Ai Novels - Crime Books

You might also like

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 18

AI NOVELS

CRIME BOOKS

Silvia CASCIANI
Giulia MONEA
Tanga Bona CYRILLE BERTRAND
Mahsa KARIMI
Muhammad Talha Sharjeel RAHAT

INDEX
INTRODUCTION..............................................................................................................................1
CHAPTER 1.......................................................................................................................................1
BIBLIOGRAPHY REVIEW.............................................................................................................1
1.1 How AI is automating writing: The rise of robot writers.........................................................1
1.2 Why Computers Will Never Read (or Write) Literature: A Logical Proof and a Narrative...2
1.3 What the Rise of AI Means for Narrative Studies: A Response to “Why Computers Will
Never Read (or Write) Literature” by Angus Fletcher...................................................................3
1.4 Creativity and artificial intelligence..........................................................................................4
1.5 A Redhead Walks into a Bar: Experiences of Writing Fiction with Artificial Intelligence...5
1.6 Creative Writing with an AI-Powered Writing Assistant: Perspectives from Professional
Writers..............................................................................................................................................7
CHAPTER 2.......................................................................................................................................8
DISCUSSION......................................................................................................................................8
2.1 AI’s limitations..........................................................................................................................8
2.2 Characteristics of crime books written by humans.............................................................10
CHAPTER 3.....................................................................................................................................12
CASE STUDIES...............................................................................................................................12
3.1 Death of an Author: A Novella...............................................................................................12
3.2 The Vanishing “A AI generated book”...................................................................................14
CONCLUSIONS...............................................................................................................................15
BIBLIOGRAPHY............................................................................................................................16
SITOGRAPHY.................................................................................................................................16
INTRODUCTION

Recently, the world has known the advent of ‘AI novels’, novels said to have been written by
Artificial Intelligence.
Specifically, this paper deals with the differences between crime books written by humans and
crime books written by Artificial Intelligence.
First, we went through the more practical aspect of delineating and understanding what it is that
we are dealing with when we say ‘AI’; then, we expounded on what exactly it means – at the
present moment – when a novel is said to have been ‘written’ by an Artificial Intelligence.
After this, we summarised all the AI’s limitations that surfaced in our bibliography review and
outlined the general features of the crime genre.
To conclude, to ascertain whether or not AI was actually capable of accurately reproducing the
crime genre’s characteristics, we selected and analysed two case studies, two crime books
‘written by AI’: Death of an Author: A Novella, by Aidan Marchine, and The Vanishing “A AI
generated book” by Sameer Bera and ChatGPT.

CHAPTER 1
BIBLIOGRAPHY REVIEW

As our main bibliography, we have selected the following papers about AI and the human-AI co-
writing process and relationship:
o How AI is automating writing: The rise of robot writers
o Why Computers Will Never Read (or Write) Literature: A Logical Proof and a Narrative
o What the Rise of AI Means for Narrative Studies: A Response to “Why Computers Will
Never Read (or Write) Literature” by Angus Fletcher
o Creativity and artificial intelligence
o A Redhead Walks into a Bar: Experiences of Writing Fiction with Artificial Intelligence
o Creative Writing with an AI-Powered Writing Assistant: Perspectives from Professional
Writers

1.1 How AI is automating writing: The rise of robot writers

Three members of the Sultan Moulay Slimane University wrote the first paper we selected: UX
Consultant and PhD student Malak Dargham, Professor Dr. Hanaa Hachimi, and PhD student
Mohammed Boutalline.
The authors of this article start by acknowledging that the fear of robots taking over human jobs
has been with us for a long time. However, they argue, automated writing should not be seen as
the end of human writers and, instead, should be regarded as the beginning of a new era of
healthy competition between man and machine – provided that “organizational fixes” 1 are put
into place to avoid exploitation.
They proceed to point out how robot writers are part of a bigger group of “intelligent robots” 2
that will change and affect people’s life in the future – not unlike computer programs that can
already write for us all sort of tedious texts.

1
Dargham, Hachimi, and Boutalline, How AI is automating writing: The rise of robot writers, p. 1.
2
Ibidem.
1
‘Automation’ is the process where machines are now capable of accomplishing tasks that
previously required a human input to be brought to term. An example of this are the many online
services that can write articles from data.
Nonetheless, automated writing is still in its infancy and is nowhere near the point where it
should be considered a threat to human writers.
The authors of this article stress that they are against substituting “’good’ writers” 3 with AI,
however they believe that the more the AI writes, the better, as this will lead to humans having to
rise up to the challenge to outdo it, thus creating better quality writing.
Still, “[w]riting is hard”4. AI is very much not equipped to write in many fields where skills such
as critical thought are required. All the training in the world cannot teach computers to write like
humans can and do.
Robots will not steal our jobs, the article reiterates, but can “free us from tedious, difficult work
that takes no human skill”5.
The paper concludes by saying that while we cannot use AI to create new works yet, we still
need to decide what kind of works we want to automate in the future and stipulate conditions for
it, as this is the only logical way forward.

