Meet a professor fed up with AI slop who made her whole class use typewriters instead of computers.

Source: Fortune

Translation: A Cornell language instructor accidentally ran a controlled experiment on what thinking looks like when the shortcuts are removed, and her students noticed something the curriculum never planned to teach.

Grit Matthias Phelps, a German language instructor at Cornell, began bringing manual typewriters into her classroom in spring 2023 after growing frustrated with AI and online translation tools producing grammatically perfect assignments that her students hadn’t written. The assignment is analog by design: no screens, no spellcheck, no delete key, no online dictionaries. Students type in real time, make visible mistakes, cross them out with X’s, and hand in pages that look like process rather than product.

The outcomes the students report are not primarily about typing. They are about attention, cognition, and social interaction. Without screens, there were no notifications. Without Google, students asked each other questions. Without a delete key, they thought before typing. One computer science sophomore described being “forced to actually think about the problem on my own instead of delegating to AI or Google search” and noted that the absence of screens made him socialize more with classmates, which he observed was probably normal before the digital default and is now “drastically different from how we interact within the classroom in modern times.”

The typewriter exercise is part of a broader national trend toward in-class pen-and-paper exams and oral tests as educators attempt to create assessment environments that AI cannot enter.

The primary distortion is the framing of typewriters as the point. The exercise is not about typewriters. It is about what cognition looks like when the infrastructure of delegation is removed. The typewriter is a props department solution to a deeper problem: how do you create conditions in which a student must think, rather than retrieve? The machine is incidental. The constraint is the intervention.

The secondary distortion is the nostalgia frame the article occasionally slides into: “back in the old days,” the 1950s classroom aesthetic, the dinging bell. Framing this as a revival of old-school methods makes it easy to dismiss as a charming anachronism rather than recognizing it as a diagnostic. The professor is not advocating for typewriters. She is exposing what is lost when the default environment does the cognitive work that education is supposed to develop.

The deepest distortion is in how the AI-in-education debate is being conducted at a policy level. The national trend toward pen-and-paper exams and oral tests is a set of individual institutional responses to a tool that has entered the environment faster than any governance framework could address it. Each response is locally rational. Collectively, they are a patchwork of workarounds rather than a reckoning with the structural question the trainee teacher piece we covered identified: what is school actually for, and is the environment we have built compatible with that purpose?

For Phelps, the incentive is the one she states plainly: she wants to read what her students actually wrote. That is not a technological preference. It is a pedagogical relationship that AI-assisted work severs. When the assignment is already correct before she reads it, the reading has no purpose. The typewriter exercise is an attempt to restore the conditions under which her actual job is possible.

For students, the incentive structure the exercise reveals is more interesting than the assignment itself. The sophomore who described being forced to think on his own “instead of delegating to AI or Google search” was not describing a burden. He was describing a cognitive experience he recognized as valuable and identified as unusual. The fact that genuine independent thought feels like a forced condition rather than a default state is the finding, and it is one that the students themselves are articulating without being prompted.

For the technology industry, the incentive is to frame exercises like this as charming but marginal (i.e., a single professor’s pedagogical quirk rather than evidence of a systematic problem with how cognitive tools are reshaping the conditions under which learning occurs). If the typewriter exercise is news, it is news because it is exceptional. The question the industry would prefer not to answer is what the normal condition looks like — and whether the students who never have a typewriter day are developing the cognitive independence that the students who do are rediscovering.

For institutions, the incentive is to contain the AI problem within assessment design rather than address it at the level of how learning environments are structured. Oral exams and pen-and-paper tests prevent AI from entering the evaluation. They do not change the environment outside the evaluation (i.e., the homework, the preparation, the daily cognitive habits) where AI delegation is already the default. The assessment workaround is easier than the structural redesign, and it produces the same credential without the same guarantee of the capacity that credential is supposed to certify.

The immediate consequence of the typewriter assignment (for the students who experience it) is a documented shift in cognitive engagement, social interaction, and relationship to error. The freshman who kept her mistake-covered pages and planned to hang them on her wall had encountered something most of her education had not provided: visible evidence of her own thinking process, including its imperfections. That is not a typewriter outcome. It is a conditions-of-learning outcome that the typewriter happened to create.

The structural consequence of the broader trend toward AI-avoidance assessment is a bifurcation between evaluation environments and learning environments. Students are being assessed in conditions that require independent thought and then returning to preparation environments where delegation to AI is the default and the path of least resistance. The gap between those two environments does not produce cognitive independence. It produces students who can perform independence under exam conditions and delegate outside them, which is the exact pattern the trainee teacher piece documented in the students who resolved to develop their own thinking and then described AI-assisted thesis generation as “responsible use.”

The consequence the sophomore’s observation points toward is the most significant: he described asking classmates for help during the typewriter exercise and noted that this was probably normal before the digital era but is now unusual. The social dimension of learning ( i.e., the lateral knowledge exchange between students that has historically been one of education’s most important outputs) is being displaced not by AI directly but by the individual screen environment that AI has accelerated. Students are not just delegating cognitive work to machines. They are doing it alone, in parallel, in a room full of people they are not talking to.

The connection to the doll study we covered is direct: the cognitive capacity being undertrained in young children through solo screen engagement is the same capacity ( think perspective-taking, social modeling, collaborative sense-making etc.) that the typewriter exercise accidentally restores in college students by removing the screen. The developmental sequence is not a metaphor. It is a pipeline.

The typewriter is not the answer. The professor knows this. The students know this. The assignment is one day per semester, and no one is proposing to return to analog workflows as a general educational strategy. The calibration the exercise offers is not a technology policy. It is a diagnostic.

What the typewriter day reveals ( reliably, across multiple cohorts, without being designed to reveal it) is that the default digital environment has trained students to delegate before thinking, to correct before committing, and to work alone when surrounded by people who could help. None of these are technology problems. They are habit problems produced by an environment optimized for frictionless information retrieval rather than genuine cognitive development.

The calibration for educators is not to ban AI but to deliberately create conditions in which it is absent, regularly enough that students can distinguish between their own thinking and the thinking they have outsourced. The trainee teacher piece ended with a similar conclusion: not that AI should be excluded from education permanently, but that students need sustained experience of the friction that genuine thinking involves. The typewriter is a delivery mechanism for that friction. So is reading aloud. So is a handwritten essay. The technology is not the variable. The presence of genuine cognitive demand is.

The calibration for the broader AI deployment conversation is the one the computer science sophomore provided without being asked: being forced to think on his own felt unusual enough to name. In a world where AI is being embedded in every workflow, every assessment, every tool, and every interface, the conditions under which a person must think without assistance are becoming exceptional rather than normal. The typewriter exercise is news because it is rare. The question worth asking is what we are producing at scale in the environments where it never happens.