Abstract

What reasons might we have to continue to engage with human-authored fictions once algorithmically generated fiction becomes indistinguishable from the best human works of fiction? I argue that there is a social value to our engagement with fiction that algorithmic fiction will inevitably lack. Thus, even as algorithmic fiction continues to develop and improve, there will be good reasons to preserve our existing practices of producing, financing, and engaging with human-authored fictions.

1. Introduction

Generative AI is poised to disrupt the artworld as we know it. Soon, anyone with a computer and an internet connection will be able to, with a few keystrokes, use generative large language models (LLMs) like ChatGPT to write novels,1 poetry,2 and screenplays3 or prompt a deep learning algorithm like DALL-E and Midjourney to create paintings,4 photorealistic images,5 music videos,6 or even entire movies.7

But in such a world, what will become of human-generated art and the traditions built up to facilitate the training, skill, time, and effort to produce it? Although current AI-generated art still bears telltale signs of mechanization,8 this question has prompted handwringing,9 soul-searching defensiveness,10 and protest11 among those working in such creative industries. And for good reason, since, unlike (for instance) the “threat” posed to painting by photography,12 generative AI raises the possibility of a future in which little or no human input is needed to create works of art that, at least by appearances, rival any we’ve ever seen.

‘How bad would this be?’ So says the pragmatic futurist. ‘If computers can create art, fiction, and music that is as compelling and looks and sounds as good as any that could be made by humans, all at a fraction of the cost and effort, so much the worse for us and our creative endeavors. Just as the rise of personal computers obviated the need for dictationists, perhaps so too would some future LLM obviate the need for authors and artists.’

My aim in this paper is to push back against the pragmatic futurist, focusing on algorithmically generated fiction in particular. Generative AI will likely be the source of many objects of great value, inspiring fascination and curiosity. However, in this essay, I will identify a dimension of value to human-authored fiction that algorithmically generated fiction will inevitably lack, no matter how good it otherwise gets. Given this, something of real value would be lost were algorithmic fiction to completely overtake the artworld, displacing human authors entirely. Even as algorithmic fiction continues to develop and improve, there will remain good reasons to preserve our existing practices of producing, financing, and engaging with human-authored fictions. Fiction, and art in general, is not something that can be automated without serious costs.

2. Distinctions and Definitions

My focus is on fiction, and for the purposes of this paper, I’ll assume the standard view that fictions are texts (which include strings of words, either printed or spoken, and also performances, such as plays or films) that tell a story and whose function is to prompt imaginings (Currie 1990; Walton 1990). In other words, fictions are texts that are to be engaged with by imagining that things are (roughly) the way they are as the text describes or represents them. I will be talking about texts produced with the aid of algorithmic generative processes like ChatGPT and others, often in response to a seed or prompt. I distinguish four degrees of algorithmic integration in the creation of fiction:

• Zero degree (human-generated fiction): Every part of the work is produced by a human being, perhaps with the assistance of various mechanical intermediaries, including the use of a computer or typewriter rather than pen and paper. This includes mechanical aids like backlighting, spell-check, grammar-check, dictionary/thesaurus, and the internet (as used for the purposes of research).

• First degree (algorithmically inspired fiction): Every part of the work is produced by a human being, although some parts are directly inspired by the output of an algorithmic generative process, such as using story ideas generated with ChatGPT.

• Second degree (algorithmically integrated fiction): This includes works that are partially algorithmically generated. Distinguish two subvariants:

  • Supplementative: Some parts of the work are algorithmically generated and other parts are human-authored.13

  • Comprehensive: Every part of the work is algorithmically generated but then subject to human editing and rearrangement.

• Third degree (algorithmically generated fiction): The entire work is algorithmically generated without being subject to human editing or input (beyond prompting).

