Ink That Never Dries

In February 1882, Friedrich Nietzsche, whose eyesight was failing, began using a typewriter: a Danish “writing ball” invented by Rasmus Malling-Hansen. Writing to his friend Heinrich Köselitz, he made an observation that the media theorist Friedrich Kittler later made famous: Unser Schreibzeug arbeitet mit an unseren Gedanken. Our writing tools are also working on our thoughts.¹

The word processor is the writing tool that has worked on more people’s thoughts than any other. Most of the world’s writing, from emails and school essays to contracts, novels and government reports, passes through one. For six decades it held to a consistent philosophy, rarely stated because it seemed obvious: the person writes the words, and the machine makes those words cheaper to store, revise, arrange and share. Every major development in its history made one more part of the work around writing easier, without touching the writing itself.

Since 2023 that philosophy has broken. The major word processors now offer to produce the words. This essay follows how the word processor came to be what it is, describes what it has become, and examines what changes when the tool that once recorded our thinking begins to propose it.

When the page was an object

For most of the history of writing, a text was a physical thing. A manuscript was a stack of handwritten pages; a book was a printer’s fixed impression of type. The typewriter, commercially available from the 1870s, made writing faster and more legible but kept the essential property: a typed page was an object. A mistake could be erased, covered or struck through, but a significant change meant typing the page again, and a change on page three could mean retyping everything after it.

This shaped both how writing was done and who did it. Authors often wrote by hand and handed drafts to typists. Offices built typing pools, staffed largely by women, to turn dictation and handwriting into clean copies. Revision had a cost measured in hours of someone’s labour, which encouraged writers to settle their thoughts before committing them to the page. A reviewer of Matthew Kirschenbaum’s literary history of word processing described the drudgery that word processing ended: writers no longer had to “type and retype clean copies of multiple revisions, or make sense of tousled penmanship, or master the art of smoothly obscuring liquid-paper brushstrokes.”²

Words become data

The first machine to change this had no screen. In 1964 IBM introduced the Magnetic Tape Selectric Typewriter, the MT/ST, an electric typewriter attached to a tape drive that recorded every keystroke. A typist could correct the tape and have the machine print a clean copy automatically. IBM marketed it as a “word processing” machine. The term is usually traced to the German Textverarbeitung, attributed to the IBM engineer Ulrich Steinhilper in the late 1950s, though the historian Thomas Haigh has argued that the English phrase may have come first.³ Either way, the name reveals the original theory. Words were treated as data to be processed efficiently, and the problem being solved was the office’s volume of paperwork, not the writer’s imagination.

Writers found the machine anyway. Kirschenbaum identifies the British novelist Len Deighton’s Bomber, published in 1970, as “a strong candidate for the first novel significantly created on a word processor.” Deighton drafted on an ordinary Selectric; his literary secretary, Ellenor Handley, then retyped and revised the novel on an MT/ST whose tape cartridges held about 24,000 characters, “remembered revisions,” and printed at 170 words per minute.⁴ The first word-processed novel was processed by someone other than its author. The tool did not yet change who wrote. It changed who retyped.

The writer takes the keyboard

In the 1970s, screen-based word processing systems from companies such as Wang spread through offices. At the same time, microcomputers brought the possibility to individuals. Electric Pencil appeared in 1976; WordStar, released in 1978, became the standard for a generation of early personal computer users; WordPerfect dominated IBM-compatible PCs through much of the 1980s. By the late 1970s a few novelists, many of them science fiction writers such as Jerry Pournelle and Larry Niven, were composing directly on personal computers.²

With these programs the writer typed and revised their own text with no intermediary, and the cost of revision fell to almost nothing. A sentence could be rewritten ten times in a minute; a paragraph could move to another chapter; an entire draft could be restructured without retyping a word. Writing no longer had to begin from a settled thought. A writer could start rough and discover what they meant by changing what they had said. The philosopher Michael Heim, in Electric Language (1987), one of the first book-length philosophical studies of the subject, examined how this fluidity altered writers’ relationship to their own texts.

