1. Just the Shape
All A are B, all B are C, therefore all A are C.
Nothing in those three sentences is concrete. No one has said what A or B is. The conclusion still follows, without room for doubt.
To reach it you need not know anything about the world. All you need is the shape of the first two sentences.
People noticed this early. For over two thousand years they sorted out the shapes of arguments like these and set them down in fixed forms, called syllogisms. They went through the possible arrangements one by one, listing which yielded a conclusion and which did not.
Cataloging them was one thing. Drawing the conclusion from the form was still, by and large, somebody's work. The sentences had to be read, compared, and answered with a conclusion.
Arithmetic had long had a procedure of its own. Move the numbers around by fixed rules. You need not stop at every step to think why it works.
In the middle of the nineteenth century someone began to ask in earnest whether reasoning could be done that way too. Move the marks around, and the conclusion comes out by itself.
Eventually someone put it in a wooden cabinet with twenty-one keys.
In 1854 a book was printed in London. Its title page read: An Investigation of the Laws of Thought, on Which Are Founded the Mathematical Theories of Logic and Probabilities.
The grammar of that title matters. The object of investigation is the laws of thought. The mathematical theories of logic and probabilities are not the objects under investigation; they are said to be founded on those laws. The author did not mean to build a mathematical system. He meant to find the rules thought follows, then show that mathematics grew out of them.
Put another way, mathematics is not what the book sets out to build. Mathematics is what it brings forward as evidence. What it puts on the table as an object of study is the rules people follow when they think something through.
The author was George Boole, professor of mathematics at Queen's College, Cork.
At the start of Chapter I he says that the aim of the treatise is to investigate the fundamental laws of the operations of the mind by which reasoning is performed. He goes on to say he will express these laws in the symbolic language of a calculus and use them as the foundation for a method.
One sentence takes in a whole territory. Reasoning is an operation of the mind. Its operations have laws. The laws can be written as symbols. There is no gap left between the three steps. If all three hold, then getting the calculus to work is getting the mind's reasoning to work. The ambition of the book sits in those joins. The first step is a premise, not something the book sets out to prove. If reasoning is not wholly an operation of the mind, or if there are operations of the mind that do not count as reasoning, the joins are not so tight. Both questions would come back in different forms for more than a century. In 1854 they looked less like questions than like common sense.
Seven years earlier, in 1847, he had printed an eighty-two-page pamphlet, The Mathematical Analysis of Logic, Being an Essay Towards a Calculus of Deductive Reasoning. The 1854 book rearranged the whole calculus and widened its scope to probability, logical method, and a more general discussion of the constitution of the intellect. The text ran to 424 pages.
In the same month as the 1847 pamphlet, Augustus De Morgan published Formal Logic: or, The Calculus of Inference, Necessary and Probable. By November he had read Boole's pamphlet and begun drafting a review. Some of their ideas, he said, lay very close together.
De Morgan was already caught up in another dispute. In March 1847 Sir William Hamilton wrote to accuse him of plagiarism. What they were fighting over was who should get credit for quantifying the predicate. They kept it up in letters and in print. The matter was a specific technical change, and they disputed where it came from and who had made it first. The sentences in a syllogism could say “all” or “some”; the change would allow a qualifier like “some” to apply to the latter half of the sentence too, making more shapes of argument writable. De Morgan pointed out in reply that their systems differed. In Formal Logic and in public he took Hamilton's proposal apart point by point. Most later histories of logic do not accept Hamilton's charge.
Boole wrote his 1847 pamphlet in haste under the stimulus of the dispute. The first book in what would come to be called the algebra of logic was dashed off while two other people were arguing over priority. The beginnings of a field are not always on the scale of the importance later assigned to them.
2. Allowing What Cannot Be Interpreted
In 1847 Boole called his signs elective symbols.
Here is what he wanted to deal with. You have a collection of things, all sorts of things. The whole collection is his universe of discourse. A letter stands for a selection. The letter for white says to select the white things from the collection. Multiply two letters and you first pick out the white things, then pick the round things from those. The result is the things that are both white and round.
