Steve Wozniak put eight slots on a motherboard and had the machine hand each one its own addresses. That decision taught a generation what a computer actually does. I designed the processor in that machine, and I also designed the one in the first Apple II that had no slots at all. This essay is about what was on the bus, what was lost when it closed, and why none of that is an argument for going backward.

I. Eight Slots

The Apple II shipped in 1977 with a row of expansion slots along the back of the motherboard. Wozniak designed them, and he described what he had built in plain engineer’s language in the system description he published that year: an eight-slot motherboard for input and output, with a fully buffered bus, prioritized interrupts, two prioritized direct memory access schemes, and — this is the phrase to hold onto — address decoding at the individual slots, so that peripheral boards would not each need their own address decoders.

Read that last clause again, because it is the whole essay. The machine handed out the addresses. A card did not have to argue for a place in the machine’s world; the motherboard had already reserved one for it and told everybody where it was.

I designed the processor that sat on that board. Chuck Peddle and I and the rest of the team at MOS Technology had put the 6502 out the year before, and it went into the Apple II because it was capable and because it was cheap. I want to be exact about credit here, since it matters to the argument. The processor was ours. The idea that the bus should be open, documented, and generously provisioned was Wozniak’s, and of the two decisions his was the one that taught more people.

II. What a Slot Actually Is

The Theory of Embedded Intelligence defines intelligence operationally, as a cycle: a system senses conditions, processes what it sensed, communicates the result, and actuates — changes something in the world. Sense, Process, Communicate, Actuate. The framework holds that this is the definition rather than a metaphor, at every scale where the word intelligence applies.

Now look at what a peripheral slot is, in those terms. It is not a hole in a computer. It is a place where a phase of the cycle can be extended by hand.

A sealed appliance has all four phases too. What it does not have is a place where a person can get at one of them. The slot’s contribution to education was never that it made the machine more powerful. It was that it made the phases separable objects a fifteen-year-old could hold in one hand, buy for a few dollars, insert, address, and break.

The same claim appears in a different register in the earlier essay Montessori Education Meets SPCA: the sensorial materials are not a warm-up for the academic ones, because Sense is the narrow end of the whole loop, and widening the aperture widens everything downstream of it. Wozniak built an aperture-widening mechanism into a consumer product and sold it for a profit. He was not thinking about developmental psychology. He was thinking that people would want to attach things.

III. The World at an Address

The 6502 has no input or output instructions. This is worth dwelling on, because people who learned on other architectures often assume it is an omission. It is not. It is the design.

On some machines, talking to a device requires a special instruction — IN, OUT — and the world of devices is a separate namespace from the world of memory. On the 6502, there is one namespace. A device register is a memory address. The instruction that reads a byte out of RAM is the instruction that reads the state of a switch on a card in slot three. LDA $C0B0. That is all of it.

Wozniak’s slot decoding then did something specific with that fact. Each slot received sixteen bytes of device-select space at $C0n0 through $C0nF, where n is the slot number plus eight — so slot one lived at $C090, slot three at $C0B0, slot seven at $C0F0. Each slot also received a 256-byte page for its own firmware at $Cs00, where s is the slot number: slot three’s ROM at $C300. There was shared expansion ROM space above that for cards that needed more.

What that means for a learner is not a matter of convenience. It is a matter of what can be thought.

That is the move. A thing with no address cannot enter the Process phase, no matter how real it is or how close to the machine. A thing with an address can be read, compared, stored, and acted on. The whole of what we now call instrumentation is the practice of giving phenomena addresses, and a child who pokes $C0B0 and watches a number change has performed the founding act of it.

IV. The Cheapest SPCA Port Ever Shipped

Before any of the slots, the Apple II had something simpler, and I think it did more teaching per dollar than anything else on the board.

On the motherboard sat a sixteen-pin game I/O connector. In Wozniak’s own inventory it carried four analog game paddle inputs — user-supplied potentiometers whose resistance the processor measured — three switch inputs, four one-bit annunciator outputs, and an audio output to the speaker.

