AI-written fiction: a solution in search of a problem?

(Promotional video from Sudowrite)
Do AI writing programs fulfill any real market need?

So-called artificial intelligence (actually a misnomer, but we’ll let that slide for our purposes here) has recently been turned to the task of fiction writing. It is now possible to use AI writing tools like Sudowrite to create large bodies of text that can meet the minimal requirements for stories.

This is especially true in genres like romance, fantasy, and fanfiction. Self-published works in these genres have been posted online in large quantities, and are therefore readily accessible to AI bots that scrape and mine content for AI predictive modeling. These genres also rely heavily on repetitive tropes, which further facilitate AI predictive modeling. It isn’t hard to predict what comes next, if the same thing always comes next.

Let’s put aside the question of whether or not AI-authored fiction that meets some minimal standard is possible. Let’s allow that it already is possible.

Instead, let’s consider the question from another angle: does AI-authored fiction meet any discernible market need? Or is this a solution in search of a problem?

There are some tasks that humans don’t perform as well as software, and which humans don’t even want to do.

Take, for example, calculations involving many rows and/or columns of numbers. Even with a calculator and an adding machine, this is a tedious chore that is highly prone to error. That’s why we’re all grateful that we have programs like Microsoft Excel and Apple Numbers.

The first mass-market spreadsheet program, VisiCalc, hit the market in 1979. While not as good as the latest, 21st-century version of Excel, by the early 1980s VisiCalc had largely eliminated the need to perform tedious calculations by hand.

No one—before or since—has complained about “the machines taking over” where spreadsheet software is concerned. Nor has anyone ever claimed that spreadsheet programs don’t fulfill a genuine market need.

Now let’s pick an example closer to the hearts of writers. There was a time when writers who didn’t want to type their own manuscripts had to hire someone else to do it for them.

That meant hiring a typist. The most expensive (and comparatively rare) option was for a typist to sit in a room with the writer while he dictated aloud. More commonly, the typist worked from tape recordings or handwritten documents.

Prior to 1990 or so, manuscript preparation was actually a real problem for most writers. Imagine, in 1985, preparing an 80,000-word novel manuscript for submission to a half-dozen publishers. Self-publishing wasn’t an option in 1985. Amazon was a decade away; and a quarter-century would pass before the Kindle would  be invented. Everything that was published had to be submitted to a publisher first. On paper.

I prepared college term papers of a few thousand words back in the 1980s, using nothing but paper, a typewriter, and correction fluid. I found that difficult enough. I can’t imagine preparing the manuscript for an 80,000-word novel that way.

The digital revolution, which includes many AI components, changed all that. Everyone now has convenient access to word-processing software that would have been astounding in 1985. And if the writer doesn’t want to type, speech-to-text software programs like Dragon Dictate can even obviate the need for hitting the keyboard. Dictation software would have seemed like alien technology in 1985!

I’m sure that back in 1985, there was a small cohort of typists who made their living preparing manuscripts for novelists. I’d also bet that such work was low-paying, tedious, and no one’s idea of a dream job.

Software, in other words, does a better job than humans of getting words from a writer’s head into a manuscript format that can be consumed by others. Nor are many people clamoring to serve as low-paid scribes for writers. Once again, you’re going to have a hard time finding Luddites or complainers. No one would deny that digital tools improved the process of manuscript preparation.

But now let’s consider the prospect of turning over the conception and writing of a novel—the actual creative work—to a software program.

Let’s return to those AI-written novels and stories. And—once again—let’s leave aside the question of whether or not a minimally passable product is possible with AI.

The first relevant question is: can AI do the job better? I think the answer to that would have to be a resounding no.

I’ve read some of the fiction texts written by AI in recent months. I’m mildly impressed that software can regurgitate and recombine human-generated stories scraped from the Internet, without completely bungling grammar and syntax.

