From 2003: The Wetware of the Digital Divide: HOISTED FROM THE ARCHIVES
A very wrong prediction about how things would go down from a quarter-century ago...
A quarter-century ago back in the High Weblogosphere Summer, I thought the big story was that some of us—those with NYRB subscriptions, fast T‑1 lines, and a taste for “unassigned surfing”—were going to weaponize the internet for learning much faster than others. Boy was I wrong!…
The hardware curve seemed straightforward: Moore’s Law would do its work, the pipes would widen, and cheap devices would diffuse globally. The interesting action, I thought then, lay in the wetware: who knew how to search, how to classify, how to hop from Balkin to Varian to Bowerstock to Millar and back again. The Digital Divide problem was ging to be not “access versus no access,” but unequal command of the skills that turned access into understanding. Google and the NYRB archives would be force multipliers that would widen the gap between infovores and everyone else. The librarians would rule the world. Those of us who understood library systems would prosper. The rest would be locked outside, shivering virtually underdressed in the virtual infofimbulwinter outside of the virtual Lamont Library.
Boy! It has been a long, strange trip since then.
Here is the historical artifact from almost exactly 23 years ago:
2003-07-08: The Wetware of the Digital Divide
<https://web.archive.org/web/20031019130907/http://www.j-bradford-delong.net/movable_type/2003_archives/001723.html> <http://www.j-bradford-delong.net/movable_type/2003_archives/001723.html>
Eszter Hargittai worries not about the hardware or the software but about the... wetware, I guess, of the “digital divide”:
Eszter’s Blog Entry: “It is not Google but search savvy that may make the Web God, and only for some” (07/08/2003): ...Results from a study I conducted on average users’ ability to find information on the Web suggest that there is great variance in whether people can locate different types of content online and their efficiency in doing so. These findings imply that simply offering an Internet connection to those without access will not alleviate differences or the so-called “digital divide”.... Referring to Google has become the high-culture status symbol of Web use. When presented with an information-seeking task, the supposed savvy searcher quickly suggests the use of Google. However, just like simply referring to the latest opera at the Met should not be equated with expertise in the genre, a throwaway comment about Google should not make us think that people know how to find information online...
Others discuss the same issue, but from a different perspective:
Alcibiades: I would recommend, for that, Jack Balkin at Yale and taxpolicy.blogspot.com.
Thrasymachus: Amazing. What else useful is there on the internet that I do not know about?
Alcibiades: A great deal. This is why Hal Varian has been known to say that in the future the librarians will rule the world: in the future the answers to all questions will be on the internet--but nobody except librarians who understand search and classification will be able to find anything.
Thrasymachus: Google?
Alcibiades: The rewards to gaming Google are just too great, and it is just too small. Think of the Treasury Department lawyers in the Office of Tax Policy versus the collective private-sector tax bar of America.
Thrasymachus: Hmmm... I suspect the advantage to the well-connected, those in the intellectual “know”, is increasing rapidly. For instance, observe how useful the website for the New York Review of Books is for quickly getting an undergrad-level understanding of almost any field of academic inquiry in the humanities. The gap between (i) a bushy-tailed undergrad w/ a subscription to the New York Review website w/ archives and (ii) an average undergrad in section must be pretty darn large--significantly larger than in our day.
Alcibiades: But in our day the back issues of the NYRB were on the second level of Lamont, easy to consult *if* you wanted to spend time in the library. There *is* however, a big difference between slogging up to the library through the February snow and then laboriously searching through issue cover page after cover page in the hope of finding something useful on the one hand, and forty-five keystrokes from the comfort of one’s dorm room on the other.
Thrasymachus: There is indeed. Only sheer intellectual curiosity tuned to a fever pitch, or a more-than-undergraduate realization that you need to really impress at least three of your section leaders in order to get three superb letters of recommendation can drive one outside into the Cambridge winter to make the cold, wet trek to Lamont to do unassigned reading.
Alcibiades: And isn’t it the same for unassigned surfing?
