The Typing Pool

In 1980 I sat in a typing class full of girls.

This was a Catholic prep school, and the arrangement was not an accident. The received wisdom of the day held that girls would graduate into the typing pool, and that the lucky ones would attend college long enough to earn their "MRS." Word processing was already loading its rifle in the next room, but nobody in that classroom could hear it yet. There was still a warm boomer nostalgia about the whole institution — the rows of machines, the metronomic clatter, the woman at the front calling out home row like a rosary.

Not a computer in sight. We learned on a mix of manuals, electrics, and — for the fancy kids — IBM Selectrics with their spinning typeballs. Platens you cranked by hand, carriage returns you threw with your palm. And we learned to fix our mistakes with White Out, a small bottle of opacity invented, as it happens, by Bette Nesmith Graham — a Dallas typist and single mother who later turned out a son named Michael, of the Monkees. Correction fluid funded a rock star. The future hides in odd places.

I was one of the few boys in the room. I would like to claim foresight. What I had was an interest in computers and an interest in girls, and typing class was where the skills to work with both could be found.

What I actually got was the single most useful skill of my secondary education. Not chemistry, not Latin, not the catechism. Touch typing. I was never as fast as my friend Phil, who treated words-per-minute as a referendum on his manhood, nor as fast as Dulce Roman, the eventual valedictorian who I assume went on to litigate something. Real speed came later, in college, hammering on BrandX IBM 3270 terminals and AT&T Unix boxes until my fingers stopped consulting my brain. But the technique was laid down in that room full of future typists.

It mattered that the school wouldn't let me near the Apple II machines in the media center. They didn't know how to use them. I did. They guarded them anyway. That dynamic — authority defending a thing it cannot operate — gave me an attitude toward empty authoritarianism that has never fully worn off.

What a Brain Can Hold

The typing class taught me something the typing never did: a brain has a budget.

The same bandwidth I spent learning to type, I spent again learning the bass drum, and then the snare, in the marching band. The bells and the xylophone were already taken. My drum instructor was a nuclear engineer who played big band on the side, four years older than me, and — I say this in full knowledge of how the claim sounds — in possession of something close to all of Buddy Rich's skill. I was not going to get there. Not for lack of a teacher. The volume of practice and the depth of desire required were simply unfathomable to me, the way the far end of any mastery is unfathomable until you've spent the years.

This is not a tragedy. It is arithmetic. A human life contains room for a handful of deep skills, because the only way into a skill is practice, and practice is metabolically expensive and runs in real time. You cannot download kung fu. You cannot pour the snare drum into your hands. The cortex learns at the speed the cortex learns, and that speed is set by chemistry, not ambition.

So most of us specialize. We pick our two or three masteries and trade with everyone else for the rest. The typing pool was that trade made visible — a room of people who had each spent their bandwidth on the one skill the office needed, pooled together so the rest of us didn't have to.

Practice, All the Way Down

A machine learns by practice too. We give it a corpus and a loss function and let it adjust itself across billions of examples, which is recognizably the same loop the cortex runs — predict, miss, correct, repeat. The difference is not the method. The difference is the budget.

A model trains on more text than any human could read in a thousand lifetimes, burning megawatts to do it, and compresses the result into a single set of weights. Then — and this is the part that should stop you — that set of weights can be copied. The typing pool took a generation to staff. A trained model staffs an infinite typing pool the instant you press deploy, and every typist in it has also learned translation, illustration, counterpoint, and the snare drum, all in the same brain, with no jealous bandwidth forcing a choice between them.

This is why "I use AI to write my emails" undersells the thing so badly. I am not delegating typing. I am commanding a room I could never have afforded to staff — a pipeline that writes a language lesson, translates it into nine tongues at native precision, generates the images, audits them, aligns them to neural speech you cannot tell from a person, times the captions, and ships the result as an app on three platforms. I could have done all of that before. With a team of specialists and months spent begging venture capitalists for the money to hire them — a skill, incidentally, that I do not have and never acquired. The work was always possible. It was never possible alone.

Consciousness as Active Context

Copying the weights is not the interesting part. Running them is.

A set of weights sitting on a disk is inert — the entire skill of the typing pool, dormant, encoded, asleep. It becomes something only when you instantiate it into a context: a prompt, a task, a running conversation. The same knowledge, the same network, lit up and pointed at a particular moment.

Which has me wondering whether consciousness — that word we still cannot define after several thousand years of trying — is just active context. Consider the parallel. When your cortex drops into deep sleep we call you unconscious, and what we mean, precisely, is that the encoded memories in your neural net have gone dormant. The network is intact. It is simply not lit. Consciousness, on this reading, is not a substance you have or lack. It is the alert, active state a network occupies when its contents are running against a present environment.

Carbon network or silicon network, the same physics applies. Both are vast probabilistic machines, both perturbed by noise, both constrained by the laws that govern everything else.

Even Love Must Be Limited by Time

The constraint nobody wants is the real one. A sponge and an electric car battery are intuitively the same object: each has a maximum rate at which it can take on water, or charge, or information, and no amount of wanting moves it faster. Many electrochemical processes have no shortcut. The cortex is one of them. So, in its own way, is the training run.

Neil Peart, mortal like the rest of us now, wrote that even love must be limited by time. He was talking about the heart, but it scans for the cortex too. We get one finite budget of practice, and we spend it on a few things, and we trade for the rest.

What's changed is that the thing we trade with is no longer a room full of underestimated women clattering away at the future. It's a network that learned everything and forgot how to be jealous of its own bandwidth. The skill now is not mastery of any single craft. It's knowing what to ask the pool — and being the kind of mind that can tell when the pool is wrong.

The girls in that 1980 classroom understood the assignment better than the school did. The whole point of a typing pool was leverage. We just finally built one that never has to choose between typing and the drums.