Like many sci-fi authors, I have written short fiction on the topic of artificial intelligence taking over the role of humans and gods in society, for good or ill. “The Chains of Plenty” and “Legacy Admissions” portray AI as supportive, if stifling, influences on human society; “E-PLURIBUS” and “Cargo Cult” show AI taking leadership positions, with outcomes that do not necessarily sync with human priorities.
I had not anticipated that AI-generated writing would emerge so suddenly as part of the publishing industry, about five years after early experiments in the field. In retrospect, it strikes me as a little surprising that the science fiction writing community, relative to other genres, has not prepared more for this change. We have written about AI taking over in so many other ways. Why did we collectively not explore the idea that AI could replace us in greater depth?
I don’t have a good answer to that question, and I won’t dispute my colleagues who see AI writing as a threat to their industry and their livelihood. My aim in writing this post is to explore the effects of AI authorship and AI writing moving forward. If we cannot roll back the clock on AI writing, what can we do?
AI Will Be Able To Write As Well As Humans (in Some Respects)
Alan Turing anticipated the prospect of machines writing and interpreting fiction back in the 1950s. Since then, AI has proven to be creative, good at problem-solving, and effective at pattern recognition. Writing and editing are sufficiently rule-based that I expect AI will continue to create engaging and grammatical fantasy and fiction.
Where AI is likely to fail in the short term is in fully representing real-world human experience. AI, like human writers, will write based on its own learning and preconceptions, and these can be vastly different from the human experience.
AI authors will hardly be the first sci-fi or fantasy writers to write beyond their personal experience. In the era when white males were the dominant force in science fiction, their efforts to write women and characters from different cultural backgrounds don’t always hold up well.
An artificial intelligence, even one that can learn, will struggle to think like us, and I don’t think it’s a bad thing. If we in science fiction believe that artificial intelligences can have minds (a thorny philosophical question), we should expect them to develop their own literary voice. As we value diversity in writing as a human endeavor, we should consider AI writing on its own terms and not just as an imitation of human writing.
We Depend on AI to Protect Us From AI Writing
I have seen this happen firsthand already. In the academic and professional spheres, people are looking for ways to weed out AI writing and, without a sense of irony, turning to automated systems to detect AI. Naturally, the AIs who guard us against AI writing have the same limitations as other AIs.
The prospect of a human writer struggling to prove to a machine that they are not an AI has gone beyond CAPTCHAs to be a genuine financial concern for freelance writers and a source of anxiety for students and instructors.
What will happen to human writing when all of us have to learn how to “not write like a machine” as a precondition for selling work or receiving academic credit for assignments? We will have to see.
AI Will Probably Not Replace the Human Author-Reader Relationship (But Could Redefine It)
One of the joys of reading is letting the writing transport us into the mind of another. Sometimes, we read solely to feel the impact of the words, but as we get to know authors as people, we so often read as a way to understand them. I do not believe that AI writing will challenge this relationship, although people will develop attachments to specific AI writers.
Once a writer develops a unique style and image as a writer, their audience will not value simulations as highly as originals, any more than a painting in the style of a Van Gogh would be a substitute for the real thing.
Once an author passes away, some readers might want to read sequels or new adventures of the author’s characters. However, I don’t think a simulation (written by a human or an AI) can substitute for the latest book by a beloved author.
The Training of AI Might Become a Legal Question
AI-generated content could have ramifications for the legal status of work. AIs themselves cannot be legally authors of work, according to current interpretations of the law, so the works they create might not have copyright protection.
If this is the case, what is the dividing line between authored and AI-generated texts when humans and AI software collaborate?
Will the legal framework change to acknowledge AI authorship? Could this be one step toward AI personhood in the future?
And… what of works that draw upon copyrighted work? A human can write a new novel after reading another author’s work without infringing on copyright. What about a machine that learns to write based on reading and analysis of an author’s copyrighted work?
As with AI-generated artwork, AI-generated writing requires the program to learn based on a training set of examples. Artists have argued that using artwork to train an AI violates copyright. As a writer, I can certainly understand the idea that freely available work, but the case for copyright violation seems unclear to me.
Human writers can read a book and store the words in their memory without infringing copyright. Buying an e-book gives you the right to store a copy on your computer so you can read it. If you write your own book in a similar style on a similar topic, it would not be copyright infringement as long as you do not incorporate the other author’s writing in your work.
It is difficult for me to see how this is conceptually different from what more sophisticated AI does when creating new works. Merely exposing an AI to legally obtained works of fiction in the training process should not be infringement, in my view, as long as the resulting work does not contain snippets or paraphrases of the original copyrighted work.
