It has come to a point, now, that if we do not sincerely protest against the rising tide of surveillance targeting so-called AI-generated prose, we will be rolled roughshod over pretty soon.
I say this not because people aren’t using AI to generate their words — they clearly are. And there you have it: I have just there dropped the m-dash red-flag, there, (haven’t I?), just so you can detect it as generated by some chatbot!
So, please consider the sentence you just read. It is neither short nor unnecessarily long, and yet it will not trigger, in any casual reader, the suspicion of having been machine-made. Well, it was not! You can accuse it of lacking in sincerity, wit, and originality. But can you really accuse it of not being human? Which is to say: sometimes we try to be absurd; sometimes we reach out for metaphors; and occasionally we try to be rhetorical. Before writing this essay, I read something that left in my mouth the taste of having bitten a bullet. What was worse, it left the aftertaste of arsenic — fermented arsenic, which I had initially mistaken for beer, it being summer and hot where I’m writing from. But it turned out to be poisonous. Xenophobic, in fact! And that is precisely what I want to say. There is a growing xenophobia being normalized against suspected AI-generated prose.
I understand — and as an educator, I genuinely empathize — when one tries to correct the ethical, besides the grammatical, mistakes of students remorselessly submitting AI-generated academic essays and, in some cases, even examination papers. But the culture I am describing is one of hostility, that is just as harmful as the devices it purports to criticize.
In many cases, it is criticism for the sake of criticism. Detecting em-dashes, the rule of three, or the alternation of short and long sentences is no feat of Sherlockian detection. It is plain to see. Some metaphors are designed to be comprehensible; others are not. And AI does a perfectly passable job of producing what human beings are already capable of producing. This became patently clear to me when I ran my own book — Indians in London: From the Birth of the East India Company to Independent India, written in 2018 — through AI detection tools, and found several passages flagged as AI-generated.
I am, of course, implicitly alluding to a recent publishing scandal that appears to have seized the literary world, with people proudly displaying their skill at telling AI writing from human writing. Let me add two personal instances of having been mischaracterized — by people geographically and culturally proximate to me, though not in any emotional, intellectual, or ethical sense.
Since AI entered parts of the world less developed than the first, I have been a strong votary of understanding what generative AI is doing to writing, reading, and cognition. My reasoning was twofold. Even if you consider AI your archenemy — which I do not; I see it as a tool, a knife, a dagger, a shotgun at worst, and perhaps a guillotine, if one finds it worse still, though one that must be wielded by a human hand — even then, why would you not wish to know more about it? And the only way to know is to use it. This position has been straw-manned by people in my vicinity as a defense of AI; which is balderdash!
And, it is voices like mine, I daresay, that get caricatured, much to the peril of those doing the caricaturing, only for our hopes of trying to understand the nuances — or the insurmountable challenges — of the problem.
The second instance involves an experiment I conducted with a ghazal, written in response to a famous Kashmiri poet who is credited, rightly or not, with having introduced the ghazal form to English. Given my admiration for this poet, I wrote two ghazals. One was written entirely by human hand — I have been reading Ghalib since the age of seven or eight, and my engagement with poetic forms is longer than the years of several of today’s poets, for that matter. The second was assisted by generative AI but successively shaped by human hands, eyes, and ears. I sent both to a group of readers to see what they actually thought, and how.
What I received were not, as I had half-expected, accusations of AI manufacturing. Instead, I received responses laced sometimes with the xenophobia, I mentioned earlier, besides utter misrecognition of the human as artificial, and the artificial as humane. Almost all the recipients were originally denizens of developing economies, though several had since settled in the so-called first world — and the more revealing responses came from precisely this group.
Their opinions were not, interestingly, about the ghazal I had openly written with AI assistance. They were about the ghazal I had written entirely myself — in which, I should confess, I had consciously imitated certain metaphorical conceits I had noticed in AI generation. On the AI-assisted ghazal, the comments were either broadly complimentary or substantively engaged, questioning the use of Indianized metaphors where the Kashmiri poet in question typically used Arabic and Middle Eastern ones. Fair enough! But the critics were speaking of this ghazal as if it were human-written — assessing its deficiencies as the deficiencies of a human author, not that a machine. Interestingly, they were comparing it to some imagined tradition in their minds, and when matched against that, my AI-ghazal, which appeared human, didn’t seem to live up to that standard. Groovy!
Now, to the role reversal!
On the ghazal that was, in fact, entirely human-written — mine — the response was the opposite. Without explicitly accusing it of being AI-generated, several respondents treated it as if it were. They said it was almost too perfect and therefore not socially accurate; that it was a little too musical, getting carried away by its own music — as if, I would add, it were some nightingale virtually invented by John Keats. They added that this kind of writing would not connect with humanity. What clearly disturbed them, beneath all this, was that someone had dared to write a ghazal addressing the Kashmiri poet, and had apparently used some kind of artificial tool to do so — when, in fact, they had done nothing of the kind.
It is difficult to extract a clean thesis from these somewhat disjointed experiences. But what is as clear as the case for global warming is this: we cannot say with certainty what AI-generated writing looks like or is supposed to look like or not look like. We can certainly claim to successfully detect it — with AI-detection tools, or with the trained naked eye of an educator or writer — when there is a genuine epidemic of it, as there is in universities and in the blogging arena. But the culture of criticism directed at AI is deeply flawed, deeply exclusionary, and persistently xenophobic, because it deploys elitist yardsticks, and anecdotal ones at that, that are often bereft of logic, insomuch as we will soon need a fallacy-detection tool.
According to this culture of surveillance, any form of perfection is now deemed inauthentic. Aspirations towards good prose — whether Latinate, Victorian, or simply correct in punctuation — are being read as the work of a machine rather than a human hand and heart, while the latter are left to languish in precarity. Meanwhile, an overwhelming majority of the critical commentary against AI-writing rests on flimsy grounds: on hearsay, on the authority of because I say so, on appeals to the sordid fact that “I have been an editor in some sort of Anglo-American publishing house or university press.” Question begging is rife, straw-mans are liberally deployed, red herrings run galore, and there is no universal principle by which to distinguish AI-generated writing from human writing — except, of course, an AI detection tool, which is often likely to interpret the human as artificial, and vice versa.
Which is to say, the self-appointed sleuths are detecting machine-work through the use of a machine, and then claiming, as a human being, to be laying down the humane ethical principles of what good human writing is and what it is not.
So tell me — who is actually using generative AI? The author, or the detector?
The lines are blurred — or dare I say, blurred, smudged, and all too obscured, exactly as between the arsenic and the beer, whose taste I still cannot seem to overcome, anytime soon.
Arup K. Chatterjee 