This year’s Nobel Prizes may have been some of the last to be awarded to humans, and even then, many of them went to humans who were instrumental in inventing modern artificial intelligence (AI). Eventually, AIs may win the prizes directly. In the meantime, perhaps the best transitional Nobel Prize for Literature next year should go to Lee Child.
Lee Child, the pen name of British author James Dover Grant, is most known for his wildly popular and best-selling Jack Reacher novels, of which the 29th is out this week. The Reacher series are often classified as thrillers but they could also be found in the mystery section or the fiction section or other categories. They can also be read in any order. While not conventional humor books, they are peppered with comedic moments, including long-running and charming character quirks. For example, in earlier books in the series, the hulking 6’5” 250-pound Reacher often shrugged. In later books, Reacher said nothing. In fact, Reacher said nothing so often that it became the title of a fascinating behind-the-scenes expose by British author and scholar Andy Martin exploring Child’s unique writing method. Martin watched, recorded, and reported as Child made “Make Me,” the 20th Reacher book.
Child writes in a way that almost no other human does, but in a way very similar to how an AI would. He does not have an outline. He doesn’t know where the story is going. When he writes a paragraph, he often has no idea what will come next. Child has said that editors sometimes suggest the story might be better if this event happened before that one, to which he replies, yes, that would have been great, but that’s not how it happened.
He also rarely revises. The first paragraph he ever wrote as an author was written in pen on lined yellow paper and appeared without modification as the first paragraph of the printed version of the first Reacher book. This is exactly how modern large language models (LLMs) work: one paragraph, one word, one syllable, one token at a time. Child also burst on the scene in much the same way LLMs did: from never having generated anything before, to being a mainstay on almost every human’s desk. Reacher books have sold hundreds of millions of copies in dozens of languages and a hundred countries or territories. Chances are you or a loved one are reading one right now.
Ironically, Child edited a volume of advice for mystery writers. Every chapter but his suggested all sorts of useful tips for plotting, character development, outlines, diagrams, and other plans. Child’s chapter basically said: don’t plan. You’ve read enough. You know what makes a good story.
This sounds exactly like the training protocol of an LLM, whose corpus has every story available. How does an AI write a story? It essentially recalls good stories it’s read before and tries to make another good one.
But do they? Just how powerful are LLMs, or how powerful can they become? Can they truly generate new stories rather than just combine old ones? Can they create? Can they think?
Illya Sutskever co-founded OpenAI, the makers of ChatGPT, and recently left to co-found a new firm, Safe Superintelligence. The meme “What did Illya see,” originally referring to the possibility that he panicked when he saw some internal AI that exhibited superintelligence, is now almost as ubiquitous a phrase as “Who is John Galt?” in Ayn Rand’s Atlas Shrugged, except that was fiction and what we are living in is probably reality.
In a conversation last year with NVIDIA’s CEO Jensen Huang, Sutskever gave this example of how AIs can write new books, make new connections, think, and generate new creative knowledge despite “merely” predicting one word ahead: imagine a mystery novel, he explained. Imagine you’ve written all the details and the clues and explained the characters and the murder and the various motives and opportunities. Now comes the big reveal when the detective gathers all the suspects in a room, and then he says, “The killer was,” and the AI has to predict the next word. That would take thought and deliberation and consistency with the clues and other facts. That would be new.
Someday, we may be reading AI-written Reacher stories. Child has already taken a writing partner for his last several books: his younger brother Andrew. As a reader, in an experiment of n=1, I am happy to report that the quality is still exceptional, the writing is mesmerizing as usual, and the only reason you’d ever put it down or refuse to turn the page, like all Reacher books, is because you want to savor it.
Others have also added their vision of the Reacher universe. A shorter but nonetheless captivating Tom Cruise starred in two excellent Reacher movies. The more physically Reacher-like Alan Ritchson now stars as Reacher in REACHER, a streaming series on Amazon set to release its third season next year. So it’s not impossible for other carbon-based humans to contribute creatively to the Reacher universe. What about silicon-based AI?
I have happily asked AI to generate brief Reacher stories set in my hometown or working in my industry. Once, the AI had Reacher investigate a hedge fund trader who himself tried to use AI to predict stock prices. The finance applications are the same as the literature ones: if it can predict the next token, it can predict the next tick. But can it in fact make good Reacher stories?
I felt a little bad the first time I generated a new Reacher story. But then I recalled two facts. First is that it may take Child three decades to write thirty books, but it only takes a month to read them. And once you read your first one, you don’t want to wait. I’ve read every Reacher book, listened to every audiobook, watched every movie and episode, and still want more. And the second fact is something that happened to Child as a child. He was enjoying a book when he noticed the last big section was missing. Then he realized he could just extend the story in his own imagination. It could “end” however he liked. So I use the AI to kick-start my imagination, and I don’t feel as bad anymore. I think Reacher would shrug.
So far, the AI stories are not quite as good, but AI is constantly improving. Students in my class on AI at Fairfield University are each writing an entire (non-Reacher) novel using AI. It’s not easy but it’s not impossible, and the books are certainly readable. Perhaps it does better when it can create out of whole cloth rather than extend an existing literary universe.
But before our species capitulates entirely, before the AIs start winning all the awards, it might be fitting for Child to win the last Nobel Prize in Literature by a human. Because when AIs do start to win, they’ll probably be doing it the same way he does. Personally, if the Child-like quality can be maintained, I would love to pay a monthly subscription to ReacherGPT.
Read the full article here