The symbol thus stands for an act rather than a thing. The same symbol can be read as an act, or, by way of its result, as a class. By 1854, classes had become the usual way the text explained the operations. Both readings remained in the system. They do not always line up.
Zero is the empty class; there is nothing in it. One is the whole universe of discourse, the entire collection. Everything outside a class is written as one minus the class. Things that are not white are all things minus white things. The one here is not a bit in a later digital circuit. It is the universe as a whole, the largest thing in this calculus.
The crucial rule is that multiplying a symbol by itself leaves it unchanged. His defense of the rule comes not from the symbols, but from thought. Select all things of a given class from the universe. Now select from that group by the same criterion. You end up with the same group. Repeating the selection changes nothing. From this act he derived a special law of exponents for the symbols.
The direction of the argument is not from the nature of the symbols to how thought ought to go. It is from an act the mind can perform to a law the symbols must obey. The whole calculus is founded on that direction of travel. With that foundation, once the signs go somewhere the mind cannot follow, this particular defense cannot follow either.
So far each step in the calculation corresponds to something one can do in one's head. From here on the correspondence gives way.
The trouble began when he started using the notation as algebra.
In algebra you can move terms, divide both sides, take an expression apart and rearrange it. Those were the powers he wanted. They would let him reduce a reasoning problem step by step the way one solves an equation. Once those operations were let loose, however, expressions appeared that made no sense as selections. There was no corresponding act for dividing one class by another, or for taking three times a class. Subtraction entered, and division; there were indeterminate symbols, and in intermediate steps there were coefficients and compound expressions that matched no class. De Morgan's draft that November picked out two things right away: the indeterminate symbol and division by a class. He called them difficulties still to be explained. The positions of the two authors are clear. One pushed the system beyond where his interpretation could reach. The other immediately pointed to the place where that happened.
Boole's way of dealing with it appears in Chapter V of the 1854 book. If the symbols had to take a directly interpretable form at every step, he said, the general method would break down. So he let forms appear in intermediate steps that could not be read directly as classes, asking only that the initial conditions and the final result be interpretable. He did not declare the intermediate forms new objects of logic. He made correctness as symbolic algebra one thing and admissibility as a logical interpretation another. Go by the rules of algebra in the middle; at the end, use the conditions for classes to sift out the parts that can be interpreted. The entrance and exit have to match the world. The stretch between them need only match the rules. The method endured. One of its consequences was that nobody needed to understand what something meant halfway through. A system claiming to write down the laws of thought had left a path inside itself on which thought did not need to be present. Nobody forced it to yield the ground. It did so to go farther.
A calculus that took itself to be writing down the laws of thought had sanctioned a passage within itself that thought could not explain. It neither denied the passage nor turned it into a new object of logic. It moved the obligation to explain to either end: the entrance was answerable, the exit was answerable, the middle was exempt.
Here, for the first time, the construction went beyond its own account of itself. Its declared province was thought, but part of the path it actually took lay outside the language of thought. The path was not an exception grudgingly tolerated. The method explicitly allowed it, because without it a general calculus would not be possible. To keep its generality, the system had to give up making sense at every step.
The heading of Chapter III of the same book says it will derive the laws of logical symbols from the laws of the operations of the human mind. In that very chapter he adds that working out and applying those laws need not wait for a decision among metaphysical theories about what the mental faculties actually are.
He named the mind and the laws of thought as his objects of study, but kept his calculus from depending on any particular claim about what a mind is. Later readers often take one half of that and decide on that basis whether or not he was a psychologist. His way into the material was neither a laboratory measure of memory, nor reaction times, nor the frequency of associations. It was how language is used in rigorous reasoning.
A chapter later in the 1854 book deals directly with the constitution of the intellect. It does not rescind the permission given in Chapter V. The book's ambition and its technique never quite meet.
Boole died on December 8, 1864.
3. Tablets and Keys
Boole's work was on paper. Someone was about to make it into something you could touch.
