My life in keyboards
From a 1920s typewriter to... THIS weird thing!
I’ve used, daily, a century and a half of keyboard advances.
Recently, on getting the newest, best thing—the Glove80, above—it struck me how far technology has come since I learned to type…
…and how lucky I have been to work with some extraordinary, historically important computers. I fell in love with just three of them.
𐫱 This post has two halves. First, my school years, the 1960s to 1990:
How I began with Victorian technology — The electromechanical era — I mistake the future for a toy — At the MIT AI Lab — The Space Cadet — Symbolic perfection.
𐫱 And 1991 to the present:
Meeting the vampire goddess with a laptop — My 1977 iPhone prophesy fulfilled — The scourge of programmers — Enabling a treadmill — A thing of beauty — The state of the art — Entering the scene.
The first, free half may be historically interesting.
The second half, for paying subscribers only, may be useful as well!
The Victorian Way
When I was ten, I learned to type on a machine very similar to this one. It had been my grandmother’s, and she gave it to me when she retired. From what I know of her personal history, she probably bought it in the late 1920s.
That was forty-ish years before I started using it. But my parents’ typewriter, from about 1960, was mechanically identical, just cosmetically updated. And the fundamental design was essentially the same as the first commercially produced typewriter:
That was vintage 1873, about a century before I learned to type. The Victorian era nailed mechanical engineering!
These were powered solely by your finger muscles. You pushed a key, which was attached to a lever, hinged to another lever, which had a mirror-image metal letter shape at its end. That whacked against a cloth tape soaked in ink, causing it to whack against a sheet of paper, so some ink came off and stuck to the paper in the shape of the letter.
This seems unbelievably crude now! It’s weird remembering I used something like that, for many years.
I grew up in a world full of Victorian technology. It was literally still the steam era. This was a shocking realization for me when I visited the British National Railway Museum a few years ago. (It’s awesome! Go there if you get a chance!) Their enormous collection of historically important steam locomotives includes the very last one built in Britain for regular passenger service: dated 1963, a bit younger than me. It too seemed unbelievably crude; the technology had barely changed in a century.
But, I shouldn’t have been shocked. As I wrote in Meaningness and Time, the Victorian era was glorious. Apart from mechanical engineering, it created most of the social systems we still live in. They’ve been updated and improved some, and also have been decaying, alarmingly.
When I was a kid, not many people could type. Most jobs were still “blue collar.” Most people in office jobs didn’t type either. Offices had “secretaries,” part of whose work was typing for more important people. Or even “typists,” whose whole job was taking hand-written documents and typing them up.
Spending a whole work day typing seems… normal, now? But, on a mechanical typewriter, it was brutal. ⚠️ You had to push the keys quite a long way, quite hard. And then when the letter hit the hard rubber roller holding the paper… for every action, there is an equal and opposite reaction. (It did make a satisfying clack though!) Secretaries and typists were mainly young women, probably partly because their hands burned out after a few years.
⚠️ I’ll mark significant plot points with the ⚠️ symbol. Like Checkov’s Gun, these foreshadow dramatic future events.
So a ten-year-old boy learning to type was somewhat anomalous. However, other people who needed to type were academics, authors, and journalists. My parents were high school teachers and dropouts from serious academia, and were planning ahead for my Important Academic Career; or if I failed at that, my becoming an author or journalist.
What ten-year-old me actually wanted to do was build artificial intelligence. ⚠️ I had read Heinlein’s The Moon Is A Harsh Mistress, and I knew my mission in life for sure.
ASR-33: Computation at last
The school my parents taught at was well-funded. Not too long after I learned to type, it got a computer. And I got access! This was the thing. I was on my way to artificial intelligence.
The computer was a PDP-8/e, which was a wonderful and extraordinary thing. Even now, in retrospect! Among computers, it was my first love, and you never get over that. This post is mostly about keyboards, not computers, but I’m going to wax lyrical about the 8/e for a bit, before talking about its keyboard. (You can see that on the ASR-33 Teletype picture above, however.)
The PDP-8/e was the least expensive computer you could buy, with a starting price of $6,500. (That’s only $53,000 in 2025 dollars!) Nevertheless, it was astonishingly capable.
The original PDP-8 was the first “minicomputer,” meaning one inexpensive enough that a medium-sized company, or a division of a big one, could afford to buy one. Earlier computers, called “mainframes,” cost millions of 2025 dollars. The PDP-8 became the best-selling computer shortly after it was introduced. (By units; the mainframe market was much larger in dollars.) The PDP-8/e update replaced the original’s discrete transistor technology with early integrated circuits: a dozen transistors on a single silicon chip. That meant a several-fold price drop, so even a high school could afford one!