The screen becomes the page

A separate line of development made the screen show what would print. At Xerox PARC in 1974, Butler Lampson and Charles Simonyi built Bravo, an editor that displayed text with fonts and formatting much as it would appear on paper, an approach later called WYSIWYG, “what you see is what you get.” Simonyi went on to lead the development of Microsoft Word. Its first version, for MS-DOS, shipped in 1983, and free demonstration copies were bundled with the November 1983 issue of PC World.⁵

Word for DOS was, in the Computer History Museum’s phrase, “an also-ran.” What changed the market was Word for Windows in 1989. “By 1993 it was generating 50% of the word processing market revenue, and by 1997 it was up to 90%,” the museum notes, crediting both Microsoft’s marketing and the technical achievement of running a graphical word processor on ordinary PCs of the time, with a few megahertz of processing power and a megabyte of memory.⁶

WYSIWYG carried its own philosophy: the screen should imitate paper, and the writer should control appearance directly. That empowered writers to do what had required typesetters. It also merged two jobs that had been separate for centuries. The writer became the typesetter, and a document could be beautifully formatted long before it was finished.

Word’s dominance had a second consequence. Its file format became, in effect, the common language for documents in offices, courts and publishing houses worldwide. When governments and institutions later pushed for open standards, two competing XML-based formats were adopted by international standards bodies in 2006 and 2008. Whatever the format, the document had become a standard object with standard parts: styles, headers, footnotes, comments, tracked changes.

The rival philosophy: structure over appearance

A different school held that what a text is should be separate from how it looks. Its most ambitious early expression came from the computer scientist Donald Knuth, who began writing TeX in 1978 after being dissatisfied with the typesetting of his own books. In TeX, the author marks structure and content, and a program applies the rules of fine typography. The same separation underlies the web, where HTML describes structure and stylesheets describe appearance, and it found a popular form in John Gruber’s Markdown in 2004. “Readability, however, is emphasized above all else,” Gruber wrote. “A Markdown-formatted document should be publishable as-is, as plain text, without looking like it’s been marked up with tags or formatting instructions.”⁷ The essays on this site are written that way.

The two traditions disagree about what a document is. In one it is a set of pages; in the other, a structure of meaning that can be rendered differently on every screen. As text moved to phones, tablets and the web, the structural view gained ground. But the most widely used word processors still present the page, a reminder of how long the shape of paper outlives paper.

The first assistants

Long before generative AI, word processors had begun to offer judgement as well as storage. Spelling checkers became standard in the 1980s, and grammar checking followed in the 1990s. In 1997 Microsoft added an animated Office Assistant that appeared when it detected that a user might be writing a letter and offered help. It was retired in the following decade. These features marked the first small break in the original philosophy: the tool was beginning to comment on the words, not just store them. But the words were still the writer’s. The machine underlined; the person decided.

The document becomes a live object

The move to the browser changed the document’s nature again. Google acquired the small company behind Writely in 2006 and relaunched it as part of Google Docs, which allowed several people to edit the same document at the same time from anywhere. Every keystroke was saved. A version history replaced the folder of drafts, comments and suggestions replaced emailed attachments, and the document became a single shared, continuously updated state. That shift, and its effects on how people communicate, is explored in the essay on communication tools on this site.

What the word processor is today

The word processor of the mid-2020s is less a program than a service. Documents live in the cloud and open on any device. They are shared by link rather than by file, edited by several people at once, commented on, suggested into, and restored from automatic version histories. The boundary between the word processor and other tools has blurred: block-based workspaces such as Notion combine documents with databases, wikis and task lists; plain-text editors serve writers who prefer Markdown; chat applications have begun to hold documents of their own.

The scale is enormous. When Microsoft announced changes to its consumer plans in January 2025, it noted that it had 84 million consumer subscribers to Microsoft 365, separate from its far larger business customer base.⁸ Google Docs is used across schools, businesses and governments worldwide. For most people who write anything at all, one of a handful of products is the medium through which their words pass.

For almost all of this history, a single philosophy held. The word processor made writing cheaper in every respect except one: the words themselves.

The machine offers to write

That changed in 2023, with remarkable speed.

