This is a really thought provoking and intellectually generous review of Generative AI for Academics by my friend Andrew Dryhurst. In one sense I think he takes the book too seriously because his review is treating a guidebook as a theoretical intervention. In another sense I think this is an insightful thing to do because he draws out the theoretical presuppositions which are obviously there. The book contains an implicit ontology of large language models which informs the practical recommendations. His agenda is also constructive rather than simply taking issue with the book:
Given that the book primarily serves as a critical guide for academics engaging with GenAI, the critique of actualism being leveled here is not a damning one. Rather, it serves as an invitation for critical realist scholars to enquire into GenAI in the domain of the real. The book consistently focuses on moments of synchronic interaction: scholars prompting conversational agents, academics navigating institutional pressures, the shared norms being developed around the technology’s usage. These are unequivocally real and important objects of enquiry. Yet, they are also expressions of deeper generative mechanisms that the book identifies but does not theorize.
The bit of the ontology I’ve since disowned (largely through prodding from Helen Beetham and Milan Sturmer) is what I realise was a voluntarism underpinning the analysis. It was informed by an auto-ethnography of incorporating LLMs into my academic work which left me with the belief that responsible practice was ultimately a matter of choice. Yet when I returned to normal routine, I realised the temptations of misuse were vastly higher than I imagined and they could not be mitigated entirely through technological reflexivity. In other words you have far less room for autonomy about how you do your job in conditions you don’t control, then you do when you’re experimenting with a new technology to see how it works.
I have a long-winded critical realist framing of this about the causal powers of occupational roles vs social actors but essentially it’s the argument Cory Doctorow makes far more clearly about centaurs and reverse centaurs. Working with LLMs looks very different if you have the freedom to control the integration into your work than it does if you’re being told how to use them at work. There are lots of interesting things shaping my initial category error (mistaking my autoethnography as a faithful representation of my actual conditions of work) which i might write about at some point, drawing on for example Jana Bacevic’s work from the late 2010s on scholarly reflexivity and critique. It’s in this sense that I think Andrew’s sharpest critique of the book is actually incomplete thinking on my part:
Margaret Archer’s notion of reflexivity is the main organizing concept of the book, and it is worth being precise about what this entails for the analysis. Here, reflexivity concerns agents monitoring their own positions and deliberating about how to respond to their circumstances. As Carrigan deploys it, reflexivity directs analytical attention toward the level of the ‘actual’, entailing a focus on contingent interaction, synchronic adjustment, and situated self-understanding, without ever engaging the generative mechanisms at the level of the ‘real’ that produce the circumstances to which agents respond
This ultimately is a disagreement about the quotidian dimension of reflexivity. I think the Erasmus Archerians are deeply sceptical of this and I only sort of understand why. For me this is a matter of what Giddens used to call “discursive penetration” into social structures. In everyday life the degree of experienced insight into the deeper structures shaping our circumstances varies widely and is obviously always fallible. To analyse reflexivity in these terms doesn’t make it an actualist concept, as much as it insists on treating the real of reflexivity in terms of the actual conditions of operation.
The problem in a diagnostic or normative mode is a real one which I (now) fully accept in a way I didn’t a few years ago. It’s why reflexivity-framed practical guidance can easily slip into voluntarism in the way I described at the start of the post. The analytical risk is clearly there when it comes to anything normative because it actualises the horizon of the subject’s own reflexivity. But the analytical restriction of treating quotidian reflexivity seriously as reflexivity is justified by the ontology. In this sense I think Archer’s reflexivity trilogy needs to be read as a correction of her earlier work, elaborating what she saw as an oversight, rather than a departure from it. She was utterly explicit about the fact this not an epistemological break and if you want to disregard the later work, the burden is on the person doing it to explain why Archer’s diagnosis about her own work was wrong.
I’m not for a second suggesting we should refuse to criticise her, given I’ve someone who is building a psychic ontology of reflexivity she would have been utterly hostile to. But this sort of work needs to take her own account of intellectual trajectory seriously and explain why she was mistaken to take the direction she did at these nodal points in the overarching project. I think this would be a really valuable conversation to have amongst those who are committed to the continued value and relevance of Archer’s intellectual project.