Even before Boole there had been a physical device that could demonstrate reasoning.
The third Earl Stanhope had made a demonstrator, probably around 1805. The one in the Science Museum is square, made of mahogany, paper, and glass, about the size of a thick book. Movable pieces laid out in space show relations between classes. It can also handle the combined probability of two events.
It was not presented as a finished device in Stanhope's lifetime. He died in 1816. Not until 1879 did an article in Mind describe it, more than seventy years after it had been made. It existed. It could be used. Nobody picked up where it left off. In the record, a thing made and not carried further leaves the same mark as a thing nobody could make: a blank. The demonstrator had to wait for someone coming by another road to get near enough to see it again. By the time the article appeared, Jevons's machine had been presented to the Royal Society nine years earlier.
In the meantime William Stanley Jevons had made something else. His first move was not a machine but a set of tablets. The tablets listed terms. The things spoken of in a syllogism, white, round, made of wood, each count as a term. Given four terms and their negations, there are only so many ways the world can be. Jevons put them on separate tablets that could be taken away. Give a premise, and pull out the cases that do not fit it. Once all the premises are in, the answer is what remains on the table. The procedure is deletion. You do not derive the conclusion directly. You take away the impossible cases one by one, and what is left is what you have.
The tablets moved some of the remembering and copying onto the table. Deciding which tablet to remove was still the work of a person's hand. Someone had to look at the premise, tell which cases contradicted it, and reach out to take them away. The tablets saved memory, not judgment. Judgment was what the next step would touch.
On June 23, 1868, he wrote to his brother Herbert that he had started another machine and had got it to work fairly well, with a view to showing it at the Royal Society. In the same letter he described its appearance. It looked like a large concertina, or a very small piano. It had twenty-one keys resembling the white keys on a piano. It could deal in a few minutes with any logical problem involving no more than four distinct terms, he said. He wrote down the four-term limit himself, before the machine was finished.
The finished version dates to 1869 and was built in Manchester. He hired a clockmaker from Salford to make it to his drawings. The clockmaker's name has not come down to us. Jevons did not name him in his acknowledgments.
That machine went on to be exhibited, photographed, and written into dictionaries and histories of computing. Someone made every rod in it to a drawing. Of that person all that is left is an occupation and a place. The drawing was Jevons's. The hands were the clockmaker's. Between the drawing and the machine lay a whole craft of wood, brass, and levers. It was not on the page. Only one of the two makers of the machine presented to the Royal Society left his name.
It was a wooden cabinet about three feet high, closer in shape to a small upright piano. There were twenty-one keys on the front. The display above could show the sixteen elementary combinations of four terms and their negations. The cabinet stood there looking like furniture. Until then, what it did had happened only in people's heads.
Take off the removable cover and you do not find the gear trains of a decimal calculating machine. You find a great many rods, levers, and pulleys linked to the display tablets. Press a key and the mechanism shifts the corresponding combinations. It does not calculate with numbers. It moves things, taking what should not stay out of view and leaving what should stay where it can be seen.
4. Four Terms
Using the machine does not mean typing in an English sentence.
The operator first translates the premises into Jevons's symbolic form, then presses the function keys for subjects, copulas, and predicates. When a premise has gone in, pressing the full-stop key moves the display and withdraws combinations incompatible with the premises entered so far. After the last premise, the display shows every case the conclusion allows. A reset key returns the machine to the start.
The machine is doing the same thing as the tablets. The difference is whose hand is moving. With the tablets a person has to tell which case conflicts with the premise and reach out to remove it. With the piano a person keys in the premise and the levers do the removing. What crossed to the machine's side was the judgment the hand used to make. It is a small step: a few dozen rods perform a deletion in place of a hand. Until then, the act being handed over had counted as reasoning. A thing often has one name before the machine does it and another afterward. The moment the name changes is often the moment the machine takes the work.
The machine has a hard boundary: four terms.