Distinguishing degrees of algorithmic integration into the creative process is important, for the first and second degrees of integration are merely a reshaping of the tools with which human creativity can be exercised. In this way, the first and second degrees are analogous to how film cameras allowed humans to more easily create realistic representations of visual scenes, which led to the development of photography as an independent art form.14 Therefore, my focus will be on the third degree—what I call ‘algorithmically generated fiction’ (or algorithmic fiction, for short). This is the point at which no act of human creativity need be involved in the creation of fiction (other than whatever creativity is involved in choosing which prompt to run the algorithm on, and even such choices may eventually be unnecessary) and, thus, at which AI becomes not merely a new tool to be used in the process of writing fiction but rather a potential replacement for human authors entirely.

Today, algorithmic fictions are cheap imitations at worst15 and strange curiosities at best.16 However, plausibly, it will soon be easy to produce algorithmic fictions indiscernible from human-authored fictions. I assume that many such algorithmic fictions will exhibit characteristics of excellence in fiction, containing well-developed characters, moving, suspenseful, or dramatic stories, and even offering profound insights into the human condition. Call such fictions really good algorithmic fictions. My question today is how we should think about the value of really good algorithmic fictions, especially in relation to the best human-authored fictions.

3. Why Fiction Matters: The Individual Dimension

Fiction matters for many reasons. It’s fun to engage with, which is part of why we spend large portions of our free time reading novels, watching television shows, and going to the theater or cinema. And engaging with fiction often has many other desirable outcomes—helping to garner empathy and teach valuable moral lessons. Let’s call this dimension of the value of a work of fiction its individual engagement value.17

Could a really good algorithmic fiction ever attain similar levels of individual engagement value as some human-authored fiction? Consider an alternative possible history of the world in which Huckleberry Finn is not written by Mark Twain in 1884, but in which an algorithm in 2024 happens to produce a text that’s identical to Huckleberry Finn. To avoid confusion, call the original Huckleberry Finn ‘H1’ and the algorithmically generated fiction (whose name is also Huckleberry Finn) ‘H2.’ Our question is whether H2 would have a similar individual engagement value for its readers in that alternative history as H1 does for its readers in our world. And plausibly, it would. At the very least, it’s plausible that reading H2 would yield similar pleasures as reading H1, whatever moral lessons are imparted by H1 would also be imparted by H2, and so on.

Of course, there are also many relevant differences between the two texts. H1 has a certain historical significance, our knowledge of which might affect our enjoyment reading it today, and this would be lacking for H2. And knowing that H1 was written and labored over by Mark Twain, and bears certain connections to his broader body of work, might also affect our experience reading it (and the lessons we take from it) in ways that would not be shared by our counterparts reading H2. But it’s not obvious that these differences would always lead to a diminishment of a work’s individual engagement value. It might be that some algorithmic fiction may be more pleasurable to engage with, and more morally insightful, than any human-authored fiction precisely because algorithms can generate texts that no human would otherwise think to. It seems rash to rule out such a possibility at this stage, and thus, I think we should be open to the possibility of algorithmic fictions that are as enjoyable and morally impactful as any human-authored text. When it comes to individual engagement value, algorithmic fiction seems likely to end up doing quite well.

However, individual engagement value isn’t the only reason why fiction matters. Engaging with fiction is not just a matter of passive enjoyment. We also evaluate, discuss, and think about fiction, exercising our own sensibilities in doing so and sharing our thoughts and feelings with others. As such, fiction can also connect us as human beings, making us into more than simply individuals separately pursuing our own pleasures. There is something valuable about finding such connections with one another. But what exactly does such value amount to? In the next section, I offer an answer to this question. Afterwards, we’ll turn to consider whether algorithmic fiction can be a source of such social value.

