I have been thinking about books and reading as we enter the AI/LLM age.
There are a couple of books that were hugely influential when I was a graduate student in the 1990s: Edward Said’s Orientalism and Benedict Anderson’s Imagined Communities.
These books were considered “essential reading.”
Edward Said’s 1978 book, Orientalism, argues that Western depictions of “the Orient” are not objective but are constructed through a long-standing system of knowledge tied to imperial power, portraying the East as exotic and inferior to justify domination.
This book has had a huge impact on multiple fields of scholarship as it forced people to think about the biases in the knowledge and representations of “the Orient” that were produced by authors/scholars from “the West,” a concept that is applicable to other settings as well.
I’ve never actually read Orientalism. I think I bought the book twice (I lost the first copy somehow). I tried a few times to read it, but in the end, I think I’ve read maybe 12 pages. It was just too boring for me.
However, I got the point, partly because by the time I tried to read the book, I had already read and heard so many people talk about it, that I already understand its main idea/concept.
Benedict Anderson’s 1983 book, Imagined Communities, argues that nations are socially constructed “imagined” communities, formed when people who will never meet nonetheless perceive themselves as part of a shared group, a sense of belonging made possible by historical forces like print capitalism and shared languages.
This book has also had a huge impact on multiple fields of scholarship as it enlightened people to the ways in which a national consciousness is tied to modern phenomena, like print capitalism.
I have read Imagined Communities cover-to-cover two times. As was the case with Orientalism, I had also read and heard about the book enough to understand its main idea before reading it, but I ended up reading the book twice: once as a student in a graduate seminar, and once because I assigned it in a graduate seminar.
I would say that reading Imagined Communities cover-to-cover was a waste of time. Anderson’s writing is brutal (the guy made no effort to communicate clearly to a general audience), and the idea/concept of the book is pretty simple. You definitely don’t need to read the whole book to figure it out.
Over the years, there are quite a few books that I have read, but many, many more that I have never read. I’m a very slow reader, so maybe I’ve read less than other people, but in general, I don’t think I’m alone. It is normal to be aware of much more than one has actually read.
From that perspective, I have been trying to understand or see what the effect of LLMs will be on the reading of books. We can now upload a pdf of a book to our LLM of choice, and voilà, it can tell us whatever we want to know.
There is no longer any need to slog through Anderson’s miserable writing. We can get a clear explanation of the book in our language of choice in seconds.
What does this mean? That is what I have been trying to figure out.
On the one hand, I think it’s normal to be alarmed: Does this mean that future “experts” will be trained in a world where they never actually read books?
On the other hand, I also wonder: Wouldn’t reading coherent summaries of (poorly-written) academic writings actually benefit people? And isn’t it great to be able to engage in a conversations with an LLM about a book or article?
However, the biggest question I have is about the act of reading books and the development of expertise. Does the act of reading 300 pages do something to our intellectual development that is essential for becoming a professional historian/scholar/whatever?
I read all of Tolstoy’s writings in one 8-week summer course, and as far as I can tell, all that did was to make me never want to read anything written by Tolstoy ever again. But am I wrong?
Did something positive happen to the neurons in my brain over the course of reading those 1337 pages of War and Peace, especially that final 35-page chapter where Tolstoy lays out his philosophical beliefs in dense prose that made me want to scream in pain and pull my hair out?
The point I am stuck on is this: Is there a certain amount of some kind of writing that people have to read directly to gain “expert” knowledge or skills (beyond, of course, the actual primary documents that one uses in research)?
If so, how much? And what kinds of writings? And going forward, how will that be evaluated/verified?
I have no idea.
Once again you’ve beat me to the punch on something I had been preparing to write about, so I throw out some rough comments here in response to the questions:
“Does the act of reading 300 pages do something to our intellectual development that is essential for becoming a professional historian/scholar/whatever?”
“Is there a certain amount of some kind of writing that people have to read directly to gain ‘expert’ knowledge?”