For each of four terms, either it holds or it does not. Two possibilities for each, multiplied together, give sixteen. With respect to these four things the world can take only sixteen forms. The positions in the display are made for those sixteen. Enter a premise and some combinations cease to be live. Once all the premises are in, the ones still standing are the entire answer. What the machine gives you is not a sentence but a list: these are the ways the world can be, given the premises. To read a sentence out of that list in ordinary language, somebody still has one more step to take.
Beyond four terms, the machine does not merely get slower. It has no spare display positions or mechanical columns to represent another term. To add one you need a larger piece of hardware. No amount of pressing the keys will put it inside this cabinet. Jevons considered extending the design to around sixteen terms. It would have taken up close to a whole wall, by his reckoning, so he did not build it.
This is a limit different from paper. To put another term on paper, write a few more lines. To put another term into this machine, build it again. Its powers did not grow with use. They were fixed in the wood the day the drawings were finished.
He stated another boundary himself. The machine cannot look at a sentence in ordinary language and decide what is the subject, what is the predicate, where there is a negation, or which word is the copula. That work of grammar and meaning stays with the operator. If the person gets the structure wrong, the machine will dutifully process a bad encoding. It has no way to ask whether what it is doing corresponds to what the sentence means. It attends to whether the form is right, not whether it is the right form for those words. That part has not been handed over. The machine takes the middle, the formal part. Turning words into form and then reading words back out of it both remain on the human side. However correct the calculation, the machine cannot tell whether it is calculating the thing that was meant.
“On the Mechanical Performance of Logical Inference” was read to the Royal Society on January 20, 1870. The full paper appeared in the Philosophical Transactions.
In private letters Jevons worried that the machine might just as easily draw ridicule as praise. He put the two responses side by side. He did not know which he would get on the day. What would people make of a machine doing in public something that only a person had done before? There was no precedent to go by.
5. For the Most Part
The title of the paper already makes a claim: the mechanical performance of logical inference. Mechanical performance says that the thing can be done by motion, without stopping to think. To describe the machine's movements Jevons uses terms for carrying out, inferring, and deducing, alongside terms for comparing, indicating, answering, and working. One group comes from reasoning; the other from mechanics. He uses both in the same paper, without choosing a side. The wavering is not vagueness. There was not yet a set of words equal to describing what was happening.
His strongest statement is that a mechanism can “for the most part” replace the mental action needed to carry out logical deductions.
The qualification is in the original: for the most part.
The words are often dropped. Once they go, a carefully bounded technical statement becomes a general verdict on machines and thought. Dropping a qualification and adding one send the meaning in opposite directions, but the maneuver is the same: a statement about things as they are becomes a statement about all time. On the very same page the man who wrote the sentence names the part the machine has not replaced.
In the next passage he leaves two jobs to the person using it: interpret the grammatical structure of the premises correctly and understand what the machine's answer means. The person does not leave the process of reasoning altogether. What becomes mechanical is the deduction after it has been put into formal terms.
The boundary in his account is clear. A person first puts a piece of reasoning into formal terms. The machine gets the formal part. A person then reads the meaning back out of what the machine produces. The machine takes the middle. The two ends are still in human hands. This division keeps recurring, each time with the middle spreading a little and the two ends drawing in. What can go in the middle depends on the technology of the time. How much still belongs at the ends is for the people of the time to say.
In The Principles of Science, published in 1874, the machine is not a curiosity tucked into an appendix. The chapter about it stands right alongside his general theory of reasoning. That chapter does not do away with the two jobs left to the person in 1870. The machine is visible evidence for the generality and power of his method, not a substitute for its operator.
A sentence often attributed to him has the machine doing the work of a rational being. That wording does not appear in the Jevons texts used here. The sentence usually put beside his name comes from the logic entry of a 1911 encyclopedia.
The more a machine is talked about, the more sentences stick to it. Some were never said at the time.
Not everyone who lived then was persuaded. In Symbolic Logic in 1881 John Venn said plainly that he did not think much of the significance of these so-called logical machines. One reason was the narrowness of the part they could take over.
