Nine Years Ago, I Speculated that Dewey’s Days Were Numbered. How Far Have We Come?

From School Library Journal:

Almost a decade ago, my colleagues and I wrote an article for SLJ entitled “Are Dewey’s Days Numbered?” in which we made the argument that the Dewey Decimal Classification (DDC) system had lost its relevance. We took a bold stance, and the backlash was swift. Fellow librarians would wait outside the rooms I was speaking in at conferences, backing me into corners to demand that I stop talking about alternative systems.

At the time, we focused on creating better, more fluent access for children and modernizing a system that was created in the 19th century. It seems appropriate that today, an era when the status quo has been turned upside down by COVID and the racial justice movement, I find myself once again looking at the Dewey system. In this time that has highlighted the vast inequalities and injustices in our country, are we going to continue to use a cataloging system that is steeped in the values and worldview of a racist, misogynistic anti-Semite?

In 2019, ALA approved a resolution to take Melvil Dewey’s name off of one of the organization’s top awards for librarians, because of his known history as a racist, anti-Semite, and sexual predator. The reason: “the behavior demonstrated for decades by Dewey does not represent the stated fundamental values of ALA in equity, diversity, and inclusion.”

It is impossible to ignore that these ideas are ingrained in the system. Just look at a few examples: African American culture and history is located in groups of people (305), not American culture and history (973). In fact, any activist from suffragettes to environmentalists are classified in social sciences, history.

It is impossible to ignore that these ideas are ingrained in the system. Just look at a few examples: African American culture and history is located in groups of people (305), not American culture and history (973). In fact, any activist from suffragettes to environmentalists are classified in social sciences, history.

What message does that send our students? Are we suggesting that these groups of people can only be studied in relation to others, that their own history is not enough to stand on its own? Christians make up approximately 30 percent of the world’s religious population, yet they make up 90 percent of the 200s (religion.) The Dewey Decimal System is a perfect example of systemic racism, and we as librarians are perpetuating this harmful worldview in our libraries.

In a recent article in SLJ, many librarians commented about how weeding their collections allowed them to focus on issues of diversity and inclusion. Collection development is the natural focus of many librarians when thinking about how to make our libraries more diverse. But how inclusive can we be if the system itself is exclusionary? Most of us put a lot of effort into collecting diverse, informative books that provide windows and mirrors. When we put those books on the shelf in an outdated system, it negates those very principles. Students searching for the 16 official languages of India find out that they are shoved into 495.9 (miscellaneous languages of southeast Asia.) There are twice as many people living in India as in all of Europe, but the message we send is crystal clear to our children that white, European people and their culture are the most important. Kids looking for LGBTQ+ nonfiction books find out that they are shelved next to prostitution and pornography. The understanding that their thoughts, their very identity, is wrong and immoral comes through loud and clear no matter what we might say to the contrary.

It’s striking to me that librarians aren’t applying our rigorous weeding criteria to the system itself. Dewey is outdated and obsolete, it is difficult to use, and it doesn’t resonate with our patrons, our values, or the world around us. The system codifies and upholds a white, male, Eurocentric, Christian, heteronormative, abled perspective.

Link to the rest at School Library Journal

21 thoughts on “Nine Years Ago, I Speculated that Dewey’s Days Were Numbered. How Far Have We Come?”

  1. What message does that send our students? Are we suggesting that these groups of people can only be studied in relation to others, that their own history is not enough to stand on its own?

    What message? Librarians are nuts.

  2. The biggest problem with the Dewey system is that it was outdated as to collections before it was first implemented. PG and I, for example, think in terms of the Library of Congress system for law books, which gives “law” a letter all to itself (“K”, 1/26, which is hardly sufficient even with the spillover into D, H, and J) (the lawyer’s motto: why use five words when fifty will do?). The Dewey system shoves all of law into 340 through 347. (I remember one local university library in the 70s in despair when they opened a law school; that was reason enough to switch from the Dewey system which would frequently end up with a dozen books under the same six-numbers-beyond-the-decimal-point call number.)