On 14 March 2023, Google announced its vision of “how Workspace is evolving to incorporate AI as a collaborative partner,” including features to “brainstorm, proofread, write, and rewrite in Docs.”⁹ Two days later Microsoft introduced Copilot for its Office applications. Its description of the feature in Word stated the new philosophy with unusual directness: “With Copilot in Word, you can jump-start the creative process so you never start with a blank slate again. Copilot gives you a first draft to edit and iterate on.” It added: “You’re always in control as the author, driving your unique ideas forward.”¹⁰

Microsoft made Copilot generally available to business customers on 1 November 2023, at $30 per user per month. In January 2025 it brought Copilot into its consumer Microsoft 365 Personal and Family plans and raised their US prices by $3 a month, the first increase since their launch, while offering “Classic” plans without the AI features for existing subscribers who preferred them.⁸ Google made a similar move the same month, including its Gemini AI features in Workspace business plans and raising their prices. Within two years, generating text had gone from experiment to a standard, priced-in part of the mainstream word processor.

The convergence also ran in the other direction. On 3 October 2024 OpenAI introduced canvas, a document-like workspace inside ChatGPT for writing and coding projects, where a user and the model could edit the same text side by side.¹¹ Other AI assistants added similar spaces. The word processor was acquiring a chatbot, and the chatbot was acquiring a word processor. The two were becoming the same kind of thing: a document that you can talk to.

What the evidence shows so far

The effects of these tools on writing are only beginning to be studied, but a few careful experiments point in consistent directions.

In a study published in Science in July 2023, the economists Shakked Noy and Whitney Zhang gave 453 college-educated professionals realistic writing tasks from their occupations and randomly gave half of them access to ChatGPT. The average time taken fell by 40 percent and the quality of the output, as judged by evaluators, rose by 18 percent; the gap between weaker and stronger writers also narrowed.¹² For the kind of routine professional writing that fills most documents, the gains were large.

A 2024 study in Science Advances by Anil Doshi and Oliver Hauser looked at creative writing. Participants wrote short stories, some with access to story ideas from a language model. Access to AI ideas made stories “evaluated as more creative, better written, and more enjoyable, especially among less creative writers.” But “generative AI–enabled stories are more similar to each other than stories by humans alone.” The authors described the result as a social dilemma: “With generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced.”¹³

A third study, a 2025 preprint from the MIT Media Lab led by Nataliya Kosmyna, used EEG to measure brain activity while participants wrote essays with a language model, with a search engine, or with no tools. Connectivity in the brain scaled down with the amount of external support, and participants who had used the model showed lower engagement even when later asked to write without it. The study is small and has not been peer reviewed, and its lead author has cautioned against reading it as evidence that the tools damage thinking.¹⁴ Its value is in the question it raises rather than in any settled answer: whether writing with a machine is the same mental act as writing alone.

How AI is changing the philosophy of the word processor

Taken together, the product decisions and the evidence suggest that the word processor’s underlying philosophy is changing in several connected ways. These are interpretations, built from the sources above, rather than claims made by the companies involved.

From instrument to collaborator

For sixty years the word processor was an instrument, like a pen with memory. It did nothing until the writer acted. Google’s chosen phrase, AI “as a collaborative partner,” names the change. A collaborator has initiative. It proposes, summarises, rewrites and continues. The relationship between the person and the tool is no longer that of a hand and a pen but closer to that of an author and an assistant, the arrangement that Len Deighton had with Ellenor Handley in 1970, now available to everyone and inverted: the assistant drafts and the author retypes.

The abolition of the blank page

“You never start with a blank slate again” is a philosophical claim disguised as a convenience. The blank page has always been where writing is hardest, and also where much of the thinking happens: deciding what to say, in what order, from what angle. The tradition of note-taking associated with Niklas Luhmann, discussed in the essay on note-taking tools, rests on the claim that “it is impossible to think without writing.” A tool that removes the blank page removes the moment of beginning, and with it, potentially, part of the reason for writing at all.

There is also a subtler effect. A proposed first draft is an anchor. Even when a writer rewrites heavily, the structure, examples and framing of the suggested draft shape what follows. The writer edits within a space someone else has outlined.

