The Demise of Reading?
“Americans are probably reading more words than ever before. What has changed is what they read, and how. People are bombarded with emails, text messages, X posts, Reddit threads, Instagram captions. This explosion of textual fragments has come at the expense of devoting sustained attention to longer written works that convey rich and complicated information. ”
That’s a bit from “The End of Reading Is Here” by Rose Horowitch, recently published in The Atlantic.
The article chronicles the general decrease in reading (of books, articles), the general decrease in sophistication of what people are reading, and the general decrease in literacy skills. People are “losing the higher-order abilities of comprehension and synthesis” and in her view, “things are about to get worse, and fast.”
Kids watch more and more videos, and read less and less. While “video contains more information than text,” watching videos is “a more passive form of engagement than reading” and generally “does not stimulate deeper thinking.” Student reading and writing scores on standardized tests are “at their lowest level in more than three decades.”
What does that mean for instruction at the college level? One avenue is remedial: “When these students get to college, their professors find that they have to teach them how to comprehend a text.” But remedial education for reading skills makes sense only if students need those reading skills in order to graduate, and I worry that increasingly, they won’t. The consumer model of higher education paired with a highly instrumentalist picture of the value of education points to a further relaxation of literacy demands on students, not greater student preparation to meet existing or more stringent demands.
Along those lines, here is one of the more striking passages of Horowitch’s article:
Reading has come to seem extraneous even to some of the best-educated members of society. Margaret Rennix, Harvard’s assistant director for humanities and social-sciences support, told me she’d spoken with a student who was struggling to read a book written in Old English. The culprit: Anthony Burgess’s 1962 novel A Clockwork Orange. (The student used ChatGPT to “translate” the book into easier language.) Not long ago, a Harvard sociology professor, troubled by course evaluations in which students said they resented the amount of dense reading they were assigned, asked Rennix to speak to his class in defense of reading. She had to explain—to students at America’s most elite university, taking a course in a discipline rooted in written observation, argumentation, and analysis—that excerpts and summaries cannot capture the depth and sophistication of a complete primary text. Rennix told me that some students now view reading as an unnecessarily burdensome way of acquiring knowledge. “By asking them to read,” she said, “professors are arbitrarily withholding information from students by forcing them to get it through this more difficult medium.”
A course’s being difficult will, more and more, be understood as a professor being “difficult”. When that is the prevailing interpretive framework, it is education that will be increasingly difficult.
There’s a nice reply piece by Cal Newport:
https://calnewport.com/why-reading-matters/
“Student reading and writing scores on standardized tests [are] ‘at their lowest level in more than three decades’.”
😉
There’s also evidence that proofreading abilities are on the decline.
(Thanks for catching that. It’s fixed now.)
No prob. Just to be clear, I didn’t mind the typo; I just liked the coincidental irony of where it occurred.
This is a true, but highly misleading statistic. In math and reading we saw huge gains on tests like NAEP for nearly twenty solid years and then a consistent and accelerating decline in roughly the last ten. The kicker though is that even after the decline they are on still on par with student score in the mid to late 90s or even slightly above. So I would remind profs in their 40s-50s like myself that you may be dealing with the worst group of students in your lives but on average they are as good or a little better than we or our classmates were. I’m not saying that the decline isn’t highly worrying– it’s honestly terrifying– but at the same time most of us have absolutely no cause to [poop] on current students or whine too much about our own lot.
Ok but FYI I’m not asserting the stat (insert joke about reading comprehension, irony, etc. etc.).
Sorry if I implied you were. I’m just really tired of people drawing quick easy lessons here or making a worrying trend into the end of civilization as we know it and so very touchy on presenting these figures without context.
Your source for this? The NEAP chart I’m looking at shows very little change in average reading scores from 1980 through 2020, hovering around 260. The 4 percentage drop in 2023 — from 260 to 254 — is by far the largest drop recorded in that time.
https://www.nationsreportcard.gov/highlights/ltt/2023/
But the main problem may not be lower than expected reading scores on standardized tests, but the poor reading habits and skills our students bring to the college classroom and into adulthood.