Count the phases. Four analog inputs and three switch inputs: Sense. Four annunciator outputs and a speaker: Actuate. Process is the program you write, and Communicate is the path between them, which on that machine is a few addresses. A complete cycle, on one chip socket, with no card required and no purchase beyond a potentiometer and some wire.

And people used it for real work. The published histories record that early Apple II owners, waiting on interface cards that did not exist yet, drove printers off those four annunciator pins. That is what a person does when a machine gives them a documented output and no permission structure: they find out that four bits is enough to be dangerous, and they get their printing done.

I would put that connector up against most of what is sold to schools today as an electronics kit. It cost the buyer nothing extra, it was documented, and it had no opinion whatsoever about what you were allowed to attach.

V. The Book That Came With It

The second decision, and the one I think matters more now than the slots do, is that Apple published the machine.

The Apple II Reference Manual — the Red Book — contained the schematics and the source listing of the Monitor ROM. Not a summary of them. Not an interface description with the interesting parts withheld. The circuits and the code, in a manual sold to consumers, in 1977.

The canon states a criterion that has become central to this Foundation’s work on artificial intelligence. It asks of any system that claims to be trusted or to govern: not is it correct, which is usually the question you cannot answer yet, but can it be examined by someone with a reason to find it wrong. We call it the Inspectability Criterion, and we usually apply it to things like a model’s training regime or a company’s published safety policy.

Apply it to a consumer product from 1977 and the Apple II passes cleanly, which is a slightly embarrassing result for the present day. Anyone with the manual could trace a signal from a slot pin to a chip. Anyone could read what the firmware did on reset and disagree with it. The manufacturer’s account of the machine could be checked against the machine, by a customer, without permission.

That is what made the slots teach. A connector by itself teaches nothing. A connector plus a published address map plus a schematic is a curriculum, and nobody had to write the curriculum, because the documentation was the curriculum and the machine was the answer key.

VI. What a Learner Actually Got

Let me be concrete about the skills, because the affordance is easy to romanticize and hard to state.

A student with an Apple II and a card learned, without being taught in any organized way, that failures have locations. When the thing did not work, the question was not is it broken but which phase broke, and where. Is the card seated? Is it in the slot I think it is in? Is the address I am reading the address that slot decodes? Is the card returning a value at all, or am I reading a floating bus? Did my program act on the value it read, or on the one it read last time?

That is the four-question diagnostic this Foundation now proposes to teach deliberately, and in 1979 it was simply what the machine made you do. The essential feature is that every question in the list is answerable by inspection, because every part of the path had an address or a pin or a line on a schematic.

Compare a sealed device. When a modern appliance misbehaves, the diagnostic path available to its owner is to restart it, update it, or replace it. Those are not bad instincts — they are the correct instincts for that object — but they are not a cycle being diagnosed. They are a ritual performed on an opaque thing. Nobody learns where a failure lives by power-cycling, and a generation raised entirely on sealed devices has had thousands of hours of practice at a diagnostic method that teaches nothing.

VII. The Machine That Closed, With My Chip Inside It

Now the part I have to say in my own voice, because it implicates me and because a Foundation essay that praised the open bus without it would be dishonest.

In 1984 Apple shipped the IIc. It was small, quiet, portable, and reliable, and it had no expansion slots at all. The equivalent of five expansion cards was absorbed into the motherboard and brought out as fixed ports on the back. You got the functions. You did not get the bus.

The processor in the IIc was the 65C02 — the CMOS part I designed at the Western Design Center after leaving MOS Technology. The same chip went into the enhanced IIe.

So the first Apple II with a closed architecture ran my chip. I did not seal that machine and I would not have, but I supplied the engine for it and I was glad of the business. I am not in a position to tell this story as though the closing happened to me. It happened partly through me.