That allowed, the imitative nature of these stories is clear within a few paragraphs, and immediately recognizable as something written by AI. Every AI story I’ve ever seen is a pale imitation of one that you’ve seen a gazillion times on sites like WebNovel and Wattpad.

Why? Because this is where the AI-generated stories are coming from.

Not every reader will be familiar with Wattpad or even the process of writing fiction. But everyone has experienced the following:

You’re having a problem with your electricity, cable TV, or Internet service. You call the electric company or your cable/Internet provider for help.

A pleasant-sounding female voice answers your call. Then you realize, with a sinking sensation, that you have fallen into the virtual clutches of an AI chatbot.

The AI phone attendant can respond to your questions, after a fashion, but she cannot really understand your problem. She is simply following a series of pre-programmed routines. For that matter, there is no “she”. There are lines of computer code written to fool you into believing there is a “she” there.

Software is incredibly useful for certain intense, narrowly constricted tasks, like adding a list of one hundred four-digit numbers. What software can’t do is tell you what those numbers mean, how you should use them to make a decision about your family’s budget, or the next product line you should introduce at your company. For that, you need human intervention—probably your own knowledge and experience.

Software is largely incapable of integrating and interpreting complex data from diverse sources, and completely incapable of engaging in anything resembling the human process of thinking. My guess is that this will always be the case.

Want proof? Is there any one of us who would rather talk to an AI chatbot than an actual person when we call a company with a pressing problem? Have you ever called the cable company, gotten on the line with an actual human attendant, and asked to be connected to an AI chatbot instead?

Of course you haven’t. Because you know that the AI chatbot is ultimately a sham. Although the phone-answering AI chatbot can fool you initially, it really isn’t adding much value to the process. To get anything done, you’ll need to talk to a person.

Nor is the AI phone-answering chatbot really adding much that is even new, beyond the most superficial level. Back in the 1990s—possibly even the 1980s—there were automated phone menus that enabled you to push “1” for the sales department, “2” for technical support, and so on. The phone-answering AI chatbot is yet another example of an AI solution in search of a real-world problem.

If artificial intelligence cannot improve a process, in the same way that spreadsheet programs can improve the process of large-scale calculations, then what is the point of involving AI at all? More often than not, AI simply gets in the way.

Some people are infatuated with artificial intelligence, and believe that it always adds value. But technology can be superfluous, and even a hindrance, if it isn’t suited to a particular task. How many of us would seriously want to take an F-16 to work, rather than commuting in a comparatively low-tech Honda or Toyota?

At present, the world faces many acute shortages. There is a shortage of cheap, environmentally friendly energy. In many areas, there are shortages of medical personnel. Since the COVID pandemic, there have even been shortages of willing restaurant workers.

But one shortage the world does not face? The world faces no shortage of mediocre, barely passable fiction. Just check the Internet. The Internet is already overflowing with imitative, uninspired fiction that no one wants to read.

This is about economics as much as it is about art. As anyone who has ever tried to make a living with novels can tell you, the marketing/selling of fiction is an uphill battle. Even novels by traditionally published, brand-name authors require dedicated and expensive marketing campaigns. The same is true of the best self-published fiction. It doesn’t sell itself. It has to be marketed.

Is there any point, is there any fulfillment of a genuine market need, in filling the Internet (and perhaps, in some cases, the virtual shelves of online bookstores) with AI-knockoffs of Wattpad novels?

This is why I say that AI-written fiction represents a solution in search of a problem. Yes, I understand that some techno-utopians will consider the idea of an AI-written novel to be irresistibly cool for its own sake. The journalistic class, moreover, is presently infatuated with the concept. Various online publications have been hyping AI writing tools for months now.

But how many of us are interested in consuming the resultant products—the novels themselves? Which would you rather read: the next John Grisham or Stephen King novel, or something scraped from the Internet by a bot, and recombined with a software program?

Until you can come up with a satisfactory answer for that question, you can’t convince me that AI-written fiction fulfills any real market need.