😞 This blog begins with a big sigh because, yes, these days, you can now use an AI to summarise papers for you.
➡️ But should you?
That is another question entirely.
You’ll make your own decision on this one as the glorious, sovereign individual you are. 😊
⚠️ But speaking for myself, my decision on this one is a very firm NO and in this blog I outline three reasons why.
📚 If you get an AI to summarise a paper for you, you are not reading it for yourself. It doesn’t matter how good the summary is, or how thorough, or how shiny the new AI is.
If you yourself don’t read the original, you can’t really know what it said.
You remain ignorant of the paper’s actual premises and claims because you don’t have any real knowledge -- you only have what Shane Parrish (2023: 165)* calls the ‘illusion of knowledge’. Your job in your literature review is to evaluate previous work to make the case for why your research question should be asked, and you can’t make any good decisions about that if you haven’t actually read any of that previous work.
➡️ If you want to develop a secure understanding of what a source says and why it matters, you simply have to read it for yourself. There is no substitute for this.
📚 When we report on our research, we are responsible for the claims we make. We cannot stand by what we have said if we don’t really know anything.
Let’s say you have used an AI to summarise papers for you and have written your literature review based on those summaries. What will happen when a reader – your examiner, perhaps – says something like ‘I see here you have dismissed X’s finding about Y. On my reading of X, this dismissal is perhaps hasty. Tell me more about your decision here.’
😟 What will you say? You can’t really account for the decision because you haven’t made it – an AI has. And you can’t have a wider discussion of X’s work because you haven’t read X yourself.
➡️ Summarising is not a neutral act. It is a series of value judgements. When we are doing our research, we need to be the ones making those judgements so that we can take responsibility for them.
📚 We diminish our own experience and de-skill ourselves if we outsource our reading of original papers. To demonstrate this, let me tell you about a little experiment I did. I took a foundational paper in my field, H.P. Grice’s Logic and Conversation **, that I had not used in a while, re-read it thoroughly and carefully as the me I am today (different from the me when I last read it), and wrote my own summary of it.
🤖 I then got ChatGPT to do a summary of the same paper and compared the two reading experiences – the reading of the AI summary, and my reading of Grice’s original.
🟥 The AI version was a summary of summaries, a pre-digested compilation of tertiary sources. It was the consensus view of Grice, what most people remember about the paper. The problem with this is that the ‘remembered’ version drifts away from the original over time, so if we depend on that we aren’t really engaging with the original.
Reading the AI summary was flat, perfunctory, soon forgotten, and meant little. There was nothing in it I didn’t understand, so there was nothing I had to struggle with or learn. It stimulated no curiosity, no sense of agreement or disagreement, and led to no questions. There were no ‘Hang on a minute, what does this really mean?’ moments.
🟩 Reading Grice himself was a deep and valuable learning experience that will go on to inform other learning experiences I will have. There were things I had to grapple with, even on a re-read, because the original is complex and multi-layered.
🔑 But what struck me the most in my comparison between the AI summary and the original was Grice’s inimitable style. Something of the man Grice was reached me through his style. It was a style I could learn from, a language that enriched my own. In my reading of the original, I was in touch with the living, breathing soul who wrote it.
🔴 What am in touch with when I read AI generated text?
➡️ We sometimes think of style as inconsequential, some sort of optional extra when write. But that is not the case. Style is expression of our humanity.
🤔 Will you use an AI to summarise papers for you? That’s up to you. But consider carefully what the actual gains and losses are as you make that decision.
* Parrish, S. 2023. Clear Thinking: The Art and Science of Making Better Decisions. Penguin Random House UK
** Grice, H.P. 1975. 'Logic and Conversation', in Cole, P. and Morgan, J. (eds.). Syntax and Semantics: Volume 3: Speech Acts. 41-58. New York: Academic Press.
✳️✳️✳️✳️✳️✳️✳️✳️
Susan Mandala, PhD | Writing and Language Awareness Consultant | Founder and Director, Writing Works Consulting
Susan is dedicated to helping everyone she works with achieve greater personal and professional success through more effective writing, thinking, and use of language.
➡️ She specialises in dissertation literature review workshops for doctoral candidates and academics new to research.
➡️ Interested in workshop? Contact Susan to discuss a workshop or short course.
🟢 Leave a comment here on the blog.
🟢 Visit her website: www.writingworksconsulting.co.uk
🟢 Connect on LinkedIn and drop her a message: www.linkedin.com/in/susan-mandala-phd-6a94b7290
