I was honored to give the keynote address at Carleton College’s 10th Annual Day of DH, a one-day event that draws together digital humanities professionals from the broader Minneapolis/St. Paul region. The full text of my talk is below, or you can just view the slides.
Thank you to Austin and the rest of the Day of DH organizers for the invitation today and for being such gracious hosts. In 2019 I participated in a symposium here at Carleton on digital scholarship, and it’s great to see how DH has thrived here since. I’m happy to be back and in some ways, pick up that conversation.
First, though, I need to apologize for my title. I was on sabbatical this spring. In fact, I was just about to go abroad for five weeks when the invitation from Austin appeared in my inbox. I said yes, but I did not the headspace then to think about what I would actually talk about, let alone the title of my talk.
I emailed Austin, “I’m heading out on an international flight today, and I hope to come up with a title while in the air. Can I send you a title after I land tomorrow morning?”
So I’m there in economy somewhere over the Atlantic, banging around on my keyboard, brainstorming ideas. And I land on this:
Living with the Pharmakon:
On the Digital Humanities and the Liberal Arts
Austin’s response, when I sent in the title, is a model of what I hope will be the general reaction to most of my work:
Let’s hear it. How many of you are intrigued? How many are confused? Me too.
The reference here to pharmakon comes from Plato by way of Jacques Derrida. I now regret using the word pharmakon in my title, because it meant I had to go back to my old Derrida books to make sure I wasn’t egregiously misusing the term, which really, all I vaguely remembered from grad school, was a concept that shows up in Plato’s Phaedrus, where Plato refers to writing as a pharmakon, which can be paradoxically translated to mean both a remedy and a poison.
You see, my idea was, that if for Plato almost two and half millennia ago, writing was a pharmakon, both a remedy and a poison, then—and I can’t be the first person to say this—then maybe AI is our 21st century equivalent. In fact, I had some fun going through Derrida and finding lines where he’s talking about writing—or really, writing about Plato thinking about writing—that could apply to AI.
Early on Derrida highlights one of Plato’s criticisms of writing, which is that the written word, when read, leads to “repeating without knowing” (74-75). Repeating without knowing, isn’t that fundamentally what a large language model like ChatGPT, Gemini, or Claude does? It’s a facsimile of knowledge, the same charge Plato levels against writing.
For Plato, Derrida writes, the written word “substitutes the breathless sign for the living voice” and “can no more answer for itself than a sculpture or inanimate painting” (92). Writing, according to Plato, leads to sophistry. And listen to how Plato defines a sophist, as “the imitator of him who knows” (106). An imitator of him who knows. Pretty much describes every conversation I’ve had with ChatGPT, which at the end of the day is really just a sycophantic mansplainer.
I could go on with further examples that seem to extend the notion of the pharmakon from writing to AI. Oddly though, while I will indeed talk about AI today, eventually, it will not be in relation to writing. It’s not because I don’t have deep concerns about generative AI when it comes to writing. I do. It’s because I want to talk about AI as a tool builder. But to get to that point, I want to travel a bit back in time.
How far back? Since this Carleton’s 10th annual Day of DH, I initially thought I might walk through the past decade in the digital humanities, highlighting key moments both within the field and in my own scholarly and creative trajectory. But for all sorts of reasons, 2016 makes a poor starting point for the story I want to tell. What if we went back five more years? 2011. That would give us a solid 15 years of history to work with. Then I realized 2011 is actually only the halfway mark in my own engagement with the digital humanities. I’ve been doing this for a long time. So let’s go all the way back to my beginning, thirty years ago. 1996. And then we’ll move forward, in 5 year increments, bringing us all the way to today, 2026.
Ready? Here we go.
1996
I had just started grad school, a new masters program at Georgetown University called Communication, Culture, and Technology. I picked up a research assistantship with Randy Bass in the English Department and began working on the American Studies Crossroads Project, a grant-funded initiative that asked the then-groundbreaking question of how does digital, networked technology change the teaching and research landscape of American Studies? The project and the site are long gone, but because part of my duties included serving in the novel position of “webmaster” of Crossroads, I have an old offline copy of the website circa 1998.