4. Why Fiction Matters: The Social Dimension

To better characterize the social value of fiction, consider two characters: Max and May. Max is a consummate follower:18 He defers whenever possible on matters of taste—on questions such as whether a novel is compelling or a movie scary or a character sympathetic, he aims to bring his judgments into concordance with the judgments of his friends. It’s not that he merely defers without thinking about things himself; rather, he listens carefully to the reasons for the judgments articulated by his friends and then does his best to see things their way. For example, after watching The Shining, he discusses it with his friend Jones, who remarks that the film is not very scary because it doesn’t contain much in the way of overt violence or terror. Max is initially confused by this reaction—to him, the scenes of Jack Torrance chasing his family with an axe are deeply upsetting. But Jones observes that Shelley Duvall’s performance in those scenes isn’t very compelling, and this detracts from being fully immersed in the excitement. After considering this perspective, Max finds himself in agreement—he overrides his initial assessment and comes to judge that the film isn’t scary.

Our other character, May, is a consummate gratifier: She is a filmmaker who makes movies aiming only to satisfy the desires of her audience. Often, in the process of making a movie, May will face a choice about how to portray a certain character or event. She has her own view about the character: Perhaps she finds them obnoxious and uninteresting. But she also believes that her intended audience will find the character heroic and inspiring, so she chooses to portray them as such (shooting the character triumphantly and guiding the actor’s performance of them accordingly) to please her audience. Moreover, all her creative filmmaking choices are made with such anticipations about audience reactions in mind.19

I hope you’ll agree that something has gone wrong with Max and May. It’s not that Max and May are bad people for doing what they do but rather that each is falling short of an ideal; they could be doing better. Max could be exercising his own judgments in evaluating fictions he engages with. And May could be making films that express her own distinctive individuality and sensibility. Each could stand to be more authentic in their aesthetic lives.

Why does such authenticity matter, though? It’s not that Max or May is lying or deceiving others through their actions. Alexander Nehemas, writing about the nature of beauty and judgments of beauty, provides a crucial insight. Immanuel Kant thought that judgments of beauty demand that others judge similarly and thus aim at universal acceptance, which marks them as distinct from preferences, which carry no such aim.20 But Nehamas observes that a world in which everyone has the same judgments of beauty would be “a desolate, desperate, world” because it would mean the obliteration of individuality (Nehamas 2007, 83). By individuality, Nehamas means something more than the mere distinctness secured by nonidentity—rather, one’s individuality is something that must be constructed and maintained through an exercise of agency. Beliefs about matters of fact, being constrained by evidence and, thus, relevantly not up to me, are thus not (usually) aspects of my individuality in this sense. Being an individual in Nehamas’s sense means exercising your agency to judge about matters of discretion (things that are ‘up to you,’ such as matters of taste).

Nehamas’s depiction of the ‘Kantian nightmare’ of universal convergence on matters of taste dramatizes the point that there is something valuable about ‘constructing’ your individuality through your judgments of taste. And this value is reflected in what we care about too. We value being ‘unique,’ ‘authentic,’ and ‘true to oneself,’ and this motivates us to look for ever more obscure works to love or to pursue unique ways of appreciating mainstream works.21 We recognize that by being individuals in this sense, we make the world a richer, more interesting place. And this helps us see what’s gone wrong with Max, who refuses to exercise his agency and stand by his own judgments of taste, and May, who defers her creative decisions to the judgments of her audience. When it comes to aesthetics, conformity is boring.

But iconoclasm is only one side of the picture. Individuality is good, but individuality pursued solipsistically—that is, without connection to other human beings—is lacking. Being in community with others is valuable too—in particular, being in community with others who recognize each other’s distinctive individuality, who see each other as the individuals they are.22 This is reflected again by what we value: We desire not just to be unique, but also to express ourselves and to be heard—that is, for others to join us in appreciating the world as we do. Such shared appreciation can be a source of intimacy and a feeling of togetherness (Cohen 1999). Moreover, it is often through communal interaction with others that our individualities are formed and reformed—just consider the phenomenon whereby by expressing your judgments to another makes them feel more settled and part of who you are (Nehamas 2010, 290).23

Thus, we recognize two, sometimes conflicting, desires: a desire to be unique and a desire to be seen (and appreciated) by others for our uniqueness. Fiction, like all art, is created via an exercise of discretionary judgment and, thus, an opportunity to publicly develop your individuality and style.24 Authors have near total control over the fictions they create, and thus, fictions are the product of countless creative decisions: choosing to tell one story over another, to tell it this way rather than that way, focusing on these elements and characters, who are characterized in this way, and so on. When such choices are the result of an author’s own judgment about which narrative is worthwhile to attend to, the author expresses this judgment by inviting others to attend to it as well.25 Such an invitation presents an opportunity for others to join in community with the author, exploring her perspective, judging it, and relating their own individualities to hers.