Reading hasn’t been quite as dramatic as the rise and fall of math, though both have the same pattern. But the point is scores were 257 in 1990 258 in 1994 and 256 now. That’s not much real difference. However, dramatic you take the rise or fall to be at any rate doesn’t undermine my larger point that hysterics of the Atlantic crowd that act as though we’ve descended into some sort of pre-printing press illiteracy don’t fit what the data says. It’s the trend that’s deeply worrying– though extrapolating huge continuing trends from a sharp decline in a few years is statistical malpractice– not the raw scores but the Atlantic and other media outlets act like the raw scores are something to absolutely panic over which is dumb.
It’s also worth pointing out that the NAEP is supposed to be hard. It’s is not a basic literacy or numeracy test. When we hear about students missing the mark in mathematics it’s more that kids can’t identify the graph of a linear function or figure out the area of a trapezoid than it is that they can’t make change.
Except that students can do relatively well on reading tests without reading novels or engaging with philosophical arguments in complex texts. The whole Standardized educational system of rubrics and benchmarks coupled with our digital, anti-intellectual culture encourages instrumental reading/learning and discourages any passion for reading and the deep learning that comes as a natural result. Of course, this isn’t exactly a new trend in our culture, but the pace and extend of the problem evident even in our college classrooms is especially troubling.
For antifragile systems like us humans, Difficult = Growth. So, if Reading = Difficult, then keep insisting on reading, even at the cost of being deemed a curmudgeon.
The problem is that student evals are closely tied to (perceived) difficulty, and retention/promotion/tenure are closely tied to student evals. So there is tremendous pressure to make things easy to get the high student evals. Their use to evaluate faculty performance has been incredibly detrimental.
I have, frustratingly, also had grad students complain about difficult prose in historical sources. I’m in a history-focused grad program.
Bleak… (future commenters please make any objections to my pessimism brief, not gonna read all that)
That’s wild. They don’t have to be there (philosophy higher ed is a choice) and there are plenty of people still willing to read who would rather be there!
I mean, they are willing to read at least enough to complain about it. And some of them figure it out. (This is true of many undergraduates, too. Some complain and give up – some complain and work through it.) I didn’t mean to suggest that, by complaining, they were giving up or weren’t willing to learn. I just wanted to point out that this is a trend that has ramifications for incoming graduate classes.
On the plus side, think of how much easier it will be to be lauded as a genius–just read a few books properly and for yourself, like in the good old days of Newton!
There is no lauding. (If you read the article…) the readers are marginalized.
a student who was struggling to read a book written in Old English. The culprit: Anthony Burgess’s 1962 novel A Clockwork Orange. (The student used ChatGPT to “translate” the book into easier language.)
I wonder about this example. There are parts of A Clockwork Orange that are hard to read – but not because they are written in “old English” (let along Old English, which would be really hard!) but because of use of “nadsat”, a sort of pidgin version of Russian (with idiosyncratic transliteration, too.) The speech of Alex and his “droogs” (that’s an example) are all full of it, and it does make it hard to know what they are talking about. If this was the problem with the book, then the example is almost an example of what it’s complaining about – a person not knowing what is going on in a literary text.
I think the OE comment was a joke. I can imagine Nadsat being harder to parse so long after the Cold War.
Riddley Walker, now. That takes a lot of getting used to.
This is an interesting piece, but I want to canvass a different perspective, or at least one I think is worth putting on the table.
We currently treat declining reading as a problem because we take the relevant contrast to be learning less and being less informed. I suspect that contrast won’t hold for much longer given brain–computer interfaces already exist and are becoming more sophisticated.
So one-way BCIs already exist. It’s possible right now to issue commands to a computer by thought alone (look it up if you don’t believe this, people can already type like 70 words per minute using just thoguht commands). Two-way BCIs, where information flows back into the brain, remain more speculative, but proof of concept is already in place. And given the pace at which increasingly capable AI is accelerating scientific progress, it seems pretty naive to assume we won’t have working two-way BCIs within ten to fifteen years.
Now, if that’s right, reading may soon cease to be the medium through which we take in information and form beliefs. That matters for how we assess the present trend I think.. What makes declining reading look like such a loss is a contingent fact about our situation, which is that reading less means acquiring and retaining less. the point being that as two-way BCIs plausibly become commonplace, reading less simply won’t carry the cost we currently assign to it vis-a-vis getting information and storing it.
