The humanities have never been about avoiding the technologies that shape how people read, write, and think. They have been about understanding those technologies better than anyone else and bending them toward human ends.
From January to September of 1508, Erasmus lived and worked in the Venice household of the printer Aldus Manutius. He had come with a collection of classical proverbs he wanted to expand, and the shop gave him what no single library could: a circle of Greek scholars, their manuscripts, and a press that could carry the result across Europe. That September, Erasmus published a new Adages with more than three thousand entries, up from about eight hundred in the first edition. It became one of the most widely read books of the century.1
Aldus was a humanist before he was a printer. He spent years as a tutor to young princes before opening his shop in 1495, and he treated the press as an instrument of learning. He published Greek texts that had barely circulated in the West, and in 1501, with a Virgil set in a new italic type, he began producing small editions of the classics that a student could carry in a pocket rather than consult at a lectern. On the title page of the Adages, the note Erasmus wrote to readers addresses them as studiosi, those who study, and begins: Quia nihil aliud cupio quam prodesse vobis—because I want nothing more than to be useful to you.2
.jpg){kind=link}
The print shop was not a place humanists visited in order to complain about it. It was where many of them worked. As the historian Anthony Grafton has shown, early modern printing houses were full of scholars reading proof, checking copy against manuscripts, and catching the errors that a new machine could multiply by the thousand.3 In the engraving below, from a series celebrating the new inventions of the age, one of them stands a few steps from the press, spectacles on, checking a freshly printed sheet.
Aldus’s own printer’s mark, the dolphin wrapped around an anchor at the top of this page, was an emblem of a motto attributed to the emperor Augustus: Festina lente, make haste slowly. In the new Adages, Erasmus wrote a long essay on the phrase and used it to praise his printer.4
Festina lente
The motto yokes two opposites, and Aldus’s mark draws them: the dolphin, swiftest of creatures, wrapped around the anchor that holds a ship in place. Speed and steadiness at once. It is not a bad description of what good work with AI requires.
Five centuries later, the humanities face a new machine for producing text, and many of us have chosen to stand outside the shop. We ban it from our syllabi. We run student essays through detection software that researchers have found consistently misclassifies writing by non-native English speakers as machine-generated.5 We reassure one another that AI can’t really read, or write, or think. The scholar Leif Weatherby calls that last reflex “remainder humanism”: the habit of saying that “machines can do x, but we can do it better or more truly.” It turns every encounter into a John Henry contest, and it keeps us from engaging seriously with the technology itself.6
We understand the impulse. But turning our backs on AI is the least humanistic response available to us.
Humanists are, in fact, unusually well prepared for this moment. Large language models are made of language: of human writing in all its genres, registers, biases, and borrowed voices. Weatherby argues that these systems are best understood not as artificial minds but as machines that produce culture, which puts them squarely in the territory of people who study how culture works. The central issues of AI use—authorship and credit, evidence and truth, bias and representation, consent and privacy—are home ground for the humanities. So are the questions that matter most when a student or a scholar sits down with one of these tools. Who is speaking? On what authority? What has been left out? Is this source real? What does this text want from me?
Those are also questions about the people inside the machine. In August 2026, Amazon announced that it would close Mechanical Turk, the crowd-work platform Jeff Bezos once called “artificial artificial intelligence,” which for two decades paid people small sums to label data, transcribe audio, and do the other tasks computers couldn’t. The platform took its name from an eighteenth-century chess-playing automaton that concealed a human player in its cabinet.7 As Lauren Goodlad, the editor of Critical AI, has written, what gets marketed as “AI” is the product not only of technology companies and investors but also “of the many millions of people and communities” subject to copyright infringement, nonconsensual use of data, and low-wage labor.8 Humanists know how to look for those people. It’s much of what we do.
None of this means dismissing the technology’s promise, which is real, especially in education. In a randomized trial at Harvard, students in an introductory physics course who worked with an AI tutor learned more, in less time, than students in an active-learning classroom.9 The tutor worked because teachers designed it around how people learn: it offered hints one step at a time instead of handing over answers. A field experiment with nearly a thousand high-school math students showed the other side. Students given unrestricted access to GPT-4 did better on practice problems, but once the tool was taken away they performed worse than students who had never used it. A version designed to coach rather than answer largely avoided the damage.10 The lesson of both studies is the same: what makes AI good or bad for learning is how its taught.
