Ten arguments about scholarly use of LLMs

Summarised by Claude 3.5 from my knowledge base:

1. The framing of conversational agents as mere tools misses their real potential as intellectual partners in our work. When we approach them as interlocutors – participants in an ongoing dialogue about our ideas – we create opportunities for genuine insight and creative development. This isn’t about outsourcing the mundane aspects of academic work, but rather about finding new ways to think through our ideas by articulating them to a responsive (if limited) partner. The reflexive dialogue which emerges helps us clarify what we’re trying to say, often in unexpected ways.
2. The distinction between thinking with GenAI and using it to substitute for thought is crucial. When we thoughtlessly outsource tasks to these systems, we risk amplifying their inherent weaknesses – hallucination, common knowledge problems, and the perpetuation of biases. But when we approach them reflectively, being clear about our intentions and expectations, we can harness their capabilities while maintaining intellectual control over the process. This means accepting that using these systems well often requires more rather than less cognitive effort.
3. The threat of automation isn’t primarily about a sudden replacement of academic labor, but rather about how individual choices made under pressure could normalize practices that make systematic automation seem inevitable. If we respond to productivity pressures by using GenAI to accelerate our work without reflection, we demonstrate that key aspects of academic labor can be automated. The real risk is that we participate in creating conditions which make our own obsolescence more likely.
4. While GenAI offers both quantitative improvements (doing more in less time) and qualitative improvements (enriching how we work), privileging the former over the latter is dangerous. The academy already suffers from hyperproductivity and metric fixation. Using these tools primarily to increase output will only intensify these problems. Instead, we should focus on how they can help us work in more varied, creative and fulfilling ways.
5. What I call an ecology of ideas represents an approach to GenAI that prioritizes its role in nurturing and developing our intellectual work. Rather than seeing these systems as labor-saving devices, we should understand them as part of an environment which can support creative and analytical thinking. This means being attentive to how different tools and practices interact, creating conditions conducive to scholarly thought rather than just efficient production.
6. The remarkable versatility of conversational agents is precisely why they require thoughtful engagement. Unlike template-based systems, they reward careful articulation of context and intent. The more precisely we can explain what we’re trying to achieve, the more effectively they can contribute to our work. This creates a virtuous circle where the effort we put into formulating our requests leads to more sophisticated and helpful responses.
7. The way we write rarely follows a straight line, even if the final product might suggest otherwise. Our ideas develop through surprising connections, sudden insights, and meandering explorations that eventually cohere into something meaningful. When we approach GenAI primarily as a tool for efficiency, we risk imposing a false linearity on this inherently non-linear process. Instead, these systems are most valuable when they support the productive messiness of thought – helping us explore connections, test ideas, and gradually work towards clarity without forcing us down predetermined paths.
8. Writing serves as a way of decluttering our minds, moving thoughts from where they consume our energy into a form where we can work with them productively. This isn’t just about getting things done – it’s about maintaining intellectual and psychological equilibrium through the process of articulation. Conversational agents can support this process by providing an ever-present interlocutor with whom we can think through emerging ideas. But if we reduce this to mere productivity, we lose the deeper value of writing as a way to manage our relationship with our own thoughts.
9. Our individual decisions about how to use GenAI in our writing practice won’t remain individual for long. Like social media before it, these choices will create ‘network weather’ – shifts in professional culture that emerge from countless small decisions about how to work with new technologies. We can’t control this process entirely, but we can try to shape it by being conscious of how our choices might aggregate. This means thinking carefully about the precedents we set and the practices we normalize.
10. There’s an unavoidable tension between the enriching potential of GenAI for individual writing practice and its possible collective consequences. I’ve found these tools genuinely helpful in developing and articulating my ideas. But if most academics primarily use them to increase their output, we risk creating conditions that make thoughtful scholarship increasingly difficult. The challenge is to find ways of working with these tools that support rather than undermine the deeper purposes of academic writing.