I had to read this twice – I’ve got a 5 year old kid. are you actually saying that by the time my kid graduates college people will get information by just giving a thought command and getting the answer in their brain ? Is that effectively omniscience ??
It’s definitely not omniscience, since the sources you are communicating with are no more omniscient than libraries or the internet right now!
exactly
It’s hard to know before we actually have BCIs and see how they work, but what likelihoods would you assign to the following possibilities?
1) It turns out to be very doable to hook brains up to computers, but extremely difficult to get high density information of certain kinds out of a computer into a brain, because the brain can only handle incoming information in certain forms. So for instance it turns out to be impossible for the BCI to stuff an accurate atlas of the entire world into your brain. This ends up applying to most of the stuff you would learn by reading: it turns out to be impossible for the BCI to stuff Making It Explicit into your brain because that information is not the sort of thing that your brain can get when it’s sent down the pipe, so to speak (or your brain could get it if we knew how to package it correctly, but we don’t).
2) It turns out there is stuff you get from reading that cannot be assimilated by a computer and thus communicated through a BCI. So for instance it turns out that whatever happens to people when they read War and Peace is something that doesn’t happen when they ask the computer to tell them about War and Peace, even if the computer can tell them lots of things very fast through the BCI. And it turns out that some of this stuff that you can only get by reading is important.
3) It turns out that the development of certain skills is tied to reading and is not facilitated by information acquisition via BCI. So for instance maybe learning to put yourself into other people’s shoes and see things through their eyes happens in part by reading books, and it doesn’t happen when the book is fed to you via a BCI.
4) BCIs turn out to be a divisive technology, with some people lauding their value and others objecting to what we lose by becoming reliant on them. In order to shore up the defensibility of their position, pro-BCI people start to disvalue things that you can’t get via BCI, to ignore evidence that there are things you can’t get via BCI, and so on. To the extent reading has any value that is not captured by BCIs, the pro-BCI forces (including the businesses that make money selling BCIs) wage a concerted campaign to get people not to believe in or care about this value.
It is also maybe worth noting that although I think BCIs are a million times more likely than the singularity, they share with the singularity the tantalizing feature of always being just around the corner, according to the futurists, but never actually here…
I suspect a big part of this is not that BCIs are better or worse than reading, but just that they’re *different*. And for an example of a medium with drastic differences, we don’t have to go that exotic – that’s already what we have with video.
There are lots of things you can get from video that you can’t get from reading – not just things like an easy explanation for how to tie a tie or how to change a tire, but also things like what it was like to visit certain places in the 1950s or 1970s. Language is just an inherently limited medium for many kinds of information.
But as you note, there are other things language is a better medium for, and if text-based media are largely superseded in the population by other media, then these kinds of knowledge will be less widely distributed.
(Also, I think it’s slightly ironic that Making it Explicit is what you chose – I suspect there are nearby works in the Pittsburgh tradition, maybe not that one in particular, but some of the more pragmatist aspects of Sellars and Haugeland, for which language may be an inherently limited medium and something like a direct brain connection may be much better.)
The story-telling part is left out of this scenario along with all that goes with it, like that it is your great uncle telling the story, or a little girl during an upheaval in her country, or an adventurer from an exotic land, etc.
Or are you imagining that through this interface an LLM chatbot tells you stories? I am assuming not. In which case: who wants information dropped in that comes from nowhere? An LLM has no standpoint. Even if you think of yourself as a person who has always been consumed by curiosity about how things are and how they work, and want more than anything else to know more and more about everything, do you think this would be satisfying? What if delivered to you were answers (including justifications and evidence) to questions you had never asked or wondered about? And not through time, i.e. in a story, but in a flash. What do you become? A storage device. How is this an exciting prospect? Very interested to know whether you do think this. Maybe I’m missing a joke.
fair q, and yeah it wouldn’t tell you a story, it’d just pipe the info in. but the story isn’t really doing the work. what narrative adds on top of content is a significance tag, and the tagging machinery is pretty well understood: amygdala/noradrenergic salience, vmPFC for personal relevance, insula for the felt bit, all modulating hippocampal consolidation. a story is just one roundabout way of driving that.and we can already do a crude version …. brief amygdala stim right after you see a neutral image makes you remember it better the next day, with no subjective emotional response at all (inman et al 2018). the BCI just does it on purpose, time-locked to the content.
only fiddly bit is binding: the affective signal has to land on the delivered content, or it’s just vague arousal you end up misattributing. engineering problem, not a mystery.
so you get the info plus the mattering. seems fine tbh.