But Venn did not just make fun of them on paper. He made one himself, five or six inches square and three inches deep, unnecessarily large by his own account. It worked more like a three-dimensional diagram assembled from pieces than like the key-driven device with which Jevons automatically eliminated cases.
One line of his objection is this: organizing the premises into a form a machine can accept already takes a great deal of logical work.
It is the same point as the two jobs Jevons left to his operator. The judgment attached to it is different. Jevons reported it plainly as a boundary of the machine. Venn took it as a reason not to make so much of the machine. They looked at the same fact from opposite sides. The work they meant was the same: turning a sentence into a form the machine can recognize. That work has not gone away, though it has changed names several times. For more than a century, who does it and how much counts as doing it have had to be answered over and over.
6. The Mill
In November 1887, the first issue of the American Journal of Psychology carried an article called “Logical Machines,” by Charles Sanders Peirce. An American who had worked on coastal surveying and written on logic, he was one of the few people of his time who had both made instruments and changed notation. The journal was new. It called itself a journal of psychology.
He began with an old satire. In a travel novel written more than 160 years before, the scholars of a floating island make a machine that turns and puts words in random order, hoping it will produce learning. The author meant to laugh at an idle fantasy. Peirce starts with the scene, then turns it around by setting alongside it the actual machines of Jevons and Marquand and the machines that did arithmetic. He compares the devices of Jevons and Marquand to a mill. Put in the premises, turn the handle, and out comes the conclusion. There is no contempt in the comparison. A mill really does turn wheat into flour, and nobody therefore says the mill does the farming. The comparison gives him the boundary he wants: the flour comes from the mill; the wheat does not. The mill cannot decide what it ought to grind today, still less go out and reap it. In the same passage he includes Babbage's Analytical Engine, saying that its reasoning is not of the simple kind. The engine was still unbuilt, but it was already on a list of machines that could reason.
The name in the article is Babbage's. Lovelace does not appear.
At its center is what Peirce calls the secret of all reasoning machines. In physical terms, if a relation among the objects involved supplies the hinge of an inference, then the same general relation must actually be possible among some of the parts inside the machine. What matters is the correspondence in structure, not whether the machine has subjective experience. He moves the question from what is inside the machine to how its parts relate. After that shift, whether it is reasoning is a question you can open the cabinet and look into.
He then gives two inherent incapacities of the machine. His word is inherent. The trouble is not that this one was badly made. These are two things that this kind of thing does not do in the course of what it is for.
The heart of the first is that it has no originality or initiative. It cannot find its own problems or feed itself. Nor, he says, can it decide for itself which of several possible procedures to take.
This is the second time in the series that the remainder has been named, again by someone inside the construction. Its name is longer this time. It cannot originate, cannot find its own question, cannot supply its own material, cannot choose its own path. Forty-four years earlier the sentence had said the machine could not originate. This one says it cannot initiate. The former draws its line at the point of production, the latter at the point of setting out. Neither says the machine is not fast enough. Neither derives the limit from the make of the machine. Both find it in the relation of the machine and the person: whose work does it do, and who gives it work? The sentences of 1843 and 1887 make the same point. A person stands on this side of the boundary.
His next sentence often goes unmentioned. In a machine used as an instrument, he says, this need not count as a defect. The person means it to deal with that person's own concerns.
That sentence changes the character of the passage. What preceded it was not an indictment of the machine but an account of what a tool is. Not looking for work on its own is a condition of being a tool. Defect and proper function are two readings of the same words.
The second incapacity is that every particular machine has an absolute limit on its capacity. A machine is made for a particular kind of thing; it cannot extend beyond that particular construction. His example was that the logical machines then available could handle only a finite number of letters.
He sets the machines then in use beside a mind equipped with pencil and paper. In the machines the number of variables is fixed in the hardware. In algebra on paper one can go on extending the notation. He singles out parentheses, because you can put one pair inside another and let the structure grow indefinitely.
He did not work a claim about all time out of the two limitations. At the start of the paper he still holds open as a question for research how much of the work of thought a machine can perform. What he gives is an absolute limit of this machine, not an absolute limit on all machines. Those claims are far apart. Later quotations often use the first as if it were the second.
