A group of skilled thieves plot to pull off the biggest heist in Indian history, stealing Rs 100 crore from a high-security Mumbai bank. The group includes a mastermind (Akshay Kumar), a tech expert (Ranveer Singh), a safe cracker (Vicky Kaushal), a getaway driver (Taapsee Pannu), and a con artist (Deepika Padukone). The plan goes awry when a corrupt politician’s son arrives at the bank at the same time, and the thieves are forced to improvise. What follows is a thrilling chase. The director: Rajkumar Hirani. The producer: Karan Johar.
This is an answer to the prompt, “Write me a Bollywood movie that will gross Rs 100 crore”, as generated by ChatGPT, an artificial intelligence chatbot launched in November last year. It can churn out stories based on one-liners, scripts of existing films written in the style of another director, and alternate endings to released films. It can also produce an infinite number of responses.
Currently, however, the platform’s limitations are apparent. Its knowledge of the world post September 2021 is limited. Ask it who the President of Brazil is, and it replies, “Jair Bolsonaro”, unaware that he was defeated by Lula da Silva in October last year, in the country’s closest election in three decades. When it comes to story suggestions, ChatGPT shows occasional flashes of inspiration (which is how we came up with our April Fools’ prank), but for most part its scriptwriting output seems generic, clichéd and a little silly. Arguably, the same could be said of the bulk of commercial cinema, which is why ChatGPT and mainstream Indian movies seem like they’re made for each other.
Mary Kom (2014) and Neerja (2014) screenwriter Saiwyn Quadras doesn't view ChatGPT lightly. He proposes an eventual worst-case scenario: Streaming platforms will have writers who, working according to the information provided by their sites’ algorithms, prompt ChatGPT to provide similar ‘hit’ storylines. The software itself will then break these down into episodes and later, seasons. “You’ll be eventually eliminating the writer entirely. In this business, we’re always looking for the easy way out, and what’s easier than an AI that does your bidding?” he said.
Cargo (2019) writer-director Arati Kadav, who has a Masters in Computer Science specialising in AI, agrees. “When you pitch to networks, they say things like, ‘Our data mining tools tell us the story should start with a bang.’ So ChatGPT will be able to tell a story that will happily tick all the boxes of your algorithms,” she said. This, however, is precisely the reason she believes the software won’t gain popularity — its output is so formulaic that the lived-in personal and cultural experiences of a human writer will come to be sorely missed.
Last week, the Writers Guild of America proposed that while writers may use ChatGPT in the way people usually look up Wikipedia pages — as a reference source — they may not be assigned AI-generated material to adapt. It also stated that AI scripts could not be considered “source material”, a move aimed at protecting writers’ compensation, residuals, rights and credits. According to Bollywood writers, however, the potential for exploitation is much higher in Indian film industries.
“A lot of directors want to take credit for being writers. If they see potential in the seed of an idea, instead of hiring a few writers, they’ll have an assistant feeding in prompts to ChatGPT and forming the answers into scenes or dialogues or even a whole script,” Quadras explained. “At the end of the day, the industry might not care because they require productivity in terms of pages churned out.” Actors known to interfere in the scripting process could use a mashup of ChatGPT-provided scenes and storylines to put together a pitch for producers, who are then likely to greenlight it because of the star’s involvement alone. As the software evolves, producers could also point to its increased use as justification to pay their writers less. “We won’t be competing with AI, we’ll be competing with people who use AI,” said Quadras. Instead of giving writers working on a web show four months to hand in three to four drafts of 10 episodes, producers might ask for a first draft and use ChatGPT to make substantial tweaks, reducing the writers’ credits. On the other hand, Kadav points out that while five years ago, a studio would give a writer or director notes directly, there is now an “army of people” between the studio and the director combing through data and providing inputs on what the film’s opening should be or what should happen within the first 10 minutes. If anything, the evolution of ChatGPT could put these middlemen’s jobs at stake.
And don’t discount the rate at which the software is evolving — Kallachirippu (2018) and Kaiyum Kalavum (2022) director Rohit Nandakumar said that while he began work on a feature film with ChatGPT as one of the plot points, he had to abandon the script given the rapid rate at which it was growing. “We would complete our research at night and then meet again the next morning to discover that some of ChatGPT’s features had changed in the meantime,” he said. “When we say that ChatGPT might impact the industry in the future, that future is not five-six years from now, but five-six weeks.”
Of the many Indian writers and directors I’ve spent the past few days asking ChatGPT to imitate, the only one whose style it seems to have an uncanny grasp of is Rohit Shetty’s. Andhadhun as re-envisioned by the director becomes the action thriller “Andha Dhoom”, which is an inspired title. The opening shot: “A sleek black sports car zooms through the bustling streets of Mumbai, its engine roaring like a lion on the prowl.” The climax: Akshay Kumar and Ajay Devgn face off in a “tense piano duel as the fate of Mumbai hangs in the balance.” Admit it, you’d watch that.
However, put in a prompt for a Shankar film as an intimate chamber-drama, and ChatGPT flounders. It serves up a generic conversational scene between childhood sweethearts whose parents don’t approve of their relationship. There’s the chamber drama, but where is the Shankar? Vinnaithaandi Varuvaayaa (2010) in the style of Lokesh Kanagaraj produces a conversation that could be taken directly from the film, with none of the Kanagaraj tropes of surprise weapons or hidden drugs. Ask for a list of Sriram Raghavan film tropes and ChatGPT’s answers lack the wickedness of his writing, but have the key ingredients — amoral protagonists, unpredictable plot beats, non-linear narratives, dark humour and homages to classic films. All fair, but can the essence of a filmmaker be distilled to just the tropes they play with?
“A story is not a flowchart,” said Kadav. “We can use a flowchart, a pattern, a three-act structure to build upon, but the story itself needs to be innovative, transcendental, spiritual. Writing a story is not a product-solving exercise, it’s a soul-searching one. That translates to the viewer.” She also points out that with the glut of streaming films and shows, ChatGPT-generated content will only add to the clutter, prompting the need for human curation and appraisal.
And there’s always the software’s potential for plagiarism since ChatGPT regurgitates existing data instead of creating original material. Kadav, who’s tried to use the software to proofread her emails and structure her statements of purpose, finds it unreliable. She often has to rewrite the generated text, doubling her work.
At the moment however, Quadras sees potential in ChatGPT as a research tool. “It’s the ideal assistant,” he said. “Whenever I’m discussing a story with my collaborators, we always discuss ‘what ifs’ and take a few days to come up with the best angle. But ChatGPT can give you a thousand what-ifs in a few seconds. If I type in, ‘Give me options for types of crimes that can happen within this story,’ I’ll have those in a fraction of a minute. It’s preferable to spending days on research.” He’s played around with the software, asking it for action, crime thriller and horror prompts to see if it does one genre better than the others. Still, he wasn’t impressed by the “unexciting” opening scenes and preludes to films that ChatGPT delivered, reasoning that he could have done much better.
Likewise, Nandakumar plans to use it for what he calls “vomit drafts”, the first drafts that he finds the hardest to write. Give ChatGPT a bunch of ideas, he reasons, and the screenplay that it generates could be a useful foundation on which to build a story. “It makes me anxious, but this is a good anxiety to have,” he said. “Writers can be lazy. But when you know there is a machine that will never be lazy, you will work to keep up. That fear is a motivator rather than a destroyer.”
With inputs from Harshini SV.