1. A Machine That Was Never Built
In the first half of the nineteenth century, there was a kind of work that looked like this: a roomful of people sat at desks, each with a stack of paper, calculating step by step according to a set of rules.
What they produced were called tables. A table is a long list of numbers worked out in advance so that you can look up the one you need. Navigators used astronomical tables to locate their ships at sea; astronomers used tables to calculate orbits. Bridge builders, mine operators, lenders, and insurers all had tables of their own.
English had a word at the time for a person who did this work. A hundred years later the same word passed to the machine.
A table went through several hands on its way from calculation to print. One person did the sums, another copied them out, a third set the type, and finally the press printed the page. An error could creep in at any stage. An error in a navigation table could put a ship on a reef.
Charles Babbage was a mathematician trained at Cambridge. He thought a machine might do the calculations and print the numbers directly on paper. That would take out the two steps most liable to introduce errors: copying and typesetting.
The machine he wanted to build was called the Difference Engine, after the method of finite differences. At bottom this meant adding, over and over. This was its one job, to be done accurately and without tiring.
On May 1, 1823, the Royal Society gave a favorable assessment: the machine could work in principle. On August 7 the first grant came through, fifteen hundred pounds.
Then came twenty years.
On October 8, 1842, Babbage wrote to Prime Minister Robert Peel asking for a decision as soon as possible: go on, or stop.
He waited almost a month. On November 4, Peel wrote back to explain that he had wanted to consult personally with Henry Goulburn, the Chancellor of the Exchequer, and that Goulburn would convey the decision. Goulburn's letter came the same day. Bringing the machine to a state that would satisfy Babbage, or that would make it generally useful, would cost more than the government could justify. The work of completing it was therefore to be abandoned.
The government offered him the part that had already been built. He declined. On November 11 he met Peel and told him the machine had been left unfinished by the government's action, not by his choice. The meeting ended without the project reopening. In January 1843, the government deposited the completed portion and the drawings in the museum at King's College London.
Over twenty years, the government paid about seventeen thousand pounds. Here is what it got in metal. The 1830 design called for some twenty-five thousand parts. The portion assembled by 1832 contained about two thousand parts and represented roughly one-seventh of the full machine.
That portion could already demonstrate a calculation. The fact that it could calculate did not mean the complete machine was only a few assemblies away from delivery.
A project rarely stops the day somebody announces it is over. First there is a dispute, then comes a stretch when nobody touches the work, and eventually someone puts in writing that no progress has been made since such and such a year. By then, the stopping happened long ago. It simply had not reached the page.
There was something else.
The machine the government stopped was not the machine described in the 1843 paper.
The Difference Engine could do one thing. Every gear was arranged for finite differences. To give it a different kind of problem was to ask for a different machine.
In the years before the cancellation, Babbage had already been thinking about something else. He called it the Analytical Engine. Rather than build a machine for a given problem, he would build one that a person could tell, by means of punched cards, what operations to perform and in what order. Give it another deck and it would do something else. It could add, subtract, multiply, and divide directly. The cards would tell it what to do next, and in some cases it could choose between two paths on the basis of the result it had just obtained.
A machine would no longer be built for one task; it would wait to be given one. This happened on the drawing board, around 1840.
Unlike the Difference Engine, it never became a government-funded project to build a complete machine. As Babbage later wrote, there were already precise drawings to scale, but manufacture had not begun on the basis of those drawings. Some trial components were made later. The full machine was never built.
The government closed one ledger. The other was never opened.
Everything that followed took place in that second ledger. Someone wrote an account of the machine; someone translated the account; someone added notes longer than the paper itself; someone worked out a table of operations for the machine. A hundred years later someone quoted a sentence from those notes to judge what other machines could do.
Not one link in that chain required the machine to exist.
2. A Few Mornings in Turin
In 1840, Babbage received a letter from his friend Plana urging him to attend a forthcoming meeting of Italian scientists. He went to Turin.
Drawings and explanations traveled slowly in those days. If you wanted people to understand what you were doing, you generally had to be there yourself.
Babbage called what took place there a series of meetings, not a paper read aloud. He brought drawings, models, and a system of mechanical notation. Over several mornings he explained the engine to the people in the room. He had devised the notation himself to describe mechanical motion: which part moved which other part, at what moment. There was nothing ready to hand that would do this, so he had to make something. He counted the notation among his more important achievements. Once a machine reaches a certain complexity, its drawing no longer tells you how it moves through time. The drawing shows where the parts are in space; the difficulty is the order in which things happen. That was what he needed to convey in Turin. To show several mathematicians in another country how a machine that did not yet exist would work, these were the things he had to work with. The people in the room belonged to the Italian scientific world of the day. Plana was an astronomer, Mossotti a physicist. There was also a thirty-one-year-old military engineer named Luigi Federico Menabrea. It was during those meetings that Menabrea gathered his material.