Note that, when I talk about judging and engaging with another’s individuality, I mean this both from an appreciative and critical perspective. Encountering another’s judgments, many of which will inevitably differ from my own, is an opportunity to refine and re-evaluate my own judgments, see them in a new light, and recognize the differences in outlook that make our world an interesting place. When we disagree, I learn something about you, that you see things differently from me, and, if I’m curious and open-minded, this can lead us to explore these differences together. Why is it that we have different perspectives? Are there ways in which your perspective is more interesting or satisfying than mine? Are there ways we can unify our perspectives and come to see things alike? By considering such questions, we cultivate what Nick Riggle calls volitional openness—the disposition “to set aside the typical modes of action encouraged by our cherished values and moral convictions and act from a generous, curious, exploratory, or spontaneous place” (Riggle 2024, 21).26 Expressing our judgments and finding where they diverge from others’ is a way of forging bonds of mutual understanding and appreciation out of the differences that make us individuals.

When made via an exercise of an author’s discretionary judgments, which are appreciated by readers as such, fiction puts author and reader, and reader and reader, into a community of individuals that recognize and appreciate each other’s unique perspectives and sensibilities. Thus, the social value of fiction arises from our creating and engaging with it in the right kinds of ways. Fiction made without the exercise of one’s own judgments (such as May’s, whose fiction is created only by deferring to the judgments of others) lacks such social value, since it presents no opportunities to engage with its author’s individual sensibilities. Similarly, when we do not engage with a work of fiction by exercising our own capacities for judgment, we are not appreciating its author’s sensibility (as with Max, who defers his evaluations to others), nor are we relating our own sensibilities to theirs and others’.

5. Fiction in the Absence of Judgment

I have assumed that the algorithms generating really good artificial fictions will continue to operate largely in the way current machine learning algorithms operate—basically, by drawing on statistical regularities from large datasets to randomly generate outputs (text, images, sounds) in response to prompts. But stochastically generating text using statistical regularities extracted from some dataset is not the same thing as making a creative judgment. Whatever creative judgments are, they issue from an individual who has a particular perspective on the world, one formed through an exercise of agency rather than merely randomly remixing the information they have gained throughout their lives. Artificial fictions are created more in line with how May created her film: by recognizing what the audience (prompter) wants and giving it to them as best as they can.

My argument that there would still be reasons to engage with human-authored fictions in the face of really good algorithmic fictions should now be easy to spot. There is a social value to appreciating fiction created through an exercise of creative judgment; when a reader engages with an author’s work, she is forced to come to terms with their perspective on the world (at least as it is expressed through the choices made in creating that work). Whatever her own judgment on that perspective, the reader is put face-to-face with the author’s sensibilities, positioned to recognize the author’s individuality and, as a result, form a very intimate bond with them, sometimes provisionally taking on board and other times merely reflecting upon aspects of what makes them a unique individual.

Algorithmic fictions, unlike human-authored fictions, are not products of creative judgment.27 And, as such, they lack the social value human-authored fictions can have. Engaging with an algorithmic fiction does not put us in contact with an individual’s unique perspective on the world. Algorithmic fiction can satisfy our desires for an enjoyable experience or provide an insightful moral lesson, but it can’t connect us as human beings.