From producing words to choosing them

As machines produce more of the text, the writer’s work shifts toward selection: accepting, rejecting, reordering and correcting. The US Copyright Office’s January 2025 report on AI and copyrightability reflects this shift in legal form. It concluded that prompts alone do not give a person enough control over the output to count as its author, while human authorship can still arise from a person’s own expressive contributions, meaningful modifications, or the selection and arrangement of material, assessed case by case.¹⁵ In law at least, the author is the one who controls the expressive choices, not the one who typed a request.

From the writer’s voice to the average voice

The word processor has always standardised some things. Templates, default fonts and style sheets gave documents a common appearance. Generative AI extends standardisation from the look of a document to its language. Doshi and Hauser’s finding names the risk precisely: each writer’s work improves, while the collection of everyone’s work becomes more alike. A tool used by hundreds of millions of people that proposes the most plausible next sentence will, unless deliberately designed otherwise, pull prose toward the centre.

From writing for readers to writing for machines

The same tools that draft documents also summarise them. When a long report is condensed by an assistant before a colleague reads it, and the colleague’s reply is drafted by another assistant, a growing share of the text in organisations is produced by machines and consumed by machines, with people reading the summaries. Kirschenbaum warned in The Atlantic in March 2023 of a coming flood of machine-generated text, which he called a “textpocalypse.” Inside the word processor the concern is quieter but related: if documents are written to be summarised, the purpose of careful prose changes.

From a product to a metered service

The business model has changed too. A word processor was once bought once, then subscribed to; now AI features are bundled into subscriptions with higher prices, or sold as additional seats, and some plans count usage in AI credits. The philosophy of the tool is visible in its pricing: the most expensive part of the product is no longer storing or formatting text but generating it.

What the history suggests

None of these changes is simply good or bad, and the history offers some perspective. Every earlier stage of the word processor provoked worry. Writers feared that free revision would make prose sloppy, that WYSIWYG would turn authors into amateur designers, that collaboration would dissolve individual voice. Some of those worries were partly right, and writing adapted anyway.

What is genuinely new is that the earlier changes all redistributed labour around the words. The MT/ST removed retyping. The personal computer removed the typist. WYSIWYG removed the typesetter. The cloud removed the attachment. Each left the composition of sentences with a person. Generative AI is the first change in the word processor’s history that offers to take over the composition itself.

That is why the philosophical question matters more now than at any earlier stage. A tool can be designed to make writing easier by doing it, or to make writers better by supporting them while they do it. The research so far suggests both are possible: the same technology that saves 40 percent of the time on a routine memo can narrow the range of stories people tell. Which of these a word processor emphasises is a design decision, made in defaults, prompts, interfaces and prices, by the companies that build them and by the institutions and individuals that choose how to use them.

Understanding the philosophy

The changes described above are easier to judge when the ideas underneath them are laid out one by one. Each of the ideas below is explained in plain terms, connected to the tradition of thought that has examined it most closely, traced in the history of the word processor, tested against its strongest objection, and related to writing today. The links to philosophers and historians are interpretations offered to make the ideas clearer. The engineers and companies in this history did not, as a rule, cite them.

1. Is the tool part of the mind?

The idea in plain terms

People often assume that thinking happens in the head and writing merely records it. Many writers report the opposite: they find out what they think by writing it. If that is true, a writing tool is not outside thought. It is one of the places where thought happens.

The tradition

In Plato’s Phaedrus, Socrates tells a story in which the Egyptian god Theuth presents the invention of writing to King Thamus as an aid to memory and wisdom. Thamus replies that it will do the opposite: people who rely on written marks will stop exercising their memories and will seem wise without being so. The worry that a tool for thinking might replace thinking is therefore as old as writing itself. In 1998 the philosophers Andy Clark and David Chalmers gave the question a modern form in “The Extended Mind,” which opens: “Where does the mind stop and the rest of the world begin?”¹⁶ They argued that when a person relies on an external device as reliably as on their own memory, the device is, in a meaningful sense, part of their cognitive system.