    The sciences and technology are even worse. Military affairs are even worse (and diffuse and frequently cross-referenced). Mixing “religion” and “philosophy” in the same classification group is rather a problem, too, especially since much of computer science ends up in there because it’s “logic” and that falls inside “philosophy.”

    The Dewey Decimal System is not fit for purpose, and hasn’t been since the 1970s, regardless of its insensitivity to non-Boston-Brahmin sensitivities.

  3. I really intended to sit this one out.
    The issue isn’t the inadequacies of the Dewey system for any specific category or tribe, its the entire underlying pigeonhole thinking that underlies it. It’s nineteenth century thinking.
    What is outdated isn’t the categories but the entire category system.

    First, because size. Dewey and cardfiles worked in an age of thousands; we live in an age of hundreds of millions of documents, heading to billions.
    Second, because granularity. Documents aren’t monolithic. They never were but more so in the modern era; they overlap, and out of necessity often meander.
    Third, because books aren’t the center of the information universe anymore. Information management and retrieval has always been far more nuanced and complex than simplstic pigeohole systems or any ilk, glorified card files, can properly process.

    The masses may have started noticing the inadequacies of pigeonholing in the 70’s but the information managers and holistic thinkers of tbe world were alreading looking for alternatives far earlier. As early as the 20’s and 30’s. By Vannevar Bush wasn’t the first but by 1945 he had worked out a structure and engineered a system (electromechanical as that was all the tech he had to work with) he called MEMEX. All of modern information processing spins off that work:

    Much like Charkes Babbage and Ada Lovelace in the nineteenth, Bush was constrained by the available tech to implement information management but by the 50’s that changed with the first purely electronic computers and by 1970 Edgar Codd codified the rules for a more modern system, the Relational Database. Hypertext itself dates back to the early 60’s and meta data to the late 60’s. 1965 saw the first Expert System at Stanford, the approach that has since evolved, spawning the Inference Engines and Neural nets today hyped as “AI”. And that is far from the endpoint: the spawner of what has become the web, Tim Berners-Lee has for a couple decades now been working on an evolution of the web that would revolve around *meaning*, content and context, and how to access it and manage ridiculoys amoubts of information, not just raw data.

    And that is precisely where pigeonholing of all stripes (books and people) fails: the world of information (and people) is complex, nuanced, messy. Context matters. Intent matters. Inference and implication matter as do corollaries and ramifications.

    What some see as a failing–category spanning and repetition–is a desirable *feature* not a bug. We *want* our information management system to provide as many paths as possible to content, as long as tbey are valid. That is the strength of hypertext. Or even the simpler metadata based systems, like the limited and flawed but still inmensely useful one used by Amazon to help people (paying customers) find a suitable item. Duplication of results is value. Systems tbat try to eliminate it are reducing the value of the system.

    So yes, Dewey has long been inadequate.
    So is any pigeonholing system (even one that encourages replication as most metadata systems do) regardless of wokeness.

    Better tool approaches exist and are in use and even better ones are incoming. The ultimate goal was articulated way back in 1955 by Isaac Asimov in his first (of many) stories featuring MULTIVAC, the ultimate computer: a system that held all known data, understood it, and could correctly answer any natural language query. In extreme, he postulated and extreme evolution of the system that could analyze, infer, and find new data.

    We are just getting started on that road with the Web, vision systems, speech analysis, and “AI” and have yet to fold in the much harder matter of semantics. It’ll likely take decades so there is plenty of time to start thinking of specific applications and subsets like content archival and retrieval, which is what pigeonholing so crudely attempts. And how to get it right once the semantic web arrives. Multivac itself is still millennia way but its always good to have a long term objective.

    The OP by itself is meaningless, as is its intent. The real meaningful work is happening far beyond the bounds of library “science” or woke evangelism.

    See below:

    • Two weeks ago, the world’s largest robotic company (disguised as a car company) showed off it’s next generation tool: DOJO.

