The Human-to-Human Principle

Or: as an AI researcher, why I don't use AI to write.

Welcome to Earthly Fortunes! 😄Helen here. If you've received this, then you either subscribed in the past (thank you❤️), or a friend forwarded it to you. I'm honored you are here! I write about The Unseens and fortunes that come from the Earth: geography, medieval farming, music, and more on the way!

This is a new essay of the series Quid ex Machina, about the Unseen impact of AI (and more broadly, technology) on our society. Here’s the rest of the series .

Today’s theme song is What a Wonderful World, by Louis Armstrong.

---

Reader friends,

Every word you read on Earthly Fortunes, I’ve written it personally, either stroke by stroke with a fountain pen, or tap by tap on a keyboard. I don’t let Claude or chatGPT or DeepSeek or other gazillions of AI language models touch my Substack essays. I call this the H2H — short for "human-to-human" — principle:

I write as a human with idiosyncratic and personal experiences, not as a machine.
I write for my fellow human readers, not for machines or algorithms or AIs.

“But!” Some writers will object. “AI makes you so much more productive, so much more efficient as a writer!”

Not so fast. The objection could be true, but let me ask first: in the idiosyncratic and personal writing we do, why does productivity or efficiency even matter?

Before we dive in, let me be clear: I am not against efficiency, productivity, or AI tools in language or other areas. In fact, for my work projects, I set up and fine-tuned open source models on my computer to speed up programming and business writing. I’m not scared of AI, either. For years, I’ve been working in different areas of AI research and applications, from computer graphics to cyber-security to air-gapped RAG systems. Perhaps my proximity to AI removes its magical appeal: all the whiz kids, in the eyes of their child-minder, are just booger-eaters.

Here’s the thing: not everything has to be about efficiency and productivity!

Historically, efficiency and productivity only rose to superb importance starting in the mid-18th century, when the First Industrial Revolution rolled around. So the push for efficiency and productivity is shorter than 300 years — and 300 years is a blip in overall human history.

Presently, after ~300 years and 10+ human generations of push for efficiency and productivity, we still carve out time and energy for activities with zero efficiency or productivity value. You know, useless things like hobbies, friendships, poetry, prayers, pranks, adventures, afternoons naps, or aimless long walks in a park.

For me, writing on Substack is not about getting to the final essay faster (efficiency) or publishing as much as I can (productivity). It is about investing into each essay my life experience and writing skills, so I can give meaning to each word, say what I mean, and mean what I say. I write because I have something to say, not because I have to say something. And I enjoy this writing process a lot.

This is reason #1 I don’t use AI to write: I enjoy writing.

A Lady Writing (Johannes Vermeer), ca.1665. Public domain. This human lady is surely writing to another human.

By “enjoy writing”, I mean the whole package, warts and all: dirty first drafts, knotty revisions, failed experiments, bad re-writes, misfired turns-of-phrases...Listen, I’m not a masochist. I just love taking on challenges. Sometimes I beat those challenges, I’d pat myself on the back and say a self-congratulatory “Good job!” Sometimes, I barely escape from being swallowed by them, I’d feel gratitude because survival experience still counts. Or in other cases, I’m bulldozed by Really Tough Challenges, then I learn some humility from my defeat. It’s all part of the deal.

But why must I personally do all the hard-work, when I could just ask Claude to give me 19 different metaphors in 3 seconds? This leads to my reason #2:

Reason #2 I don’t use AI to write: I want to actually learn the craft of writing.

The Dance Class (Edgar Degas), 1874. Public domain. Teachers can show you all the moves and poses, but you don’t learn how to dance until you actually dance a lot.

A vertical conveyor belt can move the barbells up and down as you wish, but if you want to build real muscles, there’s only one way: doing the actual power-lifting yourself. After each training session, you’d feel stiff or sore, but that’s a good sign: after the soreness passes, your muscles will start to grow back, and become stronger.

Learning the craft of writing works exactly the same way. I could opt for comfort and outsource all the heavy-lifting in writing to Claude or DeepSeek, but then I’d miss out building real writing muscles. If I didn’t fight all the challenges in each essay’s writing process, I wouldn’t know my first drafts were disorganized, or my metaphors were out of place, or I took too long to make a point. In short, I wouldn’t learn any writing craft, if I let another writer — AI or human — to do the hard work for me.

