The AI future from a 70-year-old Roald Dahl story is here – are you afraid?

It was the mid-’'90s when I became obsessed with Roald Dahl. His twisted humour, dark allegories, and grand imagination hooked me, a lonely, bullied child, struggling to learn the ways of the world, and an alien language in Bengaluru. Dahl helped me escape reality. I became James Henry Trotter, flying across the world in a delicious giant peach – a fruit I had never seen, let alone tasted. I was envious of Charlie Bucket, and dreamed of going into Willy Wonka’s fabulous chocolate factory. I cheered on Matilda Wormwood against her bullies, knowing fully well that I would never be a genius like her.

Then, very briefly, he just vanished from my life. Instead of the works of Dahl, joyless science tomes written by Morrison & Boyd (Organic Chemistry), SL Loney (Plane Trigonometry), and the much-dreaded IE Irodov (Problems in General Physics) took over my bookshelf and imagination, as I tried to enter the hallowed halls of the Indian Institutes of Technology. That misguided attempt, of course, did not work. I licked my wounds, went to a lesser engineering college and went about life in a student hostel.

Like the relapse of a malignant disease, Dahl again entered my adult life. His yarns, written for readers of Playboy and The New Yorker, were even more perverse than before. They were not called Fantastic Mr Fox or George’s Marvellous Medicine, but had more evocative titles like Switch Bitch and Lamb to the Slaughter. There were no euphemisms – he wrote of swingers and of swindlers, of passion and of revenge, of war and of colonial intrigue. Each story seemed exquisitely crafted to give me a delicious chill or a sleepless night. Sometimes I shrank in fear thinking of dying abandoned in a stuck elevator (The Way Up to Heaven), and sometimes I fantasised about plotting and mentally executing a diabolical revenge over my latest bully (Nunc Dimittis). Then of course, I snapped back to reality and the next thermodynamics assignment.

One of Dahl’s short stories particularly resonated with me, an engineering student who fancied himself a writer. It was called The Great Automatic Grammatizator. Published in 1954, this short story is about the titular fantastic machine that can write entire stories, newspaper articles, and even novels. The brainchild of a genius misfit called Adolph Knipe, the machine takes in rudimentary stylistic and content inputs from its operator to churn out a piece of writing instantly.

The storyline, the tone of the piece, the target media outlet it would go to (Readers’ Digest, The Saturday Evening Post or the Ladies’ Journal), and even the sentimental and emotional content can be chosen by the operator of this fantastic machine, just by adjusting a few switches and levers. The Great Automatic Grammatizator works because the English language has a set structure that is almost mathematical in nature. Thus, a machine can be programmed to write human-sounding output based on various pre-specified styles.

Knipe and his collaborators then churn out articles for various media houses at industrial rates, destroying independent writers who are unable to keep up with their sheer scale; they either go out of business, or are reduced to becoming mere operators of the Great Automatic Grammatizator, thus further entrenching Knipe’s monopoly.

Dahl’s tale was an allegory: it alluded to the commodification of writing in media outlets, where independent writers were treated like pawns by publishers. Was the name Adolph Knipe a snipe at publisher Alfred Knopf? Possibly. But that is not the point of this essay. Neither is this an ode to Roald Dahl, much of whose work is today viewed as misogynist and racist.

The Great Automatic Grammatizator was eerily prescient of the future, something seen a full seven decades after its publication. Generative artificial intelligence and its associated large language models work much like Dahl’s imagined machine. I do not know if Dahl was a linguist, but many linguists do posit that languages have a mathematical structure, which, if properly understood, can be used to automate the generation of text.

Trained on millions of human-written essays, stories, technical documents, these large language models, the cornerstone of generative AI tools such as ChatGPT, Bard and Perplexity, can generate computer code, consulting reports, legal briefs, academic literature surveys, and even creative outputs like poetry, stories, essays and art. Generative tools like Dall-E, Grok and Meta AI go one step further, generating graphics and videos from ordinary English-language prompts.

Generative AI has its fair share of both enthusiasts and opponents. Large-scale enterprises naturally see the potential to automate some human grunt work, and, in business consultant language, streamline processes and rationalise costs. Ironically, the world of business consulting is one of the greatest proponents of this technology. Not only is this the new wave on which multi-million-dollar contracts can be landed, but these large language models are ideally suited for the boilerplate yet jazzy slide decks that the best-paid management consultants regularly generate.

Publishers are being inundated with AI-generated book proposals and academic journals have published much work with tell-tale signs of AI-generated content. In a delicious twist of fate, the children of Glasgow were treated to a shabby spectacle called Willy Wonka, with AI-generated characters and farcical dialogue based on Dahl’s own classic – Charlie and the Chocolate Factory.

I landed my first job in the software behemoth called Oracle in 2005. My centrally air-conditioned workplace was in stark contrast to college. I had a shiny Dell desktop and dedicated landline – unlimited local calls – all to myself, along with access to a pool table, carrom boards, table tennis, free tea and coffee. Unlike the lesser software companies that restricted employees’ internet usage, Oracle did not care what we surfed on its high-speed leased line, who we chatted with on Yahoo! Messenger, or whom we “scrapped” with on Orkut.

We had team-building exercises in luxury resorts, and work lunches at fancy restaurants. There were no strict clock-in or clock-out processes, or timesheets to fill. Fresh out of campus and poverty, Oracle’s shining tower seemed no less than Dahl’s wonderful chocolate factory, with the mysterious, yet flamboyant CEO Larry Ellison being my new Willy Wonka.

