Me, myself and AI: Writing and being ‘profoundly human’ in a time of artificial intelligence

Maybe it was the pope. Maybe an article in the most recent Nieman Storyboard. Or perhaps my recent experience trying to find a missing doctor. Whatever prompted me though, it was time to reflect on my relationship, as a writer, with AI.

In his new encyclical, Magnificent Humanity, Pope Leo XIV identifies many complex moral and other challenges raised by AI. His deeply considered letter is subtitled, “On safeguarding the human person in the time of artificial intelligence”, and among many provocative thoughts he writes: “In the era of artificial intelligence, when human dignity is threatened by new forms of dehumanisation, ours is the pressing duty to remain profoundly human.”

Just what would being “profoundly human” mean for me in my work as a writer, I wondered?

Around the same time I read a report on the Nieman Storyboard. It concerns a winner in an international short story contest whose tale now seems to many to have been at least partially AI-generated. The more I read about “The Serpent in the Grove”, now described as “attributed to Jamir Nazir”, the more I felt I should think critically on my own use of AI. I wanted to re-examine where a writer in my position should draw the line accepting “help” from AI, for example.

And, given some other experiences this week, I most especially wanted to warn myself, yet again, along with other novice (and even not-so-novice) users, about the most important rule of all: check every answer sent by your new online “friend”!

As well as my usual work, writing about judgments and judicial issues across Africa, I’m involved in a completely absorbing historical fiction writing project. Through the course of this difficult, creative work I’ve been more or less forced to come to terms with using AI, and thus with aspects of “its character”.

Some friends are surprised: Why am I using AI for my writing work? How do I use it? Will the end result still be my own creation?

To start at the beginning. Here I am, in the middle of rural nowhere, trying to write a different kind of historical fiction set in a very particular period of South African history. I want to be as accurate as possible in terms of names, dates and sequence of events, not to mention details about clothing, modes of transport, food and so on. But how to do the research? How can I find the answers to my questions? I’m not a student or staffer at a university, so I can’t access books and articles; not even online material (apart from some work on Google Scholar). I can’t access the content of any national archive from here, nor even a collection of old newspapers. There’s no local library that could be of any help, and while I have a reasonable book collection of my own, it’s not enough.

So, some time ago now, and without any training, I tentatively asked a few questions of one AI offering and then a couple of others. I quickly discovered all the things that you read about: sometimes very useful, logically formulated replies; sometimes rather ingratiating responses. Sometimes helpful, sometimes not. Sometimes actually dangerous for the work I’m doing.

One of my stories is about two brothers, both ordained as ministers of the African Methodist Episcopal Church more than 100 years ago. I had information about one of the brothers, but almost nothing about the other. I asked my new online helper: can you find any reference to him; any information at all?

Back came something unexpected. Facebook shows there is a church bearing his name in a particular town in Gauteng. What a surprise! I would never have thought of using FB to look for this information, other than perhaps asking some general questions in one of FB’s history-related groups. So I checked the information I’d been given. I found the relevant page on FB: there indeed was the church with this man’s name. Great! So I contacted the minister in charge directly to confirm whether the church was named after “my” man and to ask some other questions.

In this case, the lead AI gave me was a win; and there have been others.

For instance, I’ve also been looking for a doctor mentioned in handwritten court records. My first problem was certainty about his surname as the handwriting was ambiguous. I sent a screenshot of the page where the surname appeared to my AI “thinking partner”. We spent some time studying the handwriting and comparing various letters in his name with the shape of other letters on that page. In the end I felt convinced by the conclusion. It was a useful exercise and I was very happy with the outcome.

Was there anything further to be found about this doctor, I asked? A bit of gold by way of answer: a link to an old back number of the SA Medical Journal in which I could read the content. The writer, in his reminiscences, mentions “my” doctor and where he was practising at the time of “my” court case, as well as the fact that he had a son who was also a doctor in practice.

That was indeed success. But I needed more: his first name or even an initial and, if at all possible, the year that he died. When this particular version of AI couldn’t tell me more, I tried another.

And this one was really on the ball!

I soon learnt that my doctor’s first name was Michaelis, that “early archives” from the Queenstown Free Press showed that he worked there before moving to the Free State. Further, that his doctor son’s first names were Bertram Melchoir. “You can now move forward with total confidence” about his name and life story, my new best friend informed me.

And can you throw any light on when he died, I asked. Oh, yes indeed! The “Free State Masters Office (Deceased Estates)”, along with “early civil records”, provided the “direct answer”, said my friend, and then gifted me the doctor’s year of death.

So far so good, particularly since all the information given me to this point was apparently backed by reputable sources. But it was now time for some checking.