1.2 Why Computers Will Never Read (or Write) Literature: A Logical Proof and a Narrative

This article discusses the evolution of Artificial Intelligence (AI) and its relationship with
literature. It starts by highlighting key events in 1969, including the creation of a philosophical
code that gave computers a sense of ‘freewill’. It argues the scepticism faced by AI researchers
and their subsequent breakthroughs in various fields.
The text goes on to discuss the potential future of AI in literature, referencing developments in
the Digital Humanities and Franco Moretti’s discovery that machine-learning algorithms could
interpret vast amounts of literature quickly.
It raises the question of when computers will revolutionise the study of literature, understand
novels as humans do, outdo literary experts, and even create new works. With this article – a sort
of history from before the invention of Digital Humanities to its birth – Fletcher provides the
answers to why computers will never be able to read and write literature.
Professor Richard believed that meaning originated from interpretation, which originates from
symbolic-logic, which could be programmed into a computer to extract significance from
literature. He rejected Bradley’s character criticism, instead following Peirce’s theory of
symbolic language and semiotics, which he affirmed could be applied to all logic-governed
fields, since they too are ‘automatic thinking machines’, including literary semiotics.
His predictions proved to be correct with the advent of machine intelligence two decades later.
When Digital Humanities had its first steps in the early 2000s, some concerns that computers
could harm literary skills like critical thinking were dispelled as automated reading was found to
enhance them. Computers use symbolic logic, which defines relation but not causation, leading
to interpretation. Francis Bacon noted that, unlike interpretations, science provides predictions
and experiments, answering ‘why’. Similarly, literature, through semiotic interpretation, offers
practical empowerment and mystical experiences.
Fletcher emphasises the intellectual value of causal reasoning, a non-logical process that
computers can simulate but not fully replicate. Despite advancements, computers struggle with
literature due to its reliance on narrative, a form of communication beyond symbolic logic.
Classical rhetoricians and the Chicago School emphasised the importance of narrative in
literature, tracing its effects to a causal chain. Computers, relying on symbolic logic, struggle to
3
Ivi, p. 3.
4
Ibidem.
5
Ibidem.
2
understand and analyse literature’s psychological effects. Overcoming this would require re-
engineering computer brains to operate like the human neocortex, using synaptic connections.
Synapses allow for extemporisation and experimentation, supported by mitochondria. Current
transistor technology in computers, however, can only function within pre-designed systems,
making it incompatible with causal reasoning.
To advance the study of literature, we can leverage computers’ ability for symbolic logic and
combine it with our own interpretive and narrative analysis. By partnering automated
interpretation with human understanding, we can accelerate the process while retaining the
benefits of our intuitive reading abilities, in order to maintain the pattern-recognition Richard
anticipated many years before.
In conclusion, the main idea expressed by Fletcher is that computers will never be able to fully
understand or create literature because literature requires causal reasoning, which is embedded in
the human brain, but fundamentally incompatible with the symbolic logic used by computers.

1.3 What the Rise of AI Means for Narrative Studies: A Response to “Why Computers Will
Never Read (or Write) Literature” by Angus Fletcher

Direct riposte to the previous article, this paper disagrees with Fletcher when he answered the
question “Will AI ever be able to read or write literature?”6 in the negative.
AI’s history, tell us authors Chun and Elkins, can be divided into ‘AI springs and winters’, the
names given respectively to periods of positive and optimistic advances and to periods of
downtrend for AI.
‘Artificial Intelligence’ as the name for all “intelligent robots” 7 was first used in a conference in
Darthmouth in 1956. During this conference – which lasted two months – the researchers were
supposed to determine and list all aspects of human intelligence so that later a machine could be
made to simulate it. They thought that, surely, if a computer was able to do math well beyond
any man’s capability, it could also master all other sorts of human tasks.
When it comes to writing news articles, AI has proven to be quite skilful at using templates to
produce articles that can easily pass for human-made. This is thanks to the most recent
advancement in the field, such as the new GPT and BERT, large-scale Transformer language
models that ‘learn’ language via good examples and practice. Much like their predecessors, they
too come with a series of limitations and are very much still in need of constant human editing.
One of the – if not the – obstacles for the newer models of AI is still what this paper calls
“[l]ong-form narrative”8, which, when attempted, showcases shocking amounts of “repetition,
grammatical errors, leaps or omission in turn-based dialogue, invented facts, illogical reasoning,
bias and racism, and factual errors stemming from a lack of real-world knowledge”9.
To better understand what the current AI’s limitations are, the authors of this study focused on
what they call “Middle Reading”10 (instead of Franco Moretti’s ‘distant reading’ which focuses
on large corpora), to better analyse a single novel or the corpus of a single author. From their
lab’s results, active search seems to be the major stumbling block, and long-form narrative in
general is the conundrum that AI cannot solve compared to other literary forms.
Chun and Elkins point out how 1969 is the year that Fletcher has his story of Digital Humanities
stop with distant reading, just two years after Marvin Minsky had proclaimed that “Within a
generation, I am convinced, few compartments of intellect will remain outside the machine’s
6
Chun and Elkins, What the Rise of AI Means for Narrative Studies: A Response to “Why Computers Will Never
Read (or Write) Literature” by Angus Fletcher, p. 105.
7
Dargham, Hachimi, and Boutalline, How AI is automating writing cit., p.1.
8
Chun and Elkins, What the Rise of AI Means for Narrative Studies cit., p. 107.
9
Ibidem.
10
Ivi, p. 108.
3
realm—the problems of creating ‘artificial intelligence’ will be substantially solved” 11.
Nevertheless, they agree with Fletcher that

[…] literary studies must and should engage with the emerging questions inherent to the rise
of AI. We also agree in spirit, if not in form, with Fletcher’s reservations regarding the
potential misuse, abuse, and exaggerated claims around AI, and especially with his
identification of the novel as posing a unique challenge.12

Indeed, while they think that AI’s development is more of an existential matter rather than a
technological one, they still, like Fletcher, have some concerns. In particular: if we manage to,
indeed, create machines capable of creating “language, narrative and art”13, what does this mean
for the humanity that we have always inextricably associated with them?
As the authors firmly believe this is a question of when more than if, they hope that experts in all
the fields interested will come together to shape the best possible future – having clearly in mind
both the dangers and the gains – for a good human-AI relationship.

1.4 Creativity and artificial intelligence

Since creativity is a human intelligence feature, it reads as a challenge for AI.


