So machine learning artificial intelligence has really been blowing up this past month, probably because of ChatGPT and all of the fascinating things that people are doing with it. I’ve been getting into it myself, using it to help write or improve my book descriptions, and also experimenting with it for writing stories.
At this point, any original fiction that ChatGPT writes is about the same quality as something written by an overly eager six year-old (minus the grammar and spelling errors), but I can see how that could change in the future, especially on a language learning model that’s trained on, say, Project Gutenberg, or the complete bibliographies of a couple of hundred major SF&F writers. The technology isn’t quite there yet, but in a few years it could be.
But apparently, that hasn’t stopped hordes of amateur writers and/or warrior forum types from using ChatGPT to spam the major magazines with AI-written stories. In fact, Clarkesworld recently closed to submissions because they were getting flooded with “stories written, co-written, or assisted by AI.” Neil Clarke wrote an interesting blog post on this problem, saying that this is a major growing problem for all of the magazines and that they will probably have to change the way they do business to deal with it.
So will AI eventually become so good that it replaces writers altogether? I don’t think so, and here’s why.
Replacement vs. collaboration
The gap between an AI that can do 100% of what a fiction writer can do and an AI that can do 90% is actually much wider than the gap between an AI that can do 90% and an AI that can only do 50%. That’s because both the 90%-effective AI and the 50%-effective AI require collaboration with a human in order to do the job. Neither of them can fully replace the human, though a human-AI team may be able to do the work of many humans working alone.
If we ever get to the point where AI replaces storytellers completely, we have much bigger problems than a few out-of-work science fiction writers. Storytelling lies at the heart of what it means to be human: we call ourselves “homo sapiens,” but we really should call ourselves “homo narrans,” since story is how we make sense of everything in our world. If an AI can replace that, then we as a species have become obsolete.
But I don’t think we’re going to ever reach that point. My wife is currently getting a PhD in computer science—specifically in machine learning and language models—and she believes that there is an inherent tradeoff between intelligences that can specialize well, and intelligences that can generalize well. AIs are master specialists, but humans are master generalists. If we ever build an AI that’s a master generalist, we may find that it’s actually much less intelligent than an average human, because of the tradeoff.
But all of that is purely speculative at this point. Right now, we really only have AIs that can do about 20% of what a fiction writer can do. In the coming years, we may ramp that up to 50% or even 90%, but anything less than 100% is not going to fully replace me.
Tools, force multipliers, and the nature of writing
However, that doesn’t mean that the thing we currently call “writing” isn’t going to change in some pretty dramatic ways, much as how the internal combustion engine dramatically changed the thing we call “driving.” And with these changes, we may very well get to the point where the market just can’t support as many professional writers, and the vast majority of us have to find other lines of work.
Conversely, it may actually expand the market for “reading” and create new demand for “writers,” as “reading” becomes more interactive and “writing” turns into an AI-mediated collaboration with the “reader.” Kind of like a Choose Your Own Adventure that writes itself, based on the parameters set by the “writer.”
I have no idea, but the possibilities are fascinating, and the writers who are sure to lose are the ones who fail to confront the fact that their whole world is about to change—indeed, is already changing.
I think what it’s going to come down to is who owns the tools: not just who can use them, but who can modify them, personalize them, and use them to create original work. If copyright law decrees that the person who owns the AI also owns anything created with the AI’s assistance, that is going to be a major buzzkill… unless we get to the point where everyone can have their own personalized AI, which would be pretty cool. It would also solve a lot of the problems emerging from all of the super-woke filters that are getting slapped on ChatGPT.
Personally, I’m looking forward to the day where I can use an AI model to write fifty novels across a dozen pen names in a single year. What an incredible force multiplier that would be! But only if those novels are “mine,” whatever we determine that means.
So really, instead of arguing about whether AI will replace authors, what we really ought to be talking about are the aspects of writing and storytelling that drive us to create in the first place, and how those aspects can translate into a world where the nature of “writing” looks radically different than it does right now.
The “but I already know how it ends” problem
There is one problem that is unique to the written word, and it’s something that every writer has to confront when making the leap from amateur to professional (or even just from an amateur who dabbles in prose to an amateur who finishes what they start). The problem can best be summed up by this:
Why should I bother writing this story if I already know how it ends?
Unlike visual media such as TV, movies, video games, or illustrations, the art of the written word exists 100% in the reader’s head. These things we call “words” are really just symbols that convey thought from one mind to another, and have zero meaning outside of the head of the person reading. If you don’t believe me, try picking up a classic novel written in a foreign language that you don’t understand, and see how well you enjoy it.
But when we read, we like to be surprised on some level. There is something about the novelty of the story that appeals to us—indeed, that’s why we call them “novels.” The trouble is that the very act of creating a novel kills the novelty of it. At some point, you know how it’s going to end, and after that point the act of writing becomes a chore—or rather, it can be, unless you find something else about the process that fulfills you.
Some professional writers deliberately put off that moment for as long as they can, never figuring out their ending until it comes as surprise, even to them. Others look for fulfillment in something else, like the artfulness of their prose, or the dramatic suspense built up by their use of language. Still others just plow ahead, accepting this loss of novelty as a cost of doing business.
But however they choose to deal with it, every writer has to confront this problem in some manner before they make the leap from amateur to professional. And this is perhaps the biggest reason why I’m not too worried about AI replacing me as an author: because even an AI model that can do 90% of what I do will still require its human collaborator to address this problem.
Fanfiction and derivative works
Of course, the amateur vs. professional problem will affect some genres more than others: “write me a romance just like ____ where the male love interest has black hair instead and works in my office” is going to be just fine for a romance novel addict who just wants their happily-ever-after without any uncomfortable surprises. But we already have this: it’s called fanfiction.
Which is not to say that all fanfiction is formulaic and predictable. But the thing that sets fanfiction apart from original fiction are the things make it a derivative work: things like characters and settings that are already well-established, or a rehashing of storylines that were created by someone else.
This is an area where I think AI shows the most promise, and will turn out to be the most disruptive: not in creating original works, but in creating derivative works. Imagine if you could plug a novel into ChatGPT and tell it to rewrite the ending so that the girl ends up with your favorite character, or your favorite villain wins in the end. ChatGPT can’t do that very well right now, but I don’t think we’re far from building an AI language learning model that can—especially if it’s trained on actual books, instead of online content.
What I foresee is a world where AI blurs the line between fanfiction and original fiction so much that it becomes normal to read a bunch of these derivative works after you’ve read the original. Indeed, it may become a game to see who can make the most popular derivative work, and the popularity of some of them may very well exceed the popularity of the original.
Or it might become normal to run everything you read through an AI filter that removes offensive language, or the sex scenes that you were going to skip anyway (or conversely, an AI filter that adds offensive language and sex scenes). Taken to an extreme, this could lead to some really dystopian outcomes that further divide our already polarized world. We’ll have to see how it shakes out.
But all of this derivative content is only possible if there’s original content to derive it from. And while AI may lower the barrier of entry somewhat to creating original content (or not, since there really aren’t any barriers to entry right now, aside from the time and practice it takes to become proficient at your craft), the problem of “but I already know how it ends” will keep most dabblers and amateurs in the realm of creating derivative works, not original ones.
The act of “writing” and “reading” may change dramatically based on the force-multiplying effect of these tools. We may even get to a point where “writing” and “reading,” as most of us understand it, bear little resemblance to how we understand it today. But unless our very humanity becomes obsolete, I’m confident that I will still be able to carve out a place for myself as a writer.