Where it shows up

Nietzsche’s remark that “our writing tools are also working on our thoughts” belongs to this tradition.¹ So does the Luhmann principle discussed in the essay on note-taking tools, that thinking without writing is impossible. The early personal computer word processors let writers start rough and discover what they meant by revising, which made the screen part of the process of thinking, not just its record.

The strongest objection

Thamus was wrong in an important sense. Writing did weaken some forms of memory, but it made possible philosophy, science, law and history as we know them, none of which could exist in a purely oral culture. Every tool that changes how people think also lets them think things that were impossible before.

What it means today

If the tool is part of the mind, then a tool that proposes sentences is not merely helping a writer express a thought. It is taking part in forming the thought. The extended mind view makes AI writing assistants more significant, not less: they extend the place where thinking happens to include a model trained on other people’s words.

2. From fixed text to fluid text

The idea in plain terms

A printed or typed page is fixed. A document in a word processor is never finished, and can change at any moment. That difference changes how people relate to what they write.

The tradition

The historian Elizabeth Eisenstein, in The Printing Press as an Agent of Change (1979), argued that printing gave texts a new fixity: thousands of identical copies made it possible to compare, cite, correct and build on texts reliably, which she connected to the growth of modern scholarship and science. The scholar Walter Ong, in Orality and Literacy (1982), titled a chapter “Writing restructures consciousness,” arguing that each new technology of the word, from writing to print to electronic media, changes how people think and remember.

Where it shows up

The typewriter kept the page as an object, so writers settled their thoughts before committing them. The MT/ST began to dissolve that fixity by recording keystrokes so pages could be corrected and reprinted. Personal word processors made revision nearly free, and Google Docs made the document a single shared state that changes continuously, with a version history in place of a drawer of drafts. The ink never dries.

The strongest objection

Fluidity has clear benefits. Writers can explore, experiment and restructure without penalty, and the ability to revise endlessly is arguably what made word processing so popular. Fixity, meanwhile, is not lost: documents can still be exported, archived and published in stable forms.

What it means today

Generative AI adds a new kind of fluidity. Not only can the writer change any sentence; the tool can regenerate it, offering a different version each time. A text that can be produced again with a click is fluid in a deeper sense, and that raises the question of what makes any particular version worth keeping.

3. The medium shapes the message

The idea in plain terms

How a writing tool presents a document, as a page, a structure, or a conversation, shapes what people write and how they think about it.

The tradition

Marshall McLuhan’s Understanding Media (1964) began with a chapter titled “The Medium Is the Message.” His argument was that the most important effect of a medium is not the content it carries but the change it makes to the scale, pace and patterns of human life.

Where it shows up

WYSIWYG, beginning with Bravo at Xerox PARC, made the screen imitate paper and turned writers into their own typesetters, so a document could look finished long before its thinking was. The structural tradition, from TeX to HTML to Markdown, treated the document as a structure of meaning that can be displayed in any form. Gruber’s requirement that Markdown be “publishable as-is, as plain text” is a statement about which matters more.⁷ Each design carried a theory of what a document is, and each nudged its users toward a way of writing.

The strongest objection

McLuhan’s slogan can exaggerate. Content still matters, and good writers produce good work in almost any tool. People are not simply shaped by their media; they choose, adapt and resist them.

What it means today

The new medium is the document you can talk to. When writing happens in conversation with a model, the natural unit of work becomes the request and the response rather than the sentence and the paragraph. That changes what writing feels like, whatever the content.

4. Who loses the skill when the machine gains it?

The idea in plain terms

Each stage of the word processor took a task away from one group of people. Sometimes that freed them; sometimes it removed skills and jobs.

The tradition

In Labor and Monopoly Capital (1974), the writer Harry Braverman argued that twentieth-century management and technology tended to deskill work, breaking crafts into simple, controllable steps and moving knowledge from workers into machines and procedures. His book drew close attention to clerical and office work, the world of the typing pool.

Where it shows up

Typing pools, staffed largely by women, turned drafts into clean copies. The personal computer let the writer take the keyboard, and the typist’s job largely disappeared. WYSIWYG absorbed the typesetter. The history in this essay can be read as a sequence of skills moving into software, until generative AI reached the one skill that had been left with the writer: composition.

