When I started writing this post, I built a few prototypes of what something called “hypographic media” could look like. I played around with them, decided they were all paltry, and concluded that it was better to describe the thing I was trying to understand in words rather than prototypes, because I am a creature of limited imagination and because it exists in a way that I don’t believe a web demo can access.
Tom Mullaney, who has so generously let me interview him and given me some of his time, made a point over a phone call that I haven’t been able to stop thinking about. He said: “We have hypographic mediums, but no hypographic media.” Let me explain.
In the Introduction to his excellent 2024 book, The Chinese Computer: A Global History of the Information Age, Tom discusses a crisis affecting a billion Chinese people: they are forgetting how to write. This isn’t a phenomenon affecting those without access to educational resources, but rather stalks China’s elite. Wealthy, urbanized, high-income Chinese are forgetting how to write even the most basic characters. The culprit is digital writing.
Tom introduces the word hypography to name how writing is occurring in this digital era, and argues that this new “technology of the intellect” has been taking shape over the past half-century. While orthographs, the results of conventional writing (or orthography), are formed by patterns of letters that we recognize as words bearing particular meanings, hypographs are a class of writing whose job is to help search for and retrieve orthographs from memory. The sequence of letters diantmvv is a hypograph used by a Chinese “Input Method Editor” (IME), which intercepts the keystrokes and figures out what Chinese characters the user who typed diantmvv wants to produce. The letters diantmvv that our user typed do not mean anything on their own, and only serve the purpose of directing retrieval. It is only the orthograph, the Chinese characters proposed by the IME and selected by the user, that perform the labor of meaning something. In short, orthographs are writing that says something, while hypographs retrieve or produce orthographs.
Hypography needn’t be meaningless. In his conclusion, Tom states that the prompts we give to LLMs are a hypographic sequence. While Chinese IMEs merely predict a preexisting intention—a writer already knows the characters they intend to write and merely inscribe the sequence of letters used to retrieve those characters—someone prompting an LLM may not have thought of the words or passages presented to them by an LLM when they prompt it to ideate or produce an essay. In this case, it is not necessary that the hypograph has no meaning of its own (prompts mean something!), but it is true that in the production of a final artifact, the hypograph vanishes.
Working definition: hypographic media are socially realized systems where retrieval or selection traces (prompts, queries, IME codes) are preserved and read as communicative objects, rather than being discarded as inputs.
Returning to Tom’s statement, what does it mean to have hypographic mediums, but no hypographic media? Mediums are often thought of as conduits; a television or a book is a medium for the transmission of information. Generations of media scholars have discussed how mediums also actively shape what and how we perceive and organize social life. That form and environment matter more than “content” and influence our reception of the content is well-summarized in McLuhan’s dictum “the medium is the message.” Media, on the other hand, is defined by Lisa Gitelman as “socially realized structures of communication,” where the technologies and protocols come together. John Durham Peters advocates for a more expansive notion of mediums and media: everything can be a medium, and the concept of media should not be confined to channels or institutions. Rather, media are infrastructures and environments we live within.
What, then, are hypographic media? We’re putting together two concepts: hypography, a form of writing that exists at a register removed from conventional writing and, in my reading, is the immediately erased evidence of decisions made in the process of producing an artifact; media, a socio-technical system that organizes communication. Hypographic mediums, which we do have, are the technical channels or tools that transmit hypography: IME interfaces, search boxes, prompt windows. Hypographic media, by analogy, could be thought of as socio-technical systems that make hypographic traces into communicative artifacts in their own right, as opposed to disposable inputs. Rather than vanishing once the orthograph appears, the hypograph is preserved, circulated, interpreted, and institutionalized.
I think numerous forms of proto-hypographic media already exist. “Process artifacts” can be used to describe the tangible byproducts of the development of a project or system: in software development, this might take the form of design documents, meeting notes, and other reference material. This gives insight into the choices made, tradeoffs considered, and other thinking that might have factored in the production of a final artifact but do not, themselves, exist as part of that artifact.
On the other hand, Wikipedia edit histories and git commits seem related because they are parts of the production of an orthograph, but are better thought of as paratexts of orthography because they are revisions of an orthograph as opposed to inscriptions whose instrumental role is to search for or produce orthographs.
While we can spend plenty of time quibbling over boundaries, it’s more interesting to consider the stakes. Why should we care about whether hypographic media exist?
In education, we often look for evidence of thinking and understanding in legible formats: the grade, the essay, and so on. The orthograph—an essay with no accompaniment—may no longer be sufficient evidence to conclude that understanding has occurred unless produced in specific circumstances. In The New Yorker, Princeton historian D. Graham Burnett describes how he assigns students to work with ChatGPT, then reads their dialogues; he treats some of the students’ rich exchanges with the model about theology and other subjects as serious intellectual work. In reading the process (which may or may not have led to an orthographic output such as an essay), Burnett is modeling hypographic media in education: the conversation that makes an essay or other output possible is itself a communicative object used for learning and assessment.
If writing is a mode of thinking (in fact, the dominant mode of thinking in our literate culture), hypogaphy is the part that evaporates—the choices disappear while the final artifact remains. For decades, many Chinese speakers have been using hypographs to write; now, many of us seeking to write or ideate are also using hypographs to produce a final artifact that is a piece of writing or an articulation of our own thinking.
Writing, as Tom suggests, has attained a new form in hypography. Thinking, and the evidence we use to evaluate it, has changed with it. I don’t want every keystroke immortalized, and (of course) private chatbot conversations shouldn’t be public by default. The alternative is selective publicity: keep the under-text private most of the time, but make particular hypographs citable where they serve learning, review, or safety (in classrooms, science, or safety-critical systems). In these zones, hypographic media are evidence and, sometimes, the work itself.