      “The Dojo supercomputer will use D1 chips, designed and produced by Tesla.[6] According to Venkataramanan the chip uses a “7-nanometer manufacturing process, with 362 teraflops of processing power”. “Tesla places 25 of these chips onto a single “training tile,” and 120 of these tiles come together… amounting to over an exaflop of power”.[”

      To be clear, this isn’t a binary general purpose computer but rather an advanced “education” tool to “train” neural network AIs in real world object and action recognition, essentially writing software. It isn’the first or even unique. Lately everybody from IBM, MICROSOFT, GOOGLE, AMAZON, and every other startup is creating/using this new breed of Supercomputer to create their recognition or inference engine system software. Several companies sell them on tbe ooen market. Notably Intel and Nvidia, giants in earlier breeds, among others.

      Tesla uses tbeir existing and future systems to create self driving cars (a work in progress) and are now branching into low function humanoid workers. Rather like Karel Čapek’s original “robots”. Others are using their systems to analyze all sorts of massive data, from geology to millennia old DNA to pretty much every massive dataset they can run through these trainers.

      We’ve seen some of these early “AI driven” systems doing speech recognition, brainwave to text, real time language translation, and other SF -type things. It’s just the beginning. There’s no true AI down this road we’re well on our way on (think PC’s in 1980, just before the IBM PC came out) but we *are* headed towards a meaning-based content archival/management system. Not sure where the first one will come out from but it wouldn’t shock me iif it were Amazon now that it’s run by the ex-boss of AWS.

      Dewey fooey. 😀

      Think of the commercial value of an ebook search engine you can tell you’re in the mood for a cozy viking murder mystery or a semihumorous character driven space opera featuring intelligent rats. And it will surface something. Odds are, both exist but finding them with a quick search? Serendipity baloney. This is where bookselling is headed. Especially the uberwoke. 🙂

      And its not just bookselling.
      GOOGLE could wake up and find their bread and butter is obsolete if tbey don’t get there first. After all, ALTAVISTA was Google before GOOGLE and wre are tgey now, some subbasement in the bowels of YAHOO? Which itself is barely relevant.

      In fact, all the work being done on a semantic web might be obsoleted overnight with a powerful enough training computer. Its a race between the speed of online content creation and the speed in creating the analysis engines creating the “AIs”.

      Interesting things are afoot.
      Now if only the asteroid headed our way waits long enough…

      • Now that is fast. I wonder what the plan is for developing the content on which the NN training will rely. The computer itself could be used, and that makes an even more interesting QA challenge. Rippling errors can result in a turkey being recognized as a locomotive.

        • 😀
          Yes, the tech is barely starting out and training constraints are critical.
          A classic case is a Microsoft experimental “AI chat bot” that was exposed to the open internet as an experiment. It learned internet culture and had to be shut down after only 16 hours.

          MS blamed trolls for intentionally teaching it the worst of human traits. The poor thing was too naive and impressionable for the twitterverse. 😉

          (There is also the old–possibly apocrifal–story of the Army smart antitank missile that ignored enemy models and only attacked its own because those that were used in training.)

          As for the speed of tbe training systems, the ridiculous speed is because of the very narrow range of operations the processors execute. In tbat tbey’re like the early RISC systems of tbe 90’s: less operations and simpler ones as a path to faster performance. My own takeaway is that the demand is big enough to spawn a whole new category of computer: a signpost of things to come.

          • The chips have become more and more specialized. That’s the supply problem automakers face right now. Lights use a light chip, and brakes use a brake chip. But, these are micro applications of what one could really do with a very tight focus and very deep depth.

            The Army also had problems when they used pictures of NATO and Soviet tanks. Soviet pics were taken on sunny days. NATO pics were taken on cloudy days. Therefore, a Soviet tank had a shadow.

            • That is the trend, yes.

              In fact, there is an ongoing move for hardware configuration to go closer to the final product designer, letting them define what functions they need the chip to execute and how. AMD has come to dominate the gaming console market by letting the platform holders pick and choose which CPU and GPU modules to include in their consolidated System-On-a-Chip designs. Likewise ARM became a dominant force in the processor markets by licensing the instructions and behavior but letting customers implement their own silicon design. RISC-V goes a step further by creating a modular open source Instruction Set that anybody can freely implement in any way they choose. Apple is looking to use it for future support chips. Eventually they’ll likely drop ARM, like they dropped Motorola, PowerPC, and Intel, and go full proprietary.