The strongest objection

The evidence also shows upskilling. In Noy and Zhang’s experiment, AI narrowed the gap between weaker and stronger writers, helping the people who struggled most.¹² Tools that do part of a task can let people do more, not less, and many writers welcomed the end of retyping.

What it means today

Both effects are real, and they may fall on different people. A tool can raise the floor for those who find writing hard while weakening the practice of those who no longer need to do it. The Kosmyna study’s questions about engagement are a version of Braverman’s, asked about the individual mind rather than the workforce.¹⁴

5. What is an author?

The idea in plain terms

When a machine drafts and a person edits, who wrote the result? The answer matters for credit, responsibility and law.

The tradition

In 1967 the critic Roland Barthes published “The Death of the Author,” arguing that a text’s meaning is not fixed by its writer’s intentions but produced in reading, from a web of earlier texts and conventions. In 1969 Michel Foucault asked “What Is an Author?” and argued that the author is a function: a role that a society assigns to certain texts for purposes such as ownership, responsibility and classification, which has varied across history.

Where it shows up

Len Deighton’s Bomber was processed by someone other than its author, and nobody doubted that he wrote it. Microsoft’s Copilot launch promised: “You’re always in control as the author.”¹⁰ The US Copyright Office’s 2025 report is Foucault’s author function written in law: it concluded that prompts alone do not make a person an author, while expressive contributions, meaningful modifications and the selection and arrangement of material can.¹⁵

The strongest objection

Barthes and Foucault were writing about literature and the social role of authorship, not about legal liability or professional accountability. Organisations and courts need a person who stands behind a document, and that need does not disappear because a machine helped. Someone still signs.

What it means today

Generative AI makes the author function visible by stretching it. Authorship is shifting toward choosing, directing and taking responsibility, and away from producing each word. Whether that counts as writing will be decided by readers, institutions and law, not by the tools.

6. The pull toward the average

The idea in plain terms

A language model proposes the most plausible continuation of a text. A tool used by hundreds of millions of people that offers the most plausible sentence will tend to make writing more alike.

The tradition

In Being and Time (1927), Martin Heidegger described das Man, usually translated “the they” or “the one”: the anonymous public way of doing things, in which people say what one says and think what one thinks, without ever making those views their own. He connected it to Gerede, “idle talk,” language passed along without being grounded in a person’s own understanding.

Where it shows up

Templates and default styles made documents look alike. Generative AI extends that standardisation to the language itself. Doshi and Hauser’s finding states the result precisely: “writers are individually better off, but collectively a narrower scope of novel content is produced.”¹³ What one would write becomes what is written.

The strongest objection

Shared standards have real value. Clear, conventional prose is easier to read, and many people, including non-native speakers and people who find writing hard, benefit from help that brings their work closer to accepted norms. Not every document needs a distinctive voice; most need to be clear.

What it means today

The question is not whether standard language is good or bad but where it belongs. A memo can be average. A personal essay, a piece of research or an argument that challenges received opinion loses something essential when it is pulled toward what most people would say.

7. We become writers by writing

The idea in plain terms

Skills are built by practice. If a tool does the hardest part of writing, the writer may not develop the abilities that the hard part used to build.

The tradition

In the Nicomachean Ethics, Aristotle argued that virtues are acquired by practice: people become just by doing just acts and brave by doing brave ones, as a musician becomes skilled by playing. Character and skill are not received; they are formed by repeated effort.

Where it shows up

Copilot’s promise that “you never start with a blank slate again” removes the moment where much of the effort of writing lies: deciding what to say and how to begin.¹⁰ That moment is also where writers practise the judgement that makes them good. The early spelling and grammar checkers commented on words the writer had produced; the new tools offer to produce them, which changes what the writer practises.

The strongest objection

Societies already accept that some skills fade when tools take them over. Few people do long division by hand or write in fine penmanship, and few miss them. Freed from some tasks, people practise others. It is possible that judging and editing machine drafts will become the core skill of writing.

What it means today

The useful question for any writer, teacher or institution is which parts of writing are worth practising for their own sake. Where writing is a means of thinking, as in education and research, removing the effort may remove the point. Where it is only a means of delivering information, the effort may safely go.