Thrasymachus: No. It is not. Only the real weirdos--excuse me, I meant the future Ivy-class university professors--would trek to Lamont then. The barriers to information access are so much lower now. Back then the barriers to information access were so high that only the truly demented future academics made the library their friend. Now the barriers are low--and so you merely have to be clued-in and have a little extra cash.
Alcibiades: But is the difference really that great?
Thrasymachus: Yes, it is. The key is the ability to search. Remember, you do not know ahead of time if there is anything useful on The Roman Revolution in the New York Review of Books, or even in what decade it might have been published. How long would it have taken to go through every cover page since 1970? One hour? Two hours? And even then you might well miss it.
Alcibiades: That is a good point, Thrasymachus.
Thrasymachus: By contrast, the time elapsed just now in pulling up “The Emperor of Roman History” by G. W. Bowerstock from the March 6th 1980 edition of the New York Review of Books? 25 seconds! (Admittedly w/ a fast, well-maintained T-1 line.) It looks as though I could have read this review, quite thoroughly, in 10 to 15 minutes. Follow that with 30 to 40 minutes perusing the reserved reading in The Roman Revolution and I would have been far, far better prepared for my sophomore history tutorial than I ever was in reality--even when I had spent 4 or 5 hours going through the reading laboriously.
Alcibiades: And this is a good thing, no? People find it easier to learn. They become better educated.
Thrasymachus: Children sent to private school and children of infovore parents who get clued-in and well-connected to how to make the internet their friend find it easier to learn and become better educated.
Alcibiades: Hmmm...
Thrasymachus: If you go to public school and have non-infovore parents, it will never occur to you that the online archives of the NY Review of Books will make your humanities courses much easier.
Alcibiades: It’s a force multiplier. But a force multiplier that further stratifies our society. Is that what you are saying?
Thrasymachus: Yes. Exactly. And so it is obvious what we should do.
Alcibiades: Obvious?
Thrasymachus: Our children are still young. We have plenty time to teach them...
Brad DeLong today back: What are the implications that this is how we—or at least I was—thinking nearly a quarter-century ago, in the High Classic Weblogosphere Age?
I thought I knew what the big story was going to be. The hardware curve seemed straightforward: Moore’s Law would do its work, the pipes would widen, and cheap devices would diffuse globally. The interesting action, I thought then, lay in the wetware: who knew how to search, how to classify, how to hop from Balkin to Varian to Bowerstock to Millar and back again. Hence my Alcibiades-and-Thrasymachus dialogue back in 2003, which is now merely an interesting historical artifact about how badly DeLong needed to think smarter and do more back then to mark his beliefs to what the information reality was already starting to become.
I made very wrong predictions about how things were going to go down—not because the wetware ceased to matter, but because I implicitly assumed something that has, perhaps, not held: a broad, stable floor of deep-reading plus fast-skimming capability on which these search skills would rest. But right now we fear that that has crumbled beneath us—sustained attention (and skimming a literature productively requires as much sustained attention as deep reading), comprehension, synthesis—on which any kind of digital enlightenment, egalitarian or aristocratic, would have had to have been built.
Thus I see three big, linked assumptions embedded in my 2003 dialogue.
First, that the key constraint was finding good text, not staying with it.
Eszter Hargittai and I back then both treated the ability to coax the right result out of AltaVista-or-Google as the scarce skill. Once you had Bowerstock’s review on the screen, we assumed, the mental machinery to read, digest, and connect it to Ronald Syme’s The Roman Revolution was there (at least for anyone admitted to a reasonably selective university).
Second, that attention was essentially exogenous. The barriers I imagined were physical and institutional: February snow to Lamont; lack of a subscription to the NYRB archive; lack of infovore parents to point you toward it. Lower those barriers, I thought, and curiosity plus mildly competitive ambition (“three superb letters of recommendation”) would do the rest.
