The morning my AP English midterm paper was due, I woke up early enough to read it one more time.
I had put more work into that paper than anything I had written before, and there was a paragraph on page three that made no sense. With fresh eyes, I could see that it dragged down the whole argument. It may even have contradicted it.
This was December of 1983, and I had recently made the transition from turning in final papers in cursive to delivering them on a typewriter, a huge step up in professionalism that came with terms attached. White-out could handle a mistyped word. Anything bigger meant retyping the page. For a perfectionist, almost everything meant retyping the page.
The bus would be out front in less than ninety minutes. I could think fast enough to fix the paragraph, but I had not taken my high school typing class yet. My fingers were never going to reproduce five double-spaced pages before the bus arrived. So I turned to the only technology I knew that could save me.
My mom.
She was an outstanding typist and a fine writer in her own right. Together we fixed the paragraph, and then she sat down at the typewriter and produced the entire paper again, from the first sentence to the last. We worked as a team. I owned the argument. She owned the means of production. I did not think it was cheating then, and I do not think it is cheating now.
That was the deal I made with the machine I learned to write on. No pain, no gain. Every meaningful change meant somebody typed the page again, sometimes the whole paper.
At the time, none of this felt like a feature.
A typewriter made changing your mind expensive. The farther you got into a page, the more costly a substantial revision became, and you learned to let things stand that you knew deserved better. Most weak paragraphs survived because fixing them meant retyping everything after them. Without my mom, page three would have been one of them.
Then word processors arrived and a paragraph could move without consequence. Structure could be revisited an hour before a deadline, thoughts rearranged on the fly, and writing got better, for the least glamorous of reasons. Changing your mind had become affordable.
Spell check and grammar tools worked the same side of the street. They caught mistakes in words we already knew how to spell and flagged conventions we already knew we were breaking. The red line pointed. Whether the sentence changed was still up to me. Grammar software has never approved of a sentence fragment. Several are standing in this essay anyway.
Later, search engines collapsed an afternoon in the library into twenty minutes at a desk. Each improvement removed a small piece of the work, and we loved almost all of them. I certainly did. Nobody wants the typewriter back.
That is what makes the current argument about artificial intelligence so strange. We talk as though writers spent centuries doing everything by hand until ChatGPT arrived and offered to cheat on their behalf. The history is longer than that. We have been handing off pieces of the work for generations, and the first assistant I ever used was not even a machine. The question worth asking is whether we know which work we are handing over.
None of those tools caused much existential panic once the novelty wore off, because we decided the jobs they took were adjacent to the real work. They supported the writing without becoming the writing.
Artificial intelligence makes that distinction harder to maintain because it does not stay at the edges. The same tool that corrects a sentence can supply the sentence. It can summarize the material before you have decided what matters in it, offer the counterargument before you have felt the weakness in your own. It can suggest what is interesting about a subject before you have had the chance to become interested in anything yourself. And the output can be good, which is what makes the problem difficult.
The purist says the line is obvious. Real writers write their own words. All of them. The purist is usually sitting in front of software that corrects his spelling, preserves every draft, searches thousands of pages in seconds, and lets him reorganize an essay without retyping a sentence. He does not experience any of this as technological assistance. These are simply the conditions under which writing happens.
The tools that were present when we learned the craft tend to disappear into our definition of the craft. The line we defend as principle is often just the year we happened to arrive.
The enthusiasts have a different blind spot. They point out that AI makes writers more productive, which is clearly true if the thing being measured is production. We can now generate competent prose in nearly any quantity we like. But competent prose was plentiful long before the machines arrived. The hard part was noticing something worth saying.
For a while, I thought the question was how much artificial intelligence a writer should use. I have come to think that is the wrong unit of measurement. Line up the tools in the order history delivered them, and something else comes into view. A list of jobs, changing hands one at a time.
So here is where I have landed. Give away the mechanical work. The typing, the punctuation, the citation hunt, the transcript cleanup, the two hundred words that need cutting. I will hand those to a machine all day and feel nothing. But deciding what you believe is not labor. Neither is recognizing that the obvious explanation is wrong, or sitting with an idea long enough for it to become less convenient and more true. Those jobs are the writing.
That December morning, the typing changed hands but the noticing never did. The division held because both of us knew which side of it we were on. If my mother had been the one to catch page three, to decide what the argument needed instead of producing it, the paper would have been hers, whatever name sat at the top of it.
The trouble is that the same tool can sit on either side of the line. I can give an AI five pages I have struggled through and ask where I am repeating myself, and that may save me an hour while leaving the essential work untouched. Or I can open a blank window, name a broad subject, and ask what is interesting about it, and that may save me an hour too.
Those are not the same hour.
I have been trying to figure out what test I actually use for myself, and the best one I have found has nothing to do with whether a machine touched the words. It asks whether I did the choosing. Could I defend every sentence in the piece, not merely as accurate, but as mine to put there? Is there something in it that required me to notice? Did the tool help me get somewhere I was already trying to go, or did it quietly choose the destination?
I do not always know.
That uncertainty is inconvenient, which is probably why we keep reaching for rules. Never use AI. Always disclose it. Use it for editing but not drafting, research but not ideas. Rules are attractive because rules let us stop paying attention, but each of these turns a judgment problem into a policy problem. None of them would have known what to do with my mother.
Writing has never been that tidy. A machine-generated sentence can be perfectly faithful to something I already thought, while a sentence I type entirely with my own hands can arrive borrowed and barely examined. Authorship lives in the decisions, which is a much harder place to inspect than the keystrokes.
Which brings me back to the typewriter, and not because I miss it. What I miss, or what I have only recently realized I was getting from it, was everything hidden inside the inconvenience.
All that retyping forced me to reread. A revision meant passing through the sentence again, then the paragraph, then the page, encountering my own argument over and over because the machinery gave me no choice. Most of that repetition was inefficient, and some of it was maddening, but things happened inside the wasted motion. A bad transition announced itself on the third pass. A sentence that had seemed fine began to sound false, and a detail I had treated as decoration turned out to be the point. The paragraph on page three did not announce itself while I was writing it. It waited for the reread.
I thought I was retyping. I was also thinking.
That realization changes how I look at every tool that promises to save me time. Of course I want the time, and the machine is welcome to the tedious parts. Genuinely. Please, take the tedious parts. What worries me is my own ability to know which parts those are. Because sometimes the sentence you are tired of rewriting is the sentence that has not told you what it means yet. Sometimes the half hour of wandering that feels like a failure to get to the point is how the point gets found. The inefficiency can be doing work you cannot see until it disappears.
The typewriter was not the last thing that made me do the work. At Apple, we refined every word of anything that mattered before it left the building, rehearsed past the point where rehearsing felt ridiculous, and took whatever time perfect required, because there was a standard to meet. We even had a phrase for it: Apple Caliber.
It meant that if something mattered, you kept going until it was right, long after “good enough” would have been perfectly defensible. Nobody offered you the extra passes as an option. The standard demanded them, the way the typewriter once had, and the work got better inside the demand.
For most of my life, something imposed the wandering on me whether I wanted it or not. First the machinery, later the standard. Nothing imposes it now. I can skip the rereading, skip the bad first version and the hour of not knowing what I think, and arrive almost immediately at something coherent, polished, and finished-looking. But the only standard I can enforce is my own. The wandering has become something I have to choose.
And increasingly, I have to choose it precisely because I no longer have to.
You might be thinking, “If the misbehaving paragraph was on page three, why did his mom need to retype all five pages?” Perhaps the perfectionist author also didn’t like the two globs of White-Out on page one. That sounds about right, if memory serves me.