My favorite professor, Swarthmore’s Bernie Saffran, used to say that learning is a function of studying time and calendar time. No matter how intensively you study, it takes time for some ideas to sink in.
Bernie’s hypothesis, if true, is an example of what I mean by meta-learning. That is, learning about how we learn.
With AI, the topic of meta-learning comes to the fore. How do humans learn, and how will this be affected by using AI? Will AI make learning more efficient, or will it have adverse effects by making us lazy?
How do AI’s learn? Are Large Language Models condemned to mediocrity by the nature of the content on which they are trained? Or are they destined to surpass us, because their capacity to learn feeds on itself?
For the rest of this essay, I will stick to questions about how humans learn.
Why Books?
In Reading with AI, I suggested that one can absorb the contents of a book by querying AI. Not just one query, but a sequence of queries, including testing your understanding of ideas by putting them in your own words. One cannot absorb the style of writing that way, but I find that only a small fraction of nonfiction books impress me with their style.
I claim that what you retain about a book is much less than the entire book. For example, scholars who love Adam Smith have pored over The Wealth of Nations. Yet if you were to take all of the passages that any of them have quoted from the book and laid them end to end, how much would they amount to? My guess is about 30 pages, at most.
This raises the meta-learning question of why books are so long. As an aside, I would say that textbooks tend to be particularly bloated. Only a masochist would try to read a contemporary textbook straight through.
One cynical view is that the costs of printing, marketing, and distributing a pamphlet are not much less than those for a book, and people feel that they are getting more for their money with a book. Another hypothesis is that low information density is a feature of books, not a bug. Reading the book keeps our minds occupied, even if we only learn a few things from it.
Perhaps we only have the capacity to absorb a few new ideas in a day. If we were to try to read ten 30-page pamphlets instead of one 300-page book, our minds would be overwhelmed and we would learn nothing. Spending all day with a book keeps us from getting distracted by other things, so that we come away from the book with a few insights.
Even if we are constrained in how many insights we can absorb in a day, I still think that reading an entire book is inefficient. The capacity constraint means that I cannot read ten books a day by discussing them with AI. But I could have a focused discussion of one book with an AI, taking an hour rather than an entire day.
Another possibility is that the author “sets up” the key insights. That is, if the key insight were simply presented in its most condensed form, the reader would not grasp it.
Yet another possibility is that writing concisely is challenging, and few authors are up to it. I am aware of this when it comes to making presentations. On any given topic, it takes me much more time to prepare a 10-minute presentation than a 30-minute presentation. Making your points quickly and clearly is difficult.
Why Classes?
We have had the ability to listen to recorded lectures for more than one hundred years. “Educational television” was around in the 1950s. Since 1990, you have been able to buy videos of “great courses.” We all know how to use Zoom.
Why are there still live classes?
I speculate that the answer has something to do with attention. One possibility is that in a live classroom, a lecturer “reads the room” and subtly adjusts tone and pace in order to keep the students listening. Another possibility is what I call Girardian. We pay attention to what other people are paying attention to, so that seeing other people around us looking at the speaker makes us want to focus on the speaker, also.
Or perhaps it is what students teach one another after class that matters. Gathering them together in a classroom gets the process started.
It seems that there is some sort of barrier that makes learning from a screen difficult for many of us. Identifying and overcoming that barrier would be truly revolutionary. What are the meta-learning insights that would enable us to teach students without incurring the costs of assembling them into classrooms?
Riffing on your oft-repeated "we choose what to believe by choosing who we believe", I would suggest books tend to be longer than absolutely necessary because the author has to convince the reader, who he knows is highly likely to be somebody unfamiliar with him, that he is somebody we should believe. That means a lot of otherwise extraneous information is included to demonstrate believability. A book, unless it is from an established author, doesn't have the institutional backing of a newspaper or other regularly published media that can provide a shortcut to establishing (or disproving, as the case may be) that believability. That doesn't mean authors can't also be disorganized, wordy, or poorly edited as well.
I don't think we'll get away from in-person face-to-face classes without super-immersible 3-D virtualization of ourselves as avatars. As you suggested, there is just too much information we communicate nonverbally from students to the instructor, among students, and of course from the teacher to make Zoom or any current chat platform a viable alternative.
"This raises the meta-learning question of why books are so long."
Depends on the subject and the time period.
My personal experience is that it's easy to be concise with people who are more or less friendly and open-minded and can be expected to have a sufficient intellectual foundation such that the "inferential distance" from their starting point to my claim is fairly low.
But it is impossible to be concise when facing a "hostile peer review" of critical adversaries always interpreting everything in as malicious a manner as possible, posturing as willfully obtuse, and looking to exploit every little mistake or poke any possible hole in every possible argument or point out any omission of detail. A lot of books - especially on controversial topics likely to get one in hot water unless phrased and supported just right - end up spending more time and effort on throat-clearing disclaimers and forestalling disagreement and larding things up with countless endnotes than they do on actually preparing and presenting the core of their content.
This is what happens in a modern adversarial trial, which is part of why trials are now so extremely burdensome and the records of cases are now so extremely long. The total length of text in all the concurrences and dissents in SCOTUS and appellate court opinions has got to be two orders of magnitude larger than it was a century ago. There is also the typical kind of too-clever word-twisting legal argument that has to be long by necessity because it is, at heart, completely incoherent nonsense. In the law it's easy to be coherent and brief and usually unnecessary to be coherent and crushingly comprehensive. You can often get away with incoherence buried under a mountain of confusing phrases, but it's hard to get away with the curtly incoherent. So, given that the law has been a total mess for a long time, you get a lot of LOOOONNNNNGGGG incoherent opinions.
Another reason books used to be long was because arguments were presented verbally instead of graphically, and a picture's worth a thousand words. Speaking of "The Wealth of Nations", the late P.J. O'Rourke read through the whole thing and in his attempt at summary made that exact observation, that to the modern impatient reader it was obvious that several pages of Smith's text could have been captured in a simple cartesian diagram or chart, but that was not easy given the publishing tech and costs of the time, and also, readers of that era were experienced in and used to the verbal, not so much the graphical, especially regarding topics distant from geometry and pure mathematics. NB: Internet comment tech also doesn't let me make arguments in pictures, so I'm forced to give you a thousand words, sorry.
For History, it's because history is really complicated and full of both mysteries and important details and and interactions, and adequately "setting the stage" for anything requires a lot of previous history. It is hard to narrow things down in a way that doesn't lose sight of the big picture and how all the puzzle pieces fit together, and it likewise hard for authors of history to avoid the procrustean exercise of forcing the messy facts to serve a narrative and fit into the presentation of a particular ideologically-flavored theme or lesson. You can read 50 huge and high quality books with mostly independent content about WWII and still get important new insights from the 51st.