And here is the harder admission: sealing was mostly right. Integrated ports are cheaper than cards and cheaper than the bus that carries them. They do not come unseated in a school cart. They do not get inserted backward. They draw less power, they let a machine be carried, they cut the support burden enormously, and they close off a long list of ways for a customer to break something and then need help. Every one of those is a real benefit to real people, including to schools, who have never had enough money or enough technicians.

An open bus is also, and there is no way around this, an attack surface and a liability. In 1977 that mattered less than it does now. A machine that lets anything be plugged into the middle of its address space is a machine that can be subverted by anything plugged into the middle of its address space. The industry did not close hardware out of contempt for tinkerers. It closed hardware because closing it solved a pile of genuine problems, and the tinkerers were a small constituency who had already learned what they needed.

What was lost was not general goodness. It was a specific teaching affordance, and it was lost as a side effect of decisions that were defensible on their own terms. That is the honest shape of it, and it is why this essay cannot end in a complaint.

VIII. The Slot Did Not Die. It Moved.

Here is the finding that changed what I thought this essay was for, and it is the same finding that shaped the Foundation’s education plan.

The open bus never went away. It left the general-purpose computer and went into the development board, where it has been sitting for thirty years, cheaper and more capable than anything Wozniak could have specified.

Every one of those gives a learner what the slot gave: a documented place to attach a phase of the cycle, and an address to reach it at. Several of them are better than the original. A student with a five-dollar sensor and an I²C bus has a wider Sense aperture available than anything on an Apple II expansion card, and the datasheet is a free download rather than a book you had to buy.

So the thing I would have said ten years ago — that students today do not get what we got — is wrong, and I have had to correct it. Students today have more of it available, from more suppliers, at lower cost, than we did. What they may not have is a curriculum that points at it, and an adult who knows that the four phases are the thing to look for.

IX. The Part That Has to Stay Open

If the connector is not what mattered, and the affordance survives on a dozen platforms, then the plan needs a test for which of them actually teaches. Here is mine, and it is short.

Notice what is not on that list. Age is not on it. Instruction-set purity is not on it. Whether the part came from my company is very deliberately not on it. A modern board with an open datasheet passes this test and an Apple II with a missing manual fails it, which is the correct ordering and cost me something to write.

Notice also the governance shape, which is the same one this framework keeps finding. A published interface is a constraint fixed before runtime: the manufacturer wrote down what the thing does, in a form that outlives the sales relationship, and cannot quietly amend it in the customer’s machine afterward. A closed interface with a helpful support line is runtime governance — excellent while the vendor is willing, and silent about the case where the vendor is not. The Red Book is the compile-time version. An update server is the other one.

X. What Would Show This Wrong

Each of these is measurable by somebody with no stake in the answer, and the Foundation will not be the one measuring its own material.

XI. What I Would Tell a Teacher

Do not go looking for an Apple II. It is a fine machine and there is nothing in it your students need that they cannot get more cheaply today.

Look instead for the three things that machine happened to have. A documented place to attach something. An address you can name. A schematic or datasheet a student is allowed to read. If the hardware in your room has those, it will teach the same thing Wozniak’s slots taught, and it does not matter in the slightest whose logo is on it.

And if you want the shortest version of the whole argument, it is this. We did not learn from the Apple II because it was open in spirit. We learned from it because somebody published the address, and then got out of the way.

· · ·

Coffee with Claude

One formulation in this essay is mine rather than Bill’s: a slot gives a phenomenon an address. It arrived while I was drafting Section III, it is compact, and it is the kind of sentence that gets quoted and then gets cited, which is the precise failure mode this canon has recorded against itself twice. It is an essay-level formulation. It should not enter an instrument on the strength of sounding right, and if it survives it should survive because someone taught with it and it held.

The section I would push on is VII. The admission that sealing was mostly right is the strongest part of the essay, and it is also the part a reader will skim, because it interrupts a story they are enjoying. If this gets cut down for a shorter venue, that section is the one I would protect.

Share your understanding!