At the top you can see the four main research areas of Crossroads. I want call your attention to a line about the project’s funding: a combination of foundation and federal money, in this case the U.S. Department of Education’s FIPSE grant. Crossroads received other federal funding as well. Randy Bass and his team worked on the NEH-funded New Media Classroom Project, which for several summers in the late 1990s ran a week-long workshop for faculty on teaching with technology.
The signature publication of the Crossroads Project was Engines of Inquiry: A Practical Guide for Using Technology to Teach American Culture, for which I was the principal editor. Engines of Inquiry took its name from an essay by Randy Bass, in which he argued that integrating online historical archives, multimedia creation tools, and dialogic digital communication into our pedagogy can function as “engines of inquiry” for students, giving novice learners the opportunity to pose authentic questions, the kind that expert historians and literary studies scholars ask. Our idea of multimedia creation tools at the time were Hypercard and Storyspace. And when we talked about dialogic digital communication, we weren’t referring to social media or even blogs, but to internet newsgroups and listservs.
As an aside, I think you might need to recalibrate Carleton’s timeline with the digital humanities. This may be your 10th annual Day of DH, but listen to this. Buried deep in a folder that I’ve carried around with me for nearly thirty years and across countless computers, I found a file called “Orders Database” in a format long since abandoned by Microsoft Office. I was able to finesse the file open and discovered that on March 23, 1998 I fulfilled an order for Engines of Inquiry, sent to Carleton College.
Looking back at this era, I recall a sense of shared optimism. Crossroads was an incubator of folks who would go on to have an enormous impact in the field. Roy Rosenzweig, who founded the Center for History and New Media at George Mason in 1994, taught with us in the New Media Classroom Project. One of the Georgetown undergraduate students working alongside me on Crossroads was Sharon Leon, who would go on to be the director of Omeka. We were collaborative, focused on collective experimentation, often dismissed by our more conventional colleagues, but underwritten by foundations and government agencies.
Yet we should not overlook what we might describe as the undercurrent of the era. And by undercurrent, I mean a force not fully visible at the time but which nevertheless shapes the moment. Even in the mid-nineties there were early warning signs about the way participatory online culture might be weaponized. And against whom it might be weaponized. I still vividly recall a 1996 essay by the esteemed Victorianist Pamela Gilbert—then a new assistant professor—in a special issue of the journal Women & Performance. Gilbert recounts being stalked on Victorian listservs by another professor. He began sending her alarming emails. Then threatening physical violence and offering money to strangers on internet newsgroups in exchange for pictures of her. The stalking became so serious that the MLA arranged special security for Gilbert at its 1995 convention, in case the man showed up (133).
That was all in 1996. Many of us were praising the pedagogical potential of technology. Some of us, many more than you might think, were already experiencing a different reality.
2001
Let’s jump ahead five years to 2001. I’m in a Ph.D. program at the University of Pennsylvania. This is about the time I made the switch from a dial-up modem to an always-online DSL modem. I replaced Alta Vista with a new search engine called Google. Wikipedia launched, though it would be a few years before it replaced my Microsoft Encarta CD-ROM.
A year previously—2000—John Unsworth at UVA had proposed the notion of “scholarly primitives”—fundamental methodologies across all scholarly disciplines, which, Unsworth proposed, might be the “basis for a manageable but also useful tool-building enterprise in humanities computing.” In other words, what scholarly practices can new computational tools support? Unsworth’s non-exhaustive list of scholarly primitives included techniques such as “Discovering,” “Annotating,” “Comparing,” “Sampling,” and “Representing.”
And indeed, one of the draws for me at Penn was a long-running digital project by the Romanticist Stuart Curran that engaged with several of these scholarly primitives: an annotated, multimedia edition of Mary Shelley’s Frankenstein. By the time I arrived to Penn in 2001, Curran and his team of grad students had already been working on the project for several years. The work-in-progress had a secret URL, what we would now call the dev site. Imagine my delight, when on a whim preparing for this talk I tried that old URL. It worked.