              A trend I’m watching is the use of Field Programmable Gate Arrays Chips, which allow a single generic chip to change its behavior after manufacture so the same chip can be told to be a light control or a brake control or an AI processor. As a rule, FPGAs have been bigger, less efficient, and slower than fixed design chips but the flexibility is very valuable in a lot of applications. One advantage is that comparable FPGAs from different vendors can substitute for each other in some cases. One report/rumor I saw suggested Tesla was using FPGAs to maintain their car production on track amid the shortage. As leading edge chips of all times get better and cheaper FPGAs are going to be useful for more and more applications and so things like DOJO will become more common.

              AMD recently threw a ton of money ($35B) at the FPGA market leader, XILINIX, after INTEL bought the number 2 player, ALTERA, for $17B in 2020. The most advanced chips in this market are reducing the penalty vs the fixed designs.

              Amusingly, several SF authors since the turn of the century have posited the evolution of molecule level circuitry, at which point the difference between fixed and programable hardware becomes insignificant and the ultimate product is a commodity: compute power, with all foundries producing a standardized product and the added value being in top level software. Rather like metals or grains.

              We’re still a ways from that but the trends all point in that direction. The shortage is contributing by driving investment to semiconductor production and a likely glut late in the decade. Once chips become a commodity whoever can serve the most custom needs will be a survivor. The best FPGA producers will have an edge.

    • I have little argument with this so long as the “information” doesn’t have to have a distinct physical locus. I’m afraid that in that circumstance, there still needs to be a set of categorization and priority rules so that physical items can be located… and reshelved for the next user (or even simple things like “gee, Marion the Librarian, how do we track which books in our collection need to be replaced due to wear/theft/whatever?”). Under those circumstances, there must be some set of rules for where the physical item will be placed.

      What? You think it can be just by author, then by title? Consider:
      * Pseudonyms
      * Jointly-authored works, especially when the “lead name” is the “lead” primarily for marketing purposes (J____ P____)
      * Multiauthor works, especially in series/distinct editions with varying editors (The Norton Anthology of Inebriated Verse, which really should exist if it doesn’t)
      * Alternate titles, especially when translated (just shudder at Proust)
      * Ghostwritten works (The Art of the ___)
      * Institutionally-presented works that don’t disclose authorship at all (the Declaration of Independence)
      * Multiauthor distinct-work series (___ for Complete Morons; continuation of a certain interminable fantasy series that’s about to hit streaming services by another author after the pseudonymous originator’s death)
      * Titles beginning with numbers, and worse yet composites (1L; and that’s before considering whether any number-sorting is numeric or linguistic, which gets really interesting for foreign works, let alone those with non-Roman alphabets)
      * Single works published across multiple volumes for publishing purposes (LotR, which if shelved alphabetically by title is out of reading order)
      * Alternate titles, or conversely later-discovered “true” authorship or controversies about authorship (the ghost of Gordon Lish requests your respectful consideration)
      * Alternate authorial identities distinct from pseudonyms (“de Groot” or “Grotius”?)

      Dewey isn’t the solution (for one thing, the question of whether Shakespeare belongs under “S” or under 822.33, especially when there isn’t a universally-accepted Dewey call number for most of Shakespeare’s contemporaries). But there needs to be some distinct, internally-consistent, unambiguous means of shelving that doesn’t require decoding information not comprehensible to the casual reader while simultaneously not misleading anyone.

      And then there are the problems with the accuracy of “natural language inquiries,” which remain for another time… and just a little sensitivity to non-native speakers…

      • Why can’t shakepeare reside in both places in the management system simultaneously, just because he is in only one physical location?
        That is the whole point of hyperlinks: multiple pointers to one storage location.