This is sort of cool. Thanks for understanding my question and your reply.
To restate to see if I understand: you’re saying that amygdala stimulation can make people remember things otherwise unremarkable to them (this has been shown), and that, in principle, brain processes could be triggered so that there is emotion felt, attention recruited, perspective formed, context found – i.e. meaning made/mattering happening. And then, (maybe) this mattering can be associated with very specific downloaded informational content (pending a solution to a binding problem).
I do now see how we could do without written texts (narrative) or reading. I still wonder about the role of time, and also, the role of effort.
A new question (but maybe takes us too far from the topic of this thread): who gets to decide which parts of the downloaded content should be made meaningful to the person who is receiving it?
On the question of who decides: presumably some versions of this would be like super-video, where in addition to images and sound, you get smells and sensations, and stimulations of different types of brain systems. Some people would use this artistically, others would use it educationally, and probably a lot will be generated on the fly by successors to LLMs. So there will be a lot of different patterns to who decides, though it’s hard to say if some will turn out to be dominant.
That sounds more attractive than what I was thinking. I was imagining the transfer of information to be below the level of conscious awareness, not even dreamlike.
I don’t think this is true, and I don’t imagine many philosophers believe it to be true. Reading (on its own) is a really bad to obtain and retain knowledge. If all you want is to obtain and retain knowledge, then lectures, handwritten notes and daily flashcard practice are by far the best method (as any student in an knowledge-heavy program like medicine or engineering will tell you!).
The problem instead is that:
yeah but this could all be done by a fancy BCI (i’m talking about an AI optimised BCI in 10 to 15 years), that’s the point. including higher order stuff… there’s no barrier to BCIs auto-cognitively-integrating info . so lack of reading now seems to come at the cost of at least some of the epistemic ‘good stuff’, my point is that’s highly contingent
When you say a fancy BCI could do all this, i.e., “the higher-order abilities of comprehension and synthesis,” what do you mean? Does the optimized AI just drop the synthesized and comprehended information into our brain – which we then use to do stuff (make decisions, make further connections, identify weak points, etc.)?
The whole point I think is that these higher order skills are not just intermediaries giving us nuggets of information. That’s not what reading does either. It’s the process which produces thinkers and critical citizens. If your BCI can replace that, you are just replacing the sort of person reading in theory helps make.
All of this is contingent on a process that works in such a piecemeal fashion which I doubt. How we think critically is a lot closer to being a blackbox which makes replacing it with a BCI speculative, at best.
Information access and retrieval is not particularly interesting–or difficult. We’re there already, in that so long as they recognize relevant keywords, anybody at all can find the relevant SEP entry and (struggle to) read the words out loud to you. But that’s not really what we’re interested in, here. That’s not at all the same thing as Neo downloading the kung fu module. If you have the SEP in your head or at your fingertips and can recite it without even reading, that’s cool, but it’s really not the same as being able to connect the dots and explain it (or knowing kung fu).
Well, reading is not primarily about ‘learning more and being more informed,’ ‘taking in information’ or ‘acquiring and retaining more.’ The main goal is precisely not maximally efficient information extraction. The starting assumption about human cognition is fundamentally flawed here; we are already letting AI define what knowledge and cognition entail, and that is dangerous circular reasoning.
Nope, we’re not letting AI define what knowledge and cognition entail. Take knowledge. Knowledge is probably (if analysable) a justified true bleief plus an anti-gettier codicil, perhaps it is something like true belief through virtue or some such thing… AI doesn’t change anything to do with that. AI is absolutely orthogonal to what analysis of knowledge is correct, if indeed there is a correct analysis. if unanalysiable per williamson, then the rise of AI has nothing to do with knowlege being the most general FMSO. so basically: no!