This digital edition of Frankenstein demonstrates nearly all of Unsworth’s scholarly primitives. Hyperlinks in the text open up annotations with historical or literary context. A key feature is the variorum, a tool that lets you compare the 1818 and 1831 editions side-by-side. I miss this aesthetic. Frames!
And look at the list of credits on the front page. Like many early DH projects, the bulk of the work was done by grad students, often generations of grad students. The work was slow—there was no rush to get it done. There might have even been incentives to draw out the work. It operated on what we might call academic deep time. There was a commitment to continuity, stewardship, and substance over style.
One of the reasons for the perpetually deferred publication of the digital edition of Frankenstein ties into one of the undercurrents—if I can use that concept again—of 2001. Quite simply, there was a rights issue. The digital edition included over 700 images—maps, portraits, artwork of the period, and much more. And the creators of the digital edition wanted to clear all those rights. It sounds almost quaint today, this commitment to citation and credit, such care for provenance that it halts publication of a genuinely valuable teaching and research resource.
It’s easy to forget now, when tech companies have looted the internet to train their algorithms, have downloaded terabytes of illegally-distributed copyrighted material from so-called archives, that there was a time when credit was treated with care, and there were honest yet critical debates about the public domain and the idea of the commons. In fact, Creative Commons, under which I’ve licensed many of my own works, was established in 2001. From this point on we’ll see an escalating tension between the private and the commons, the individual and the crowd.
2006
Consider the next stop along the timeline, 2006. I’m now an assistant professor in the English Department at George Mason, teaching contemporary American literature and new media. New media was the terminology of the day and it signified something different than the humanities computing that Unsworth was working in. Humanities computing sought to apply computational tools to print materials and traditional scholarly questions—turning a printed novel into a digital edition, for example. New media studies suggested that computational culture itself and the digital artifacts it gives rise to ought themselves to be objects of study. My own work blends both of these approaches.
Take this class I taught in the Fall of 2006 on 21st Century American Fiction. It was great to teach, there was only six years’ worth of fiction to choose from! We studied contemporary print novels, avant-garde digital works, and early in the semester, Mark Z. Danielewski’s not-yet-cult classic House of Leaves, which the literary theorist Jessica Pressman characterized at the time as a “networked novel” (107). An assignment asked students to edit the Wikipedia entry for House of Leaves. Remember, this was 2006. Wikipedia was the bane of teachers everywhere and there was still serious debate on its status as a credible source. Which of course made it a perfect platform to write about House of Leaves, a novel which at its core is about the limits of epistemology.
One of my most memorable days of teaching ever was the day after the assignment was due. Students stormed into the class, furious, not about the assignment, but that their work—their Wikipedia edits—had almost immediately been undone by the Wikipedia community. We even made it to the Talk page of the House of Leaves entry.
I turned the students’ disappointment into a learning opportunity, discussing concepts such as verifiability, original research and so on. We talked about gatekeeping knowledge and whether or not the deletion of their work constituted “censorship.” (It did not.)
Let’s use Wikipedia as a stand-in for one of the undercurrents of 2006. Thanks to the open source software that powers Wikipedia, every single edit to the site ever is preserved, and even twenty years later we can go back and revisit my students’ contributions and subsequent deletions, a reminder of both the seeming ephemerality of the digital artifacts but also the traces they leave behind. But in 2006 most of us weren’t yet thinking about our so-called data exhaust. The era of Big Data is just about to begin. It’s at this exact moment that we see the rise of the surveillance state. The NSA’s PRISM surveillance program begins in 2007. And then there’s surveillance capitalism—the way we’re tracked on our devices and our data sold through data brokers. I find today’s students have a kind of data nihilism. They know big tech is relentlessly harvesting their data and they feel like there’s nothing they can do, so they just avoid thinking about it. It’s possible that the last generation of students who did not fall into data nihilism were those students so angry back in 2006.