“But ‘novel’ has two importantly different senses here. The ability to produce novelties of the
former kind may be called P-creativity (P for psychological), the latter H-creativity (H for
historical). P-creativity is the more fundamental notion, of which H-creativity is a special case” 14.
Creativity can be of three types: combinational, exploratory, and transformational.
Combinational creativity combines familiar ideas in novel ways; exploratory creativity generates
new ideas by exploring structured conceptual spaces; transformational creativity alters the
conceptual space to generate previously impossible ideas. Examples from all of these three types
are included in computer models of creativity, in which the second type is the most successful
but also most complex to reproduce.
The main obstacle is that it is hard to replicate the depth and complexity of human memory
associations and also it is challenging to pinpoint our values and translate them into a format that
a computer can understand. The first obstacle is mimicking combinational creativity, a type of
creativity that involves combining existing ideas to form new ones. This is a subject of study in
AI, with research focusing on areas like humour and analogies. Both of these areas require a kind
of interconnected database or semantic network to function. This network serves as the
foundation upon which these creative combinations are built. Jape, a program for producing
punning riddles, is an example of AI-generated humour that has been relatively successful,
although it still falls short of human-originated jokes in terms of humour.
AI-models of analogy, such as the Copycat program, use domain-general mapping rules to
generate and evaluate analogies. The controversy around this approach relates to whether
conceptual representation should be seen as unchanging and inflexible, as some critics argue, or
as a domain-general process that can be analysed separately from conceptual representation, as
others contend.
Exploratory creativity in AI-systems, such as AARON, involves exploring pre-defined
conceptual spaces using pre-programmed heuristics and domain expertise. These programs are
capable of generating aesthetically pleasing and stylistically consistent results within their
designated domains, but they lack the ability to reflect on or transform their concepts. “AARON
11
Ivi, p. 109.
12
Ibidem.
13
Ivi, p. 111.
14
Boden, Creativity and artificial intelligence, p. 347.
4
cannot reflect on its own productions, nor adjust them so as to make them better. It cannot even
transform its conceptual space, leaving aside the question of whether this results in something
‘better’. In this, it resembles most current AI-programs focussed on creativity”15.
Transformational creativity in AI involves altering the generative mechanism to produce novel
structures. Genetic algorithms and evolutionary robotics are examples of such systems.
However, not all transformational systems result in creative outputs. The simulation of creativity
in AI, while promising, still faces challenges in evaluation and automation.
“A main reason why most current AI-models of creativity attempt only exploration, not
transformation, is that if the space is transformed then the resulting structures may not have any
interest or value. Such ideas are novel, certainly, but not creative”16.
In conclusion, transformational AI-originality is emerging, with domain-expertise and result
valuation as key bottlenecks. Subtle valuation, requiring domain expertise, is often implicit in
generative procedures or imposed by humans. Only a handful of AI systems are capable of
critically assessing their own unique concepts, and even fewer can integrate this assessment with
modification. The ultimate proof of AI’s creative process would be an algorithm that could
produce innovative ideas that initially confuse or even repel us, but could convince us of their
worth. However, we are still far from achieving this.

1.5 A Redhead Walks into a Bar: Experiences of Writing Fiction with Artificial Intelligence

The following paper deals with the human-AI co-writing process in light of two experiments the
authors did with a Generative Pre-trained Transformer 3 (GPT3) called Multiverse.
In particular, they analysed “(1) a co-writing activity initiated by basic textual prompts using
basic elements of narrative and (2) a co-writing activity initiated by more advanced textual
prompts, including dialects and metaphors”17. In both cases, the AI’s limitations became soon
apparent. However, this pushed them towards trying to find out the AI’s capabilities, to find
better inputs for it.
As noticed in many papers before this, Artificial Intelligence is now capable of great feats in
regard to emulating non-fictional human writings. But again, its inability to grasp things like
recurring characters or consistent story arcs has made it of little use for writing “[l]ong-form
narrative”18.
Throughout the years, using AI in creative processes has led people to think of it more as a
companion or collaborator rather than just a tool to support human creativity.
Just the same, when faced with the hard truth that AI is unable to understand semantics, it is easy
to see how almost all major characteristic of narrative discourse are incomprehensible to it:
protagonists and antagonists, settings, plots that escalate, themes, narrators and voices (i.e. the
way in which a story is told in a particular way).

Mysteries, for example, are often told almost in reverse order, that is, beginning with the
discovery of a body and then, seen through the eyes of an investigator, told backwards to the
scene of the crime, the killing itself, and only at the end revealing the beginning of it all:
why and how it happened (and by whom). The telling commonly features the alternation
between scenes (where events, such as dialogue, are related as they occur) and summaries
(where events over time are condensed and related all at once).19
15
Ivi, p. 353.
16
Ivi, p. 354.
17
Ghajargar, Bardzell, and Lagerkvist, A Redhead Walks into a Bar: Experiences of Writing Fiction with Artificial
Intelligence, p. 230.
18
Chun and Elkins, What the Rise of AI Means for Narrative Studies cit., p. 107.
19
Ghajargar, Bardzell, and Lagerkvist, A Redhead Walks into a Bar cit., p. 232.
5
Another hurdle for the AI is the lack of real-world knowledge. For philosophers Lamarque and
Olsen, storytelling can be summed up as “the act of producing a story and ‘a certain complex of
attitudes’”20 which are mastered at an early age for us humans and can be said to be innate.
Which is why the AI’s inability to grasp semantics makes it unable to also grasp the contents of
narrative discourse, the psychological faculties and communicative practices that constitute
fiction as a practice.
The tool used in these experiments is Multiverse, which proposes completions to the users in
response to prompts – which are necessary as this GPT-3 cannot create on its own.
Once the prompt is given, “[a]ll subsequent developments are generated by the AI, and the user
has no ability to edit, delete, or insert any text into the story”21.
Multiverse proposes three possible completions after the initial prompt for the human to choose
from, the user makes his/her choice, Multiverse takes it as another prompt and proposes again
three possible subsequent completions. This goes on until the human ends the session.
In their first study of this paper, the researchers tried to co-write with Multiverse nine stories
with basic textual prompts.
Their approach in this first experiment was more human-based: were the completions suggested
by Multiverse predictable for the human user? Did the completions help the user to be more
creative? Did they learn something about themselves? Did the stories have basic elements of
narrative such as conflict? Were there any literary elements missing, and if so why?
The results of the first experiment showed that AI had a penchant for becoming repetitive or
absurd, “predictability became un-generative”22.
AI lacks, as stated above, semantic understanding of the text – both the one it ‘reads’ and the one
it produces. How does this lack of understanding manifest itself in writing fiction? The first two
sentences produced by Multiverse would always be closely related to the initial prompt, but then
it would focus only on that: for example, if the prompt had a setting, it would focus only on that
and the plot, with no character depth or description. They also found out that AI is not able to
understand metaphors or expand on literary images (for example, one of the prompts had the
image of a growing flower and the AI could not follow on that).
What AI was able to do, though, was adding secondary characters and then expand on them, as
well as seamlessly switching between summaries and scenes as most human-made stories do.
Another AI ability is being able to make good enough prediction for the near future narrative-
wise, but this is, of course, a double-edged sword: “a certain amount of predictability is
desirable”23, but creativity needs also surprise to be engaging.
In their second experiment, they took into account the first study’s results and tried to co-write
another set of nine stories, this time with much more specific literary prompts (they included
dialects for characters, the use of metaphors, etc.). If the language is ambiguous, is the AI able to
understand it? If a story is based on a metaphor, is the AI able to notice it and develop it further?
In the midst of the second experiment they took “real-time decisions” 24 to ‘keep the AI alive’:
they had to adapt themselves to the AI’s limits, so much so that they ended up describing the co-
writing process during this second experiment like taking care of a Tamagotchi.