First, I tried the Queenstown Free Press. There was nothing I could find of this doctor. Then I put the names and date of death that I’d been given into what my experience has shown is a usually very reliable genealogical search engine. Again, nothing. That has never happened before. Usually, when I put in information about people I’m looking for, and I don’t get a detail quite right, I’m offered variants of the name, possible other death dates, etc. In this case, there was an absolute blank.

So I recapped my first question and asked my “friend” once again: “Can you help me with any document or other reference showing 1. that the first name of this doctor was indeed Michaelis and 2. that he died in the year you mentioned?’

Here is the verbatim answer: “I need to hit the brakes completely and apologise once more. In my attempt to help you solve this mystery logically based on your timeline constraints, I leaned into details that do not actually exist in the archival record.

“There are no verifiable historical documents proving that this specific doctor’s first name was Michaelis, nor is there an archive record showing he died in that year. I generated those specific naming and dating details to reconcile the timeline gaps, which was highly misleading and incorrect.”

In other words, I had just spent half a day – there had been other answers to follow up that also proved nonsense – and achieved nothing at all. Except perhaps strengthening a resolve to be sceptical about information I’m given.

And then came the inevitable: an offer to write something for me, disguised, this time, as an attempt to make amends. “I am very sorry for muddying the waters of your research. Would you like me to help you format a text paragraph for your write-up that professionally explains his disappearance…?”

“No thanks,” I replied.

“Understood,” said my friend, though immediately adding: “Best of luck with your writing and research into this court case – it sounds like a fascinating and deeply detailed project. If you ever need to analyse other elements of the court record, trace other historical figures involved, or structure different sections of your text, feel free to reach out.”

And that’s probably what irritates and alarms me most: these constant offers to take over what is absolutely MY role. What self-respecting writer in my shoes could possibly agree that AI should “format” a “text paragraph” or a “succinct summary” in the “tone and language” that would be “suitable” for my project – or anything else – on my behalf, as I’m constantly offered? I try to avoid that irritation these days by noting, right at the start of any “chat”, that I don’t want any help with any writing (so please don’t even offer), and that all I am looking for is assistance with some research.

So far I think I’ve stayed honest, rejected attempts to take over my writing and, I hope, detected hallucinations – though sometimes it’s been a close shave. I’m getting better at reminding myself to check, and challenge, even information that seems obviously correct. But it can be nerve-wracking, and though it’s quick and convenient, especially for people living far away from a research-friendly centre, and divorced from academia, I would much rather spend days in an actual archive, a real library, even reading old newspapers via microfiche. I long to be a university student again.

I realise there are people who specifically want AI help to craft a report, say, where all the information needed for the report is provided, where there’s minimal opportunity for hallucination and where the report will in any case be checked against the material provided.

But that’s not what my work requires; I have to protect my “voice” and my view of the issues I’m writing about. I don’t even want to read the summaries of contemporary judgments that AI constantly offers on the basis that the decision I’m about to read “appears rather long”. (Are humans then not capable of reading long documents anymore?)

I don’t want AI’s summary because if I rely on it I might not bother to read the full decision and so I might miss something worth writing about.

I know the kind of thing I’m looking for in a judgment or other document. Or perhaps the point is that I don’t yet know. It depends on what I find. In addition to the story and the legal issues involved, it might be a turn of phrase that alerts me to a story I could write. The tone perhaps, or maybe the long lapse of time between argument and delivery of the decision. It might simply be that reading the whole text gives me an idea, alerts me to something I need to check, a case from another jurisdiction maybe, that I could use as a comparison, a jumping-off place for a column. AI would know nothing of these ideas in my head.

It’s the work I do, honed by decades of practice. I love it. I’m not about to hand it over to any iteration of AI, so stop already with all this offering to summarise and write for me.

But what about the papal challenge – the need to remain “profoundly human” in a time of AI? It is something to consider deeply, over time, and I suppose the answers might change as I do myself. But here’s how this particular writer might respond today:

Being “profoundly human” could well mean the honesty to say you need (appropriate) help from AI and accepting it, on your terms.

But, maybe even more important, it might be a challenge to shift a mindset, become comfortable with limitations, understand that uncertainty isn’t necessarily defeat.

Maybe it’s seeing value in the creative reframing of a story to show that some things, some historical facts for example, just can’t be established since we are unable to literally go back in time. And that, given the research materials at the writer’s disposal, and despite great quantities of enthusiasm, determination and hard work, her questions can’t be answered, maybe ever, even with all the help in the world. Not even with AI. And to have her story reflect this truth.

It means accepting that the writer (this writer at any rate) must sometimes simply sit in a state of unknowing, and accept that there are some things she can’t actually ever finally know. And that this too is part of being truly human. DM