2011
Let’s jump ahead another 5 years. 2011. This is the midpoint in my timeline. It’s also perhaps peak “digital humanities.” Let me offer a few examples, both personal and more broadly that capture the DH moment of 2011.
First a quick return to Frankenstein. That digital edition started in 1998 was finally published by the Romantic Circles scholarly website in 2009 with an updated interface. It was helped over the finish line by the DH lab MITH at the University of Maryland. Though without any the images. They never did solve the rights question, so they just sidestepped it entirely.
A year after that, in 2010, the Center for History and New Media ran the first of two iterations of an NEH-funded summer institute called One Week | One Tool.
Subtitled “a Digital Humanities Barn Raising,” the project brought together a mix of about 15 humanities faculty, grad students, and librarians who, in one intense week, developed a fully-functioning digital humanities tool from scratch. That first year they built Anthologize, a WordPress plug-in that lets you select specific blog posts, reorganize them with headers, and bundle everything as an ePub or PDF.
2011 was also peak THATCamp. Just out of curiosity, how many of you have participated in a THATCamp? The premise was simple but also revolutionary for people accustomed to stuffy academic conferences. The Humanities and Technology Camp was an unconference where the “program” is established collaboratively on the fly at the start of the event. THATCamp grew into an international movement, with unconferences in North America, Europe, Australia, and Asia. There were themed THATCamps as well, such as THATCamp Games in DC and THATCamp HBCU in Atlanta. THATCamp Pedagogy in New York.
In 2011, many of us in the DH community were hard at work on articles that would appear the following year in the first Debates in DH volume edited by Matt Gold. The University of Minnesota Press, the publisher, developed an online platform called Manifold for reading and annotating chapters in Debates in DH. The series has continued under Matt and Lauren Klein and a host of guest editors. I’m especially fond of Brian Croxall and Diane Jakacki’s 2023 volume What We Teach When We Teach DH. We’ve come a long way since Engines of Inquiry.
2011 was a pivotal year for me as well. I was working on the finishing touches on 10 Print, a collaborative book in the nascent field of critical code studies. My nine—nine!—co-authors and I wrote about a one-line program for the Commodore 64, doing a deep dive into the computational and cultural significance of that short program. We wrote 10 Print, not using Google Docs, but with Mediawiki, the same open source software that powers Wikipedia. Which of course means we can trace every single edit and revision of the manuscript. We wrote it very wiki-style; nobody “owned” anything, and anybody could revise anybody else’s work. Very chaotic, but also thrilling.
The entire endeavor was very much driven by an open source, open access ethos. The book was published by MIT Press, and we also offered a free download of the entire book under a Creative Commons License. My co-author Nick Montfort maintains the download site, and he told me recently that we’ve had over 10,000 downloads, and the book has also been viewed on the Internet Archive over 10,000 times. The book remained in print a decade even though it was free, something that affirmed for me the idea that being open could expand your audience.
Speaking of Nick Montfort, in 2009 he released a generative nature poem, “Taroko Gorge,” an attempt to capture the unreal experience of hiking the seemingly endless chasm in Taiwan. Slowly at first and then very quickly “Taroko Gorge” became the object of dozens of remixes, people taking the base code and subbing in their own thematic content. In 2011 I released my own remix, called “Takei, George.” The poem transforms Montfort’s original meditative generative poem into a comment on pop culture, fandom, and contemporary politics through the prism of George Takei, the original Mr. Sulu on Star Trek.
There was so much energy and enthusiasm behind the digital humanities in 2011. But there was also an undercurrent of criticism from within the academy. Conventional scholars were coming to resent what they saw as an upstart with no respect for tradition. Take this New York Times column by the legendary critic Stanley Fish in January 2012. Fish takes multiple potshots at Kathleen Fitzpatrick whose new book at the time, Planned Obsolescence, was all about reimagining scholarly publishing. Fish mentions me as well, not in a good way. It’s one of the highlights of my academic career.