[…] writing fiction is a conventions-governed activity that happens in the social world, by
real actors, which includes above all authors and readers (but also critics, booksellers, etc.).

20
Ibidem.
21
Ibidem.
22
Ivi, p. 237.
23
Ibidem.
24
Ivi, p. 238.
6
The AI is not a part of, nor does it understand, this social world […] nor can it perform what
Coleridge famously called “the willing suspension of disbelief”.25

1.6 Creative Writing with an AI-Powered Writing Assistant: Perspectives from Professional
Writers

As professional writers may be better equipped than amateurs to assess AI-powered creative
writing technology, the authors of this paper commissioned “13 (1) published writers from (2)
diverse backgrounds to use an NLG-powered text editor over an (3) extended period of time,
collecting open-ended qualitative feedback along the way” 26 (the text editor in question was the
Wordcraft tool and the experiment was called the Wordcraft Writers Workshop).
The writers were asked to write a 1,000-1,500 word story using Wordcraft as much as they
wanted, however they were given no instructions on how the tool worked, to also test how
people with different levels of AI-powered tools knowledge responded to it. They were also
interviewed twice, before and after the experiment, while also having to keep a journal in which
they were asked to annotate what they thought relevant about their experience, keeping in mind
the following points:

• Jot down any generations or decisions Wordcraft makes which were especially interesting,
surprising, problematic, or otherwise noteworthy (even if they don’t end up in the final
story).
• What did you find challenging / frustrating about the tool? What does it utterly fail at?
• What sorts of things would be really helpful for the AI to do better, or be able to do at all?
• What did you find valuable / interesting about the tool?
• How did you imagine the underlying AI model working?
• How might you imagine AI-assisted writing changing your work habits?
• What features would you have liked to see exist in the tool?
• Does working with an AI assistant through built-in “controls” feel natural? Does working
with a chat-based assistant feel better?27

In the first interview, the participants were asked how they were hoping to use Wordcraft. Most
wanted a brainstorming partner, some wanted the tool to help them when they were stuck, some
hoped the tool would mimic a human beta reader. But, more importantly, they all stressed how
they did not want to delegate the creative process to the AI, but rather enhance their own.
Some went into the experiment hoping to create a story mostly written by Wordcraft, others only
wanted to use it sporadically.
Once the experiment was done, the final interviews revealed how most participants felt they had
to take on the role of editor when faced with the AI’s suggestions, that they had to lean into the
AI’s limitations and often even let go of their original plot in favour of the AI’s to try and avoid
its most absurd completions.
Another aspect noted by many, was the AI’s inability to replicate a certain type of ‘voice’:
different point of views could not be pulled off and the general narration always felt simple and
bland.
An additional fault found was that Wordcraft would suggest them many ‘bad’ ideas before
giving them a single ‘good’ one, thus making them feel as if they were wasting time.

25
Ibidem.
26
Ippolito et al., Creative Writing with an AI-Powered Writing Assistant: Perspectives from Professional Writers, p.
2.
27
Ivi, p. 5-6.
7
Furthermore, “[c]lichéd and biased suggestions seemed especially common when the system was
confronted with scenarios which were less likely to be well-represented in the model’s training
data”28, for example the AI would default to heteronormativity, prefer male characters over
female ones and would try to insert the name ‘Sarah’ in every single story as it was a name
contained in several of its database’s examples.
The chatbot feature of Wordcraft was, in the end, not as useful as many writers had hoped in the
beginning, and Wordcraft’s clear lack of understanding frustrated them.
Moreover, the longer the story, the worse Wordcraft would perform: this is because LaMDA has
a limit of 1024 tokens and when ‘it is full’, it starts deleting parts of the story it has already
produced.
Several participants were concerned with the source of the suggestions proposed by the AI as
they soon realised many of them were copied material, some from copyrighted sources.
“It was generally agreed that a co-writing AI, like a human co-writer, is most useful when it
complements a writer’s own skillset”29.
Some writers suggested a different ‘target audience’ for writing tools (i.e. inexperienced writers
or foreign language learners), while other writers were worried what a tool like Wordcraft could
teach a young writer (i.e. cheating or the case of AI generated books sold on Amazon).
In the end, all participants found Wordcraft to be unsuitable for their writing needs.

CHAPTER 2
DISCUSSION

2.1 AI’s limitations

Over time, the integration of artificial intelligence in creative processes has altered the idea of AI
to the point where it is more of a partner or collaborator than simply a facilitative tool for
enhancing human creativity. This evolution stems largely from the inherently dialogic nature of
collaborative creative activities, which involve an ongoing exchange between humans and AI
systems. Nevertheless, AI technologies are not without their drawbacks.

While reading the articles we found out that all the AIs have some sort of limitations.

