Some thoughts on writing with AI.
I predominantly see writing as a method of communicating ideas to someone else. The process to get there carries a whole host of benefits to you as the writer; to structure and deepen your own thinking, to formalise your ideas and arguments, and to come up with counter arguments. The friction you have to overcome to draft and revise your words to communicate your point is where the magic happens. There’s an inherently creative side to writing, whether you’re writing a technical proposal or something else.
It’s that creative process that is important to hold on to when we’re using AI to support our writing. By no means is this me saying I don’t think we should use AI at all to help with writing. I actually think AI for brainstorming and editing our words is enormously valuable, as they’re the parts of the process that we can make more efficient without losing anything really important or inherently human. They’re the parts where a co-pilot, rather than a driver, can help us write better. But delegating the process of thinking and writing completely to an AI agent? No thank you.
I was recently reading about the displacement of human alignment which started my thinking behind this post. The concept comes from evolutionary mismatch theory, which basically says that traits that were once crucial for our survival as humans are becoming increasingly detrimental in today’s world. As a non-AI related example, our ancestors had bodies that were built for endurance, regular low-intensity movement, and moving across different terrains. Today a growing population leads lifestyles that are increasingly sedentary (hello from me writing this hunched over my laptop), leading to an increase in various health related and musculoskeletal issues.
The link to writing here is that by increasing our use of AI when writing, we risk losing something that is incredibly important to us today. Maybe that’s creativity, or maybe it’s just creating the headspace to properly think about a problem we’re trying to solve. Some of it is certainly showing the “human” side of writing — language models are optimised to output statistically probable, predictable sequences of words, but as humans we don’t write or think like that (at least I don’t 😅). If we’re outsourcing our thinking or the entire process of writing to AI, we lose our personal voices and the emotion behind our writing. As a result, I think we start to run the risk of losing important human traits. You know that feeling when you’re reading something that you think has been generated by Claude? Do you want to read that, or do you want to read something that might be a little less polished but is representing the actual thoughts of another person?
In software engineering, given the capabilities of models today, a large chunk of the implementation and code writing can be safely delegated to agents to complete on our behalf. But I think we’re in dangerous territory if we’re also delegating both the thinking and the explaining of the intent behind the implementation to agents. Most of that side of engineering is done via our writing. If we’re handing the keys to an agent to cognitively guide the direction of a solution, I think it eventually leads us towards the point where it directs the strategic direction of our companies too. And I’m not sure I like that…
Anyway — I sat down to write this thinking I’d take 30mins to structure some of my thoughts (the power of using writing to prompt thinking 😉) and I’m here an hour later having gone way deeper on this than I planned to.
Over the next few weeks I’m intending to come up with a set of principles that I’ll hold myself to when I’m writing with AI. I’ll share them here as they might help you too. It’s very likely they’ll be along the lines of “don’t outsource your thinking to AI”, or “by putting your name on this piece of writing you are confirming that these thoughts are yours, not Claude’s!”. If you already have thoughts on this (or even better you already have a set of principles you’ve come up with!), please share!