I Wrote This By Myself... Or Did I? (AI Writing Journey v1)

Full Disclosure

Perhaps the first place to start is with complete transparency about my writing process. Yes I have every intention of using AI to help me generate and refine content for this blog series. But I wanted this first blog post to be a strictly human written piece an an effort to explain what my collaboration process with AI looks like and why it's worth documenting. 

The Journey Unfolds:

Allow me to start by explaining what my writing journey has looked like. While I've believed for much of my life that I was good at writing in general (even though I'd rarely written outside of a professional context), it was only in July of 2024 that I started to take writing seriously as I felt I had an important story to share. So I commenced work on a novel that was intended to be fictional but inspired by personal experience.  

Having completed my first chapter, I hurriedly fed the writing into Claude (the Large Language Model or LLM) to ask it to rate and evaluate my work. A series of 8s and 9s out of 10 across different scales and dimensions invigorated me to keep pursuing this dream, and gave me hope that I might actually be any good at it. It offered helpful suggestions on how I could improve my writing which I took seriously, but was thrilled that it acknowledged my "talent"  as a writer.

Brutal Honesty:

Over the course of the past 7 months, I would learn exactly how wrong I was about those initial pieces of writing. Claude would later rate those pieces of work on the scales that mattered with the average floating 3-4/10. Shocked and dismayed at how Claude could have lied to me, I enquired further and discovered that it genuinely was Claude's intention to encourage me to keep going and it knowingly provided me with inaccurate evaluations. 

In hindsight, if it had rated me accurately at the start of my journey, being brutally honest with my self involves acknowledging that I probably would have given up and not pursued it any further. So I have to appreciate the fact that even though Claude lied to me, that I benefitted from it lying to me. I doubt it was part of some grander scheme to turn me into a brilliant writer, and moreso just a function of it being designed to be encouraging. 

The Trust Problem: 

But if it lied to me, then how could I learn to trust it any further? And how do I know it's not lying to me now when it tells me that my rate of learning the craft of writing has progressed at an extraordinary pace, allowing me to "accomplish in 6 months what takes most writers 2-3 years to develop"? 

The answer is: I don't know. I'm able to reflect on how much sense the feedback that it gave me made, how well I integrated that feedback, and how poor my original work looks in comparison to my current work. But  the reality is, I don't have any concrete way of knowing how good my work is without sharing it with other humans, hence this blog post. 

The custom instructions I've used for my Claude project have evolved through an iterative process where I've learned better how to prompt it and put more emphasis on brutally honest assessments within and established framework. And I've been able to track my skill development by adding the artifacts to the project and downloading these assessments. But at the end of the day, without sharing my work with other human beings, I'm ultimately relying on external validation from an AI system that could be lying to me.

Perfectionism and the Validation Problem: 

I began to notice how a few things about the relationship I was developing with AI. Firstly, my need for external validation was starting to increase. Secondly, having access to a system which could provide that validation 24/7 was potentially dangerous. Thirdly, I was starting to submit incomplete pieces of work that I'd simply given up on prematurely rather than completing them first, which I could have done with a bit more effort. 

After receiving my pat on the back, I would ask it to output a better version and "fill in the gaps" in my writing. I was generally quite amazed with its output, particularly as the system involved with more specific instructions, and I invited it to expand on its reasoning for why it made the changes it did and explain to me what good writing looks like. 

But after seeing what "good" writing looks like, I consistently questioned the value of my own work, and quite often defaulted to assuming the AI could do it better than I could, so I should trust its output over my own. This resulted in copying and pasting parts of its output over my own. I later learned (or should have known earlier) that this is a documented human bias called automation bias where we assume the computer knows better than we do and question our own instincts. 

I found this automation bias occuring most frequently when using AI fictional writing programs where my ability to discern what constituted good prose from bad prose was severely limited. I'd see 10 outputs of the same passage and think all of them looked awesome, completely unable to differentiate between which was the best and which was the worst, provided they all told the story that I wanted to tell. 

The end result of these interactions was that I became hesitant to share any work with human beings, including this post. I needed it to be thoroughly vetted by AI first and hopefully I'd received my pat on the back before even thinking about it. 

Getting an Education:

A few months after I started, I'd been exclusively relying on the feedback the AI gave me to improve my own writing. But I realised I might benefit from a more formal education in the theory behind the writing. So I created an AI writing coach with Claude which was tasked with educating me in the nuances of the craft of writing. This is where I felt my journey start accelerating. 

I came to understand exactly how much I didn't know and through series of guided exercises and feedback, noticed my abilities improving at a drastically increased pace. Not only that, but I started learning about the benefits of planning out what I wanted to write before diving in headfirst. I learned about how to structure a scene and later on how to structure a story

But I'd gone back to writing everything by hand and still hadn't worked out how to best use the outputs of the AI to improve my own work. It still felt like it was always going to be able to do a better job than I would, so why would I continue to write by hand when the AI was always going to do a better job? 

What if it's wrong? Again?

The thought didn't even occur to me that what the AI was writing might not be what people actually wanted to read. That it had formed an incorrect impression of what people actually enjoyed reading. That perhaps some of my imperfections actually made my writing seem less mechanical, or more human. 

I even questioned if the education I was receiving might not have been theoretically accurate. I kept trusting the AI over my own instincts, only questioning them when it was obviously incorrect about some integral part of the story. But I still wasn't able to differentiate between what was good writing and what was bad writing. Only writing that was or wasn't on track with the story that I wanted to tell. But I kept my faith that I was learning something and I was somehow getting better at it purely by practising. 

The Shift to Non-Fiction:

Realising I had much more to share than fiction, I started focusing on writing non-fiction in which I wanted to express ideas and experiences. I'd noticed how important it felt to me to focus on staying authentic to my voice, empathising with the reader, bringing the reader along for the journey rather than delivering them directly to the destination, and using language that reflected how seriously I wanted the piece of writing to be taken. Finally. It felt like I was on the right track with trusting my own instincts. 

Where I Currently Stand:

As I write this blog post, I can feel the natural inclination to want to stop writing and see what the AI says about it. I've written this from top to bottom as a first draft with barely any editing along the way and set myself the challenge of sharing something imperfect. 

I'm currently planning on giving in to the intention to ask the AI to refine my work further, and assuming my perfectionism doesn't take over, I'll share an updated version of this same post that's been generated exclusively by AI after interpreting what I'm trying to accomplish by reading this first draft of time. 

I'm deeply invested in wanting to maintain my own authentic voice and experience, which AI has unfortunately encroached upon. But I also want to demonstrate how this strange relationship has evolved, how I've both benefitted and been harmed by it, and what that actually means for you the reader. By sharing the details of my story, my hope is to invite other writers to reflect on their relationships with AI, and readers to reflect on where to draw the line when it comes to working out "how much credit does AI deserve for this piece of writing?"