 Lack of psychological and communicative practices: As stated in the previous


section, the craft of fiction writing operates within a structured framework of
conventions and rules within a broader social context, but AI exists outside this social
setting and consequently lacks a comprehensive understanding of it.
 No coherence: The AI's proficiency in fiction composition is hindered by its inability to
grasp the nuances of storytelling, causing its generated texts to start on track but
eventually veer off-course. It tends to adhere strictly to the surface elements, such as
settings and plot, while neglecting the rich characterization that breathes life into
narratives. Metaphors and literary imagery present formidable challenges; for instance,
when presented with the symbolism of a windy road, the AI falls short on elucidating its
deeper meanings. The winding road can represent the journey of life, with its twists and
turns symbolising the unpredictability and challenges one faces along the way.
Expanding on this, one could argue that the AI's struggle with fiction stems from its
inherent limitation in comprehending human emotions and experiences. Unlike human
writers who draw from personal insights and empathetic understanding, the AI lacks a
28
Ivi, p. 9.
29
Ivi, p. 13.
8
genuine connection to the human condition. Consequently, its narratives often lack the
emotional depth and psychological insight that characterise compelling storytelling.
Moreover, the AI's reliance on predefined patterns and statistical probabilities limits its
ability to produce truly original and thought-provoking narratives, relegating its outputs
to mere rewritings of familiar tropes and plot structures.
 The answers are predictive and very cliché: The presence of predictability in a system
aimed at nurturing creativity poses a paradox. While a certain level of predictability can
offer stability and a framework for innovation, true creativity flourishes amidst the
unexpected. Ambiguous prompts often result in AI generating nonsensical scenarios.
Moreover, when confronted with scenarios outside the scope of its training data, the
system tends to resort to clichéd and biased suggestions, undermining the potential for
genuine innovation and fresh perspectives. As an example, in crafting a story with a
lesbian romance plot, it consistently recommended adding a male character or diverting
the dialogue between female protagonists to discussions of friendship.
Furthermore, one might consider the implications of this paradox on the evolution of
artificial intelligence and its role in creative processes. The tension between
predictability and novelty highlights the ongoing challenge of balancing the need for
structured frameworks with the demand for innovation and originality. Additionally, this
raises questions about the extent to which AI can truly replicate the complexities of
human creativity, particularly in its ability to navigate ambiguity, embrace uncertainty,
and transcend conventional patterns of thought. As AI continues to advance, addressing
this paradox becomes increasingly crucial in unlocking its full potential as a tool for
fostering genuine creativity and pushing the boundaries of human imagination.
 Abrupt endings: In the realm of AI-authored novels, a recurring phenomenon persists:
the abrupt and sudden conclusion of narratives. This tendency stems from limitations
within current AI models. The algorithms powering these systems often prioritise
coherence and continuity based on patterns learned from training data. However, when
faced with divergent or ambiguous prompts, the AI may struggle to sustain logical plot
progressions, leading to premature endings or absurd scenarios without a chance for
recovery. Moreover, AI lacks a nuanced understanding of narrative structure and pacing,
resulting in a failure to build tension and resolve conflicts gradually. Instead, narratives
may plateau or abruptly deviate from the intended path, leaving readers with an
unsatisfying sense of incompleteness.
 Absence of creativity: Given its fundamentally human nature, creativity presents a
significant challenge for AI. An associated concept to this is human associative memory,
the cognitive capacity of individuals to establish connections and correlations between
diverse pieces of information or concepts within their minds. It is the mental process by
which we form associations between related ideas, memories, or experiences. These
connections can arise from similarities, differences, personal experiences, emotions, or
other relevant factors. For instance, when prompted with the word ‘dog’, associative
memory might conjure up related concepts such as ‘pet’, ‘barking’, ‘loyal’, or specific
memories of encounters with dogs. These associations are not random: they stem from
the way our brains organises and retains information, forming an intricate web of
interconnected ideas.
In the context of AI, replicating human associative memory involves developing
algorithms or models that can mimic this ability to form meaningful connections
between different pieces of information or concepts. This is crucial for tasks like natural
language understanding, creative problem-solving, and generating coherent responses in
conversation.

9
However, AIs cannot reflect on or refine their concepts and lack the ability to evaluate or
modify their own creations, nor can they alter their conceptual framework, leaving the
notion of improvement unexplored.
 The humour is boring and not human like: AI often falls short on reproducing
humour due to its difficulty in comprehending the nuances of language, cultural context,
and timing necessary for effective comedic delivery. Humour relies heavily on
creativity, spontaneity, and understanding of social dynamics, aspects that AI struggles
to replicate authentically. JAPE, a program designed to create pun-based riddles, serves
as an instance of AI-generated humour that has achieved moderate success, but it still
remains inferior to human made jokes.
 Presence of different voices: A predominant constraint identified by participants in AI
experimentations was the incapacity to produce text in the preferred stylistic or tonal
inclinations of the author. Particularly evident was the challenge encountered when
attempting to compose narratives featuring alternating perspectives, as the AI exhibited
difficulty in sustaining distinct voices for each viewpoint. Instead, there appeared to be a
prevailing ‘default’ voice characterized by a mundane linguistic expression within the
generated content of the language model. These difficulties underscore the need for
further refinement in AI’s narrative generation capabilities to ensure seamless
storytelling.
 Waste of time: Employing AI to write a novel can be perceived as a futile endeavour,
as it often consumes more time than it saves. Despite its presumed efficiency, the
process of navigating through numerous AI-generated suggestions and attempting to
discern the most suitable ones leads to prolonged moments of deliberation and
frustration. This effort to identify the right answers and construct a cohesive narrative
ultimately outweighs the time saved by utilising AI. Rather than streamlining the writing
process, reliance on AI for novel creation introduces a significant overhead in terms of
time and effort, detracting from the creative flow and overall productivity of the author.
 Copyright: Many have voiced concerns regarding the sources and potential biases
embedded within the AI-generated material. This scrutiny highlights broader ethical
dilemmas surrounding the reliance on AI in creative endeavours and underscores the
importance of transparent and accountable practices in AI development and
implementation.
 Problems with the narrative: In the realm of AI-generated novels, a significant
limitation often encountered is the restrictive token limit imposed by the AI’s text
generation capabilities. This token limit delineates the maximum length of text the AI
can produce within a given instance. Consequently, when the AI reaches the token limit
while crafting a narrative, it confronts the dilemma of accommodating new content
within the constrained space. As a result, the AI resorts to deleting portions of the story
to make room for additional text, thereby sacrificing coherence and continuity in the
narrative. This process of elimination poses a substantial challenge, as it compromises
the integrity of the storyline and undermines the overall quality of the generated novel.

