Though people don’t read books as frequently or consistently as they used to, reading is still a popular pastime and people are still eager to consume the latest and greatest pieces of literature that emerge. Today, the bookshelves of the dwindling number of physical bookstores are stocked with copies of books written by brilliant, creative, human individuals.
Is it conceivable that artificial intelligence (AI) could one day write a bestselling book that matches, or even exceeds these books in terms of quality and popularity?
A decade ago, most people would have laughed at this question. A few decades ago, most people wouldn’t have even considered it. But today, we have access to AI that’s capable of text and content generation, and in some very impressive ways.
Still, there may be some limitations on what AI can do. And we owe it to ourselves to understand those limitations.
The Limitations of Current AI
These are just a few of the limitations of current AI worth considering:
- Logistics and processing. AI may be good at creating text in some contexts, but it’s not good at making important decisions that surround the work, such as decisions about the business, printing, marketing, or distribution of the work. Technically, it’s not impossible to program AI to make decisions about things like what paper stock to include or what type of binding to use. It may even help with marketing and advertising analytics. But in isolation, even a great piece of literature may fall flat in the absence of these other considerations.
- Initial direction. Current AI is entirely dependent on the inputs and guidance of human actors. This is both a strength and a weakness, as a person with an excellent idea may be able to guide the AI to excellence, but any prompt that falls short is going to artificially limit its potential.
- Opinions and original thoughts. Despite some subtle appearances to the contrary, AI is incapable of thinking or feeling on its own. It doesn’t have opinions. It doesn’t have original thoughts. It doesn’t have experiences or unique perspectives. It’s easy to argue that great works of literature are primarily built on opinions, original thoughts, and unique perspectives; AI would thus be capable only of mimicking these pillars, at best.
- Continuity and length. There are many AI tools that are currently capable of producing short articles and swaths of text. But when we start working on book-length pieces, we start to see issues with continuity and consistency. AI is perfectly capable of generating works of 50,000 words or longer, but those works aren’t necessarily cohesive or driven by a consistent direction.
- Artistic prose. Whether or not AI can produce any form of creative art is debatable. But right now, it seems like AI is simply incapable of developing artistic prose. AI-generated text tends to be somewhat simplistic, highly repetitive, and easy to discern as written by a machine. The best books available to us aren’t burdened by these challenges.
Narrow vs. General AI and Production Implications
Narrow AI and general AI are two very different things.
If you’ve ever seen a movie from the Terminator franchise or if you’ve read any number of sci-fi books related to the topic, you’re familiar with the concept of a sentiment AI that poses an existential threat to humanity. This type of AI is smarter than humans, good at everything, and capable of adapting to new circumstances. This is general AI, and it’s something beyond our current range of development abilities.
Narrow AI, in contrast, is not only feasible, but widely developed and accessible for a wide range of applications. Narrow AI is sophisticated and competent, like general AI, but it’s focused on a much narrower slice of potential applications. For example, it might be extremely good at processing and analyzing data, but completely incapable of solving rudimentary math equations or understanding text.
Currently, we have access to narrow AI systems that can process, “understand,” and even produce content that seems like it was written by a human natural language speaker. But the quality of the content produced is closely tied to the direction given to it; without adequate direction, it’s unlikely that narrow AI would ever be able to create a literary work worthy of critical praise.
The narrow versus general dynamic must also be considered in terms of audience taste. AI content generation is perhaps at its best when given an adequate number of constraints; the more detailed your instructions are, and the more specific your vision is, the better the resulting product is going to be. Because of this, it’s perhaps unlikely that narrow AI would ever be able to produce a literary work capable of appealing to a wide audience.
Conversely, if AI is wielded for the purpose of creating niche work that only appeals to a narrow target audience, other problems arise. AI is only as good as the materials used to train it, so if there’s a shortage of materials available to review, the AI is naturally going to come up short; this makes it unlikely that AI will be capable of writing truly impactful works even in a specific niche or genre, at least for the time being.
What Does the Future Hold?
Right now, it seems unlikely that AI would ever be able to write a bestselling book, but many people are fascinated with AI not because of its current capabilities, but because of its future potential. So, could this change in the future?
We tend to think of the progression of AI as being explosive and unidirectional. But this isn’t necessarily the case. Already, we’re seeing evidence that AI systems trained on content generated by themselves and by other AI systems can actually regress in ability over time and lose abilities they previously had. Accordingly, it’s far from a certainty that AI will progress to a level that allows it to greatly transcend its current potential.
Still, it’s entirely possible that engineers and scientists will dream up entirely new strata of AI in the future, and potentially even in the near future. It’s a fool’s errand to try and project this technological trajectory since there are so many uncertain variables in play.
If there’s a bottom line, it’s this. Currently, AI is incapable of producing a book worthy of widespread praise and consumption. That seems unlikely to change in the near future. But because technology can grow explosively and unpredictably, we can’t yet say for sure with the long-term limitations are.