This observation by Steven Connor in the Madness of Knowledge (loc 5976, my emphasis) feels extremely important for understanding what happens when academics start to habitually use LLMs. What I talk about in Generative AI for Academics in a practical mode as functional and expressive documents (or remixing your work for new audiences) is a matter of what Connor describes as personification here:
But something unexpected happens when one tells what one knows. For that very process, which brings knowing into focus and fulfilment, also begins to separate me from my knowledge. As a consequence, I am much more likely to forget something to which I have given written form. Understandably, and politely, people sometimes ask me to represent the views or arguments that are articulated in things I may have written. Having externalized what I know, or my processes of coming to know, in written form, I am then asked to personify that knowledge, to act as convincingly as I can as the sujet-supposé-savoir of the knowledge. But the mere fact of having articulated the knowledge in some way that makes it intelligible means that the knowledge is no longer quite mine, and must become less and less mine as time passes. This is why, when people pay me the compliment of asking me to explain what I may have meant by something I have written, I have to put myself in their position, by reading it, in just the way they could do for themselves.
Once you’ve written something, how do you personify it in order that it serves a new purpose with a new audience? I’m not suggesting LLMs should be used to endlessly paraphrase your own work in an exhausting play of psuedo-novelty (though this will undoubtedly happen) as opposed to more skilfully and quickly, say, preparing slides for a presentation based on a chapter of the book. The slides are a scaffold for Connor’s personification of our past knowledge, in order to enable it to be reinhabited in a way that brings it to life for the audience.
What matters I think is the reinhabiting, rather than the props used to do it. It creates a smoother process, in which energy can be directed to the part of it which matters the most. But it can also empty out the care for the process, creating functional artefacts which are empty shells and so don’t permit that reinhabiting of knowledge. I’m trying to make sense of my own experience of this variability, in order to better understand the point at which getting assistance in crafting the artefacts to support personification tips over into constraining how much personification can take place. Yet another example of how fascinating ontological questions occur with LLMs when you approach them through a practical register, as I was discussing with Helen Beetham yesterday on the latest episode of the podcast,
He continues on loc 5986 to talk about stockpiling knowledge through writing (which immediately feels like it’s what I do compulsively with this blog) and how this props up a fantasy of the author as the one who knows, rather than the one who has written:
I act and sometimes speak as though I thought that I were stockpiling what I know in what I write, laying up knowledge against the day when knowledge may have gone from me, or I from it. But the more knowledge I accumulate in this way, the less actual knowing there is in me, of me. The knowing will always have to be in what I have written, in the warrant it provides for the fantasy of the one supposed by the writing to know; the one, that is, that provides the imaginary support for the writing that might otherwise impossibly seem to come from nowhere and rest on nothing, but which actually is itself the support for this supposition.
This is why machine writing is so threatening. It presents us with writing that “seem[s] to come from nowhere and rest on nothing”, revealing the projection which sustains the fantasy of our own knowing. It is why, I’m coming to think, practices of personification of knowledge become crucial to knowledge production in these circumstances. We must be more reflexive, more engaged, more alive to what we are doing because otherwise the fantasy could rapidly come crumbling down. I had a recent experience of abyssal despair when I realised Claude could probably finish my new book for me once I’d given it 30k of text to work with. “Why bother?” struck me as a question which was psychic rather than intellectual or practical. I feel like I’m through the other side of that experience now, having effectively answered the question for myself. But I suspect this investment in writing (cathexis, the occupation of territory which facilitates emotional investment) will rapidly start to feel precarious for many people, if we don’t tread carefully.
Mark Carrigan 