Eleven months before the paper, on December 30, 1886, Peirce had written to Allan Marquand at Princeton. Marquand had made a mechanical logical machine in 1881–1882. Electricity, Peirce said in his letter, was what he would put most faith in. He drew and explained circuits in series and parallel and showed how different switch arrangements matched operations in his algebra of logic. The following year Marquand drew up a design for an electrical logical machine on that basis.
In 1887 the step went no further than paper. It produced no working electrical machine. Getting an idea down is not the same as having someone take it up. What was missing was an industry that needed it.
7. Pure Thought
In 1879 a publisher in Halle, Germany printed a small book of eighty-eight pages, by Gottlob Frege of the University of Jena.
The title was Begriffsschrift: A Formula Language, Modeled on Arithmetic, for Pure Thought.
The title also contains “thought.” But this is not The Laws of Thought. The grammatical relation in Frege's title means that the formula language belongs to pure thought and is for its use. It does not mean the book sets out to investigate the rules thought follows. Twenty-five years before, the object was the working of the mind. Here it is a way of writing. One man went looking for laws; the other went about making a tool. One placed his object on the side of the mind; the other placed his on the page. The same word appears in two titles, but its position is quite different. In one it is what is studied. In the other it is what the work serves.
The small book adds two things.
One is the quantifier.
A sentence in a traditional syllogism takes one “all” or one “some.” Put in two and things get tangled.
Everyone loves someone.
There is someone whom everyone loves.
The same words say quite different things. In the first sentence, people may love different people. In the second, there is one person everyone loves. The difference is not in the words. It is in which has scope over which. Ordinary language uses word order to show it, and word order is often ambiguous.
Frege's notation gives the distinction a formal shape. Each “all” and “some” has a definite scope. One can be nested inside another, and on the page one glance shows which contains which. The two sentences above have two different shapes in his notation. They cannot be confused.
The other thing is a different way to take a sentence apart.
The earlier division followed grammar. Split the sentence into a subject and a predicate: who is it about, and what does it say about that person? This works for Socrates is human, but not so well for A is taller than B. Is the sentence saying something about A or about B? It is hard to say.
Frege took it apart differently. Lift out some of the names, leaving gaps. The frame with its gaps is one thing, the names you put in them another. So-and-so is taller than so-and-so is a frame with two spaces. What the sentence is about now is a relation between A and B, not one of the two alone.
These two devices can express relations that Boole's algebra of classes cannot reach.
The person who built them detached them from the workings of thought.
In 1884, in the preface to The Foundations of Arithmetic, Frege put his first principle firmly: separate the psychological from the logical, the subjective from the objective, and keep them sharply apart.
He was opposing a fairly common view that the laws of logic were really the laws of how our minds actually work. On that view, to understand logic you had to investigate how people think.
That was just what the title of the book thirty years earlier said: investigate the laws of thought, on which logic and probability are founded.
Frege wanted to cut that connection. For him, two plus two being four had nothing to do with how human brains worked. If every person in the world began getting it wrong tomorrow, the answer would still be four. The province of logic is not how people actually think, but what follows from what. The preface to the first volume of Basic Laws of Arithmetic in 1893 went on to distinguish the laws of logic from psychological laws about what people believe or associate, or how an idea occurs to them.
So the line receded twice in forty years.
It receded first because a machine was made. A step in deduction usually done by a person was performed publicly by a wooden cabinet and its levers. That step on its own no longer told you whether the thing doing it was human. The second retreat was not because of a machine. A builder no longer wanted to claim that ground. One of its makers deliberately detached the formal apparatus from thought.
The two retreats went in different directions. The first took away a piece of ground. A machine did part of the deduction, and the mind was left with the rest. In the second, the field announced that the ground was not its business at all. The laws of logic are not the laws of how people think and should not be discussed as if they were.
Begriffsschrift did not go unremarked. Six journal reviews appeared in 1879 and 1880, by reviewers in Germany, Britain, and France.