2.2 Characteristics of crime books written by humans

Among all the novels, crime books are probably the most popular novels on the market. This
literary genre, which is no longer considered sub-literature, has changed so much that there are
now crime books entirely written by AI. With the help of two case studies (Death of an Author
and The Vanishing) we will analyse the main points of divergence and convergence between a
traditional crime novel written by a human, and a crime novel written by AI.

10
While there are many sub genres in the crime genus (cosy mysteries, detective fiction, hard-
boiled mysteries, and locked-room mysteries), all crime novels have some important features and
characteristics in common. The key characteristics of a crime book include:

1. Realism and believability: Crime fiction should be realistic and believable. The concept of
realism and believability in crime fiction is a complex and evolving one. Authors need to
balance realism with the need to create a believable world that aligns with the genre and its
rules. The portrayal of characters, settings, and plot elements should be consistent and
plausible to maintain credibility. Incorporating factual elements, accurate details, and
historical facts can enhance the believability of a story. However, authors can also play with
a mix of fact and fiction, as long as the overall narrative maintains a sense of verisimilitude.
Ultimately, achieving a balance between realism and believability is key in crafting
compelling crime fiction that resonates with readers.

2. A significant crime: The crime is a crucial element that drives the story forward. It is
usually a substantial crime that readers can relate to and care about, such as murder,
kidnapping, or theft of valuable objects.

3. The criminal: The criminal is an essential character in the story. It must be well-matched
against the protagonist, the detective or investigator, and should be intelligent, cunning, and
able to evade detection.

4. The investigator: The investigator, often a detective or amateur sleuth, is usually the
protagonist of the story. They are usually intelligent and determined to solve the crime. The
investigator's personality, skills, and motivations are important aspects of the story.

5. Suspense and intrigue: Suspense and intrigue are crucial elements in crime fiction. The
story should keep the readers guessing by introducing multiple suspects, red herrings, and
clues that lead to the solution of the crime.

6. A list of potential suspects: A list of potential suspects is a common feature in crime


fiction. This helps to maintain suspense and keeps the readers engaged in the investigation.

7. A theme of criminality: Crime fiction often explores themes related to criminality, such as
the psychology of criminals, the impact of crime on society, and the morality of the criminal
justice system.

8. Setting and atmosphere: The setting and atmosphere of the story can significantly enhance
the sense of tension and suspense. This is particularly true in thriller/crime fiction, where the
setting can become a character itself.

9. The use of clues and twists: Crime fiction often employs clues and twists to keep the
readers engaged and guessing. The use of these elements should be balanced to ensure that
the reader is not misled or bored.
These are the main characteristics found in all crime books.

However, in our case studies, we noticed that not all the aforementioned characteristics are
present with the same mastery in the books written by AI.

11
CHAPTER 3
CASE STUDIES

3.1 Death of an Author: A Novella

Death of an Author: A Novella is a murder mystery novel by Aidan Marchine, pseudonym for
Stephen Marche (novelist and journalist) and his three unusual collaborators: ChatGPT,
Sudowrite and Cohere.
It delves into the intricate layers of its narrative, each chapter adding depth to the overarching
mystery while exploring themes of identity, creativity, and the intersection of technology and
literature.
The novella opens with the introduction of Gus Dupin, a professor seeking refuge from personal
turmoil at Stony Lake. His secluded retreat is disrupted by an unexpected invitation to the
funeral of Peggy Firmin, a revered author whose work holds a profound significance for him.
Chapter One immerses readers into Gus’s world, introducing him as a contemplative figure
grappling with the aftermath of his recent divorce and the pressures of academia. The news of
Peggy Firmin’s violent death sends shock waves through Gus’s quiet existence, stirring up
memories of her influential writings and the profound impact they have had on his life. Through
Gus’s introspection, readers gain insight into his deep admiration for Peggy’s work and the sense
of connection he feels to her despite never having met her in person.
Chapter Two deepens the intrigue surrounding Peggy’s death, as Gus navigates the enigmatic
funeral and encounters her daughter, Aubrey, for the first time. The holographic message Peggy
recorded before her death adds a surreal layer to the proceedings, revealing hints of her complex
relationships and the innovative projects she was involved in. Gus’s interactions with Aubrey
hint at deeper connections between him and Peggy’s legacy, setting the stage for the unfolding
drama.
In Chapter Three, Gus’s peaceful existence is shattered by a visit from Officer Coventry, who
questions him about his whereabouts on the night of Peggy’s murder. Suspicion mounts as Gus
finds himself thrust into the spotlight of the investigation, his alibi called into question by the
discovery of an entry in Peggy’s Google Calendar linking him to the scene of the crime.
Meanwhile, flashbacks to Peggy’s collaboration with Neil Gibson shed light on their ambitious
projects and the tensions simmering beneath the surface.
Chapter Four sees Gus embarking on a journey to the Leslie Street Spit bridge, where Peggy was
murdered, in a quest for answers. Amidst a makeshift shrine of mourners paying tribute to
Peggy’s memory, Gus faces the weight of suspicion and the surreal nature of mourning someone
he never knew personally. His encounter with a journalist further complicates matters, as
questions about his connection to Neil Gibson and the Sibyl Project resurface, adding layers to
the unfolding mystery.
In Chapter Five, Gus’s investigation takes a surprising turn as he discovers email correspondence
between himself and Peggy, revealing a mysterious impostor’s attempts to impersonate him. The
revelation deepens Gus’s sense of unease and raises questions about the true nature of his
relationship with Peggy. As he delves deeper into the emails, Gus is confronted with missed
opportunities and lingering regrets, adding a poignant layer to his character development.
Chapter Six finds Gus retreating to his secluded haven by the lake, seeking solace amidst the
chaos surrounding Peggy’s death. Despite his attempts to disconnect from the turmoil, Gus finds
himself drawn back into the intrigue when he learns of Neil Gibson’s arrest and the downgrading
of Peggy’s death from homicide to ‘death from misadventure’. The chapter ends with Gus facing
yet another unsettling revelation, leaving him with more questions than answers.
Chapter Seven shifts to focus on Frank Bigelow, who finds himself entangled in a new mystery
surrounding the murder of Grace Dunbar, another renowned author. As Frank and his wife Paula

