You couldn’t quibble with that title – at least, that’s what I thought, until I was reading about Paper Digest, an AI tool that creates summaries of scholarly articles, and I was surprised to come across an article against all such AI-based tools, not because they didn’t work, but on principle: they should not be needed. When I looked at Paper Digest, incidentally, I was impressed by the quality of the summarization.
But David Beer’s article which appeared in the LSE Impact of Social Science, was not about the quality of the summaries. Beer, who is professor of sociology at the University of York, has two main complaints:
- Automatic summarization removes the opportunity for “reflection and non-core ideas”.
- Such tools as Paper Digest push students and researchers “in a particular direction”, which pushes them to work faster so they feel that are not being left behind. The growth of publications means we need to be “ever more efficient … so we can race through the materials that surround us”.
Beer suggests “The human researcher is not in a position to keep up with the expansion of knowledge”, and does not disagree that this growth in the number of publications is taking place. But for him, “the growth of literature is undeniable; how we choose to react to that growth is contentious”.
Clearly, he is not happy with using AI tools if they are a “prosthetic to our limited cognition”. But how far would he want to go? Should we ban the use of computers, which enable us to write and to communicate faster? Should we ban libraries and search engines, which are clearly a prosthetic to our very limited ability to find material? If you can’t read every article, do you try to read everything and abandon the attempt when you run out of time, or do you use all means available to simplify the selection of what you will read in full? Lisa Janicke Hinchcliffe expressed it succinctly in her Scholarly Kitchen article about Paper Digest:
Like many other people, I often find myself trying to figure out which articles are likely to be most relevant or important for a project I am working on. I use well-established heuristics such as scanning the article title, author(s), journal name, abstract, and keywords. I also greatly appreciate various services that “push” documents to me based on algorithms that use my past reading and personal publication history to predict my future interests. None of these approaches are perfect but I benefit from them all.
What alternative is there? Does Beer suggest with twice as many articles to read that we read at the same speed we do currently? This became impossible for science subjects many years ago. Within sociology, academics are publishing double the number of peer-reviewed articles per year compared with 25 years ago. There were more than twice as many sociology journals in 2016 compared with 1986. Yet Beer seems to suggest that the growth in papers is somehow linked to AI tools for discovery:
The suggestion is that the piling up of research requires us to find ways to keep up and accelerate. It suggests to us that we need to be ever more efficient at dealing with these fields of knowledge, so that we can race through the materials that surround us. Of course, this would also exacerbate the problem, by pushing all of us to also produce more on a tighter timescale, ultimately adding to this avalanche of information.
A far more likely reason for the growth in publications is from within the academy: the drive for tenure and for full-time academic positions makes researchers publish more. AI did not create the challenge of discovery, but helps to solve it. If the figures quoted by Paper Digest are correct, every academic reads the average of 21 papers per month at 32 minutes each, which means 10.5 hours of reading per month. Since there are twice as many sociology articles being published each year, does this mean that every academic will now spend 21 hours per month reading new papers? It sounds unlikely to me that researchers have an extra 10 hours per month.
As for Beer’s first objection to AI summarization tools, the opportunity for reflection and for reading non-core ideas will come when the researcher reads the articles selected in full – exactly as happens today. These tools are not a substitute for full reading; they simply enable the choice of which articles to read in depth. They provide a way to deal with, in the words of the Inside Higher Ed’s headline “one discipline’s soaring publishing expectations”.