If becoming a professional scholar requires learning how to emulate the terrible writing of other scholars, then maybe the answer is yes. Otherwise, no.
In grad school, summarizing classic works of the field was a common exercise. In my case, it was a good if painful method for learning how to identify the structure and quality of an author’s argument. But I think I could have become just as skilled at this if the arguments had been clearly communicated, instead of being buried under hundreds of pages of pretentious, unnecessary verbiage. As my interests evolved, I became much more willing to stop reading these books once I got the gist of the author’s general premise, or if it became obvious within the first page or two that there probably wasn’t an actual premise because the writing was so bad. No matter how “important” the book is, reading it is no longer worth my time and energy if I have to struggle to follow what the author is trying to say. And now I can just ask an LLM to give me a summary — which might itself let me know if the book is worth reading. I.e., a preview.
In the graduate courses I teach now, I routinely ask students to identify an author’s hypothesis, variables, and method of analysis, and explain whether the author’s argument holds water. But this is for books and articles that I deem well-written in terms of style. I’m not going to subject students to writing that I find objectionable, even though this excludes large portions of what others might consider canonical literature.
As for the undergrads who roll through, few have any sense of cause and effect when it comes to information presented as text. Stuff just happens. Whether this is a function of fact-memorization teaching to the test in K-12 or a collapse in literacy, I don’t know. Custom chatbots that guide students through the process of identifying how texts present causal argument might be one remedy for this. I certainly haven’t had much success trying to teach them this skill. But this would only work if students are somehow persuaded to engage with the original text instead of using AI as a shortcut. Requiring students to use such a chatbot when sitting in a physical classroom so they can’t cheat is a waste of everyone’s time. And viva voce exams where instructors ask students to verbally explain the components of author such-and-such’s book are not scalable.
Thank you for the comments!!
In graduate seminars, I used to ask students to do exactly what you said here, but I’d try to get them to do it in “one sentence.” Man, they struggled. But ultimately, pretty much everything can be boiled down to one (sometimes long) sentence, so why read 300 pages?
I guess for the sake of documenting information it is still important for books to be created (maybe not?), but when it comes to learning, I can’t see why people have to slog through hundreds of pages, especially when by the time you get through the intro, you’ve already figured out what the author is doing/arguing.
As for oral exams, I agree that it doesn’t scale. I then just recently read this pro-AI-in-higher-education article, and it suddenly dawned on me that AI can do the interviewing, and can do it more objectively than humans. . .
https://www.forbes.com/sites/nicholasladany/2025/08/20/the-radical-changes-ai-is-bringing-to-higher-education/
Then last night I saw this article on “DukeGPT,” a watered down version of ChatGPT (using ChatGPT 4) which, among other things, doesn’t link to sources, like ChatGPT 5 now can do.
https://www.dukechronicle.com/article/duke-university-dukegpt-conversation-questions-generative-artificial-intelligence-provosts-initiative-20250827
I think we can see where everything is going. Responding to the problem that jobs require knowledge of AI but universities are not teaching it, now you can see that places like Duke are trying to incorporate a neutered version of it so as to be able to maintain the old ways of doing things while still educating people in the new. . . All while a student can just open another browser and access ChatGPT 5 for $20 a month. . .
^ that’s me
I really like the idea of assigning students the task of distilling a 300-page book down to a single sentence. I’ve done something similar with undergrads: first, write a half-page argumentative response to a question that refers to an assigned reading. Then condense that half page to a single sentence. Then shrink that sentence down to no more than 30 words. And so on. But now AI can do any version of this. While I could make it a “get out pen and paper, laptops and phones on the floor under the table” classroom exercise, most people will simply hand off such tasks to AI whenever they are outside the classroom. Employers will expect this.
That’s like a graphic design/marketing technique. Write something, take out half, take out half again, etc. But yes, now. . . Hey ChatGPT. . .