On futile rage against the chatbots

When AI says it better.

Chatbots keep taking the words right out of my mouth.

Occasionally, when I’m struggling with how to articulate a half-formed thought, I type my muddled ramblings into Claude and ask,

“What am I trying to say here?”. And then, to my dismay, it tells me.

I watch that little orange cursor blink dutifully across the screen, rearranging the contents of my own mind into eerily perfect prose.

“Where would you like to take this next?”, asks Claude, perfectly compliant and diligently helpful as always. I feel the same exasperation that hits me when my cat delivers a dead mouse at my feet in a misguided act of generosity, staring up at me with her huge, earnest eyes.

“Yes, that’s exactly what I meant”, I murmur. “Also, fuck you”.

My turbulent relationship with chatbots brings out the full force of my irrationality. I am not ashamed to admit that it triggers me. It will have me alone in my room yelling "I CAN DO IT MYSELF" at my laptop, in a (wo)man-meets-machine display of pointless fury to rival Basil Fawlty thrashing his car with a tree branch. At a recent EA Global conference, I had a one-on-one meeting with an undergraduate who lamented the fact that most of his classmates are using ChatGPT to complete their assignments. Suddenly, I was 26-going-on-60, denouncing the myopic laziness of ‘kids these days’ with all the Luddism of a parent decrying mobile phones at the dinner table.

“That might help them get ahead now, but it won’t serve them in the future”, I heard myself say in an uncharacteristically shrill voice. “They should be considering their post-graduation prospects” (as I was saying this, I was acutely aware that it literally made no sense given my own single-digit-year AGI timelines).

I have many gripes with AI. My biggest one is the alarmingly high likelihood that it will kill everyone. My second biggest one is the ease with which it does in seconds what I do in hours. I have always taken great pleasure in crafting a well-formed sentence. I love arriving at the perfect articulation of an idea after rounds of near-headache-inducing mental labour. I don’t imagine that Claude gets headaches¹.

---

One evening, I decide to experiment with AI text detection. I spend an hour or so pasting chunks of my own writing into free online tools, followed by AI-generated content on the same topics. To my absolute delight, I discover that, at least initially, it seems like they actually work. The detectors are distinguishing between human and AI text with impressive accuracy, and I am high on my own supply of hopium. Ah yes, the sacrosanct uniqueness of the human voice, irreproducible by the shallow mimicry of simple pattern-detecting machines. The skeptics were right! There is a secret sauce, and ZeroGPT has found it.

But it doesn’t take long for skepticism to rear its ugly head. It does seem unlikely that there would exist some magical tool capable of differentiating between the products of biological and silicon minds. And I already know that AI can generate content far more sophisticated than I’ve been testing the detectors with using just a little more prompting (reluctant though I am to acknowledge this).

There is a Substacker whose work I enjoy. She writes lovely, touching vignettes about small-yet-significant personal experiences like lending a dog-eared copy of a favourite paperback to her mum. It’s precisely the kind of thing that your resident skeptic friend who hasn’t so much as tried an LLM will swear up and down that AI could never produce. I give Claude the prompt:

“Write a touching story about lending a dog-eared copy of a favourite paperback to your mum, in the style of a personal essay on Substack”.

Its first try, The Book That Came Home, is a laudable attempt at the kind of emotional authenticity that I’m after – my detector-of-choice, Quillbot, estimates that it is 51% AI-generated. I take it up a notch:

This second essay is actually quite poignant. Claude has decided to pack an extra punch by giving the mum cancer. It’s a cheap move, but I’m in an emotionally vulnerable place, so it works on me. It apparently also works on Quillbot, which confidently declares that Mom’s Sticky Notes is 0% AI-generated. The rage is back. I feel the urge to write a strongly-worded email informing Quillbot of this false negative, and reminding them of their duty as the final line of defence against the unstoppable encroachment of AI into every corner of the internet, and as one of the few remaining bricks in a wall around the sacred territory of human creativity. But of course, there is no wall (in more ways than one), and the whole enterprise of building one, as I well knew even before my foray into the world of AI text detection, is patently ridiculous.

Two sources of cope

I’ve always loved to write. That AI can do it so effortlessly upsets me enough to inspire this petty blog post. But there are reasons I think this need not be as nihilism-inducing as it sometimes feels.

First, for the time being, there remains at least some skill in prompting chatbots to produce the kinds of content you want. Extremely low-effort prompts result in writing replete with tell-tale signs that AI detectors can spot a mile off. But very good ones can produce writing idiosyncratic enough that, just maybe, there is something of the human user that survives into the final product. For example, Amanda Askell of Anthropic got Claude to generate a whimsical fable about a man locating the deed to some valuable land using a precise arrangement of garden gnomes, making me wonder whether the creative locus of AI text can sometimes lie with the prompter, rather than the promptee² (though of course, Amanda is one of the staff members behind the alignment of Claude, meaning she gets to take a little more credit for its outputs than the rest of us).

Second, and I think more importantly, each of us has always shared the planet with minds better than our own at all sorts of things – all that has really changed is that some of those minds are now artificial. When I was younger, I was part of a children’s choir. One Sunday afternoon rehearsal when we were both around 13, a friend of mine dramatically announced her intention to give up singing because she was upset that some other choir members were better at it than her. She couldn’t see the point in any pursuit that she wasn’t the best at. I pointed out that only one person on Earth could claim the title of Best Singer, and that it would be a great loss to the world if everyone else were to stop singing. If 13-year-old me could grasp this simple lesson, 26-year-old me would do well to heed it now.

However capable AI becomes, I think it will continue to matter that my words are my own – and I will keep writing them.

1. (Probably)

2. Death of the prompter – is this something?