AI & writing Part 2: A failed ChatGPT experiment and a few glad authors

AM Gautam carried out a similar exercise. He was working on a horror story set in Haridwar, “where the Ganga river is a malevolent entity.” He asked the tool to create a 3,000-word story based on the following input.

Input: The river was held sacred by millions of people, but it had always scared Aneesh a little, especially after sundown. During the daytime, the Ganga-jal reflected sunlight like mirror shards, and pilgrims splashed around within the safety of solid steel pipes fixed deep into the ground or held on to chains fixed in the ghaat’s steps when they submerged themselves completely for a dubki. The submersion, supposedly, washed away all sins, assorted crimes, and misdemeanours. He did not set much store by the religious nonsense, but he remembered how much he used to appreciate a dip in the ice-cold water on hot summer days.

Output: As dusk settled over Haridwar, Aneesh found himself drawn to the ghats despite the chill that ran down his spine. The fading light cast long shadows, and the once bustling ghats now stood deserted, the echoes of daytime rituals replaced by an eerie stillness. A distant chant carried on the cool breeze, a reminder of the day’s devotion now fading into the night. Unable to resist the allure of the river that held both fascination and fear for him, Aneesh wandered along the ghats. The moonlight shimmered on the water’s surface, turning the Ganga into a silver ribbon that snaked through the heart of the ancient city. He could almost hear the whispers of the river, a haunting melody that seemed to beckon him closer. […]

The tool rendered an 821-word response. To this, Gautam concludes, “It’s not 3,000 words, because that’s just how GPT rolls, sometimes it follows orders and sometimes it doesn’t. That may actually be the most human thing about it.”

Give and take

Tanuj Solanki, author of bestselling thriller Manjhi’s Mayhem (Penguin, 2023), says that if his books are used to train large language models (LLMs) like ChatGPT, then these models should offer him something meaningful in return too.

“There has to be a way for me to get something personal out of it. When Facebook took my data, at least I got a nice little service. Now if large language models (LLMs) want to take my writing, then they should give me some service specific to me for free. Otherwise, it’s extractive. Alternately, if big publishers get into a multi-billion agreement with OpenAI and decide that all writing will be available for ChatGPT, then that money needs to trickle down to writers. These are all legal or operational problems. I’m not scared of the technology; I’m scared of how the legislation will fail to upkeep some very basic principles of consideration for work,” Solanki adds.

ALSO READ: AI & writing Part 1: Four authors on why human writing is unlikely to replace LLM-based predictive writing

Shruti Buddharavapu, who has written a book for young readers Mother Steals a Bicycle and Other Stories (Tara Books, 2019), along with Salai Selvam, “thinks of [her] work and non-work writing identities very separately, which gives [her] an interesting outlook.”

Buddharavapu says, “Personally, as a writer, I don’t think it’ll threaten anything in the long run. This ‘long’ could be beyond my lifetime, so maybe I won’t see it. But eventually, the percolation of AI-generated content will come back to put human-written content at a premium. Something only available as an elite, premium-class marker in the future. Problematic, of course, and part of an old cycle that AI is only a blip in, but human writing will survive. Maybe alongside AI, maybe replacing it.”

“As a professional,” she continues, “who coaches analysts and researchers in business writing, I do worry about the relevance of my job. It’s just the kind of profit carrot that organisations can’t help but run behind. You know, to think of it fairly, say my work is used to inform GenAI as it is right now. Would it use my name? Would it inform people this is ‘in the style of’ Shruti? Maybe I’d reach more people than I have or will have organically as a writer. Maybe it’ll be just the thing that pulls my work out of oblivion. Without my name? I’m not sure. It’s quite like death. A lot of people worry about what they’d feel after it. I won’t be hurt by what I never find out, but how tragic when I think about it. An amputation, a loss that cleaves a part of you and codifies you forever in an algorithm and no one is to know.”

Consent and compensation

A.M. Gautam, however, sees the ChatGPT phenomenon as unethical. “Honestly, because these tools are not toys that are just going to be used for the amusement of curious Internet users,” he says. “They are going to be used for profit by their customers, and companies like OpenAI are going to be making money off of them. Everything written up to this point in human history is the raw material — the training data — used by the AI programs and it sounds absurd to me that the producers of this raw material, i.e. ‘the human writers’, should not even be asked for consent, let alone paid their due. Over the last few years, especially in corporate setups, people have been encouraged to write like machines. These efforts are in service to the all-powerful algorithms that make some content appear above other content in search result pages on the web (the principle behind SEO).”

He adds, “It has been happening in marketing departments that want their content to dominate the content of their competitors, in news organisations that want their version of the news to take priority over others, and other places that refer to writing as ‘content’. When it comes to writing like machines, i.e., following the rules of algorithms about keywords to be used, word limits to be followed, divisions to be made with a certain number of headings and subheadings, then sure, machines will do a better job than people. My particular skillset i.e. telling stories that reflect the world through lenses of defamiliarisation, however, requires me to be constantly on guard against mechanical writing and that’s not becoming redundant on account of AI anytime soon."

"That said, there is some cause for concern — if enough people start overestimating the capabilities of tools like ChatGPT, then this overestimation itself could make writers redundant. Unless AI research takes a completely different direction, which doesn’t seem probable given that the current direction is quite profitable, it is unlikely that I will be getting anxious about computer programs replicating my work. They can plagiarise it, and reword it beyond recognition, but I am not sure what, if anything, could be achieved as a result.”