If this email forwarded to you, sign up to receive weekly letters for the examined intellectual life.

All previous editions of The Letter are available on our website.

— Written by Reece Barlow

AI summaries are everywhere, and we don't seem to mind all that much. Take searching the Internet via Google, for example. Gone are the days of typing a few keywords together and browsing individual webpages looking for what you need; Google currently behaves less like a search engine and more like a chatbot because every single Google search is accompanied by an "AI overview" which has done the digging for you. Meta is even rolling out AI summaries on Reels, content which is already in a condensed form. In the context of scholarship, the prevalence of AI summaries is equally, if not more, pervasive. The information we consume, casually and for our scholarship, is collapsed before it reaches us.

Researchers are spoilt for choice when it comes to tools which offer to summarize academic literature. No self-respecting research tool would call itself such if it lacked the capacity to summarize PDFs. Even Google Scholar and Adobe Acrobat, relatively "low-tech" tools which have been in use by researchers long before the possibility of an AI summary could even be imagined, offer "quick reads" or "instant AI powered summaries." A summary, by its very nature, strips away excess context, theoretically leaving behind the most important bits of whatever was summarized. Typically, discussions around whether AI summaries are useful, or appropriate, focus on the quality of an AI summary from the perspective of the reader. How do AI models decide what is important? Are such summaries really better than an abstract? Can we trust the delegation of decisions regarding relevance to a piece of software? As I have written before, the exact nature of an individual's disposition to AI in research can be partially linked to their personal conceptions of scholarly virtue. For example, if one believes that close, careful readings of previous work make one a good scholar, then AI summaries are abhorrent; if being a good scholar means consuming large amounts of literature as fast as possible, then AI summaries are of great assistance. However, for this essay, let us leave questions regarding the quality of AI summaries - and whether they are "good" for scholarship - to one side. Instead, why don't we turn our attention to how AI summaries may affect scholars themselves? For a scholar does not only read summaries. A scholar also produces work which will inevitably be subject to summarization.

The ubiquity of AI summaries has made the collapse of hours of work and thousands of words into single sentences, or at the longest a handful of sentences, routine. Long-time readers of The Scholarly Letter will not be surprised that my concern lies with how such routine compression, flattening, squeezing, cutting and collapsing may injure the scholar who produces that work. Scholarship can be a job. However, it is a unique career path in the sense that it is, almost always, a vocation as well. To put it bluntly, scholarship requires that a piece of a scholar's soul be added to the work we produce. In the vast majority of cases, a scholar is more than willing to oblige. We pour ourselves into our work, agonizing over the smallest of details to produce something worthwhile. We do this because in giving a part of ourselves to the scholarship we do, it also becomes a part of us in the process. So when the fate of the things we have lovingly and carefully created is to be squashed and cut into a single line for convenience, it hurts. And it should, shouldn't it?