drama as the humans that are seen and described. Although the novel is
unconventional in its layout, its absence of plot and character development,
the fragmentation and dislocation of the narrative, and the fleeting images it
describes, are in synchrony with the preoccupations of postmodernist American
fiction, even if this is not intentional.
CONCLUSION
1 the Road
passes de Sautoy’s Lovelace test: it is a creative work of art, made with a
repeatable process and producing things that its creators are unable to
explain. However, in terms of literature, artificial intelligence has so far
been unable to match human ability, and this novel does not change that.
Goodwin is aware of that, saying: “This is very much an imperfect document, a
rapid prototyping project. The output isn’t perfect. I don’t think it’s a human
novel, or anywhere near it”.[34] It is not a novel written by a human
clearly, and it may be judged as inferior to human novels, but is it a novel?
It is an homage to road trip novels, it has been published by a publishing
house alongside human authors: in terms of context, the answer is yes. It has
been created by a human coder, the ‘writer of writer’, and by a neural network
– this is an unusual authorial partnership, but nonetheless it fulfils the
author-function, as described by Foucault: in terms of authorship then, the
answer is also yes. And what about the text? In relation to its form, its
style, and its production of provocative descriptions and images, the answer is
yes. By all of these measures, 1 the Road is a novel, and the AI at
least contributes to the creativity of its collaborative authorship. Although
the novel undoubtedly has its limitations, it opens up interesting avenues of
exploration for human collaboration with artificial intelligence in the writing
of fiction. More broadly, and perhaps most importantly, the work provocatively
compels its readers to ask fundamental questions about the definitions of
artist and artwork in 21st century.
an
international and interdisciplinary journal of postmodern cultural sound, text
and image
Volume 19, Fall/Winter 2022/2023, ISSN 1552-5112
Notes