Maths for writers: AI text ≠ zero value

Thinking in public: where does the value of a piece of writing actually live?

I like writing. I like sitting and playing with words, twisting and shaping a sentence until it does what I want it to do. Sometimes I’ll change something, change it again, then put it back almost exactly as it was. An absolutely excellent use of an afternoon, depending on who you ask.

Sometimes, that process is a large part of why I write. It’s a process that I, as the writer, derive value from - personally, emotionally, maybe even professionally (if I’m improving my craft). I enjoy a nice turn of phrase. I’d like to think I can produce one occasionally. At the same time, there is (hopefully) also some value for you in reading it. 

Other times the thing I want to do most is… share an idea. There’ll be something I’ve learned that I’d like you, dear reader, to get to learn too. Or there might be a connection I’ve made, a perspective I want to explain, or a question I think other people should be asking too. The words are how I get it out of my head and into yours - and while I still care about how they read - the thing I most want you to leave with is the thought.

Which is where I find the debate about ‘AI writing’ a bit frustrating. Knowing that someone used AI to help write something tells me very little about what they contributed. They might have spent months developing an argument and needed help explaining it. They might have asked ChatGPT to come up with an argument for them. Those are different contributions, however similar the resulting paragraphs might look. 

On the words themselves - well, if a piece of writing is enjoyable, engaging, even challenging, then does it matter so much if they were handwritten, typed by human fingers, or generated by a model? As a reader, a great idea, an interesting provocation - well they survive all but the MOST generic of AI drafts. 

So this got me thinking… how do you define the value of a specific piece of writing? What factors actually matter as the reader? And then, for reasons that probably made sense at the time, I decided this needed an actual formula.

I enlisted ChatGPT. Because (as you may yourself relate, I am not normally ‘a numbers person’). I proposed my idea, gave my initial list of variables, asked it to mathematical-ise my formula. At one point I had to ask what the little letters next to the other letters meant but learning-by-doing has always been how I learn.

The result: Maths for writers. A formula to explain the value of any given piece of writing.

First, whose value?

My starting point was ‘the idea’, or the writing itself, and whether the thing ever gets shared. The best idea I’ve ever had could sit in my head indefinitely, where it would be of precisely no use to you.

But I’d mixed together two things: what I get from writing, and what you get from reading. I might thoroughly enjoy spending an afternoon rearranging a paragraph. You might skim it on the bus. Both perfectly reasonable uses of our time, but if I’m trying to talk about reader value, my lovely afternoon or experience of personal catharsis doesn’t belong in the calculation.

This became clearer when we tried putting ‘effort’ into the formula. Multiply by the ‘effort required’, and something that takes more work gets a higher score.

But… writing comes relatively naturally to me. Whereas a friend of mine, a writer with dyslexia, has shared with me the effort they expend every time they want to get their thoughts onto the page. If we’ve both produced something equally clear, interesting and useful, does the difference in effort change what the reader receives? Especially when, most often, it’s an unknown effort to the reader at the point they, well… read.

So effort came out. Whatever this formula was going to measure, it was going to be value to the reader.

Taking the writing apart

The first variable I settled on - given the thinking above - was ‘idea’. This became intellectual contribution (IC). I wanted room for the things we do with ideas as well as the ideas themselves: bringing together research, making a connection, offering an interpretation, asking a question that changes how someone looks at a subject. You don’t have to be the first person ever to think about something to contribute to the thinking around it.

Then I got stuck on the ‘craft of writing’. Being able to explain something clearly and being able to write a sentence that someone wants to read twice are related skills, but they aren’t quite the same skill. A set of instructions can be beautifully clear without anyone feeling moved to quote it.

So craft became two things. Articulation (A) is how well the piece communicates: whether the explanation makes sense, whether the argument holds together, whether you can follow what I’m trying to say. Artistry (AR) is the pleasure in the language itself. The rhythm, the voice, the slightly unexpected choice of word. The bit I might happily spend that afternoon fiddling with.

This gives us something like:

Value (V) = IC + A + AR

But this was still missing something. 

I’d asked ChatGPT to try the formula on a fully automated industry roundup: AI collects articles, summarises them, sends an email every week without fail. There’s almost no new intellectual contribution in that, and very little artistry to speak of. Yet I might be quite happy to receive it.

It saves me reading forty articles.

That gave me utility (U): what the reader gets from it. Useful information, time saved, entertainment, something that helps them see a familiar problem differently. ‘Useful’ needs a fairly generous definition here; I’m not proposing that everything we read should come with an action plan. 

Then we have distribution (D). Does any of what I’ve written actually reach the people it’s meant for? This goes outside the brackets as a multiplier. If nobody receives the writing, nobody receives its reader value, however pleased I am with the draft.

So I was at a formula that looked a little like this:

V = (IC + A + AR + U) × D

The maths for writers bit

A practical guide needs to help me do something. 

A literary essay might be worth reading largely for the writing. 

An argument might stay with me for years despite some fairly ordinary prose. 

Ultimately, the value of authorship doesn’t sit in the same place for every piece of writing.