Bringing the threads together

The word processor’s history is a sequence of changes in where writing happens and who does it. If the tool is part of the mind, each change altered thinking as well as typing. Printing made text fixed; the word processor made it fluid; generative AI makes it regenerable. Each medium carried its own theory of what a document is, and each moved a skill from people into software, from the typist to the typesetter and now to the composer of sentences. The result is pressure on the idea of the author, a pull toward the average voice, and a question about which kinds of effort make people writers. None of these consequences is fixed. They depend on how tools are designed and how people choose to use them.

Ink that never dries

Return to Nietzsche’s observation that our writing tools work on our thoughts. For most of the word processor’s history, the tool worked on thought indirectly: by making revision free, it encouraged thinking on the screen; by making layout easy, it changed how documents looked; by making documents shared, it made writing a conversation. The words themselves remained the writer’s.

Now the tool can work on thought directly, by proposing the words in which it is expressed. The ink of the word processor never dried, which was always its gift: nothing had to be final. What remained fixed was the hand that wrote. The question the next decade will answer is whether the hand stays fixed too, or whether writing becomes, for most people most of the time, the act of choosing among sentences a machine has already written. The most important feature of a word processor may soon be not what it can write, but what it leaves for us to do.

Selected bibliography

• Kittler, Friedrich A. Gramophone, Film, Typewriter. 1986; trans. Geoffrey Winthrop-Young and Michael Wutz, Stanford University Press, 1999.
• Kirschenbaum, Matthew G. Track Changes: A Literary History of Word Processing. Harvard University Press, 2016.
• Kirschenbaum, Matthew G. “Prepare for the Textpocalypse.” The Atlantic, March 2023.
• Hicks, Dylan. “Type Slowly: Word Processing and Literary Composition.” Los Angeles Review of Books, 14 April 2016. lareviewofbooks.org
• Heim, Michael. Electric Language: A Philosophical Study of Word Processing. Yale University Press, 1987.
• Haigh, Thomas. “Remembering the Office of the Future: The Origins of Word Processing and Office Automation.” IEEE Annals of the History of Computing 28, no. 4 (2006).
• Computer History Museum. “Microsoft Word for Windows Version 1.1a Source Code.” computerhistory.org
• Gruber, John. “Markdown: Syntax.” Daring Fireball, 2004. daringfireball.net
• Voolich Wright, Johanna. “A new era for AI and Google Workspace.” Google Workspace Blog, 14 March 2023. workspace.google.com
• Spataro, Jared. “Introducing Microsoft 365 Copilot – your copilot for work.” Official Microsoft Blog, 16 March 2023. blogs.microsoft.com
• Microsoft. “Copilot is now included in Microsoft 365 Personal and Family.” Microsoft 365 Blog, 16 January 2025. microsoft.com
• OpenAI. “Introducing canvas.” 3 October 2024. openai.com
• Noy, Shakked, and Whitney Zhang. “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence.” Science 381 (2023): 187–192.
• Doshi, Anil R., and Oliver P. Hauser. “Generative AI Enhances Individual Creativity but Reduces the Collective Diversity of Novel Content.” Science Advances 10, no. 28 (2024). PMC
• Kosmyna, Nataliya, et al. “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” arXiv:2506.08872 (2025), preprint.
• US Copyright Office. Copyright and Artificial Intelligence, Part 2: Copyrightability. January 2025.
• Plato. Phaedrus. c. 370 BCE.
• Clark, Andy, and David Chalmers. “The Extended Mind.” Analysis 58, no. 1 (1998): 7–19.
• Eisenstein, Elizabeth L. The Printing Press as an Agent of Change. Cambridge University Press, 1979.
• Ong, Walter J. Orality and Literacy: The Technologizing of the Word. Methuen, 1982.
• McLuhan, Marshall. Understanding Media: The Extensions of Man. McGraw-Hill, 1964.
• Braverman, Harry. Labor and Monopoly Capital. Monthly Review Press, 1974.
• Barthes, Roland. “The Death of the Author.” 1967.
• Foucault, Michel. “What Is an Author?” 1969.
• Heidegger, Martin. Being and Time. 1927.
• Aristotle. Nicomachean Ethics, Book II.