In fact, the PDP-8 series’ maker, Digital Equipment Corporation (DEC), sold mainly to the education market for many years. Their PDP-10, introduced just after the -8, was the standard mainframe in research university computer science departments in the 1970s. ⚠️ The technological superiority of their products, plus university-graduate programmers’ familiarity with them, resulted in their becoming the #2 computer maker in the 1980s, behind only IBM.
The school’s PDP-8/e had 12 kilobytes of memory. That is much less than even just the text of this post: more than 50 kilobytes. It may now seem somewhat unrealistic to do AI research on a 12-kilobyte computer, but that didn’t stop me. In fact, with 12 kilobytes, the 8/e could support half a dozen simultaneous users, using timesharing and virtual memory. It ran a sophisticated operating system—a cut-down version of the PDP-10’s—and had compilers or interpreters for a dozen programming languages.
Part of the appeal of DEC’s computers was that they were interactive. You could type at them, and they’d respond immediately. That wasn’t usual at the time! ⚠️
Specifically, you typed at an ASR-33 Teletype. This was essentially an electric typewriter, except that the keys, as well as causing metal to whack an inky ribbon against paper, sent an electric signal to the computer. And when the computer wanted to respond to you, it would send an electric signal back, which would cause automatic whacking.
This was an electromechanical device. While writing this post, I looked for a video of an ASR-33 in use. I spent years using them, learning to program, so I expected the sound to be nostalgic. But man, it’s a nasty racket! It’s not the friendly clack-tick-clack-dzeep of a mechanical typewriter. It sounds like a construction site, and the narrator has to yell to be heard over it.
You didn’t have push the keys as hard as on a mechanical typewriter, though. In this post, I will often describe the feel of using particular keyboards. I can’t clearly remember the feel of an ASR-33; but since the whole thing was shaking like hell when you used it, I doubt it was pleasant.
The rows of keys on the ASR-33 are diagonally offset—probably just like the keyboard on whatever device you are reading this on now. Mechanical typewriters had to be that way, because the keys actuated levers that had to pass beside each other. If the key rows were aligned, they levers would get in each other’s way.
The diagonal offset was already unnecessary on the ASR-33. The keys were electrical switches, not mechanically connected to the print head. But the assumption was that users would have learned to type on a mechanical keyboard, and it would be a nuisance to switch to an unfamiliar layout. This is user interface skeuomorphism, and it is ⚠️ BAD. Kids these days learn to type on an iPad; why not start them out on a more sensible keyboard arrangement?
The Teletype also served as our mass storage unit. You edited your program interactively, in the computer’s memory. Then when you wanted to save it, you would turn on the paper tape punch unit. (That’s the thing at the back left, with the pale yellow tape.) You’d get the computer to send the program to the punch, which would slowly reel out a hundred feet of inch-wide paper with holes in it. You’d take that home and bring it back next time you wanted to use the program. The front left box used electromechanical “fingers” to feel the holes in the tape as they went by. Cool, right?
Here’s the guts of an ASR-33. You can see it has nearly no electronics in it; it’s a big motor and a bunch of gears and levers and pulleys:
I realized, only while writing this post, that I grew up in the electromechanical era. When I was a little kid, the only electronic device in the house was the stereo: a great hulking metal thing full of vacuum tubes. Transistorized devices mostly became affordable for consumers only later, in the ’70s. Back in the ’60s, clunky electromechanical devices were replacing the elegant mechanical engineering of the more civilized Victorian age. Even the telephone and clocks were still purely electromechanical.
But the typewriter—it was still just mechanical.
Selectric: this is not the future
The school did own a single Selectric typewriter. Electromechanical typewriters had been commercially available since the 1920s, but they were so expensive that few were sold. Surprisingly, teleprinters (like the Teletype) were in routine use from the late 1800s—not long after the invention of the manual typewriter. They were used for sending telegrams, replacing the much slower and more error-prone manual Morse code. In fact, early electric typewriters were based on teleprinters, which pre-dated them by several decades.
The IBM Selectric was a breakthrough in 1961. It was sleek and relatively affordable. Soon, every serious office bought one. That included my school’s.