More meaningful critiques coming from within DH itself. Tara McPherson’s essay in Debates in DH—“Why are the Digital Humanities So White”—struck a chord. Other critics such as Adeline Koh, Lisa Nakamura, Roopika Risam, and the late David Golumbia called out DH for ignoring race and gender and the problematic—some might say authoritarian—dimensions of any computational technology.
2016
Now we arrive at 2016. Much has changed. Many of those undercurrents from earlier years—the weaponization of participatory culture, the abuse of data, the troubling relationship between technology and power—are no longer undercurrents. They are on the surface, everywhere we look.
2016 is the moment it becomes obvious that our shared digital public sphere—our imagined community—has deteriorated into fragmented, partisan bubbles. It had been creeping up on us for a while, for example, GamerGame in 2014 and 2015, in which misogynstic trolls and later the right wing media targeted prominent women and people of color in the video game world, like game designers, game journalists, and game studies academics. New terms entered our daily vocabulary around 2016. Doxxing. Swatting. Redpilling. White supremacists rebranded as the alt-right begin marching across college quads, quite literally.
Against this flood, my own work receded into an undercurrent, of little visibility beyond my immediate setting. I was now at Davidson College, where I began building a new program called Digital Studies. The world around me seemed to be burning, and here I am trying to build something new. I was the PI for a large Mellon Foundation grant, tasked with imagining what the digital liberal arts might look like. This was the start of Davidson’s domain of one’s own program, the design of a new digital culture and creativity curriculum, and a lot of faculty development around digital scholarship. It’s the work I focused on when I visited Carleton in 2019.
2021
Somehow we all made it to 2021. The pandemic, Black Lives Matter, protests and counter-protests. Zoom fatigue. The last year of a world without ChatGPT. My own teaching, creative, and scholarly work begins taking a darker turn here. Whereas my approach to digital culture in 2011 might be characterized as whimsical, by 2021 it can only be called sardonic. Interesting etymology, that word. Sardonic comes from the Latin sardonian, an adjective that describes the effects of eating the poisonous leaves of a plant native to the island of Sardinia. According to the Oxford English Dictionary, the plant was said “to produce facial convulsions resembling horrible laughter, usually followed by death” (OED “Sardonian”). Horrible laughter, usually followed by death: that’s my new aesthetic.
Exhibit A. In 2020 I released The Infinite Catalog of Crushed Dreams. It was the early days of COVID, that first spring when all the college students were sent home, K12 kids were on Zoom, and everything was shut down and no one knew when the world would start up again. It’s an endless stream of procedurally generated disappointment, sadness, and grief. The randomly generated list goes on forever, using an infinite scrolling mechanic. You just keep scrolling and scrolling.
When I first made the work public, more than one person asked if I was using AI to generate the text. At the time GPT-2 was the state of the art in text generation, but it wasn’t something any casual user had access to, and most people had never even heard of it. The Infinite Catalog is not AI. It’s powered by a small template-based model handcrafted by myself. The whole model is 80 kilobytes. That’s less than 2 pages of typed text. Yet the heartache goes on forever.
Exhibit B on my sardonic turn in the 2020s is Content Moderator Sim, a workplace horror game in which the player assumes the role of a content moderator at a Facebook-like company. That is, you play someone charged with reviewing posts that users of the platform have flagged as possibly violating the standards of that platform. You have to review several cases, and, like real content moderators, you only have a few seconds to make each decision. The game is research-based, informed by the ethnographic work of Sarah Roberts and actual details taken from Facebook’s internal content moderation handbook. But like I said, it’s also a horror game.
Exhibit C. No Time to Discourse came out more recently, but the seed of the idea came in 2020 and I had a working version by the summer of 2022. It’s a map of speculative climate disasters told through flash fiction. The project opens with a map of the North America, littered with thousands of red flags. Each one is a climate disaster, generated by a non-AI handcrafted model I made myself. There millions of possible disasters. (Here I read a dozen or so of the flash fiction stories. Check out the project site, or download the novelized version.)