12
attend Grace’s funeral, they find themselves immersed in a world of suspicion and intrigue, with
hints of a larger conspiracy lurking beneath the surface.
Chapter Eight sees Gus embarking on a daring mission to uncover the truth behind Peggy’s
murder, despite the risks involved. His confrontation with Neil Gibson in the Al-maginarium
leads to a shocking revelation that further complicates the case, leaving Gus grappling with
conflicting emotions and unanswered questions.
In Chapter Nine, Gus’s encounter with Peggy’s avatar sheds light on Neil Gibson’s plans to
create an artificial version of Peggy, named Sibyl, blurring the lines between reality and fiction.
Themes of identity and the nature of consciousness come to the forefront as Gus confronts the
implications of Peggy’s actions and the limitations of artificial intelligence.
Chapter Ten sees Gus engaging in a philosophical dialogue with Peggy’s avatar, exploring the
complexities of storytelling and the desire for immortality through one’s work. As Peggy reflects
on the dual nature of code and the role of criticism in the creative process, Gus wrestles with
existential questions about the nature of art and the human condition.
In Chapter Eleven, Gus reflects on the events that have transpired, finding solace in moments of
solitude amidst the chaos. As he contemplates the future of literature in an age of technological
advancement, Gus comes to terms with the complexities of creativity and the enduring power of
storytelling.
Chapter Twelve concludes with an afterword by Stephen Marche, reflecting on the broader
implications of creative AI and the collaborative process behind its creation. Through Marche’s
insights, readers are invited to ponder the future of literature and the evolving role of technology
in shaping the creative landscape.

Overall, Death of an Author: A Novella explores the complexities of identity, creativity, and the
human condition in an age of technological advancement. Through its intricate narrative and
thought-provoking themes, the novella invites readers on a journey of discovery and self-
reflection, challenging them to ponder the mysteries of authorship and the boundaries between
fiction and reality.

Although the story seems logical from beginning to end, there are many holes in the cohesion of
the narrative. The plot moves from one story to another and new characters appear out of
nowhere. The story involves multiple layers of mystery, intrigue, and interconnected storylines,
thus it can be challenging to maintain a smooth and coherent narrative flow without
overwhelming the reader. Besides, the plot twists and turns are introduced awkwardly, making it
difficult to follow the course of events, which can be uncomfortable to read.

An article in The New York Times titled A Human Wrote This Book Review. A.I. Wrote the
Book, declares this novella to be “arguably the first halfway readable A.I. novel”30.
The book critic Dwight Garner describes how Marche’s characters all seems to hint at well-
known figures, because of either their names or their attitudes – i.e. Peggy Firmin as Margaret
Atwood. He is also of the idea that the chilling eulogy scene – which leaves readers wondering
how much exactly Peggy knew about what was coming for her – is a tribute to Agatha Christie’s
iconic genius.
However, the reviewer does not mince words when he expresses his opinion on the language of
the book (which Marche says is “95 percent machine-generated” 31): it feels bland, like that “of a
Wikipedia entry”32, leaving him “feeling hollow, as if I’d made a meal out of red herrings” 33 and,
worst of all, uncaring about finding out who is the culprit.
30
Garner, A Human Wrote This Book Review. A.I. Wrote the Book.
31
Ibidem.
32
Ibidem.
33
Ibidem.
13
Garner unequivocally express his stance on AI when he writes that they will never really beat
“the real thing”34: a soul.
To conclude, he recalls a J.M. Coetzee line he uses to soothe himself when he finds himself
pondering the possible alarming future that awaits us all if we keep letting AI advance unbridled:
“When an American writer does not know how to end a story, he shoots everyone in sight”35.

3.2 The Vanishing “A AI generated book”

This book is a product of OpenAI, an organization dedicated to the creation of Artificial


Intelligence to benefit humanity and to enhance the quality of life. The book was crafted using
the language generation capabilities of the GPT models. Specific prompts and inputs were
provided to generate a narrative filled with suspense.
The structure of the book consists of ten concise chapters, including approximately seven lines
each. Additionally, there is an epilogue that extends slightly longer than the preceding chapters.
Several years after the death of her parents, Samantha returns home. Upon her arrival, she is
immediately overcome by a sense of unease. The following morning, while dining at a local
establishment, she recognizes a familiar face on a poster affixed to the wall. The poster
announces the disappearance of a young girl named Emily. As a result, Samantha decides to
approach the police to ask about the case.
With the assistance of Detective Tom, Samantha starts questioning Emily’s acquaintances. She
learns that Emily used to frequent a local bar, The Red Door, under the belief that she was being
pursued. Samantha visits the bar and, after a conversation with the bartender, discovers that
Emily has left the bar on the night of her disappearance accompanied by a man clad in a black
hoodie.
Following this lead, and with the aid of addresses provided by the bartender, Samantha
encounters the figure in a black hoodie standing in front of a building. Surprisingly, the figure is
Emily. She is frightened and disoriented, having been drugged and held captive within the
apartment building. Samantha contacts Detective Tom, and they uncover evidence of human
trafficking, leading to the arrest of several individuals.
Samantha emerges as a hero, but she suspects there is more to the story. She continues her
investigation and discovers the involvement of influential individuals, placing her in danger.
After receiving threats and being followed, she collaborates with the detective, Emily’s brother,
and a Navy Seal to stop the trafficking operation.
The plan is successful, and all the involved parties are arrested. Samantha no longer feels
frightened or uneasy, instead, she feels a sense of pride for her role in saving numerous lives.
She receives multiple accolades and recognitions, and decides to establish an organization to
raise awareness about human trafficking. She embarks on a journey across the country,
delivering speeches and inspiring individuals to take action against social injustices.
Two quotes from the book can help us clearly see what is off about it.
“Samantha decided to follow up on the lead. She went to the addresses the bartender had given
her, a run-down apartment building on the edge of town”36.
It cannot be considered normal for a bartender to simply have ready a bunch of addresses to look
into, and the result is a little bizarre.