Of course, we can use algorithmic fiction to forge connections across our distinctive individualities, appreciating differences in our judgments about which algorithmic fictions are enjoyable, interesting, suspenseful, and so on. Algorithmic fiction can, in this way, be like natural objects of beauty like sunsets, gemstones, or crystalline structures. Such naturally occurring things can be delightful, beautiful, sublime, and in many other ways aesthetically pleasing. We may draw on our reactions to such things to characterize our individual aesthetic outlook on the world (maybe you’re the kind of person who appreciates a good sunrise and so gets up before dawn to appreciate the colors). But a world without intentionally crafted artifacts that are expressive of someone’s aesthetic outlook would be a massively impoverished one, in part because it would lack the fine-grained distinctions about one’s individual taste that one expresses in choosing what and how to craft something. In other words, it would be a world with less individuality. And, I’ve argued, such a world would be, other things being equal, worse than one that includes such intentionally crafted artifacts. But a world in which such artifacts existed but no one engaged with them (loving them, critiquing them, sharing them) would also be worse than one in which there was such engagement because it would be a world with less mutual understanding and appreciation between individuals. Thus, even in the presence of really good algorithmic fiction, not only are there reasons for people to continue expressing themselves by writing fiction, but there are good reasons for us to continue to engage with (that is, read, promote, discuss) human-authored fictions.

What if sufficiently advanced generative algorithms eventually become agents themselves, capable of expressing themselves through creative acts? Even if that were to happen, I still think there would be good reasons to engage with human-authored fictions. One reason is that appreciating the individualities of other human beings matters to us. But another is that the individualities expressed by artificial agents through authoring fictions would likely be wholly opaque and alien to us. Just as we now do not understand why various generative algorithms output what they do or why various chess algorithms play the moves they do, it seems likely that the actions of suitably advanced artificial agents would also remain opaque to us.28 And thus, engaging with the output of such agents would not result in the kind of appreciation of those agents’ individualities that is realized when we engage with fiction written by other human beings.

6. Onward

In light of my argument, a practical question now arises: how can we (if at all) promote the creation of, and engagement with, human-authored fictions in an artworld full of really good artificial fictions?

On this question, I think looking at analogous cases is illuminating. Take chess, for example. Since IBM’s Deep Blue defeated then reigning world champion Gary Kasparov in a six-game match in 1997 (the first time a computer had defeated the reigning world champion), and after the subsequent development of even more powerful chess algorithms like Stockfish and AlphaZero, it is now widely accepted that no human chess player will ever be able to beat such algorithms. Yet, despite this, participation in, and spectatorship of, human chess has grown in recent years.29 What sense can be made of this, given that we know that we are inferior chess players to our computational counterparts?

A plausible answer is that chess is not just a game of calculation, but it can be appreciated aesthetically, with moves expressing insights and ways of understanding the game that are particular to individuals who make them. Magnus Carlsen’s playing style has been described as “beautiful” as well as “punishing,” and he has been likened to Mozart for his ability to find winning moves in difficult positions; by contrast, the style of Fabiano Caruana has been described as “hip-hop” because he “cultivates on the chessboard a ‘rapping,’ non-stop urgency in his play, with threats maintained on both sides of the board simultaneously.”30 There is also a drama to human chess—especially speed chess, which has become popular to watch online—which is entirely absent from algorithmic games. Seeing the players adjust to each other’s threats and find new positions quickly and on the fly is exciting in part because, with the help of a good commentator, we are seeing the inner workings of another’s mind made tangible.

This suggests a few strategies to preserve a culture that values human-authored work. One is to bring more attention to the creative process of authoring fictions. Movies already do this: most recently, the Tom Cruise Mission Impossible movies have, under director Christopher McQuarrie, marketed themselves by highlighting the work needed to pull off the death-defying stunts of its main star. YouTube channels and podcasts highlight the creative process of artists, from photographers,31 visual artists,32 musicians,33 and so on. Appreciating the human effort that goes into the creation of such fictions can help us better appreciate the creative choices behind them, which are expressive of their authors’ unique points of view.