Giving each component equal importance would miss the distinction I was trying to make in the first place. Which gave me the variable ‘purpose’. How much does each of these things matter for this piece? Purpose determines how much each other variable is weighted when running the formula in a piece of text.

Which brought me to this formula:

V = [IC(P) + A(P) + AR(P) + U(P)] × D

V is reader value. The (P) after each component means ‘weighted according to purpose’. It’s shorthand: we give each component its own weighting, rather than multiplying everything by one number called Purpose. 

And here is the same thing with the weights written out properly:

The tiny letters are labels. The little IC beside P means ‘the weighting for intellectual contribution’. That’s all. Nobody has slipped another calculation in while you weren’t looking.

We score the components (IC, A, AR and U) out of ten, then give each a percentage weighting, with the percentages adding up to 100%. The purpose tells us what matters most; the scores describe how well the piece delivers it.

These are judgements, obviously. I haven’t discovered a law of literature. But the numbers are a way of making the judgement visible so we can argue about it (something we seem to be doing a lot these days). 

Testing the formula: back to that roundup

Taking the earlier example of an automated industry email, suppose 20% of the weighting goes to articulation and 80% to utility. Its purpose is to keep me informed without requiring me to spend hours finding and reading everything myself. I’m giving no weight to new intellectual contribution or artistry in this example.

Assume it selects relevant stories and summarises them accurately. Articulation gets eight out of ten, utility gets nine, and distribution gets ten: it reaches its intended readers every week.

[(0 × 0%) + (8 × 20%) + (0 × 0%) + (9 × 80%)] × 10 = 88

Eighty-eight out of a possible hundred on our invented scale. You could disagree with every score I’ve given it and still see the point: a piece with almost no new intellectual contribution can deliver considerable reader value.

The original reporting belongs to the people who did it. Summarising their work doesn’t make it yours, and whether they’ve been credited or treated fairly remains a separate question. But the reader has still received something useful. Those two things can be true together.

Now take an essay with an idea I’ve spent months working through. The weighting would look quite different. My intellectual contribution might be the main reason to read it, with articulation and artistry helping you follow the argument and enjoy spending time with it. If AI helps me articulate that thought, it doesn’t follow that the thought itself has lost its value.

There is a complication here though, which I recognise from my own writing: sometimes I only find out what I think while trying to write it. A sentence won’t work, and eventually I realise the problem is the idea. So these components overlap, and handing over one part of the process to AI could affect another. We get to the stage where the IC is in effect, ‘co-created’ with AI. But is that any different from academics discussing ideas with research partners? Authors shaping plot with a writers group? I’m not sure.

And then for the main event: somebody has to read it

Distribution is the variable that means value is determined specifically by the reader. There is plenty of personal value in writing something nobody else ever sees. But I’m talking here about ideas I want to communicate, and for those, getting the thing finished and shared with the world (or simply with 56 people on Substack) matters.

D needs to stay tied to reaching the intended reader. It would be ridiculous to suggest that a poem becomes better when its author acquires more followers, so it’s not about scale or reach. What I’m trying to capture is the difference between something a reader can benefit from and something they never get the chance to read.

This is where AI assistance can make a difference that the label ‘AI writing’ doesn’t explain. If I have something worthwhile to contribute, help getting it into a form I can share might increase the value that actually reaches somebody. The part I contributed is still there. We can look at it, question it, decide whether it’s any good. Actually sharing our work has been something that has plagued writers - amateur and pro - for decades. Editors, collaborators and writing groups have long helped with that, but access to their support has often depended on money, connections or getting past a gatekeeper. AI brings many of those capabilities to people who previously had no access to them. Help developing an idea, shaping a draft or getting it ready to share becomes available to far more of us. 

AI text ≠ zero value.

I’d like us to be able to talk about the value of writing, and who contributed which part of it, with a bit more precision. Sometimes I’ll want to spend the afternoon playing with every sentence. Sometimes I’ll want help getting an idea onto the page while I still have time to do something with it. I don’t think I need to make one permanent decision about what sort of writer that makes me.

And dismissing AI generated or AI-assisted writing as inherently value-less is reductive. This article is only here because AI let me sort through my thinking quickly, giving me draft formula to push back on and adapt. The ideas, and the formula - that’s mine. The words on this page - mostly typed by me; even where they’ve been lifted from that ChatGPT conversation, ChatGPT had lifted them from my voicenotes. 

So, where does this leave us, dear reader? 

With a formula that hopefully helps us think about reader value with nuance. And with an idea already forming in my head: could reader value divided by writer-derived value give us what I’m thinking of as the ‘absolute value’ of a text? Reader Value / Writer Value. That would need a formula for what the writer gets from the process, and I haven’t worked that out yet. One to explore after this article is published. I’m off to chat with ChatGPT. 

V = [(IC × PIC) + (A × PA) + (AR × PAR) + (U × PU)] × D

---

Thank you to Dr Sam Illingworth whose original article inspired several days’ thinking about the value of AI and how we define it.