Has anyone ever become a bodybuilder by watching other people bench-press 400 pounds? Not yet. So until vicarious power-lifting or surrogate acquisition of writing skills becomes a thing, we can only become better bodybuilders by lifting, and better writers by writing — we must do the hard things ourselves.

In popular business lingo, actually doing the hard things is our “moat”: no bodybuilder bulks up overnight, and no writer becomes good enough to give their own voice to strings of common words, if they ask language AIs to do the hard things for them “in the style of their writing voice”. Why? Because they skipped over the hard things, they never developed their own writing voice in the first place, their “voice” is whatever Claude or chatGPT or DeepSeek gives them. Also, without actually doing the hard things, they never built the taste to figure out exactly what’s wrong in AIs’ output, let alone fixing it. The hard things are truly hard.

By “hard things”, I mean more beyond challenges in the writing process. Because before we even put down a single word, something more important has been happening in the background: thinking. This brings me to reason #3:

Reason #3 I don’t use AI to write: I want to think for myself.

Far Away Thoughts (Lawrence Alma-Tadema), 1890. Public domain. Basically me while thinking, except I don’t dress nearly as fancy.

Thinking comes before writing. After all, we write to catch the flying thoughts, racing ideas, passing feelings, and brewing insights in our minds. So the root question is: where do we go to inspire our thinking?

In essence, language AIs are carefully tuned machines to put out the average ideas, opinions, viewpoints from a vast amount of training data based on existing written materials — and ~85% of the data are words scrapped from the internet. By design and by default, the chatGPTs and DeepSeeks are the Average Take Machines of the internet. I don’t know about you, but that’s not where I want to go for sparks of ideas. Defaulting to the average views is easy — it saves us the time and energy to think — but it means we would have hyped up the same business or self-help books already hyped by every other “best of” list (which I didn’t do), and come up with the same thoughts already repeated many times over (which I don’t want to do).

“But!” Some prompt engineers will object. “You can use different prompts, so AI gives you not the average take, but takes from the viewpoints you want, even in your own style! Do that! It makes your thinking easier!”

Not so fast. In bureaucracy- and business-oriented writing, I use the “in fill-in-the-blank style” prompt a lot. But for my human-oriented Substack writing, I don’t do it, for three reasons.

• Sincerity Gap: If I passed off language AIs’ “in Helen’s style” ideas and insights as my own thoughts in my own words, it would be an insincere and careless thing for me to do.

• Effort Asymmetry: If I didn't care enough to take the effort to write, but you still spent the time and energy to read the AI’s output, there’s an asymmetry of effort and care between you and me. I don’t want this to happen.

• Meaning Deficit: Behind each word of AI’s output, there’s a lot of linear algebra operations. Behind each word I write, I intend the word to mean what I say, and say what I mean. I want my words to have meanings and intentions invested into them, like real food has actual nutritional values because they take the time to grow.

All this is to say: productivity and efficiency should not be measures for the idiosyncratic, personal kind of writing we do. Writing is hard, because it matters: as a joyful activity to work on, as a craft to learn, and as a character-building exercise for our mind. Joy in the process, mastery of craft, and strength of character — would you trade these (all of which takes a lot of time and care) for productivity and efficiency gain? I won’t. They are parts of our shared human experience, and I’m a human, writing for other humans.

For technical, ecological, and historical reasons, I’m not worried about machines gain super-intelligence. Biggest threats always come from how humans use and react to new technology, be it AI, the internet, electricity, or combustion engines.

The real danger is always humans cease to treat each other with sincerity and care, and let machines do this hard work for us.
The real danger is always humans quit thinking, and let machines keep the sources and courses of our ideas firmly in control.
The real danger is always humans stop giving meaning to the process of our work, and let machines take away our joy in creating.

This is Helen, I enjoy writing as a human being for fellow human readers, and thinking far, far away from the average viewpoints. That’s all from me for now, see you in the next post!

Like what you read here? I’d vastly appreciate your sharing it!

Hot takes? Hard disagree? Resonates well? Comment below or DM me!