None of these perks were free of course – my job was to write and debug code for a well-known banking software application. This is where it got interesting. Much of the Java code could be auto-generated. I just had to tick a few boxes to specify the layout of a user interface design element, its expected behaviour, and final destination in the database, and a back-end application would generate snippets of code – functions, loops and explanatory comments – that were perfectly formatted, and production-ready.

My job as a coder was simplified to that of ticking a few boxes, very much like an operator of the Great Automatic Grammatizator. I felt less like the round-eyed kid Charlie Bucket, and more like an Oompa-Loompa (a dwarf-like figment of Dahl’s fecund but racist imagination) who slaved for Willy Wonka’s chocolate factory. I soon quit this job to join graduate school, looking for more creative ways to use my energy.

My next tryst with generative AI was with a software called Wolfram Mathematica. Graduate-level mathematics is often complicated. Calculus, linear algebra, complex analysis, vectors and tensors, matrix inversion and differential equations are daunting subjects even for people smarter than me. Mathematica is a very useful, but prohibitively expensive tool – it does a lot of complex-looking math in seconds, complete with Greek letter notation.

I do not have access to this tool today, but Mathematica opened up my eyes to the possibilities of generative AI, a term that had yet tp gain currency. The company operates a free version of its powerful engine, called Wolfram Alpha, that is arguably a precursor to the modern generative AI tools of today.

A decade passed. I went into another corporate job and quit. I acquired a doctoral degree in management, joined academia and encountered many waves of keywords. Social Media, Big Data, Machine Learning, the Metaverse, cryptocurrencies, the sharing economy, neuro-marketing, edu-tech – many trends and sometimes fads rose and fell. A pandemic wreaked havoc on one and all, and soon I found myself in a new job, in a new place called Ahmedabad. The Covid-19 pandemic had just subsided, as had the various academic fads associated with it. But people were now back on the streets, and the self-declared Covid experts had moved on to the next big trend – generative AI.

Generative AI is a literal realisation of Roald Dahl’s Great Automatic Grammatizator. Can it write better essays than the average college sophomore? You bet it can! ChatGPT is already the go-to tool for many students looking to save time and energy. Can it summarise a complex research paper into an engaging podcast – a conversation between two white people? Google’s Notebook LM does just that. Can it make you look more erudite on LinkedIn? Subscribe to LinkedIn Premium (first month free) and see the magic – your AI generated posts can “get up to 35% more distribution and 30% more reach” – LinkedIn’s claims, not mine.

In the era of generative AI, creativity is no longer a bonus. Several types of ad copy (especially in digital marketing) can be generated instantly with a few prompts to ChatGPT, Dall-E and Grok. Meta encourages you to generate images using its in-house AI tool, which you can then post on Instagram.

For every clip I see of supreme creativity by artists at Disney or Warner Brothers, a friendly tech bro on my X timeline offers a graphically rich, yet creatively vapid clip by Dall-E, that he promises is the future of this industry. He predicts that the arts are on their way to extinction, with prompt engineers replacing artists.

Meanwhile, the software industry where this tech bro works gears up to exploit generative AI to streamline processes and rationalise costs, making thousands of jobs like his, redundant.

As an academic, I despair. Many assignments, carefully designed to teach market research, critical thinking and effective business communication, are now outsourced by students to ChatGPT, with results that are barely readable. Many of them have wrong or irrelevant data exhibits. There is no evidence of any significant human effort.

And yet, there are generative AI tools that can grade my students’ work. They provide verbose, and yet largely meaningless feedback, to output that is itself verbose and largely meaningless; the large language model essentially grades its own output, like a technological Ouroboros (a serpent from Greek mythology that devours its own tail).

“Am I a Luddite?” I question myself. I grew up in a different time and milieu, where a teacher censured me for printing out the banner of our school house’s wall magazine (using a dot matrix printer) instead of handwriting it. A college professor insisted on hand-drawn graphs and regressions instead of using Microsoft Excel. In graduate school, I sometimes felt guilty for using Google Scholar to instantly download academic research articles, instead of reading the physical copies of journals kept in the library. But now, something feels different. “Generative AI is different,” I convince myself.

I mull over the future. Maybe generative AI can reproduce the wry wit of PG Wodehouse and the parochial pulp of Chetan Bhagat, but will these works be read by the Oxonians and IITians of our day? Or will they delegate it to a large language model to be succinctly condensed into bite-sized executive summaries?

Maybe generative AI can replicate the witticisms of Douglas Adams and Mrs Funnybones – the writer also known as Twinkle Khanna – but will there even be a new generation of bibliophiles to read such literature? Maybe generative AI can crunch high frequency stock market and cryptocurrency data to guide us on investments, but can it deliver financial insights with the panache of a Warren Buffet or Ankur Warikoo? Food for thought indeed.

My mind wanders further. Can generative AI prevent the Israeli massacres of thousands of women and children in Gaza? Can it bring back the Kohinoor and Vijay Mallya from the United Kingdom? Can it solve global warming? Can it cure AIDS? World peace?

The sound of music interrupts my reverie. It is garba night in my apartment complex in Ahmedabad. Someone with a microphone is belting out a lively song, but I do not understand its lyrics. I don’t need to. It is an invitation to dance. A hundred people are dressed in shimmering finery, dancing away to the beat of a drum. Sweat drips down their happy faces. Festive lights adorn the otherwise bleak courtyard of the residential society. “No generative AI can replace this,” I think to myself. “Or can it?”

Prithwiraj Mukherjee is Associate Professor of Marketing at the Amrut Mody School of Management, Ahmedabad University. Opinions expressed are his own, and do not reflect those of either his employer or of any large language model.