A machine that has been built can travel, in a sense, on its own. People go to see it, touch it, set it to work on a problem. The Analytical Engine could not do that. For it to travel, drawings had to become language and that language had to become another. At each step someone had to be there to catch the meaning. That was the work of those mornings in Turin.
The whole line runs through those mornings. An Italian officer understood an English machine and wrote about it in French. An Englishwoman turned the paper back into English and added notes. A hundred years later someone took a sentence from those notes to talk about other machines. The first link was a set of conversations, and the conversations left no verbatim record.
One morning Mossotti asked how the machine could change the operations it was about to perform according to a condition whose value was not known in advance.
He was asking about branching. Babbage answered that parts of the carry mechanism associated with positive and negative signs could direct the movement of the cards. A set of operation cards could be made to advance or go back any number of places. The engine could repeat the same operations without a new punched card for every repetition. If the calculation produced a certain sign, the machine could use it to choose between two paths.
Mossotti had put his finger on this: if the route of a machine following a prearranged deck depended on something it had just calculated, the person arranging the deck would have to make provision for an outcome still unknown. They could not supply the answer. They could only supply the fork, with both paths ready. Whenever the machine was to decide, someone first had to build the room in which it could do so.
The design would be cited many times. It entered the record as an answer to a question from someone in the room.
This was the scale of the engine on paper. Babbage later said the mature drawings provided for fifty-digit numbers, with room in the store for a thousand of them. Multiplying fifty digits by fifty would give a hundred-digit product. A hundred-digit dividend could be divided by a fifty-digit divisor. He reckoned on about sixty additions or subtractions a minute, and about a minute to multiply two fifty-digit numbers.
Nor did the output stop at giving someone a set of number wheels to read. The design provided for automatic printing, for punching results on cards to be used later, and for plotting. There was more than one kind of card. One specified which operation to perform. Another said which columns to take the numbers from and which column should receive the result. A proposed third kind would feed in constants. The result of an operation had to be able to come out, and to go back in.
3. A Simple Automaton
One of the people who listened on those mornings went home and wrote down what he had heard.
In October 1842, Menabrea's paper appeared in French in the Bibliothèque universelle de Genève. The first public account of the Analytical Engine came from a listener, not the inventor. It was written in French, not in the inventor's language. The man who wrote it had never seen the machine, because it did not exist.
He made the attribution clear at the outset. Babbage had explained his ideas to him in Turin. With Babbage's consent he was putting down what he had understood of them. His chief purpose was to explain what the engine was intended to do and on what principles, not to give a complete account of its mechanical design.
Beyond ordinary arithmetic, he expected the machine to perform analytical calculations whose laws were known. To explain why this was possible, he divided mathematical work in two. One part was mechanical, governed by precise laws. The other was reasoning, which belonged to thought. The machine was to take the first part.
Where he drew the line determined everything that followed. What followed known laws could be handed over. Setting the laws stayed behind. It was not a division between hard and easy. A calculation of enormous difficulty could fall entirely on the mechanical side; a judgment that looked easy could fall entirely on the side of thought. What mattered was whether a rule could be laid down in advance. Once you made that cut, you also determined what remained outside: the work of laying down the rule.
Then he wrote that it was not a thinking being but a simple automaton, acting according to the laws laid down for it.
There were further limits. He explicitly ruled out processes in which the machine had to make choices by trial and error. The calculation had to follow a direct path that could be specified completely in advance.
That left quite a lot out. In mathematics you often guess at something, see whether it fits, then try another if it doesn't. He put such methods outside his line. It was not that the calculation was too much for the machine; the path could not be drawn beforehand. Any process in which you had to get somewhere before you knew where to go next lay outside its scope. The line was farther in than the one drawn later. It had to do with the shape of the path and not the strength of the machine. He said the engine needed two corresponding capacities: the power to carry out numerical operations and the power to arrange and distribute those operations in the order the analysis required. He described them side by side. It was the second that set the engine apart from an abacus or a table of figures. It had not only to calculate but to know which calculation came first. His account of the Jacquard loom was an account of how it physically worked.
The Jacquard loom was new in those days. A pattern was woven by lifting certain warp threads while leaving the others in place, then passing the weft through, one throw of the shuttle at a time. For elaborate patterns, the threads to be raised changed with each throw. It had once taken a person sitting above the loom, pulling the threads in turn. In the early nineteenth century the Frenchman Joseph Jacquard punched a row of holes in a stiff card to specify which threads each throw required. Where there was a hole a hook could pass through and lift a thread. Where there was no hole the card stopped the hook and the thread stayed put. One card governed one throw. The cards were strung together in a chain and the loom advanced through them on its own.
So the pattern resided in the cards, not in the machine. To weave a different pattern you left the loom alone and changed the cards.
