Jill O’Neill wrote a fascinating piece in Scholarly Kitchen about the challenge of having too much to read. Everyone in academic research is familiar with this challenge, but all too often, our response is, as Jill O’Neill, states, “TL;DR” (too long; didn’t read).
At UNSILO, we often mention in presentations the remarkable statistic that there are over 3,000 new academic articles published in the STM (science, technology, medicine) area every day. This means that no human researcher will be able to keep up with such a torrent of content. Some kind of automated pre-selection is vital.
In her post she describes how some of her peers address the problem. One suggested a spell of binge-reading. Ms O’Neill then mentions various AI-based tools that claim to have a solution for this problem, and describes how in the UNSILO AI in academic publishing report several publishers report they are using AI tools for relatively straightforward tasks, to complete processes faster than a human alone could achieve them, such as finding related articles or carrying out technical checks on submitted articles. Her conclusion, however, is surprising: “While I see this as a sensible use of AI by content and platform providers, the pragmatic reality suggests an uncomfortable possibility. There is no magic solution. AI isn’t currently up to the task. “
She then tries two or three other AI tools claiming to solve the challenge of too much to read and finds them unsatisfactory. Does this prove that “AI isn’t currently up to the task”?
The argument employed here is curious: “figuring out what to read as well as finding the time needed to read it continues to be a human problem. It can’t yet be delegated to the machine.” In other words, because AI cannot currently completely solve the big challenges, such as creating a full summary of an article, there is no point in using AI to solve any of small-scale, intermediate problems such as identifying key terms, or (in UNSILO’s case) finding relevant related content, or checking that a medical research paper has not omitted the required ethics statement.
Ms O’Neill’s piece concludes with the exhortation: “Reading with concentration is work. Give it the time needed”. Unfortunately, for researchers attempting to read the 3,000 new papers published every day, or for the lecturer who claims to read the 60 or so new articles on AI added to the preprint repository arXiv every day, concentration may not be enough.