Third, that the “digital divide” would be a story of skill gaps layered on top of an otherwise healthy culture of print literacy. The United Nations was talking, around the same time, about connectivity gaps between rich and poor countries. I was talking about who knew Jack Balkin’s URL. Neither of us was talking about a future in which 40 percent of fourth-graders cannot reliably infer why a character in a short story did something, or in which the share of 13-year-olds who read for fun “almost every day” halves in a decade.
That world is the one we now inhabit.
What has happened, I think, is that the wetware problem has migrated inside the skull.
We still have big inequalities in search savvy and in access to “clued-in” intellectual networks. But we have layered on top of that a much more basic erosion: the breakdown of the habits and circuits of print themselves: both deep-reading and fast-skimming. The interior skills that my 2003 post took as given have turned out to be contingent, fragile, and very much at the mercy of technological and institutional design.
Consider the NAEP and PISA trajectories since the mid‑2010s: flat at best at the top, steady declines in the bottom half, with the steepest slides among the lowest-performing students. The very top has, so far, mostly held its own; the rest have not. Add to that the time-use data on reading for pleasure—down sharply across adolescents and adults alike—and the survey evidence that a non-trivial fraction of undergraduates cannot make it through a medium-long article without their minds or their thumbs wandering, and you have not a “digital divide” but something closer to a postliterate drift.
The 2003 dialogue assumes that “the web” is essentially a gigantic reference library: NYRB archives, Balkin’s blog, Varian’s essays, maybe a few crank sites one learns to discount with the librarians’ help. What we got instead is an attention-maximizing, notification-saturated, algorithmically-curated simulation layer that lives in everyone’s pocket and is paid to keep them flickering from one fragment to the next. We did not just lower the February snow barrier to Lamont. We replaced Lamont with FaceBook and Instagram carefully constructed to maximize outrage and “engagement”.
Thus reading as your window onto the world outside your brain ceases to be the default. Reading becomes, instead, a deliberately trained, socially scaffolded, and institutionally protected practice—very much like playing the violin, or distance running. You do not get it for free by virtue of having passed some seventh-grade phonics threshold. You get it because schools, families, and libraries carve out zones of quiet and duration; because the built environment enforces times without buzzing glass rectangles; because there is a culture around you that treats getting lost in a 300-page argument as normal rather than pathological.
We have not, in general, built those scaffolds.
What we have built, instead, is a school system in which teachers are under pressure to demonstrate growth on short, decontextualized reading passages, at the same time as their students live in an ecology of fractured micro-content. We have given students LLMs that can sketch an adequate five-paragraph essay on Middlemarch in 20 seconds, but have not rethought what it means to read Middlemarch in a world where an adequate five-paragraph essay is free—and this is where I confess that I have tried to read Middlemarch four times, and have not gotten more than a quarter-way through it each time. (On the other hand, reading The Magic Mountain at night in Europe in the summer of 1982, seated with my Eurailpass, was one of the most magical experiences of my life.)
In 2026, the advantage goes to children of infovore parents plus institutions:
Both need to be willing and able to protect their children’s attention.
Both need to be willing and able to insist on print.
Both need to be willing and able to insist on deep engagement with print.
Schools that teach decoding systematically and assign whole books.
Teachers who force students to wrestle with hard passages and publicly own their confusions.
Families where adults read long-form themselves and enforce some minimal screen sanity.
The rest are not just missing out on the NYRB archive; they are finding it difficult to read anything that has not been pre-digested into bullets, an explainer video, or an AI-generated “summary.” Eszter Hargittai was already right in 2003 that the divide was about use, not just access. But the use we now need to worry about, at the median student, is not whether they can find Bowerstock’s review but whether, once they have it, they can stay with it past the second paragraph before some part of their cognitive apparatus, trained by thousands of micro-interruptions, screams for the next hit of novelty.
Layer LLMs onto this, and you get a further twist of the knife.
It is still possible, of course, to use the new tools as scaffolding:
to draft by hand;
then ask the machine to suggest counterarguments;
ark up a difficult chapter of Keynes and then see what an LLM notices that you missed;
get a first-pass summary ;
then go back to the text and argue with it.