Two are worth looking at. Ernst Schröder's long review ran in a journal of mathematics and physics in 1880. From the standpoint of the algebra of logic as it had developed, he criticized Frege for not taking sufficient account of Boole's line of work. Venn's review ran in Mind in 1880. He recognized a quite distinctive symbolic logic. But from the standpoint of his own Boolean notation, Frege's two-dimensional writing seemed more like a diagram, a geometrical arrangement. He doubted it would be convenient to use.
Both reviewers measured it with the ruler they had to hand. Nested quantification and function–argument analysis were exactly what that ruler could not measure. What someone trained in the notation of the algebra of logic saw was a more troublesome way to write. Their ruler worked; the measurements it returned were right as far as they went. This little book simply lay beyond its range. To someone using a ruler, what the ruler cannot measure tends to look not new but awkward.
Frege qualified to teach at Jena in 1874 and became an unsalaried Privatdozent. In 1879 he became an extraordinary professor; in 1896 a full professor without an established chair. For twenty-two years his appointments kept him outside an established chair.
In June 1902 Bertrand Russell wrote to him pointing out a contradiction. In an appendix at the end of the second volume of Basic Laws of Arithmetic, published in 1903, Frege wrote that a scientific author could hardly meet with anything less welcome: the work was finished, and one of its foundations had been shaken.
8. An Entry in the Dictionary
James Mark Baldwin's Dictionary of Philosophy and Psychology appeared in 1901–1902 with an entry for “Logical Machine.” The definition used Jevons's device to explain. The user keyed in the premises. The levers raised the compatible combinations automatically.
By the turn of the century these devices had an entry of their own. What they were, how to define them, what example to use, all had been settled. Once a thing makes it into a dictionary, you need not explain it from scratch every time it comes up. The definition followed the movements of the wooden cabinet: a person enters the premises, a lever raises the combinations. That also settled where this definition drew the line. A logical machine was there to delete cases for a person. Who supplied the premises and who read the answer were not part of it. Those were just the two jobs the paper of 1870 expressly left to people. Once there was a definition, what remained moved out of the picture. It had not disappeared. It simply no longer counted as part of what this machine did.
There had already been connections between logical operations and switches in Peirce's letter of 1886 and Marquand's design of 1887. A relay is a switch used in telephone exchanges. A small current pulls it in; when it moves, it completes a second circuit. On or off: there is no third state. In 1937 a master's thesis used the algebra of logic systematically to analyze and design relay and switching networks. It states at the outset that the algebra came from George Boole. The idea in the 1886 letter stayed on the page. The 1937 thesis met an industry spread across the world. By then there were millions of switches like these in the world's telephone exchanges.
Jevons's wooden machine appeared again in October 1952, at an exhibition about him in Manchester. The newspaper photograph shows it next to an electronic logic device. A wooden one and an electronic one stand side by side, eighty-three years apart. Both hand a part of logical work to an apparatus, one with levers, the other with circuits. The exhibition put them next to each other. It did not say which did something more like reasoning. That question was asked in 1870, written down again in 1887, and by 1952 it still had no answer one could set in the display case. An old man in the photograph is Jevons's son, Herbert Stanley Jevons. It was through his efforts that his father's machine came out in public again. Today it is in Oxford's History of Science Museum.
Between 1847 and 1887, the work of rewriting reasoning as calculation took at least four visible forms. In his algebra on paper Boole allowed intermediate expressions that could not be interpreted. Jevons gave the possible cases the form of removable tablets and a moving display. Marquand used rods and springs for such combinations, then designed a scheme with switches. Frege put quantification and the binding of variables into a two-dimensional notation.
Four people went four ways about the same thing. They had no common account of what the mind was. What they had in common was an act: take reasoning off the person and put some of it in a form you could set on a table. From then on, someone else could examine that part, rewrite it, turn it into a physical device, or sell it. As to what was left after taking it away, they gave different answers. None stayed where it began.
The machines are still being built. Each generation says again where the boundary lies.