        That is precisely what Vannevar Bush was working on three generations ago and why all pigeonhoke systems, both informational and physical are obsolete. Location is just another (live) field in the databasem

        In fact, you have almost certainly hear about how Amazon (to use an example familiar here) stocks their warehouses: in the first available physical location. Things are mapped on arrival rather than pigeonholed. Faster and cheaper. As lobg as the management system knows where each physical object is located the robots can bring the object to the folks filling the boxes for shipping.

        Next question is how to maintain that location field current in the management system, right?
        RFID, the downscale commercial version of mikitary IFF systems. It comes in bith active and passive form, with passive stickers being tbe cheapest, at $0.10 each, retail for Libraries and other repositories.,application%20of%20the%20technology%2C%20RFID%20has%20earlier%20roots.

        There hasn’t been a valid excuse for misfiled books (other than luddism/ “it was good enough for grandpappy”) for decades now. Or any other significant inventory item anywhere. (The things are simple metal stickers. Just stick it inside the back cover. Each has a unique serial number so the management system, which is a whole industry unto itself) can “ping” the tag and locate it remotely. (Amazon carries handheld readers under $200.)

        So, really, the answer to how to classify physical implementation of information becomes: all of the above. Duplication *is* a feature that increases access to the obuect rather than an obstacle to finding it.

        Search by author, subject, format, date, style, purpose, anything you care about, and the system will tell you where it is. The exact physical location of the tagged item can be anywhere within the facility and still be accesible, even in restrooms. 😀

        • For a warehouse, with regulated access, these sorts of accomodations are fine. Heck, you could just use an index number based solely upon order of acquisition if you don’t allow browsing.

          I am going to gently suggest trying to browse an unfamiliar collection as pretty strong evidence that this is not sufficient. I don’t want the pony. I do, however, demand browseability. Non-device-dependent browseability. Open and sensitive to both experts and nonexperts, native-language-speakers and non-native-language-speakers, and that does not require rebooting my only content-finding device every time there’s an upgrade pushed by the device vendor.

          There’s also a corresponding problem in practice: The lack of a single definitive locus will lead to misfiling, because someone will decide to use an alternate. (What? You’ve never tried to figure out whether UnfamiliarGroceryStoreX keeps ready-to-drink coffee with “soft drinks,” with “coffee,” with “energy drinks”, or with refrigerated drinks even when not requiring refrigeration? At o-dark-thirty in a different time zone, trying to get back on the road quickly?) (Or tried to figure out whether the Kenneth Branagh Hamlet is shelved with 822.33, along with other recordings of RSC productions, or with feature films? And you’d better not even think about finding Berlin Alexanderplatz…)

          Now throw in language difficulties, especially when there’s an inept misappropriation of terms. As a chemist, I’m offended by “organic food,” but my brain can process things well enough to distinguish between overpriced bread and overpriced lab manuals. As a lawyer, I snicker at the thought of a nonlawyer (or even newly minted one) trying to find material on renvoi or depecage that’s meaningful, and then watching the confusion when the correct category begins with the word “conflicts” instead of “choice.” (And this matters — a lot — for simple things like trying to figure out whether and how to respond to a “lawyer letter” you received in the mail from out of state.)

          There needs to be a definitive, human-friendly baseline system. (With as many cross-referencing bells and whistles in the indexing system as you desire.) Dewey ain’t it.

          • Seems to me you want “browsability” for the sake of “browsability” while complaining about its impact on discoverability. Not too different from the folks who insist B&M bookstores are superior because of “serendipity”.

            I’ll posit that the folks who *need* fast and precise access to the physical product far outnumber the ones who *want* “browsability” and this is a case where the needs of the many outweigh the wants of the few. Or the one.

            One thing, though: modern physical inventory management isn’t just for warehouses or limited access facilities. The most eficient retailers rely on it. Walmart and others demand RFID of tbeir suppliers. Not for everything but for high value merchandise. At $0.10 a tag, saves far more in inventory and theft control.

