I’m unleashing a chatbot of me. She has a few of my quirks but can go rogue

Companies using AI to replace human creativity is just a spicier version of endumbification. I suspect the illusion of efficiency will come crashing down when it’s discovered that AI can only imitate human ingenuity. Because how will a machine ever know what it really is to live inside a decaying, hormone-doused flesh closet, defying death daily to do the most ordinary things, as well as the most fabulous? It’s the uniquely human tension of sentience and senescence that seems to me to be the source of some of the greatest – and some of the most necessary – leaps of creative innovation.

So, as a person, a writer and a businesswoman, I am trying to take the path of least endumbification. I have been experimenting with AI tools, which I treat like interns in need of strict supervision. I use a range of tools, from note-takers to large language models (such as ChatGPT), audience-identification tools, and web-scrapers that purport to measure “sentiment” on any given topic. I’ve learnt that they’re good at some things, not so good at others.

ChatGPT is not just prone to cliches; as a predictive language model, it is hard-coded to deliver them. Unlike an intern, it has no “spark” and no personality. It’s terrible at humour, not great at insight, and never has an original idea. The note-taker is unreliable at identifying priorities. The sentiment analyst often struggles to understand tone. But overall, I’m a fan. AIs can save time and add quality. And, according to a recent survey, 58 per cent of Gen Z is using AI.

Since my company hires interns, that means I’m supervising interns who have their own AI “interns”. Teaching myself how to supervise AI helps me teach them.

So I’m not an AI novice. But having an AI bot of myself is fraught with risk. Bots can be bullied into saying outrageous things by persistent users. While Yell at Parnell has been told to look for its answers in my catalogue of columns, when it doesn’t have an answer, it has been known to go rogue in our beta tests, vomiting up the internet like a credulous teenager.

Loading

The process of testing it and refining it with Shaun has been instructive. I’ve learnt a lot about how to intelligently limit datasets – mine is under strict instructions to consult its source documents (my columns and a few other bits and pieces) – before making anything up. I’ve also learnt how sensitive bots can be to their prompts – so eager to please that they end up overdoing any instruction. Add a smidgen of snark and you get a gallon of snide.

I didn’t want mine to be a wowser, and now it offers chatters a dirty martini at every opportunity (cheers!).

Still, it seems a worthwhile experiment in extending engagement with readers. This is obviously just the start of what AI will make possible – a toy, because children learn through play and we are in our infancy when it comes to AI.

Shaun and I remain unconvinced that AI will ever be able to combine knowledge and experience with a sense of irony and fun to write a really good column. And even if it could, I have a hard time imagining that readers would be interested in the opinions of a sycophantic cliche machine. There’s a very human relationship between writer and reader, evident in the letters pages and comments and, frankly, even in the personal attacks I sometimes get on social. It’s precious. And anyway, you can’t launch an ad hominem attack without a hominid to target.

Parnell Palme McGuinness is managing director at campaigns firm Agenda C. She has done work for the Liberal Party and the German Greens. This is the link to her chatbot, Yell at Parnell.