I’m increasingly aware of how odd it might appear that Anthropic’s Claude is part of my intellectual lifeworld. I talk to it on a near daily basis about my work, in a similar way to how I talk with collaborators about what I’m doing. These conversations are obviously rather different in their form and frequency, with a myopic quality to the conversations I have with Claude in contrast to the circling around issues of mutual intellectual interest which characterises conversations with collaborators. The conversations with Claude are more frequent yet shorter, less intense but not as rewarding. The risk in talking about conversational agents as part of your intellectual lifeworld, as showing up in a way comparable to human collaborators, is that it might sound as if you’re erasing the differences between them or suggesting one can replace the other. In contrast I experience it as simply increasing the diversity of the dialogues available to you, enabling you to have different kinds of conversations which suit different kinds of purposes.

The problem is that it’s hard to specify in advance how you can use conversational agents as an interlocutor. By interlocutor I mean “a person who takes part in a dialogue or conversation”. Obviously Claude is not a person, as I’m trying to underscore by my insistence on using ‘it’ in my references to this system. But my point is that you can relate to it as an interlocutor, rather than saying it is one. You can address it as a conversational partner in a remarkably similar way to a human interlocutor. With the more recent models* (GPT 4 and then Claude 3 Opus) there is now the capacity for it to respond in a way that is remarkably similar to a human interlocutor. In fact its intellectual breadth far exceeds that of any human interlocutor, even if its ‘knowledge’ (which of course is ultimately pattern-recognition and transformation) is more shallow than that of any domain expert.

I suggest in chapter 4 of Generative AI for Academics that it provides a new way of encountering ideas. It provides new modes through which you can reflect on ideas, connect them and write about them:

Treating GAI as an interlocutor can serve a similar function to Russell’s planting ideas. For example, I have spent the last few days thinking about the Whisper AI functionality which is now built into the ChatGPT iOS application, enabling voice commands with a reliability which far exceeds any other system I have experienced, reflecting on what this means for how we relate to conversational agents. The capacity to talk to ChatGPT, through the medium of speech rather than writing, changes the dynamics of the writing process. I routinely find that ideas occur to me when I am on the move: “Not with my hand alone I write: My foot wants to participate” as Nietzsche (1974: 52) once put it. But I found dictating to ChatGPT particularly amenable in the morning, often in a chunk of twenty minutes or so before I leave for work, producing a response which I would either read on the tram or return to later in the day. I have found it a quick and convenient way of planting ideas in Russell’s sense, even if it doesn’t quite match the intensity of the process he describes. It enables me to quickly record an idea, to externalise it in a way which is accessible later. When I return to it I often find new insights and perspectives, as if the seeds had been growing in my mind since I dictated it earlier in the day. The responses which ChatGPT provides are often useful but the simple act of dictation is itself the main point. It also enables the dictation to be analysed and presented in ways which match my needs, such as categorising ideas into different sets of bullet points and classifying themes using bold and italics. 

It also provides an alternative means of engaging with the project when I either could not or did not want to sit down and write. This switching between modes of writing paralleled the feeling I often find when switching between cafes and libraries every couple of hours reenergises me during an intensive day of writing. If you share the experience that introducing variety into your thinking and writing, in the sense of where you do these activities and how you do them, helps improve your enjoyment of the process and the outcomes then GAI can be helpfully understood as a way to expand the horizon of these activities. It provides new ways to think and write, as well as enabling you to think and write in context which would have previously been difficult, such as walking to work. If you haven’t tried this then I’d suggest doing so before you draw a conclusion. It might not work for you but you won’t know until you try.

My point is not that this should replace existing modes of thinking and writing. But rather that it can supplement them, contributing to a more varied and thriving ecology of ideas within which your creative work can take place. By expanding the range of ways in which you can work with ideas, it creates more space for non-linear creativity enabling you to keep in touch with the feel of an idea and pursue it wherever it takes you. It gets at the essence of what I’m trying to argue in this series: enjoying writing involves an interplay between limits and creation in which the former creates the conditions for the latter, while the latter relies on the former to give it shape and keep it sustainable. If you’re unsure where to get started, past this blog post into Claude 3 Opus or Chat GPT 4** and ask it to give you suggestions about how to get started on using generative AI as an interlocutor to support your writing.

---

*Until GPT 4 I was vaguely sceptical of the implications of generative AI. I believed it would have a significant impact through political-economic force of will, but I wasn’t entirely persuaded of the sociocultural significance of the underlying technology.

**I would honestly suggest it’s not worth trying to use GPT 3.5 or the weaker Claude models for serious intellectual work. If you’re at all interested then pay the $20 and try it for a month, in order to make an informed decision.