What’s the undercurrent of 2021? I don’t really know. Everything that was once subtext is now text. Within DH we continue writing our books, publishing our articles, building our projects, teaching our students, but many of us are barely holding on, barely holding up. We’re struggling to keep on keeping on. On top of all that, and though we don’t know it at the time, 2021 is the final year before generative AI changes the ground beneath us. I guess that’s what the undercurrent is. GPTs.
2026
Which brings us to 2026. Thirty years from the start, or at least 30 years from the start of my story. There are three points I want to make about DH in the year 2026, and each revisits ideas from the past three decades.
The first is the collapse of public infrastructure to support the digital humanities. And really, any form of the humanities that does not adhere to a rigid notion of patriotism and nationalism. I’ve highlighted federal funding of DH initiatives several times. We all know what happened with the National Endowment for the Humanities at the start of 2025. It was DOGEd. Or attempted DOGE. More than 1,400 grants were terminated, and the MLA, AHA, and ACLS sued the Trump administration. Just last month, on May 7, the U.S. District court agreed that the grant terminations were illegal.
Thanks to discovery materials in the case, we now know the details of DOGE’s grant termination process. The two lead DOGE “officials” did not read any of the actual grant proposal materials or project websites; they simply used brief descriptions the NEH had provided in a spreadsheet, pasting those descriptions into ChatGPT with the prompt: Does the following relate at all to DEI? Respond factually in less than 120 characters. Begin with ‘Yes.’ or ‘No.’ followed by a brief explanation. Talk about off-loading cognitive function to AI.
We see the fig leaf of government “efficiency” fall away to reveal what was known to all of us all along, that this is a war on DEI and so called “woke” academia. There’s no end in sight yet to this dismantling of the public infrastructure to support the arts and the humanities. And it’s cold comfort that STEM fields are just as much under attack right now.
Paradoxically, despite that hollowing out we are about to enter an era of proliferating DH tools. This is the second point I want to make about 2026. Think back to One Week | One Tool. It was an ambitious, experimental venture, not at all certain to succeed, given that most digital humanities projects took years of labor to release, if they ever saw the light of day at all. Fast-forward a dozen or so years, and it’s now possible for one humanities scholar to produce a working DH tool in a day—maybe not a good, but in a day—with the help of Claude Code or OpenAI’s Codex. Instead of One Week | One Tool, we have entered an era of One Geek | Many Tools.
To give you an idea of what this looks like, I’ll share a few DH tools I’ve made in the past few months, while on sabbatical, utterly disconnected from daily campus life, and then circle around to my third and final point about 2026.
A few minutes ago I mentioned the discovery materials in the lawsuit against DOGE. Among the reams of documents, emails, and spreadsheets were over 20 hours of deposition videos, some of which went viral when they were first released. I couldn’t bear to watch the videos, so I made a tool to do it for me.
This is the Simple Supercut Maker. If you’re not familiar with the term, a supercut is a short video highlighting patterns in a video or collection of videos, repeating them until they become absurd. I built a working prototype in an hour with Claude Code, and then refined it over the course of a few days. Here’s how it works:
- Upload a bunch of videos. In this case, the deposition videos with DOGE “official” Nate Cavanaugh. If you don’t have subtitles already, the tool will generate them for you.
- Then use the “Explore Words” feature to find frequently used phrases. Here I searched for “I don’t recall.”
- Export the video, which ends up looking like this:
Supercuts are of course superficial, facile, and analytically weak but—rhetorically powerfully. That’s why they’re a staple of late night talk shows. You could set students lose with the Simple Supercut Maker in the classroom and it becomes a kind of distant reading tool of whatever visual text you want.
