The Messy Writer and AI the Gaslighter: Are we losing our voice?

Let me start with a foundational experience. A few years ago, I read my Masters research after finishing my PhD, and what I saw was a completely different writer. What a shift in confidence and authority in writing style! That journey, pre-AI, was about dealing with liminality, impostor syndrome, and the slow, sometimes painful process of finding my voice. That philosophical question of “who am I?” as a writer is the connection between language and consciousness. And I’m not sure AI really gets that.

This is the core of what I’ve been wrestling with as I see AI being integrated into writing and assessment. I want to take a side step from the technical debates and discuss the lived reality of it.

Assessments Designed to Sit in the Unknown

The process of drafting, thinking, learning to become a self editor, receiving and taking on feedback to grow in your writing style, polish a thought, or make clear your sentence structure, is the joy and frustration of learning to write. You get to learn about yourself while sitting in the discomfort and messiness of your creativity. You get to disagree or challenge your mentor when they’ve provided feedback on your writing. Maybe that was just me! Importantly, you learn about how to become that self critic when engaging in a constructive discourse about your literary expression or thoughts. An early writer like a five year old also engages in this process as the teacher tries to decode text that is still in its rudimentary literacy stages. The five year old tries to put together not only thoughts and ideas, but how to use correct semantics, graphemes, phonemes, and morphology. This is the most complex time of a writer’s life long journey, and the iterative nature of that development keeps going even as I write this word. Isn’t that beautiful?

I find AI doesn’t do that process well. Maybe it wasn’t designed to, but educators are being asked in new curriculum requirements and assessment designs to use AI and try to train it to; and teach students to train it also.

The Goldilocks Problem: Why Training AI Feels Like a Waste of Time

So when I look at my experiments with AI to review drafts or chunks of ideas, I found it took away my tone, my style, and changed the lexical choices I made that reflected my voice. This made me irritated and a little disempowered to say the least.

I’ve spent a lot of time and tears learning to craft my writing across many genres. The cocky confidence that AI presents and the justification of why my work needed its refinement was challenging. It holds a confidence that gaslights you into thinking you’re either a contender for a Nobel prize in Literature, or that you could write something perfectly in their recommended way. When I reflected on the way it provided feedback, it went against the foundations of how I taught pre service teachers and school teachers about effective feedback.

Sure AI looks for patterns like I do when I give students feedback, but the patterns I look for when working with students extend to the writer themselves. The relationship I’ve built with the student, navigating those hard decisions about how I can explain it to them in a way that balances critique, while building on their learning where they are at; and most importantly how and what they wrote last time. Also I correlate data on what they said in class and how it connects to their writing. When you conference writing with a student of any age, you are engaging in a discourse that gives the student a chance to explain what they thought, the way they attained it, and how they tried to express their thinking in the way they did. All the while modelling and facilitating that process, waiting for the a-ha moment.

I discovered that unless you instruct the AI first, the feedback was contrived. I worry that this feedback just feels like a hollow form of operant conditioning, dressed up as mentorship. This mode of reviewing drafts will take away from the messiness that we are, and that puddle we muddle through when we create and write.

You see AI doesn’t know me, it confidently claims it does, but it really doesn’t. There’s a weird ‘trust’ being conflated there. Here, writing is still seen as a polished product, not the nitty gritty learning moments or dipping my toe in the water to try expressing an idea in a new way. I’m not trusting AI with being vulnerable to my process, I’m trusting AI as a source of truth about my capacity and how others might see me in their learned standard of AI writing. There is a new standard of genre being created here, and I’m not really sure what it is, but you can tell.

Can I say I have tried many AI tools that Goldilocks has nothing on me. No amount of blueberries or cinnamon sugar is making that porridge palatable. You can give it feedback to try and improve this interaction…it doesn’t really like it. You can also give it the ubiquitous thumbs down. As a result, you get a weird apology, but it just fixes the perceived error, and often does it again. Apparently this is called training your AI.

The Loss of Our Mental Models

Using AI to brainstorm writing, and turning it into a component of assessment is something I’m seeing teachers across sectors dabble with. When I used non AI brainstorming tools with students, it was a diagnostic tool that captured a snap shot of learning. It allowed me to see the words students used from their own connotative and denotative bank, it’s their own mental model. My research in pragmalinguistics and meta text really solidified what I practiced in classrooms for years - our words and the relationship of those words to others creates our voice, and communicates to others how we think. And maybe a bit more about who we are and our experiences.

This also leads to a critical problem I call the 'Empty Vessel.' We can scaffold the process of giving feedback to an AI, but if a student hasn't yet developed the foundational literacy and self-awareness to recognize their own voice, the prompts will be empty. They won't know what to ask for because they don't yet know what they're trying to say. They are being asked to critique a voice they have not yet discovered.

Using AI to brainstorm takes away this vital information. It polishes your work and ideas, makes it sound better, which a writer still learning their craft can be challenged with in the tension of confidence and voice. So whose work is it? Ownership of our ideas and thoughts and words is powerful, even for a five year old.

When I have used AI to ideate, and then try to train it to refine my writing so it sounds like me and my thinking, it’s just more wasted time and more work. It can’t adapt to my voice and my thoughts at the rate I can. How AI thought I wrote in my last chunk of text isn’t me now, because I’ve grown through each piece I create. It holds me back to who it thought I was. Also, it challenges my thinking not on how to think, rather on filtering and sorting really low level processes, rather than help create neural pathways to ideate, be iterative and make connections. This process develops how we think about the world and who we are. That human voice and consciousness is embedded in the choice of words we use, they’re my words, my polysemous words in sentences, for better or worse. And that takes time and practice. I want to learn, not pretend to learn.

The Data Becomes Skewed and Not Credible

Building on those ideas of mental models and language use, I turn to my work with middle leaders in developing education policy, curriculum or accountability related documentation. I see these writing processes as teaching opportunities with a drafting process for feedback. This takes time, and I build this time into the process of the school timelines strategically so the impact of overwhelm and workload are balanced. I’m teaching middle leaders that if we value good work, we need to make time for it.

The way teachers and leaders write about improvement of programs or curriculum is their voice as professionals or as a team, and it contributes to the culture of their teams and the school. The words we choose are ours and reflect our thoughts and create the culture we are in - these are our mental models. As an education strategist in these spaces, it helps me see gaps in their knowledge, or writing skills, or thinking. It is through the drafting and feedback of these documents that I build on and develop improvement. Importantly, it helps build relationships with the people I work with. They know I actually read what they say, and value their ideas and authentically want to help them grow in their expertise. AI takes those signposts away, and we don’t get a real idea of where we are at as a school. The data becomes skewed, and not credible. As a strategist who uses evidence and research in education for improvement, this is key information for how I improve and build systems and grow the people in the organisation.

I think the beauty of writing is that it is a deeply human act. It is how we share our humanity. I urge you to think about how your choices to use AI in education will influence that.

When was the last time you chose the messy, imperfect, but truly ‘yours’ version of your work over the polished, easy alternative? What did you learn in that discomfort?