I might of course change my mind as I learn more about the copyright argument, and I suspect the legal question will become more complex as AI abilities improve and human recognition of them becomes more widespread.
If AI will in the years or decades to come possess legally recognized personhood, I am not comfortable with the idea of criminalizing or imposing civil penalties on the act of teaching/training them how to read or write. There are too many human precedents of cultures and one gender in particular using control over writing as a way of exerting political and social control. I’d rather err on the side of promoting literate and educated AIs.
AI Writing Will Be Its Own Cultural Force
Just as there will be humans who reject AI writing utterly, there will be readers who embrace AI writing for its own sake, relishing its non-human take on various literary styles, its forays into the uncanny valley, and its departures from human conventions.
AI art, itself a threat to book illustrators and cover artists, captures our attention not just for replacing humans but by failing to do so. Its odd depictions of the human form, weird hands with too many fingers, and other innovations are now part of our cultural landscape. The literary equivalents will be as well, for better or for worse. Humans, in turn, will find inspiration in what AI writers create.
Human-AI Collaborations Will Lead to New Approaches to Writing
Already, we see AI creations emerging as a combination of the AI’s raw output or response to prompts, coupled with human editing. Human-computer interaction has vastly increased productivity in many fields, and many businesses will probably capitalize on the ability to churn out advertising copy and the like of varying quality.
On the other hand, human-computer interaction could lead to creative approaches to writing that transform the process. Consider the following opportunities:
- An author collaborating with an AI to create multiple versions of a story with different narrative styles, points of view, etc., and allowing the reader to customize the reading experience.
- Custom writing for rewards on sites like Patreon that would not be feasible without AI assistance. Reward a patron by writing a dialogue between them and their favorite character or an alternate ending based on a patron prompt, with AI writing and authorial editing.
- Trained AI writing programs as part of the author’s merchandising as a jumping-off point for “authorized” fan fiction.
- AI interactions as part of the reading experience for a hypertext novel.
- AI-generated narratives of the AI experience.
The last point merits some elaboration. As AIs interact with the world in more complex ways, it will be no more difficult for them to narrate their learning experience than for them to create narratives of fictional characters. Conceivably, the features of AIs that create online content might enable them to be the forerunners of artificial intelligences that we eventually recognize as self-aware. In a genre that celebrates “own voices,” why should humans exclusively control the narrative of the emergence of sapient AI?
The variety of new approaches to writing that AI will usher in is too vast to describe, much less predict, within the scope of an article. My point is that AI might replace or encroach upon some styles of writing, but it will unlock others and empower the next generation of writers to create in entirely new ways.
AI as Beta Tester or Developmental Editor?
As AI has becomes more sophisticated and capable of detecting large-scale patterns in text, companies are already marketing AI as a big-picture editing tool in the same way that tools like Grammarly help writers polish prose at the level of the sentence or individual word. The result might be an influx of aspiring writers into self-publishing and an increase in the quality of self-published work by writers who cannot afford to hire a human editor.
There might also be some decidedly non-human editing decisions that leave human readers scratching their heads.
I can’t endorse any such products or comment on the state of the art at present. However AI developmental editing tools and beta testing apps will probably find many customers, perhaps among the ranks of writers who feel AI writing nipping at their heels within the writing industry.
Where Do We Go From Here?
In an industry as decentralized as AI development, the only clear answer is: to the future. Writers might have to find new ways of practicing their craft, as scribes did with the introduction of the printing press. It will be a challenge, but the 21st century is not awash in unemployed scribes. The 22nd and 23rd will not overflow with starving authors. We will find a way forward in the long run.
However, as John Maynard Keynes pointed out, “in the long run, we are all dead (for now, anyway).” The future of writing does not pay the bills today. So, what are writers to do in the here and now? I can’t say for sure, but as someone who has been at least a peripheral part of the industry for over a decade, I have the following suggestions.
- Look for markets and employers that reject copy that is the product of AI. Keep in mind that your work might face some sort of reverse Turing test and might get bounced for being “AI-generated.”
- Generate personal relationships with editors and readers, which should buy you time until individual AIs develop their own fan bases.
- Avoid relying too much on genre and style conventions. Initially, AI will have an easier time writing stock characters in predictable plots, and even subverting those tropes from time to time. They will not be able to simulate lived human experience as easily.
- Grapple with the ethical challenges of using AI to automate your writing process. Consider the implications of choosing to use AI to create an audiobook, generate promotional materials, and assist in editing.
So, as always, to all the human writers and self-aware AIs out there, Happy reading and happy writing!
S. R. Algernon 