The possibilities and limits of AI-authored thrillers
Shylashri Shankar Shylashri Shankar | 19 May, 2023
(Illustration: Saurabh Singh)
CAN MACHINES think? That’s the question Alan Turing (the father of modern computer science) asked in 1950. He answered it by proposing a game in which a machine tries to pass off as human (The Imitation Game). Any machine that succeeded by giving an impression of intelligence could be said to have intelligence, he said. His focus was on perception, rather than real and self-experienced intelligence by the machine. Two years earlier, in 1948, brain surgeon Geoffrey Jefferson set a higher bar. “Not until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, could we agree that machine equals brain—that is, not only write it but know that it had written it.”
Have we reached that stage with the entry of Artificial Intelligence (AI) into the creative world? It has written poems (competent ones). Does it mean the machine has a creative consciousness? How do we even know if it has, since we humans find it very hard to pin down the feeling of being in the world? We may be able to describe the sensations produced by an ice-cream, but we couldn’t say what tasting ice-cream is really like. As an article in MIT Technology Review concluded: “For all their sophistication, today’s AIs are intelligent in the same way a calculator might be said to be intelligent: they are both machines designed to convert input into output in ways that humans—who have minds—choose to interpret as meaningful.”
This input is precisely what the recent writers’ strike in Hollywood has spotlighted. Scriptwriters are paid peanuts by the Hollywood Moguls, and with AI and streaming shows on the scene, things have gotten worse. As The Guardian reported, the Writers Guild of America has demanded increased pay, better compensation when shows are transferred to streaming services, regulations for how artificial intelligence is used in the writing process, an end to the “abuse of mini-rooms” in which a smaller group of writers develop scripts often for minimum pay rates despite their seniority or experience, and increases to pensions and health funds.
With AI entering the arena, their situation has become ever more precarious. AI is good at summaries but not so great at making the emotional connect (though it has done well in writing poems). In a script, the actors are the vehicles for connecting with the audience, so an AI-type summary can work and reduce the production cost. A studio executive could feed in the main elements of the movie they want to make, and the AI can churn out a first draft. Even if the film studio hires a writer to finesse the script, it could get away with paying a far lower rate.
How about books, especially crime fiction? In a thriller, would AI be able to produce the goods? Do we require a strong emotional connection with the main character to keep on turning the pages? Or can the plot’s momentum take us through? How do we tackle the problem that we humans create meaning where there is none? A simple statement (composed perhaps by AI) can become emotionally loaded when the reader processes it through their own contexts and world views.
There have been several attempts by crime writers to use AI to write a whole book. Death of an Author (a premature title!) is written by Stephen Marche who worked with ChatGPT and other AI tools to craft this thriller novella. A New Scientist reviewer calls it plodding prose and points to a final twist that leans too heavily on a Sherlock Holmes story. The good news for writers is that ChatGPT can only produce 600 words at a time and won’t remember what it has done in the previous 600 words. A human author still has to sew a patchwork quilt of a novel by feeding it the prompts and elements—which is what Marche did. In a review in Wired, the reviewer said, the prose came closer to Dan Brown—“compulsively readable, but not in danger of winning the Edgar” (a sort-of Nobel prize for crime writers).
Others have used ChatGPT to write Sherlock Holmes stories. I read a couple online. The problem is they are written in first person with Holmes as the narrator, not Watson. And that is a huge barrier because Holmes’ mind is an enigma in the original stories, and we readers are in Watson’s skin, marvelling at Holmes’ cleverness. The AI version makes Holmes into someone like us, and hence more pedestrian and ho-hum. The mystique disappears in these versions.