I only got to use the Selectric once; it was locked away in a special room. My vague recollection is that it was much nicer to type on than either my 1920s manual or the ASR-33. ⚠️ It took much less force to press the keys, and there was no mechanical connection to the print head, so there was no whack of equal and opposite reaction. Unlike mechanical typewriters, the Selectric wouldn’t jam if you typed too fast. It must have been a godsend for typists.
I wasn’t much interested, though. Computers were the future, not typewriters.
VT-05: I have seen the future
I attended the high school my parents taught at. During that time, the school bought a DEC VT05 video terminal. This was the future! Its looks were straight out of Star Trek, and it could put 30 characters per second on its cathode ray tube screen. The ASR-33 ran at just 10 characters per second, which made getting a program listing super tedious. And the VT05 was silent! And, there was zero whack when you pressed a key. Typing was smooth and fast and easy. Compared with the ASR-33, anyway!
The VT05 and its descendants soon replaced Teletypes as the standard computer “terminals.” During the ’70s and ’80s, I used many historically important ones, such as the DEC VT52 and VT100. The “Terminal” program on every current computer, which you use to give it textual commands, is actually a VT100 emulator!
029: Heavy iron for my first AI publication
The NSF had a program that sent precocious high school students to university science departments for a summer. I went between my junior and senior year.
The university had an IBM 360 Model 65. The 360 series dominated the mainframes of the 1960s and ’70s. Their innovative architecture was highly influential during that era. However, it was a dead end. Almost all current computers descend, more or less, from the Intel 8008, the first practical microcomputer chip, whose architecture was inspired by minicomputers such as the PDP-8 instead.
The 360 was also, for me, a step back because it wasn’t interactive. It ran in “batch mode,” which already seemed archaic and insane then. You hand-wrote your program on paper, then encoded it as holes punched in cardboard cards rather paper tape. You handed your deck of cards through a window to an “operator,” who put it in a physical queue of “job decks” to run on the computer. As each job finished, the operator fed the next deck to a card reader, and pressed a button that caused the computer to slurp the program into its memory. The “go” button ran the program, and its output went to a line printer. Input and output were enormously faster than an ASR-33, so the multi-million-dollar mainframe could be doing something more important, actually computing, full time.
But you would have to come back, in an hour or so, to a different window, and an operator would hand you a stack of paper along with your card deck. Usually your program would have “ABENDed,” IBM-speak for “got an error.” Then the whole contents of the mainframe’s memory would be dumped to the line printer. Interpreting that was a black art. Altogether, debugging was ludicrously slow; but in those days computer time cost a whole lot more than programmer time.
So this was ridiculous, but, the 360 had a compiler for SNOBOL4, a programming language I’d been reading about for years and salivating over. It was way ahead of its time, and incredibly powerful—but not available on the PDP-8/e. Its most striking feature was first-class pattern objects, with BNF notation that gave them the power of context-free grammars. You can think of SNOBOL4 as vaguely similar to perl, but its syntax was much cleaner, and context-free grammars are much more powerful than perl’s regular expressions.
So SNOBOL4 was ideal for sophisticated text processing… such as AI language generation. Which is how I spent the summer. I got my first AI publication from that, in some lost and forgotten volume of proceedings. (This was also around the time I invented the iPhone. ⚠️ I have located my 1977 article about that, and you can read it here.)
Anyway, to do AI, I had to punch holes in cardboard, and the cutting-edge technology for that was the IBM 029 Card Punch. Mainframes at the time were referred to as “heavy iron,” and the electromechanical 029 fit that description too. It was a monstrous chunk of metal, the size of a large desk. There’s a nice description of how it worked, and why, and what it was like using one, here. You can see one in operation in this video:
I honestly can’t remember what it felt like to type on one. From the video, it seems more stable than the ASR-33, by using five times as much steel. But the punching mechanism probably still jerked the whole thing pretty hard each time you pressed a key. And I would guess that sensation dominated whatever the key press itself felt like.
PLATO: I mistake the future for a toy
The university also had a few PLATO terminals.
PLATO was, in retrospect, utterly astonishing, decades ahead of its time. It should have astonished me then too. I completely missed the boat.
PLATO terminals had flat panel displays, starting in 1964. Those were nearly unknown elsewhere until the late 1990s! And they were touch screens, also nearly unknown until the late 1990s! And the terminals had high-resolution graphics, and sound generators. The PLATO software ran on a network of mainframes, and provided the first forums, message boards, email, chat rooms, instant messaging, remote screen sharing, and multiplayer video games.