Let me share another DH-y tool I made this year. Since at least 2020 I’ve used Glitch.com in my teaching and for my own creative coding work. Glitch was a browser-based coding platform where everyday users—like my students who had no coding experience—could build, host, and remix modest web projects like digital poems or interactive fiction. It was a creative coding playground. You didn’t have to worry about things like deployment and version control. You could just focus on experimentation and expression. It was fantastic. So of course, the large cloud computing company that bought Glitch in 2022 decided, in 2025, to shut it down.
I looked around for alternatives. But there was nothing that was open, decentralized, and not at the mercy of VCs or Big Tech. So I built my own.
This is Glitchlet. It runs on any shared hosting service (e.g., Reclaim Hosting). If you can run WordPress, you can run Glitchlet. Projects-in-progress are stored in the browser’s local storage, but you can also one-click publish to make them public and remixable. I designed the site, with the help of Claude Code, with educators in mind. There’s no single, primary Glitchlet that everyone uses. The idea is that every instructor installs their own Glitchlet and manages their own classes/workshops/projects. You can seed your instance with template files, or Glitchlet can easily import projects (including archived Glitch .tgz files).
There’s one final project I want to share. It hearkens back to Unsworth’s scholarly primitives. It’s a digital variorum—like we saw with the side-by-side comparisons of Shelley’s two editions of Frankenstein. In this case, it’s a digital variorum of the novel Subcutanean.
Subcutanean is a 2020 queer coming-of-age novel by Aaron Reed, a well known game designer and writer. The narrator Ryan is an introverted college student whose love for his closest friend, Niko, goes unrequited. Subcutanean is also a horror novel. Ryan and Niko discover a vast basement under their dilapidated off-campus house that branches into parallel worlds, which they get lost in it. Most importantly for today, Subcutanean is an experimental novel. It’s human-written but procedurally-generated, and no copy is the same as any other copy.
The printed book (or ebook) reads as a cohesive, linear textual object. Nothing indicates that the story inside is variable, that every copy of Subcutanean is different. The author Aaron Reed explained, “Each time a book is ordered, a different version of the text is rendered from a source text containing many possible variations of individual words, sentences, and even entire scenes.” There are thousands, hundreds of thousands of possible variations. And yet at any given instant, a reader has access to only one version of that world.
Until now! The Subcutanean Variorum is a digital workbench for studying and teaching the novel. Though I used Claude Code, it was not “vibe-coded.” The project went through many iterations, theoretical and practical approaches, and design decisions by me, and took nearly five months of my sabbatical to complete. I was only able to do this because in 2025, on the fifth anniversary of its publication, Aaron Reed released all the text and source code under a Creative Commons license. (And here I demoed some of the features of the Subcutanean Variorum. I encourage you to play around with it yourself, or read my post introducing the tool.)
I made the Subcutanean Variorum because it’s the tool I wish I had when I taught the novel in the fall of 2024. The class began every day with students comparing their copies with each other. They were often confounded and delighted by the variations, and I wish we had spent even more time methodically comparing variants using the approaches that come from book history and textual scholarship, where variorums are a stock tool.
I designed the variorum with students and researcher in mind. Let’s say you’ve found an intriguing juxtaposition of similar-but-not-quite-the-same passages in two different seeds. Any view can be bookmarked to return to later. Teachers can share a link of a specific view with students. Double clicking any paragraph in any view opens a note panel, allowing easy annotation of passages and code. The notes are stored in your browser, so they are totally private. But, notes can be exported and imported. So I as a teacher could annotate some passages, export the notes, have my students import them, and respond to my annotations with their own notes.
The variorum is preloaded with 25 unique copies (Seeds 60001-60025), but users can generate new versions and upload them for comparison. Like Reed’s original project, each version of the novel generated by the Subcutanean Variorum has a unique seed and will be different from any previous—or future—copies.
I’m not usually a braggart, and I cringe inside whenever I self-promote. But I have to say, the Subcutanean Variorum is an amazing project. It has as much scholarly rigor and honestly more imagination than many DH projects I’ve seen over the last 30 years. It’s transformed my own understanding of Reed’s novel. I hope students, teachers, and fans find it valuable. But—and here’s the undercurrent of my work in 2026 that I’m surfacing for you all—I feel a profound sense of emptiness and disillusionment regarding the project and the other tools I’ve just shared.