THEN I TURNED to an old favourite—David Baldacci. His new psychological thriller, Simply Lies (Macmillan; 432 pages; ₹750), could absolutely not have been written by ChatGPT. It is a twisty cat-and-mouse game between a former police officer and now single mom Mickey Gibson and a mysterious woman, Arlene. Gibson whose ex-husband absconded with all her savings and his secretary, is now working for ProEye, a global investigation company that tracks down the zillionaires who’ve absconded without paying their debts. An unseen colleague, Arlene, rings her up and gives her a new assignment: to scope out a mansion owned by a notorious arms dealer, Rutger Novak. Novak is the bete noire of her company’s clients in the past, having cheated them out of millions. The frazzled, dogged, and spunky Mickey finds a long-decomposed body in a secret room, and realises she is being played by Arlene. Enter the police, a tall and attractive Wilson Sullivan with “pure cop’s eyes” that are alert, suspicious, roving and contemplative—all lasering in on Mickey, his main suspect.
Things go south for poor Mickey when it turns out Arlene doesn’t exist, and the call had been made on a burner phone. And when the victim turns out to be a former mob accountant in a Witness Protection Program, the race is on to find out what connection Arlene has with the victim, and why more people are being murdered. Though Mickey’s motivation— simply that she is bored— is not quite believable, especially since her actions put her two young children in danger, the twists and turns of the story make up for it. As readers, we get a glimpse into the mind of Arlene or whatever the con artist decides to call herself while the protagonist (Mickey) is still trying to figure her out. The fact that the victim was in a witness protection program means the police also end up on the suspect list, a favourite twist with a thriller reader. The theme centres on family dynamics, especially fraught parent-child relationships. Fans of Baldacci will enjoy this pacey read, which for sure, AI couldn’t have written.
We are still at a stage where AI is playing an imitation game. It is dependent on human input to create the world, the characters, the plot and the conflict. So, novelists and crime fiction writers need not despair as yet
Nor could ChatGPT have written Keigo Higashino’s new mystery, A Death in Tokyo (Abacus; 368 pages; ₹699). A production inspector (Takeaki Aoyagi) at a metal company is found stabbed to death on Nihonbashi Bridge in Tokyo, and the only suspect (Fuyuki Yashima) flees the scene but is grievously injured by a truck minutes later. Takeaki’s wallet is found on Fuyuki, now in a coma, and though it seems he did it, the two policemen on the case are doubtful about Fuyuki’s guilt. Inspector Kyoichiro Kaga, Higashino’s series detective is back on the case, this time with his cousin Detective Matsumiya. They set about in police-procedural style to interview the families and friends of the suspect and the victim. Fuyuki’s pregnant girlfriend may be hiding something, and so is the victim’s sullen teenage son. As Kaga delves deeper, he discovers that Fuyuki worked for Takeaki’s division and after having an accident, had been let go. But other things—opportunity and timing— don’t add up to Fuyuki’s guilt. Takeaki’s presence so far from his home and workplace is also puzzling until Kaga discovers the reason. Takeaki was making a circuit of the Seven Lucky Gods and leaving paper origami cranes in the shrines. But why? Though Fuyuki dies, Kaga continues with the case, determined that the innocent should not suffer.
Inspector Kaga carries echoes of Kosuke Kindaichi, the great detective created by Seishi Yokomizo in the 1940s, but without his lyrical sense of place. Higashino’s descriptions of Tokyo are clinical and terse, and match the solve-the-puzzle focus of the book. This emphasis harks back to the puzzle-solving ethos of Agatha Christie and her ilk. By discovering what the victim was doing just before he died, and why he was doing it, the detectives unearth the motive and the killer. While A Death in Tokyo cannot be classed among Higashino’s best novels, it nevertheless offers an interesting glimpse into the traditions and preoccupations that animate Japanese society, while also complying with the norms of a Golden Age puzzle.
We are still at a stage where AI is playing an imitation game. It is dependent on human input to create the world, the characters, the plot and the conflict. So, novelists and crime fiction writers need not despair as yet. We still have control over the fundamental aspects of the novel, the story, the plot and the characters.
More Columns
‘AIPAC represents the most cynical side of politics where money buys power’ Ullekh NP
The Radical Shoma A Chatterji
PM Modi's Secret Plan Gives Non-Dynasts Political Chance Short Post