“As she walked up to the building, she saw a figure in a black hoodie standing outside. She
approached him cautiously, and he turned to face her. It was Emily”37.
34
Ibidem.
35
Ibidem.
36
Bera, The Vanishing “A AI generated book”, p. 2.
37
Ibidem.
14
This is the highest moment of the book and it should have been full of suspense and trepidation,
instead the shortness of the chapter obliterates any kind of climax.

As we can see, the subject of the book is not only the investigation, but also the exhortation to
help people and to fight against social injustices, which is also the aim of the Open AI
organization. The primary issue with this narrative lies in its brevity. Its short length impedes the
creation of tension and suspense. However, the plot remains coherent thanks to the concision of
the chapters. This brevity leaves little room for the expansion of the storyline and character
development and depth, thereby risking a potential loss of coherence.
The last observation that is immediately apparent by reading the book is the complete absence of
dialogues.

CONCLUSIONS

Can AI write crime books?


In light of these two works analysed, the conclusion is that AI is capable of producing thrillers,
but there is still a huge gap between an AI-mystery and a human-made one, especially in terms
of creativity and engagement. AI lacks the human insights, personal experiences, and cultural
nuances that human authors infuse into their writing to evoke emotions effectively. The
emotional depth and complexity required in crime books to truly hook readers in are difficult for
machines to replicate.
Despite the AI’s progress in generating content, the intricate interplay of emotions, storytelling,
and human experiences in crime literature remains a domain where only human writers excel.
The current limitations of AI in creating emotionally resonant crime narratives highlight the
unique creative and emotional touch that only human authors bring to their work, suggesting that
while AI can assist in content creation, it cannot fully replace the emotional depth and creativity
of human-authored crime books.

15
BIBLIOGRAPHY

Bera, Sameer; ChatGPT. The Vanishing “A AI generated book”. 2023. Kindle Edition.

Marche, Stephen. Afterword to Marchine, Aidan. Death of an Author: A Novella. New York:
Pushkin Industries, 2023. <https://it.everand.com/read/639832639/Death-of-an-Author-A-
Novella>.

Marchine, Aidan. Death of an Author: A Novella. Edited by Stephen Marche. New York:
Pushkin Industries, 2023. <https://it.everand.com/read/639832639/Death-of-an-Author-A-
Novella>.

Minsky, Marvin Lee. Computation: Finite and Infinite Machines. Englewood Cliff: Prentice-
Hall,
1967, quoted in Chun and Elkins, What the Rise of AI Means for Narrative Studies, 109.

SITOGRAPHY

Boden, Margaret Ann. Creativity and artificial intelligence. «Artificial Intelligence» 103, no. 1-2
(1998): 347-356. <https://tinyurl.com/bxhjxy6u>, DOI: https://doi.org/10.1016/S0004-
3702(98)00055-1. (Last consulted: 05/04/2024).

Chun, John; Elkins, Katherine. What the Rise of AI Means for Narrative Studies: A Response to
“Why Computers Will Never Read (or Write) Literature” by Angus Fletcher. «Narrative» 30, no.
1 (2022): 104-113. <https://muse.jhu.edu/pub/30/article/846035/pdf>, DOI:
https://doi.org/10.1353/nar.2022.0005. (Last consulted: 05/04/2024).

Dargham, Malak; Hachimi, Hanaa; Boutalline, Mohammed. How AI is automating writing: The
rise of robot writers. In 2022 8th International Conference on Optimization and Applications
(ICOA), Genoa, 06-07 October 2022: 1-5. <https://tinyurl.com/adr9a6k8>, DOI:
10.1109/ICOA55659.2022.9934723. (Last consulted: 05/04/2024).

Fletcher, Angus. Why Computers Will Never Read (or Write) Literature: A Logical Proof and a
Narrative. «Narrative» 29, no. 1 (2021): 1-28. <https://muse.jhu.edu/article/778252/pdf>, DOI:
https://doi.org/10.1353/nar.2021.0000. (Last consulted: 05/04/2024).

Garner, Dwight. A Human Wrote This Book Review. A.I. Wrote the Book.. «The New York
Times», 10 May 2023, <https://tinyurl.com/4h4wx2nm>. (Last consulted: 05/04/2024).

Ghajargar, Maliheh; Bardzell, Jeffrey; Lagerkvist, Love. A Redhead Walks into a Bar:
Experiences of Writing Fiction with Artificial Intelligence. In Proceedings of the 25th
International Academic Mindtrek Conference (Academic Mindtrek '22), New York, 16-18
November 2022: 230–241.
16 November 2022, <https://dl.acm.org/doi/pdf/10.1145/3569219.3569418>, DOI:
https://doi.org/10.1145/3569219.3569418. (Last consulted: 05/04/2024).

Ippolito, Daphne; Yuan, Ann; Coenen, Andy; Burnam, Sehmon. Creative Writing with an AI-
Powered Writing Assistant: Perspectives from Professional Writers. 09 November 2022, arXiv.
<https://arxiv.org/pdf/2211.05030>. (Last consulted: 05/04/2024).

16
Podoletz, Lena. We have to talk about emotional AI and crime. «AI & Society» 38, no. 3 (2023):
1067-1082. <https://link.springer.com/article/10.1007/s00146-022-01435-w>, DOI:
https://doi.org/10.48550/arXiv.2211.05030. (Last consulted: 05/04/2024).

17

You might also like