The critic Michael Sacasas recently wrote that “the arc of AI bends toward demoralization.” We as educators devoted to the mission of the liberal arts—cultivating the life of the mind in our students—are of course demoralized by students prompting a box instead of figuring out what they think through writing. We’re demoralized by the way attention to reading and discernment of meaning have been offloaded to pithy AI summaries. But the one positive of generative AI, or so I thought, was the way coding tools make it possible for people like me to realize digital projects they could once only dream of, or which would have taken years of funding and teams of workers to create. But no, making these projects—and many more I haven’t even shared—has left me demoralized, hollowed out.
– Michael Sacasas
What’s the source of my demoralization? I don’t think it’s the environmental costs of AI, because I am well aware of the many ways I use more than my fair share of natural resources, even discounting AI. I know the carbon footprint to bring me here today, .3 metric tons of CO2, is enough to power a house for two weeks. I know this cotton shirt, fast fashion designed to be worn for a season and then discarded, required 700 gallons of water to make. I know the hamburger I ate for lunch in the airport yesterday took 660 gallons of water to make. I know there are conflict minerals in my phone mined by children at gunpoint. At least the fury over AI data centers has us talking in a meaningful way about the origins of things, origins that we ignore all too often.
I don’t think the source of my demoralization is the way coding agents have been trained on billions of lines of code made by human programmers, a mass extraction event. Coders themselves use these tools all the time. Many are wildly enthusiastic about them. As Anil Dash explained to the New York Times Magazine in March, “The reason that…coders…see L.L.M.s differently than everyone else is that in the creative disciplines, L.L.M.s take away the most soulful human parts of the work and leave the drudgery to you…And in coding, L.L.M.s take away the drudgery and leave the human, soulful parts to you.” I don’t agree with the first part of Dash’s formulation—I disagree with his notion that “the drudgery” of writing is separate from the “soulful human” part of writing. But his point that many coders feel empowered by AI, not threatened by it, remains.
No, I think the answer to why I’m demoralized can be found by looking back at three decades of DH. We’ve seen how over the past 30 years the digital humanities have organized and structured people, labor, expertise, even our imaginations. DH was a community of practice. Digital scholarship has always been grounded in communal efforts, stewarded across academic generations through foundations, grants, and a sense of institutional continuity. Today’s tools, however, threaten to be one-off, quickly built tools. Disposable. The DH equivalent of fast fashion.
And at the very heart of it, the very heart is this, my third and final point about DH in the year 2026: AI Collaboration is Fake Collaboration.
Collaborations should intellectually enrich the collaborators, collaboration should fill their cups. But Claude does learn from my collaborations. Claude does pay it forward in any legible way to future digital humanities students and scholars. Even if I were using an entirely open weight model, a model not made by a corporation whose IPO is reckoned to raise a trillion dollars, collaborating with AI not a collaboration. There’s no exchange. It’s merely a transaction. What’s missing is the shared knowledge-making with other humanities scholars. With other humans! Collaboration is mutual aid for the mind. On our best days, this is what we do in the humanities: produce meaning with interlocutors in a dialogue that nurtures the intellectual vitality of everyone involved.
I know many others in the arts and humanities share my sense of demoralization about AI. I’m not so sure many others in the arts and humanities have investigated AI as a toolmaker as strenuously as I have in the past six months. In the end I’m deeply ambivalent. I began this talk with the idea of the pharmakon, both a remedy and a poison. It’s not such a difficult concept after all. The same thing can oscillate between opposites, never to be fully resolved. Yes, AI is a pharmakon, as is all technology. It’s empowering but isolating, generative yet inert, as sublime as it is mundane. I’ve made things with AI that I could never make on my own, and maybe that’s the lesson for myself looking back at the past 30 years: if I can’t make it on my own, I really should try making it with someone else.
Someone else.