Research limitations

Nietzsche’s German sentence is quoted as it is attributed through Kittler and later scholarship; the original letter was not consulted. Kirschenbaum’s Track Changes is drawn on through his published claims as reported in reviews, and his Atlantic essay through a republished copy and summaries. The origin of the term “word processing” is contested. Descriptions of Bravo, WordStar, TeX and the Office Assistant rely on standard histories rather than primary documents. The Computer History Museum’s market figures are revenue shares. Pricing details describe US list prices at the time of the announcements cited and have changed since. The Copyright Office report is summarised from legal summaries of its conclusions. The Kosmyna study is a small, non-peer-reviewed preprint and is presented as raising a question rather than answering it. The account of how AI is changing the word processor’s philosophy is the author’s interpretation of these sources, not a claim made by any company. The philosophical section’s connections to Plato, Clark and Chalmers, Eisenstein, Ong, McLuhan, Braverman, Barthes, Foucault, Heidegger and Aristotle are the author’s interpretation, offered to make the ideas understandable; their works are paraphrased, apart from the opening line of Clark and Chalmers’s paper and the chapter titles of Ong and McLuhan.

Footnotes

1. Attributed to Nietzsche’s letter to Heinrich Köselitz of February 1882 and popularised by Friedrich Kittler, Gramophone, Film, Typewriter (1986); see also “How Typing Transformed Nietzsche’s Consciousness,” The MIT Press Reader, thereader.mitpress.mit.edu. ↩ ↩²

2. Dylan Hicks, “Type Slowly: Word Processing and Literary Composition,” Los Angeles Review of Books, 14 April 2016, reviewing Matthew G. Kirschenbaum, Track Changes (Harvard University Press, 2016). ↩ ↩²

3. On the competing origins of the term, see Thomas Haigh, “Remembering the Office of the Future,” IEEE Annals of the History of Computing 28, no. 4 (2006), and the summary at historyofinformation.com. ↩

4. Kirschenbaum’s account of Bomber as summarised in Hicks, “Type Slowly” (2016). ↩

5. “History of Microsoft Word,” Wikipedia, on the 1983 release and the PC World demonstration disk. ↩

6. Computer History Museum, “Microsoft Word for Windows Version 1.1a Source Code,” computerhistory.org. ↩

7. John Gruber, “Markdown: Syntax,” section “Philosophy,” Daring Fireball (2004). ↩ ↩²

8. Microsoft, “Copilot is now included in Microsoft 365 Personal and Family,” Microsoft 365 Blog, 16 January 2025; on the November 2023 business launch and its $30 price, see contemporary coverage such as TechRadar. ↩ ↩²

9. Johanna Voolich Wright, “A new era for AI and Google Workspace,” Google Workspace Blog, 14 March 2023. ↩

10. Jared Spataro, “Introducing Microsoft 365 Copilot – your copilot for work,” Official Microsoft Blog, 16 March 2023. ↩ ↩² ↩³

11. OpenAI, “Introducing canvas,” 3 October 2024, openai.com/index/introducing-canvas. ↩

12. Shakked Noy and Whitney Zhang, “Experimental evidence on the productivity effects of generative artificial intelligence,” Science 381 (2023): 187–192, doi:10.1126/science.adh2586. ↩ ↩²

13. Anil R. Doshi and Oliver P. Hauser, “Generative AI enhances individual creativity but reduces the collective diversity of novel content,” Science Advances 10, no. 28 (2024), abstract, doi:10.1126/sciadv.adn5290. ↩ ↩²

14. Nataliya Kosmyna et al., “Your Brain on ChatGPT,” arXiv:2506.08872 (June 2025, revised December 2025); MIT Media Lab. ↩ ↩²

15. US Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (29 January 2025); see the summary by Sidley Austin, sidley.com. ↩ ↩²

16. Andy Clark and David Chalmers, “The Extended Mind,” Analysis 58, no. 1 (1998), first sentence. ↩