Our students are already using these tools, whether we allow it or not. The question is whether anyone will teach them to use them well. In a 2025 survey of 319 knowledge workers, the more people trusted AI, the less critical thinking they reported doing; the more confident they were in their own expertise, the more critically they engaged with its output.11 That is an argument for exactly the kind of education the humanities provide: not training in how to operate a tool, but the formation of judgment strong enough to question one.
That starts with learning the tools ourselves, and many humanists have been poorly served by first impressions. A colleague types a one-line question into a chatbot, gets back something fluent and generic, and concludes that the technology is shallow. But a question with no context gets the answer you’d get from a well-read stranger at a party. We would never hand a new research assistant a single sentence and expect a useful memo; we’d explain the project, the sources, the audience, and what we already know. We spend whole seminars teaching students that a good research question does half the work of research. The problem usually isn’t the machine. It’s the brief, and writing a good brief is a skill humanists already have.
Learning AI yourself
- Brief it the way you’d brief a new research assistant: the project, the sources, the audience, and what you already know.
- Try it on material you know cold, so you can see exactly where it’s right, wrong, or merely plausible.
- Treat the first answer as a draft to argue with and ask follow-up questions.
- Check every citation. The machine will sometimes invent them.
Teaching it to students
- Assign the prompt along with the product: have students submit their prompts and explain their choices.
- Ask students to compare their own reading of a source with an AI’s reading, and account for the differences.
- Make verification part of the grade: sources checked, errors caught, claims corrected.
- Teach students when to close the window and think alone.
Teaching AI well also means teaching students when to say no. Our graduates will be asked, by employers and clients and sometimes by their own institutions, to do things with AI that shouldn’t be done: to invent a quotation, to feed a community’s oral histories into a commercial tool without asking the community, to pass off generated text as reporting. In May 2025, the Chicago Sun-Times and the Philadelphia Inquirer ran a summer reading list that recommended new novels by Isabel Allende and Min Jin Lee. Neither book exists, and neither did several others on the list. A freelancer had used AI to put the section together, and it ran without review from the Sun-Times’s editors.12 Any English major could have caught it. The failure wasn’t a lack of technology. It was that no one in the chain had the habit, or the standing, to stop and ask whether the books were real.
The same judgment is what lets AI amplify creativity instead of flattening it. In a 2024 experiment, writers given story ideas by a language model produced stories that readers rated as more creative and better written, especially when the writers were less creative to begin with. The AI-assisted stories also resembled one another more.13 That is the promise and the problem in a single result. AI can get a beginner past the blank page.
It takes something else—taste, voice, a sense of what hasn’t been said—to make the result worth reading. That something else is what we teach.
In our own studio, AI has lowered the technical barriers that used to keep student projects small: students who have never written code can now build searchable archives, interactive maps, and public websites for their research. They need guidance about what’s worth building, and the judgment to notice when the tool is leading them astray.
All of this would be easier if the value of the humanities were obvious. It isn’t, least of all to the people who most need to see it. Over the past decade, the study of English and history in American colleges has fallen by a third, as Nathan Heller documented in The New Yorker.14 AI has handed skeptics a new argument: if a machine can produce a competent essay, why pay anyone to teach students to write one? That argument rests on two mistakes at once. It overestimates what AI can do, and it misunderstands what humanists do.
We have never been in the business of producing competent essays. We are in the business of producing people who can tell a real source from a fake one, hear what a text is doing beneath what it says, and make an argument they can defend.
AI makes those capacities more valuable, not less, but no one will take that on faith. If humanists refuse to engage with AI, we confirm the caricature that we are custodians of a past the machines have made obsolete. If we engage—if we teach students to learn with these tools and to push back against their misuse, if we show what careful AI-assisted scholarship looks like, if we’re in the room when our universities decide how to use them—we make the case in the only way that works. We have to fight for this, and the fight is the work itself.
Some of that work has begun. The Modern Language Association and the Conference on College Composition and Communication formed a joint task force that has called for “critical AI literacy” across writing, literature, and language programs.15 The rest of us should follow.
What we’re asking
Of humanists: learn these tools well enough to teach them, and to criticize them from knowledge rather than rumor. Try them on the materials you know best, and write down what happens.