Uh, yes, multiplayer video games. So, PLATO’s intended purpose was computer-aided instruction. It would teach you illustrated, animated, interactive textbook lessons, and then give you problems to solve, and sorta-AI scoring and feedback on your answers.
This seems like an obviously good idea! But for some reason, no one has ever been able to make it work. And, indeed, no one used the PLATO terminals I saw for this purpose. As an educational experiment, PLATO was a costly, decades-long failure.
But, uh, yes, video games. It was awesome for that, and many of the fundamental game genres were first developed on it. There was a bit of a hitch, which was that using PLATO for games was strictly forbidden. The hardware was extremely expensive, and altruistically funded by NSF and CDC to support idealistic educational goals. It definitely should not be used for dungeon crawls or spaceship battles! So those had to be officially covert—although they were immediately demonstrated to me when I arrived.
And I thought “this is a silly toy.” I wanted to do AI; so I went and punched holes in pieces of cardboard instead.
And so I missed the future. Famously, the Xerox PARC Alto, the first “modern” computer, took inspiration from Doug Engelbart’s “Mother Of All Demos,” the NLS system, which showed the first mouse and window system. I learned only when researching this post that the Alto team was also extensively influenced by PLATO. Famously, Apple copied many of the Alto’s innovations for the Lisa, which was the first mass-market computer with a graphical user interface. The Lisa was the precursor to the Mac. (I’m writing this on one of those. I’ve gotten to the future now.)
This post is supposed to be about keyboards. I don’t remember anything about the PLATO keyboard, though. It seems to be the only unremarkable aspect of the terminal!
Knight keyboard, and the world’s first megabyte
Naturally, as soon as I got to MIT, I worked my way into the AI lab. For a random undergraduate, the most accessible computer there was its DEC PDP-10 mainframe. This machine looms large in hacker folklore; for example, the original version of emacs was written on and for it.
The stock PDP-10 hardware was extensively reworked by the lab. One striking feature: it had the world’s first megabyte of RAM. When it was new, a megabyte of RAM was an inconceivably large amount. The AI lab insisted that, to approach human intelligence, an entire megabyte might be needed. RAM, at that time, was made from tiny magnetic donuts, stitched together with tiny wires by hand. A megabyte is eight million bits,1 so making a megabyte would entail an army of seamstresses working for a year. The Department of Defense was eventually convinced that this was somehow reasonable, and allocated a million dollars for the project. (The equivalent of many million now.) By the time I arrived, it had been discovered that not even a megabyte was enough for human-level AI; so the lab had constructed a second megabyte out of Intel’s new breakthrough 1-kilobit semiconductor memory chips.
Nowadays, an iPhone shopping list app takes at least a gigabyte—a thousand times more—and we’re still not quite at human-level AI. But, in the 1970s, two megabytes was enough to support twenty AI researchers working on the PDP-10 simultaneously.
The other striking PDP-10 modification was the TK TV terminal system. It was invented and built by the legendary hacker Tom Knight (“TK”) in 1972. At that time, and for another couple decades, nearly all computer terminals were either paper printers like the ASR-33 or “glass teletypes” like the VT05. There were also special graphics displays, which could show pictures, but those were very expensive and not routinely used. The TK TV terminals were bit-mapped, and so could display images, text, or both. Similar in effect to the PLATO terminals, but higher resolution, and the underlying technologies were quite different.
The TK TV terminals used the custom-built Knight keyboard. This is first keyboard I encountered that has legendary status among keyboard connoisseurs. (Yes, there are keyboard connoisseurs.) ⚠️
The Knight keyboard felt good. It was a pleasure to type on. Fast and smooth.
It had a variety of innovative features that were eventually widely copied, and now found in essentially all keyboards.
The ASR-33 had a CTRL (control) key, and whatever you type on now does too—copying the Teletype. On those, you held CTRL down and typed a letter, and that caused the electromechanical hardware to do something other than print. For example, control-G rang a metal bell. Software could also detect these “control characters,” and use them to do something else. For example, control-C was used on the PDP-8 to mean “abort the currently running program”; and that meaning has continued to the present in many systems.
The Knight keyboard added a META key, next to CTRL. That tripled the set of possible commands: control-C, meta-C, and control-meta-C if you held both down at the same time. Emacs, for example, makes extensive use of these.
The Knight keyboard’s layout was modeled on the Stanford AI Lab’s 1970 SAIL keyboard, which I never used.
I believe the SAIL keyboard was the first with a meta key. The “option” key on Macs, and the “Alt” key on Windows machines, descends from that.
