Human Writing & Editing in a Post-GPT World
- Pete
- Jan 3
- 7 min read
With the rise of ChatGPT and other generative AI and machine-learning models, it's not surprising that there is lots of conversation around the future of writing — and, by default, the other side of the coin, editing. Alas, it's a conversation we need to have but isn't one we shouldn't have to.
From one perspective, we should be very worried. In an age of deep-fakes and dis- and misinformation, it is all too easy for this technology to be used to mislead people. But then, people have been lying and cheating other people for as long as we've been communicating. So the issue here is not the tool so much.
And this is my point.
Machine-learning models like ChatGPT are simply tools. As is writing.
So, I'm very bad at being a 'handyman'. If there is a job I need to do around the house, I have no idea where to begin or even how to do it. I walk into a hardware store and don't even know what all these
tools do, let alone how to use them. I own a hammer and a couple of screwdrivers (and tons of those little Ikea tools). But I'm fairly certain that if I tried to use those for everything, it would possibly be a disaster.
Now, you could say, "Go watch some YouTube videos and learn how to do that thing."
Sure. I could.
(And here comes a side-learning on neurodiversity and inclusion and why you should never assume what works for you will work for others.)
I can't 'learn' by watching videos. I need text instructions. And they need to be very clear, very precise, and very succinct. This could send me now on a digression into the importance of a good technical editor ... but I'll refrain for now.
The point is, I can't follow instructions on a video in the same way I can text. And I need to be doing the thing using the text instructions, step by step. Which takes time. Because, inevitably, I'll come across something I don't understand, a presupposition in the instructions that assumes the reader will already know something. And because I am completely inept when it comes to handyman stuff, that means I need to go find out about that thing ... and inevitably, I fall into a rabbit hole of trying to understand all these new concepts and find ways to connect the information. Which takes time. It is much easier, efficient, and cheaper for me to call my handyman, who immediately knows the problem and how to fix it, has the tools required for the job, and simply does it. In far less time than it would have taken me to learn what the handyman has learned over decades of experience.
Yep, to save $100, I could have spent tens (if not hundreds) of hours teaching myself something new. But that would have taken time away from spending it on things that feed my soul and bring me joy (and also working, which also feeds my soul, as it turns out ... yes, I truly love what I do!)
Anyway, you see my point, right?
What does this have to do with machine learning and generative AI?
These are tools.
Not everyone has learned to write well. Good writers are good readers. Their command of language comes from extensive experience with many different examples of the written word. Find me a writer who doesn't have shelves filled with different books. To quote the meme: "Change my mind!"
So, when someone who can maybe count the number of books they've read on one hand is asked to write a report for their corporate boss, it's understandable they may turn to something like ChatGPT to help them write—and edit—what will inevitably be read once and then archived for shareholders down the track. A corporate report does not need to read like a Jane Austen novel. Wrong audience.
ChatGPT is to the corporate lackey, what the handyman is to me.
It will be bland, professional, and to the point. It will provide the appropriate numbers and data analyses with the wordy fluff in between that just explains the appropriate numbers and data analyses in such a way as to make sense. Hopefully. I'm not saying that to dismiss this kind of writing — it's appropriate for its audience and the purpose of conveying appropriate numbers and data analyses.
It still requires the human touch — you must feed it the data first. And the quality of what it spits out will depend on what you feed into it. For the purpose of efficiently spewing out a corporate report — great!
It won't be a best-seller. It won't be engaging (unless you find dry, humourless analysis sexy, in which case, aren't you in for a wild ride).
I started writing this blog with one very simple prompt: why is ChatGPT a useful tool for people who can't write?With that prompt, my mind came up with everything you have just read above. This is what ChatGPT generated with the same prompt:
ChatGPT serves as a valuable tool for individuals who struggle with writing by helping them articulate their thoughts clearly and effectively. It assists in overcoming challenges related to grammar, vocabulary, and organization, enabling users to create well-structured and coherent content. Whether for personal expression, academic work, or professional communication, ChatGPT provides support by generating ideas, offering suggestions, and refining drafts. This empowers users to convey their messages confidently and accurately, bridging the gap between their ideas and written expression.
See the difference?
Hopefully, mine is a little more ... interesting. Entertaining even. There are some who will say the latter is much more 'to the point' without all the hyperbole and filler. Which is true. But those people are probably not reading this blog. They're not my target audience. Because the kind of writing they want can be achieved with tools such as ChatGPT. And that's fine.
But, you — you're not that person. You love reading. You love writing. You'd love other people to read what you wrote. When I sat down to write in response to that simple prompt, my mind began to think about you. What would you like to read? Links to articles and evidence (I could've done that, I really could ... and to be fair, I did because that's just how I roll). I didn't think about SEO and web optimisation (which the Wix AI told me to do, enticing me like a succubus seeking to snatch my soul on behalf of its infernal master) — because my audience is you, not Google's search algorithm.
Generative AI is a tool, and what it produces is subject to the limitations of what it can be used for.
Like my hammer, I'm not going to use it to tighten the screws on my door hinge (I'm pretty sure that's the wrong tool for that job, but [shrugs shoulders] ... I honestly have no idea). Machine learning is never going to reproduce the human imagination. I mean, ChatGPT did not come up with the succubus metaphor, now, did it? Nor did it come up with this:
Mr. Collins was not a sensible man, and the deficiency of nature had been but little assisted by education or society; the greatest part of his life having been spent under the guidance of an illiterate and miserly father; and though he belonged to one of the universities, he had merely kept the necessary terms, without forming at it any useful acquaintance.
That's from Jane Austen's Pride and Prejudice, by the way. Only a creative human mind could think of using the phrase "deficiency of nature" to describe Mr Collins. And only a human mind would consider finding a quote from Jane Austen to illustrate the point made by their musings on the original prompt ("Why is ChatGPT a useful tool for people who can't write?")
Also, notice how deeply flawed this blog post is. Other editors (or even writers) reading this will be tearing their hair out at some of my phrasing, punctuation, and even the length and waffling nature of this post. But I did that deliberately. If you go back to GPT's response to my prompt, you'll notice the text is ... well ... perfect. It abides by all the rules of grammar. Because a robot can only ever follow the rules. Human creativity doesn't do this. It can. But more often than not, it chooses not to. Because what we find interesting is usually what is novel and unique.
And this is the part of editing I love the most. It's finding the unique voice of the writer and bringing it to the fore. And yes, sometimes we'll tweak things to make it 'more you' because we want the reader to 'see' you (and your writing) ... and also because we want you to connect with the reader.
Generative AI simply cannot do this. It can't edit for nuance; it can't edit for style. It can correct your spelling and grammar (I'd argue its ability to do that is also questionable, given there are no hard and fast rules of grammar and spelling), so long as you state the parameters, but it cannot make suggestions that align with what's inside you, because it doesn't know you. Yes, you can feed it lots of data and parameters ... but does it really know you? Does it know who you are? Does it engage in deep, human conversation that is not transactional ... like human discussions are? (Okay, I also need to acknowledge here ... not all humans ... clearly.) And this is the difference between working with a human editor who knows you and understands you ... and also knows and understands your readers.
So, do I feel threatened by generative AI? Not in the slightest. Do I use this tool in my work as a writer and an editor? Sure. But also because I know how to use this tool — unlike the hammer in my bottom drawer.
Happy writing, humans!