Of departments: teach students to learn with AI, and to refuse it when refusing is the right call. Build that into the curriculum rather than leaving it to individual syllabi. And count the work: a documented AI-assisted workflow that others can inspect and adapt is a scholarly contribution.
Of universities: put humanists in the room when AI initiatives are designed, not after the contracts are signed.
Aldus’s motto still works. Make haste, because our students are already using these tools and the decisions about how our institutions will use them are being made now. But make haste slowly, with the patience of people who check the source, read the proof, and ask what a text is for. The first humanists moved into the print shop and made the new machine serve learning. It’s our turn.
.jpg){kind=link}
-
Deno J. Geanakoplos, “Erasmus and the Aldine Academy of Venice: A Neglected Chapter in the Transmission of Graeco-Byzantine Learning to the West,” Greek, Roman and Byzantine Studies 3, no. 2 (1960): 107–34. ↩
-
Aldus Manutius, note to readers on the title page of Erasmus, Adagiorum chiliades tres (Venice: Aldus, 1508), University of Basel Library copy. Translation ours. ↩
-
Anthony Grafton, Inky Fingers: The Making of Books in Early Modern Europe (Cambridge, MA: Harvard University Press, 2020). ↩
-
Desiderius Erasmus, “Festina lente,” Adages II.i.1, in Collected Works of Erasmus, vol. 33, Adages II i 1 to II vi 100, trans. and annot. R. A. B. Mynors (Toronto: University of Toronto Press, 1991). ↩
-
Weixin Liang et al., “GPT Detectors Are Biased Against Non-Native English Writers,” Patterns 4, no. 7 (2023): 100779, https://doi.org/10.1016/j.patter.2023.100779. ↩
-
“‘Language and Image Minus Cognition’: An Interview with Leif Weatherby,” JHI Blog, June 11, 2025, https://www.jhiblog.org/2025/06/11/language-and-image-minus-cognition-an-interview-with-leif-weatherby/; see also Leif Weatherby, Language Machines: Cultural AI and the End of Remainder Humanism (Minneapolis: University of Minnesota Press, 2025). ↩
-
Alfonso Maruccia, “Amazon Is Shutting Down Mechanical Turk After More Than 20 Years,” TechSpot, August 27, 2026, https://www.techspot.com/news/113643-amazon-shutting-down-mechanical-turk-after-more-than.html. ↩
-
Lauren M. E. Goodlad, “Editor’s Introduction: Humanities in the Loop,” Critical AI 1, nos. 1–2 (2023), https://doi.org/10.1215/2834703X-10734016. ↩
-
Greg Kestin et al., “AI Tutoring Outperforms In-Class Active Learning: An RCT Introducing a Novel Research-Based Design in an Authentic Educational Setting,” Scientific Reports 15 (2025), https://doi.org/10.1038/s41598-025-97652-6. ↩
-
Hamsa Bastani et al., “Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics,” Proceedings of the National Academy of Sciences 122, no. 26 (2025), https://doi.org/10.1073/pnas.2422633122. A correction to this article was published in August 2025. ↩
-
Hao-Ping Lee et al., “The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects from a Survey of Knowledge Workers,” in Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (New York: ACM, 2025), 1–22, https://doi.org/10.1145/3706598.3713778. ↩
-
“Chicago Sun-Times Admits Summer Book Guide Included Fake AI-Generated Titles,” NBC News, May 20, 2025, https://www.nbcnews.com/tech/tech-news/chicago-sun-admits-summer-book-guide-included-fake-ai-generated-titles-rcna208325. ↩
-
Anil R. Doshi and Oliver Hauser, “Generative AI Enhances Individual Creativity but Reduces the Collective Diversity of Novel Content,” Science Advances 10, no. 28 (2024), https://doi.org/10.1126/sciadv.adn5290. ↩
-
Nathan Heller, “The End of the English Major,” The New Yorker, March 6, 2023. ↩
-
MLA-CCCC Joint Task Force on Writing and AI, Working Paper: Overview of the Issues, Statement of Principles, and Recommendations (Modern Language Association and Conference on College Composition and Communication, July 2023), https://hcommons.org/app/uploads/sites/1003160/2023/07/MLA-CCCC-Joint-Task-Force-on-Writing-and-AI-Working-Paper-1.pdf. ↩