The SAIL keyboard also had a TOP key, which was like SHIFT only more so. It let you access the weird math symbols that you can see above the letters. I would guess that these were used, or were intended to be used, in some logic programming language designed by John McCarthy. He was the head of the Stanford lab, and the foremost advocate of the logicist approach to AI. The Knight keyboard also had a TOP key that could access a similar set of symbols; but as far as I can recall, none of them were ever used for anything at MIT.
The Knight keyboard is famous among keyboard geeks, but the SAIL one seems forgotten. I don’t know why. Maybe using it felt bad? TK said that “We replaced their key switch with the Microswitch hall-effect high reliability switch.” Perhaps those felt better, as well as being more reliable!
Also, the Knight keyboard was the progenitor of the Lisp Machine keyboards…
Space Cadet: the future arrives
The PDP-10 was vintage 1966, and by the mid-’70s, the AI lab had outgrown it, even with all the add-on power-ups. And by Moore’s Law, many times more power was available at the same cost. It was time for something new. The Lab had a close relationship with DEC, but they didn’t yet offer a suitable alternative.
The Xerox PARC Alto—remember the Alto?—was inspiring. In addition to its pioneering graphical user interface, it was a personal computer. Unlike the batch-mode IBM machines, and the time-shared DEC ones, an individual researcher would get full-time use of a whole high-end machine. You couldn’t buy Altos yet, and also they were underpowered for the AI Lab’s ambitions. So, Tom Knight and Richard Greenblatt were charged with designing and building a new type of personal computer specialized for the Lab’s needs.
AI research then was all done in Lisp, a programming language invented by John McCarthy while he was still at MIT. Conventional computer hardware, optimized for FORTRAN and Algol, made Lisp run slowly; but this was not an intrinsic fault of the language, just a hardware/software mismatch. Knight and Greenblatt designed the new machine with hardware optimized for Lisp instead, which made it run much faster.
The MIT Lisp Machine was my second great love, computer-wise. The Lisp-based development environment was decades ahead of its time, and as far as I know, some of its features are still unmatched. It was also extraordinarily powerful, comparable to a PDP-10, but for a single user rather than twenty. I did my undergraduate, Master’s, and PhD research on it. I still miss it.
The Lisp Machine’s keyboard was referred to affectionately—or perhaps derisively—as the Space Cadet. You can see it above.
The Space Cadet was based on the Knight keyboard, but took it to eleven. It was probably the most extreme keyboard, in its chosen dimensions, ever made. It just had more of everything. In addition to the CONTROL and META keys, it added two more: SUPER and HYPER. That meant the Lisp Machine’s version of emacs could have sixteen different commands per printing character, with different combinations of these four “bucky bit” keys held down. In addition to the TOP key, it had FRONT and GREEK keys, with a corresponding slew of peculiar new math-y symbols. Plus a bunch of other strange keys you can find if you peer at the photo for a while.
Most of that stuff was added with a “maybe we’ll find a use for it someday” excuse, but probably it was just fun to take everything over the top. In fact, most of the new weird keys were rarely or never used in practice.
Also, adding keys made the whole thing bigger, which was cool, but turned out to be bad. ⚠️ Reaching the outer keys meant moving your hand a lot. HYPER was especially a problem; whereas you could reach CONTROL and META by stretching your outer fingers, HYPER needed a second hand.
Otherwise, the Space Cadet felt great. It was a step up from the Knight keyboard. Softer and yet more precise.
Symbolics: the Space Cadet refined
The AI Lab licensed the Lisp Machine design to several spin-out companies, of which Symbolics was the most innovative and successful. Initially Symbolics sold MIT’s keyboard, with a brand badge added, as on the photo in the previous section.
Soon they refined the layout by removing the fun but useless keys, Greek letters, and math symbols. They also made the HYPER, SUPER, and META keys narrower, so you could reach them without straining so much. And, the key feel was once again improved—I don’t know how.
This was the best conventional keyboard I have ever used. I miss it too.
For the next twenty years, I never used anything as good again.
But… then I found ones that were better.
The second half of this post, for paying subscribers only, covers 1991 to the present. The keyboards it discusses are radically more distinctive than those we’ve seen so far!
It discusses the third computer I fell in love with; surprising but forgotten innovations in mobile device keyboards; the keyboard enthusiast subculture; and the recent development of increasingly sophisticated keyboards, up to the currently state-of-the-art Glove80—the one shown at the top of this post.