An editor’s take on AI writing — and how to train the humans

Jeff Israely is the Co-founder and Editor-in-Chief of Worldcrunch. A former Time magazine bureau chief (Rome and Paris), and Associated Press correspondent, he teaches at the Sciences Po Journalism School in Paris- and offers his insights on how AI impacts writing - and teaching writing.

For humans who make their living with words, reactions to the first encounter with ChatGPT depended on where we sit along the writing food chain. Novelists brushed it off as soulless. Producers of marketing copy marvelled at its ease and ear for rhythm. Beat reporters pointed out the factual errors.

We editors and teachers came at it sideways, comparing the Machine's output not with our own, but with the wide range of human writing that we are hired to study, judge and improve.

We could thus claim to offer both a marginally more informed and more objective response - and pull no punches: "Probably better than your average hack," the news editor says of ChatGPT. "Definitely better than your average student," responds the college professor.

Having spent thousands of hours of our lives slogging through, breaking down, shining up, sending back, pushing out, rating and grading and otherwise evaluating written prose, what we noticed first about generative AI writing was that it was clean.

It was of course also deeply unsettling: How will we monitor the work coming in for traces of AI's hand? Will we have to adjust our standards and expectations of both speed and quality? And what happens when the algorithm learns to slog and shine and grade better than us?

But much has changed since that initial 'Wow!' moment of late 2022 when we clicked return and watched all the scrubbed copy come pouring out.

Yes, even if that's exactly what we're praying for from a pile of news copy or student essays, clean is not necessarily good - or relevant, or accurate.

With all of the supposed advances that ChatGPT has ushered in, once you scrape past whatever style tricks you try to train it to do, most of the prose is simply not very good at all - and far (far)from the great writing that makes all the slogging worth it.

None of that is stopping our feeds from filling up with all sorts of tips and warnings and promises about where AI will take the endeavour of writing. Attending the WAN-IFRA Congressin May of the world's top news publishers, there was virtually no panel or cocktail conversation where the topic didn't come up.

But my subject here is not what AI can or can't do, today or in the future. What if, instead, we take the ongoing explosion of generative AI (hype and otherwise) as a unique moment of clarity to better understand what goes into human writing, human communication - and why not, human, er, humaning?

ALSO SEE:How will journalists use ChatGPT? Clues from a newsroom that's been using AI for years

Trust the reader

Vernyn Klinkenborgwould say I don't need a transition sentence here. He would remind me to trust the reader, maintain my rhythm and find the next sentence.

I discovered his book Several Short Sentences About Writingfive years ago when I began a side gig teaching a class on writing and editing to graduate journalism students in Paris.

My day job editing aninternational news siterequires a lot of fixing stuff on my own, and the very occasional note or tip to a young reporter about what went wrong in a particular spot on a specific story.

Those newsroom exchanges with writers involve a lot of pointing and grunting, and "Well, in this case," and "Don't do that," and "See what I did there?"

Still, as ham-handed as we old editors may be, at least we can work with the writers who are writing. Every day.

Teaching writing is something else.

If an editor is an auto mechanic with a pen, the writing teacher is supposed to show you how to build a car - and nobody ever showed us!

Most of those who wind up in a classroom arrive with little or no formal writing training, at least a drop or two of natural aptitude and years of just doing it: the writing, the rewriting, the editing and being edited, by ourselves and others.

We try to channel some of that through a mix of examples, anecdotes, exercises and pep talks that might offer some kind of technique or framework or loose guide.

But we also know that the 12 weeks we have will not leave our students with much, and their future writing prowess will depend almost entirely on whether they can muster those same drops of inspiration and buckets of perspiration.

'The writer's real work is the endless winnowing of sentences,

The relentless exploration of possibilities,

The effort, over and over again, to see in what you started out to say

The possibility of saying something you didn't know you could.'

Klinkenborg has a way of weaving the almost-physical mental labour of the craft with philosophical and even spiritual components.

As the title tells us, no matter what story you have to tell or information to transmit, the atomic particle of all writing is the sentence. The book is a tautological wink: a series of sentences, sometimes but not always connected, to help learn how to produce sentences.

'Don't try to distinguish between thinking and making sentences.

Pretend they're the same thing.

'All writing is revision.

Revise at the point of composition.

Compose at the point of revision.'

ChatGPT is also powered by sentences. We've been told that the Machine has been fed billions of them, which are ground up and reconstituted into "new" sentences that are placed in proper order according to the commands of a human prompt and some unknowable mathematical formula. (That's a long, short sentence!)

Shape-shifting sentences

Written decades before AI writing, Klinkenborg warns us against sentences that "volunteer a shape of their own (and) supply their own words as if they anticipated your thinking."

In these last years leading to the arrival of ChatGPT, our internet-powered computers and telephones were coaxing us toward "volunteer sentences" with the auto-complete function.

'The writer's job isn't accepting sentences. The job is making them, word by word.'

The AI writing machine is just a very clever compiler of one volunteer sentence after another.

Or, as someone wrote recently about ChatGPT: it's the average of the Internet.

The course I teach is called "How to Write and Edit Stories Online." The university gave it the name, and it felt odd at first to specify "online" since the students barely know of any other kind of writing, or reading.

But I have found it useful to keep technology close by when trying to teach young people about writing.

Yes, Klinkenborg (not Zuckerberg) remains the boss in my classroom, but there is no denying that information technology, both hardware and software, have already changed basic things about the way humans write, from the delete key and copy and paste to the iPhone - and Google, of course.

Generative AI will accelerate that process, with ramifications for both the content and content producers varying along that aforementioned food chain.

Even if the tech and money people building the Machine have little interest in what it all means for writing or writers, teachers and editors, lots of other people do. I am not worried.

There will always be a demand for better than average.

Throughout this past semester, we made a conscious effort to work on our writing with minimal interruption or distraction from the AI clamour.

Instead, we set aside the last class for a direct confrontation with the electronic elephant in the room: we worked on prompts; we tried to get the Machine to match some of the best writing from the previous weeks; we asked how we could use it to organise our material, and in what scenarios it might be helpful in our work and careers.

To conclude, I asked the students a simple question about AI and writing: What can we do that it can't?

The responses from the group of mostly 21- to 23-year-olds were swift and inspiring: the Machine can't verify or ask hard questions; it contains no personal experiences; it is unable to go offline… and plenty more.

My favourite response of all: it can't have convictions.

And so, for current and future human writers, with or without convictions, Klinkenborg can help you get started:

'How do you begin to write?

Look for a sentence that interests you.

A sentence that might begin the piece.

Don't look too hard.'

WAN-IFRA External Contributor

[email protected]