Earlier this month, a short story titled The Serpent in the Grove won the Caribbean region section of the Commonwealth Prize, run by Granta magazine. Yet critics quickly accused the story, submitted by the Trinidadian writer Jamir Nazir, of not being the work of human hands.
The story, they claimed, was full of the linguistic signatures of generative artificial intelligence. AI detection tools reportedly yielded the same verdict. (Personally, I don’t see it – but then, Nazir and I share a lot of the writerly tics such as em-dashes and semicolons that AI is fond of.) The fact that Nazir had published little else, and had discussed AI heavily online, didn’t help.
So, was The Serpent in the Grove written by a robot? The prize organisers admit we may simply never know. Coming in the same week as a controversy over Nobel laureate Olga Tokarczuk’s use of AI in her writing process, this uncertainty might prove hard to live with.
Turing Tests and electric chess
If Nazir’s story turns out to be AI, we might say that by winning a literary award, AI just passed a very particular Turing Test: an encounter where a person cannot tell if they are speaking to a person or a machine.
But AI passes that test in classrooms, courtrooms and boardrooms every day. AI-generated books are nothing new, either. James Daunt, the CEO of book seller Barnes and Noble, recently said he’d be comfortable stocking AI-written books so long as they are clearly labelled.
(Daunt went on to say he doesn’t think there’s much of a market for such titles – despite the fact that Amazon contains thousands of them.)
But it’s one thing to write an essay or a legal submission, or even a self-help eBook. It’s quite another to write an award-winning piece of literature.
A better analogue might be the moment in 1997 when the IBM supercomputer Deep Blue beat world chess champion Gary Kasparov in a six-game match. At the time, Kasparov’s fairly narrow loss was seismic; today, chess engines are drastically better than even the strongest human players.
Is this where we are heading? Could AI not simply replace the best human authors in literary skill, but surpass them? To answer that, we need to stop and consider what we mean by ‘writing.’
Writing vs writing
In 1980, the American philosopher John Searle published an influential thought experiment known as the Chinese Room. Imagine that you’re sitting in a room where people outside keep slipping bits of paper with Chinese characters written on them under the door.
You don’t speak or read Chinese, but you do have an incredibly detailed rule-book that tells you things like ‘if you’re given this set of characters, send back a piece of paper with the following characters on it’. You don’t know what any of the characters actually mean, but by following these rules you’re able to reply correctly to the person outside. If they ask you (in Chinese) what 1+1 is, you’ll send back the Chinese character for ‘2’ even though you have no idea what you’re saying.
In this example, are you speaking Chinese? Many of us would say ‘no’. To speak a language you need to understand it subjectively, not just know how to manipulate symbols in the right way. It requires conscious awareness. And if that’s the case, a computer cannot speak Chinese either – or any other language. It can only manipulate symbols. That’s all that AI, which is essentially a prediction machine, can do too.
Others might disagree. If the machine is transmitting the correct information, surely that’s speaking? What does consciousness add that the computer isn’t already doing? This view (which Searle rejects) is what’s known in philosophy of mind as functionalism: if the computer produces the right inputs and outputs, it’s communicating. Shouldn’t that be good enough?
Communication and communion
If writing is simply the production of readable text that conveys information, AI can certainly do that, just like you could in the Chinese Room case. Not everything we write is high art. Sometimes we just need to leave a note on the fridge to remind us to buy more milk, or text someone step-by-step directions. Sometimes we just need a bit of text that fulfils a particular task, and GenAI can produce that for us. All that matters, on this way of seeing things, is getting the right output.
But compare leaving a note on the fridge to writing a condolence card. Here, your subjectivity is decisive. No message comes from the heart if it hasn’t passed through your mind along the way. The process matters as much, if not more, than the outputted text. It is, quite literally, the thought that counts.
For many of us, the disappointment in finding out a text is AI generated is that it robs us of that connection to another mind. When we read a short story or a novel, we are not simply aiming to be entertained, but to experience a brief moment of communion with another consciousness. We connect, in some more than metaphorical way, with Olga Tokarczuk or Jamir Nazir; we enter their minds, and they ours. Without another consciousness at the other end of the words, there is no encounter, no resistance, no genuine otherness.
Another Polish writer, Cezary Jan Strusiewicz, recently declared that ‘AI is a sin against writing’. What is sinned against is not the noun ‘writing,’ but the verb. Writing, as a process, should be hard. It should be slow. It should not be comfortable. But the reward, we might add, is the everyday miracle of knowing and being known by another mind.
A world of AI words and stories, conjured up at will from nobody in particular, would be a solipsistic disaster. However pretty it might look, it’s solitary confinement.
Read more on ArtsHub:
- You write like AI, people think it’s AI generated—what should you do?
- We’ve achieved a successful AI copyright ruling – but ‘significant’ advocacy is still needed
- Copyright win for Australian artists vs AI – but how long can we hold off the bots?
