Polishing your writing
Even though software like Grammarly preceded the generative wave, it has the characteristics we associate with GenAI. It offers feedback on and supports the revision of text, in a way that goes beyond the more basic facility of a spelling and grammar check.
Clarifying Your Ideas
Using conversational agents in this sense means framing them as occasions for thinking. In explaining an idea to a conversational agent you are practicing articulating it to yourself. A useful parallel can be found in the practice which software developers have of ‘rubberducking’: explaining a problem out loud to a rubber duck which has been placed on top of their monitor. A common experience is that verbalizing the problem, even if it is to an inert rubber duck, leads to a breakthrough which enables the problem to be solved.
The same orientation works effectively with Claude and ChatGPT, with the difference being that these rubber ducks are able to talk back. Not only can they offer commentary on the problem you have presented them with, they can actually take on complex roles to structure a dialogue with you about this problem and push you forward in your thinking.
Jump Starting Your Writing
Baron (2023: loc 4333) compares this to “jumpstarting a car battery”:
“When the battery’s dead, the engine won’t turn over. In the world of humans and writing, the equivalent is writer’s block or paucity of ideas. To give batteries or human brains the spark needed to get chugging, you harness an external assist – connection to a Good Samaritan’s battery of an AI tool.”
What does this look like in practice? I’m writing this in the mid evening, after a long and slightly difficult day. It feels important that I reach my word count for the day. But I also feel physically and mentally depleted. This phrase from Baron (2023) was on my mind following a conversation about GenAI I had earlier at a conference. I wasn’t sure exactly what use I wanted to make of it but it was clear I had something to say here, even if I felt too tired to work out exactly what it was.
I shared this extract, from my initial reference onwards, with Claude 3 Opus in the hope it would jump start my evening’s writing. It suggested four points of development: overcoming writer’s block, collaborative ideation, maintaining momentum, balancing inspiration and authenticity. The first two of these feature heavily in the draft text which I’d already presented it with and effectively involved Claude feeding my own arguments back to me in the manner suggested in the next section.
The final point captures a core argument in a way which was more concise than I had thus far managed. As Claude put it, “The goal should be to use AI as a catalyst for your own ideas, not a replacement for them” leaving writers with the challenge of finding “a balance between the generative power of AI and the need for authentic, original expression”.
In a sense this captured the central tension explored in my book, how we use conversational agents without coming to rely on them, with an admirable clarity. It provided me with a motto which I could consider in relation to Baron’s notion of ‘jumpstarting’ your writing. How can we cast conversational agents in this role while avoiding the risk of over-reliance on them? At what point does jump starting become writing for us? How much can we use them in the writing process while still ensuring that it is our process? Claude’s suggest here is extremely clear: “a catalyst for your own ideas, not a replacement for them”.
The third point embodies how I’m actually using Claude in this example. It’s enabling me to sustain a writing practice, even when my motivation is low. It’s provided the impetus to write this evening, even when I’m tired. There’s a real sense in which this writing is dependent on the catalyst it has provided, filtered through the odd reflective prism involved in writing a book about writing with GAI, which itself uses GAI as part of the writing. It is nonetheless my writing. I’m not using the text from Claude as a replacement for my own, I’m quoting it as a catalyst in relation to which I develop my own reflections.
I would treat these suggested structures with the utmost caution. Their value isn’t in providing you with an answer about organizing your writing, as much as contributing to the questions you are asking about this process. I’m conscious of students using suggested structures to jump start their writing, providing them with a sense of traction on a project too broad to get started with, only to constrain their actual writing by remaining rigidly within the initial suggestion.
Representing Your Writing
This use of conversational agents has become extremely important to me over the last couple of years. I explained in chapter 2 how using Claude to summarise the arguments of this book at the halfway point was crucial to how the project developed. What made it valuable was not an addition to my arguments but rather a representation of them. When you are deeply immersed in the writing process, it can be difficult to retain a sense of what you are trying to say, particularly as that inevitably develops through the process of writing.
In part this is a matter of the old adage that you can’t see the wood for the trees i.e. when you are deeply immersed in the details of what you are doing it is easy to lose sight of the task as a whole. In Nintendo’s popular Zelda series there is a recurring area called the Lost Woods in which you wander through a forest maze filled with fog. There is a route through the trees which enables you to leave the forest but getting there is a matter of trial and error. Until you find the winning pattern you are condemned to feel you are making progress but then suddenly find yourself thrown back to the starting position as if no progress has been made. I’ve often felt that writing resembles the Lost Woods in the sense that you are stumbling around in the fog, until you eventually find a pattern that enables you to crack the mystery. Once you’ve found it the solution seems retrospectively obvious and it’s easy to forget those confused hours in which you were unsure how to make progress.
However it goes deeper than this because in a writing project the ‘wood’ is not a static context but rather an environment which is continually evolving as you move through it. At risk of stretching the analogy to breaking point, the work you undertake with individual trees (e.g. writing a section predicated on the most terribly nerdy metaphor) accumulates into a forest with different characteristics (e.g. a book which might be disengaging to readers who don’t share those nerdy points of reference). Not only is there the ‘fog’ of being unsure how to find your way through the forest, the maze itself is continually reconstructing around you as you work.
What I’m calling representations of your writing can be seen as dynamic maps which help you inventory where the project is at, with a view to supporting its further development. The intention isn’t to give the conversational agent the final verdict on your argument but rather to use its capacity for summarisation to help you better understand what you are trying to say. This could involve different roles depending on the stage of your project e.g. a critical review attempting to find holes in your argument, a conciliatory supervisor trying to build up your intellectual confidence or an interdisciplinary collaborator familiarising themselves with a project they are joining late. There are countless roles which a conversational agent can play, if you can be clear enough with it about exactly what you’re looking for.
This is the core challenge which necessitates the reflexivity which is the main focus of the book: in order to specify what you need from the system, you have to be clear about what it is you would actually benefit from. Often this isn’t the case when we come to writing because we often improvise our way through the process, working from a strange mix of intuition and inertia which often doesn’t actually work for us in the way our adherence to it would seem to imply.
Mark Carrigan 