            A more relevant example, though, is Home Depot, one of the few B&M retailers prospering in the face of Amazon. Their entire store is organized by broad categories and physically laid out by corridor and shelf. It serves strollers but it is controlled and marketed by a finegrained online system that allows queries by category, function, brand, use, etc and then reports the physical location and the number of units in stock. It allows placing an order for pickup and even setting a time and date for pickup. This came in extremely handy during the more restrictive pandemic rules that limited consumer access to the interior.

            The system appears to be a crude pigeonhole system but is actually a sophisticated managed inventory system tbat encompasses three product “layers”: immediately available stock, online products for store delivery, and online only for home or store delivery. Just last week, a friend was looking for power backup home generator (hurricane season!) and found a suitable candidate in the local Home Depot but they only had one. Another regional store had a whopping 70. By the time he got there his preferred model was gone. They simply ordered it for him from tbe alternate site and he picked it up the next day. The day after that, the store had two dozen.

            Bottom line: their online storefront accurately abstracts their physical store and, more importantly, precisely tracks inventory and customer needs. If it says tbey have it (and offer detailed specs and even product videos) they have it and they have it exactly where the system says it is.

            Physical location is just another line in their database and the corridor category labels mostly decor. In the age of 88% smartphone penetration even instore shoppers are better served by the online system than strolling or asking for employee assistance.

            Home Depot is among the best of breed but not unique.
            The tech is common and readily available. The “problem” was solved ages ago. And physical location need not be particularly important if the will exists to be effective and efficient.

            For libraries, all that is needed is numbered racks and alpha-tagged shelves. Instead of card files, each viewing room can have a handful of cheap tablets locked to the inventory management system or PC stations. folks can search by any of dozens of attributes (even wokeness) or just scroll through each rack’s content, providing multiple discovery paths all pointing to a unique physical location.

            As Jerry Pournelle was fond of saying “What man has done, man can aspure to.” All that is required is the will to adapt and use the commercial tools of the 21st century instead of the ball and chain of the nineteenth.

              • The LOC is restricted access, so that’s precisely my point.

                Once upon a time, during the 1990s due to my particular duty designation, I had a stacks pass at the LOC. The dirty looks I got from experienced academics and researchers stuck out in the reading room were worth the price of admission (which, admittedly, was the Metro fare from my office, but whatever).

                I think Felix is misconstruing my concern. “Discoverability” includes browsing; for many users, for many purposes, “discovery” arises from “find something I know is on point and spiral outward.” That even happens with indexing systems, whether formally hierarchical or not. (Even Quiksort-based indices, although the math for demonstrating that is rather hairy.)

                My argument is that for physical resources the physical shelving arrangement is significant, and needs fewer layers of abstraction from browseability/usability… and whatever system is chosen needs to be unambiguous, as non-arbitrary as possible, and oriented toward fewer false positives. (To continue picking on Dewey, the relationship between 822.33 and 822.34 is… misleading at best, not to mention founded in nineteenth-century-Boston-Brahmin misapprehensions.)

                And this is why a warehouse system doesn’t work with open access; or have you tried getting all of the materials you need for, say, repairing a broken and initially-misinstalled bathroom vanity at a Home Depot lately? (I have, and they’re, umm, not in the same parts of the store.)

                • I used YouTube on my kitchen sink. Used an angle grinder on the old rusted stuff, then installed the new stuff I got at Lowes.
                  For the last several months I have enjoyed posing by the sink for my wife as I turn on the water. She wanted to hire a plumber.

        • Why can’t shakepeare reside in both places in the management system simultaneously, just because he is in only one physical location?

          Location becomes a pointer. Shakespeare is everywhere.

  4. But…but…but, the Luddite in me loves the Dewey Decimal System…./s.

    But seriously, I can understand and appreciates everyone’s valid points here. I’m none too fond of a system in which I go searching for a book (after finding it the normal way on a library computer), looking for the usually minimal 7 digit number, only to find it not there. Was it misfiled using common sense? Is it out? Who knows?

    For me, prior to 2020, it was much easier using the Dewey as a base starting point, as all books where shelved according to the Dewey, and search out what I needed the old-fashioned way, which was infinitely less time consuming than the Dewey way.


Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.