The point is that we need to start with the question what knowledge is and what ‘reading’ entails, and there are more minimal and more maximalist accounts. I would argue, for instance, that AI does not have ‘knowledge’; it has information and can do impressive things with that information–some very useful things, admittedly, but no more, so far.
AI or BCI can displace reading, as you claim, ONLY if we, as human beings, believe that reading is primarily about maximally efficient information extraction, that is, if we ourselves reduce ‘knowledge’ to the AI/BCI model. (From the fact that I ‘read’ some information which I need, it does not follow that the primary goal of reading is information extraction.) The latter reduction is, in effect, already happening, and the interesting question for me is why.
“ I would argue, for instance, that AI does not have ‘knowledge’; Alright, let’s have it, what’s the argument? There are viable externalist lines in mainstream epistemology that can straightforwardly accommodate AI possessing knowledge. Let’s assume knowledge entails belief. So, a constraint on AI having propositional knowledge is that it can host beliefs. One might think something along the lines that bleiefs require consciousness and AI isn’t conscious so AI can’t have knowledge. But that’s too quick. Beliefs can be occurrent and dispositional, and on the dispositional line, what matters is (somethign like) stored representaitons that are integrated with other representations as well as with behavioural dispositions. there’s nothing to preclude AI from having beliefs on this line, and by extension, knowledge, so long as the representations hosted and integrated are the result of reliably truth-conducive processes/competences. The above of course reflects some substnative theoretical choices in mainstream epistemology but these are hardly idiosyncratic ones. So, in short, if you’re an externalist in epistemology and you elect to host certain permissive views of belief already available on the market and reasonably popular, you can accommodate AI knowledge. There are other routes to this conclusion, such as the Cappelen/Dever Whole Hog line, which reaches the same conclusion by another kind of broadly abductive argument.
One way to think about Prof. R-S’s idea is this.
Instead of debating which route in contemporary epistemology to take, it is possible to question some of their shared assumptions. For example Dewey thought of intelligent activity as behavior born out of a real, present doubt or problem whose resolution involves the removal of that doubt or problem, and he thought of knowledge as the repository of such behaviors and the ability to deploy them appropriately. On this view reading, an activity that some (Ong) have argued is largely constitutive of modern self-consciousness, is a primary site of intelligence, and knowledge is manifest in the practiced ability to engage in this activity and in no way reducible to the information extracted from it.
To be sure, the conception of knowledge as justified true belief or some variation on that theme is much older than AI or information theory, reaching back at least to Descartes (if not the Theaetetus). But it has also long been challenged by thinkers as diverse as Reid, Novalis, Condillac, Peirce, Wittgenstein, Ortega y Gasset, Polanyi, Maturana, …, even Reydams-Schils.
One could point to AI’s astonishing feats as evidence that an information-theoretic conception of knowledge was right all along. Or one could see them as demonstrating just how extensively language use and information processing can be even when dissociated from any recognizable process of grappling with doubt itself. Then one would more naturally conclude that the information-theoretic conception of knowledge was too narrow all along.
Yes, thank you, now we are actually discussing the underlying assumptions.
Depressing AF. Speaking as an old person, my own reading habits have also been damaged by screens.
I’ve written lots of natural language, lots of computer code, and plenty of mathematics. And I’m far from convinced that natural language is the best way to do deep thinking. After all, it was the severe limitations of natural language that motivated the development of mathematical, logical, and computer languages.
I suspect that lots of “deep thinking” is just struggling with the limits of natural language. Writing philosophy in natural language prose, it often feels like I’m not using the right tool for the right job. Conversely, I suspect that much of the effort of reading is just effort wasted struggling to get a complex idea from a medium that didn’t evolve to encode much cognitive complexity.
Learning how to shoot and develop film aren’t skills essential to photography. They’re skills for working with a poor medium for the art. As a photographer, I’m very glad that film was replaced with digital. I suspect that writing and reading natural language prose are the same.
What is the relation between poetry and what you call ‘deep thinking’? (A genuine question.) Natural languages, also in their wide variety, are inextricably woven into who we are and how we perceive the world (sometimes maddeningly so, yes …). Just as reading is not primarily about information extraction, language is not a mere tool, and that is where the analogy with photography breaks down, I submit.