There are some suggestive experimental results that, used this way, the technology can actually increase brain connectivity and engagement relative to no-tool baselines. But that is not the default.
That, in turn, suggests a research and policy agenda that is rather different from what I was implicitly proposing back in 2003. We need:
Measurement: Develop better empirical measures of deep reading and synthesis—beyond short comprehension tests—so we can actually see what is happening to the wetware over time.
Causality: Disentangle, as best we can, the relative contributions of screen-based childhood, curricular choices, standardized testing regimes, and socioeconomic stress to the observed declines in reading for pleasure and reading proficiency.
Attention ecology: Study concretely what kinds of classroom, home, and library practices help rebuild attention stamina: device bans, time-blocked “drop everything and read,” teacher modeling of reading, physical space design.
Curriculum design: Integrate the “science of reading” emphasis on decoding and phonics with deliberate cultivation of deep reading via whole books and extended texts, especially for low-SES students who do not get that elsewhere.
Tool design: Explore how digital tools—including LLMs—can be designed and constrained to support deep engagement rather than to optimize for speed and convenience: interfaces that slow you down, require annotation, prompt re-reading.
Assessment redesign: Rethink school and university assessment so that outsourcing to AI or SparkNotes is less attractive, and so that what is rewarded is demonstrated understanding built over time rather than last-minute product.
Family and community modeling: Investigate interventions that change adult behavior—parents, teachers, peers—so that children actually see people around them reading long-form text for pleasure and for work.
Inequality mechanisms: Map in detail how the emerging deep-reading divide tracks income, race, school funding, and geography, and identify which institutional configurations manage to buck the trend.
Libraries as sanctuaries: Reimagine school and public libraries not just as content warehouses but as protected attention spaces—places where the default expectation is quiet, duration, and difficulty.
Democracy and deep literacy: Make more explicit the link between deep reading and democratic competence: the ability to follow a complex policy argument, to resist simple slogans and conspiracy memes, to live with ambiguity rather than jump to “blind moral certainty.”
In 2003 I thought we would be arguing about who would get to use the NYRB archive to ace their sophomore tutorial. In 2026, I fear we are arguing about whether there will be a large enough share of the population who can, unaided, read the NYRB at all. That is a very different kind of wetware problem.
Brad was mostly right 25 years ago: the Internet makes smart people smarter and stupid people stupider. What he missed is that the Internet also skews the smart/stupid ratio, at least where humanities-style critical thinking is concerned.
No doubt that print and reading is important but many would consider whether the real world understanding gap between those who read long form and don't is as large a discriminator as understanding and thinking mathematically through at least multivariate calculus and being numerically illiterate. And math can really only be communicated and understood not with long form reading which is at the root based on pre-written spoken language but a completely different system of symbols and ways to combine them into new knowledge. Regarding LLM AI for instance the understanding of anyone who understands what a Hilbert space is and functional basis is will be different from someone who struggles with 7th grade algebra but reads long form print about LLM AI. Another example comes from economics specifically money and wealth. Even economists (and central bankers like Warsh) who should understand high-school calculus will talk about stable money or like Krugman who now interchangeably uses income and wealth in his Substack writings. I would hope taking limits is still first semester high school calculus and that basic dimensional analysis and scaling would be at the core of any moderately educated persons understanding of the world we live in and thus what they communicate. The rapid growth and sophistication of technology over my lifetime is substantial with an increasing impact of more demanding physics and chemistry and biology and engineering which is further from most peoples math and science understanding. Deep reading helps to understand how people thought in the past with some reasonable hope that some weighted collection of principle component thought functionals will sometimes produce new human intellectual capital. But the real world and the real economy increasingly depend on math and science beyond the understanding of likely many more people than those who don't read deeply. If the tradeoff was between better thinking mathematically and scientifically for deep form reading it is probable the world would be better off. I suspect the gap in math and science knowledge and thinking skills is both larger and more important going forward than the transition to short form word input from deeper reading.