Another strategy is to broaden humanistic education about art and literature. This is necessary because, just as one needs some understanding of chess to appreciate the beauty of Magnus Carlsen’s games, one needs an understanding of the arts and their histories to appreciate how and why certain creative choices are expressive of an author’s distinctive style and personality. It’s often thought that humanistic education is about learning how to engage in such expressive acts oneself—articulating one’s own ideas and judgments through writing, thinking, and other creative acts. But this is just one upshot of a humanistic education: another is being able to understand and appreciate other people by way of their creative acts.

Our algorithmic future is not yet written. We will ultimately be the ones who choose how to integrate AI into our lives, for better or for worse. My aim in this essay has been to offer reasons to not let AI completely overtake the role of human creative endeavors. Our creative forms of self-expression, and the bonds such expressions form between us, are valuable and should not be (entirely) outsourced to mechanical generation. But this still leaves open many ways of integrating generative AI into own creative projects. And on this score, I am optimistic. Just as the mechanical recording devices (the film camera, phonograph, and so on) of the nineteenth century paved the way to the recording arts of the twentieth (photography, record albums, and so on), perhaps the mechanical generative devices of this century are already paving the way to the new artforms of tomorrow.

References

Blazek, Paul J. 2022. “Why We Will Never Open Deep Learning’s Black Box.” Towards Data Science, March 1, https://towardsdatascience.com/why-we-will- never-open-deep -learnings-black-box-4c27cd335118.

Brainard, Lindsay. 2023. “The Curious Case of Uncurious Creation.” Inquiry, September. https://doi.org/10.1080/0020174X.2023.2261503.

Cohen, Ted. 1999. Jokes: Philosophical Thoughts on Joking Matters. Chicago, IL: University of Chicago Press.

Cross, Anthony. Forthcoming. “Social Aesthetic Goods and Aesthetic Alienation.” Philosophers’ Imprint, Preprint, accepted March 22, 2023. https://doi.org/10.3998/phimp.3475.

Currie, Gregory. 1990. The Nature of Fiction. Cambridge, UK: Cambridge University Press.

Goldman, Alan H. 2006. “The Experiential Account of Aesthetic Value.” The Journal of Aesthetics and Art Criticism 64, no. 3 (Summer): 333–342.

Grant, James. 2023. “A Sensible Experientialism?” Philosophy and Phenomenological Research 107, no. 1 (August): 53–79.

Hopkins, Robert, and Nick Riggle. 2021. “Artistic Style as the Expression of Ideals.” Philosophers’ Imprint 21, no. 8 (May), 1–18.

Kant, Immanuel. (1790) 2000. Critique of the Power of Judgment. Reprint, Cambridge, UK: Cambridge University Press.

Nehamas, Alexander. 2007. Only a Promise of Happiness: The Place of Beauty in a World of Art. Princeton, NJ: Princeton University Press.

Nehamas, Alexander. 2010. “The Good of Friendship.” Proceedings of the Aristotelian Society 110, no. 3 (October): 267–294.

Nguyen, C. Thi. 2020. “Autonomy and Aesthetic Engagement.” Mind 129, no. 516 (October): 1127–1156.

“Photography.” 1855. The Crayon 1, no. 11 (March): 170.

Riggle, Nick. 2017. On Being Awesome: A Unified Theory of How Not to Suck. New York: Penguin Books.

Riggle, Nick. 2024. “Aesthetic Value and the Practice of Aesthetic Valuing.” The Philosophical Review. 133, no. 2 (February): 113–149.

Strohl, Matthew. 2022. Why it’s OK to Love Bad Movies. New York: Routledge.

Thwaites, Abi. 2024. “Knowing How, Knowing Who, Knowing What to Do.” PhD diss., MIT.

Vilone, Giulia, and Luca Longo. 2021. “Notions of Explainability and Evaluation Approaches for Explainable Artificial Intelligence.” Information Fusion 76, no. C (December): 89–106.

Walton, Kendall L. 1990. Mimesis as Make-Believe: On the Foundations of the Representational Arts. Cambridge, MA: Harvard University Press.