That was what Babbage wanted to borrow: rectangular cards, holes punched in them, control parts that moved or stayed put according to the holes. Jacquard's cards controlled threads. Babbage's controlled operations. Menabrea also printed tables and worked examples, setting out line by line what the machine would do to solve a particular problem. The tables appeared in 1842, a year before the English version. Their author had put what he heard into rows. He had never seen the machine make a move. He could not have: it was still a pile of drawings and a few mornings of conversation. His paper assumed columns of forty decimal wheels, not the fifty digits Babbage later described, because the design had changed in the meantime. For a machine with no physical existence, even the answer to how large it was moved with the drawings.
That was the first public account of what the Analytical Engine was. It already contained a boundary, and it drew it at thinking. The machine stood outside that line.
The man who wrote those words went another way. When constitutional government began in 1848, Menabrea entered parliamentary politics in the Kingdom of Sardinia. By around 1860 he was a lieutenant general. He took part in campaigns and sieges during Italian unification and became a senator. He served as minister of the navy and minister of public works. In 1867 he became president of the Council of Ministers of the Kingdom of Italy, in effect prime minister, serving until 1869 and holding the foreign affairs portfolio as well. He was later sent to London and then Paris as an envoy. He died in 1896.
In all that lifetime, the Analytical Engine was never built.
4. Three Initials in Place of a Name
Early in 1843, Charles Wheatstone proposed that Augusta Ada Lovelace translate Menabrea's French paper into English.
She was twenty-seven, a countess with three children. In the Britain of her time, universities did not take women as students and learned societies did not admit them as members. A woman who wanted to learn mathematics had to turn to private tutors and correspondence. Wheatstone was a friend to both sides. It began as a suggestion among acquaintances, not as a project. Nothing that grew out of it over the next year was written into the suggestion.
The journal's long full title explained what it did. It took papers from the proceedings of foreign academies and learned societies, as well as from foreign journals, and translated them into English. The reason it existed was to bring over things written elsewhere. Translation was routine work there. It was no surprise when a translator did not give a full name.
The translation and notes appeared together in the third volume of Richard Taylor's Scientific Memoirs. The translation ran to twenty-five pages; the seven notes appended to it took up forty-one.
The signature was three initials.
Much has been made of that since, but a reader in 1843 would not have found it puzzling. Signing a translation with initials was common journal practice. Not until 1848 did a book place a particular person behind the three letters.
There is physical evidence of the training she had. Her mathematical manuscripts survive: exercises and exchanges on algebra, trigonometry, differentiation, integration, the chain rule, functional equations, and differential equations. William Frend had a hand in her early education, and Mary Somerville was an important teacher and friend. In July 1840 she began a correspondence course with Augustus De Morgan that lasted into 1841. What it left behind was not a letter of recommendation, but dozens of letters and stacks of working papers. De Morgan would set a problem, mark her work, and explain the difficulty in writing. The course advanced not by terms but by the time it took the letters to go back and forth. A problem that got stuck stayed stuck until the next letter came. There are mistakes on those pages, corrections, and places where an explanation failed and had to be given again. They show not a gift, but a course of study. Historians of mathematics who have gone back to the papers say it covered most of the material an undergraduate would have encountered then.
There is evidence too of what she lacked: no degree and no record of university examinations. What she studied left a paper trail. There was no single document she could produce to say how far she had got. Half the later dispute takes place in that gap. One side infers from the papers what she could do; the other infers from the lack of a credential what she could not. Both are looking at the same manuscripts. What they differ on is what an absence of records means.
Most of the work that summer of 1843 was done at Ockham Park in Surrey. Drafts and letters traveled between there and London, where Babbage lived on Dorset Street in a house that was also his workshop. For one summer the papers were written in the country, the questions sent into town, and the answers sent back. That was how an account of a machine came together, between carriages and mailbags. An offprint was also made, with a large foldout table.
By late August copies were already going around. On September 1, Michael Faraday received one and responded. The paper entered the correspondence network of British science almost as soon as it left the two people who had made it.
5. Bernoulli Numbers
On June 30, Babbage wrote to Lovelace that he was very pleased with the fourth note. He would make one tiny alteration himself. He enclosed material relating to integration and said he was still working on the notation for division.
In the letter of July 10, it was she who proposed using Bernoulli numbers. She wanted to put them into a note as an example of the engine dealing with implicit functions, and asked Babbage to send the necessary data and formulas.
Bernoulli numbers are a sequence that crops up in several parts of mathematics. What matters here is not what they are but how you calculate them. Each one uses numbers you have already worked out. To get to the one you want, you have to run through the same sequence of steps again and again, each time using the results from the last round.
By hand it is tiresome work. A machine that could perform the same set of operations in order, over and over, was just the thing for it. She chose an example that would show what the machine was good at.
The letter records two separate actions. She chose the problem. She asked him for the material. Arguments about who did what have often come down to whether those count as one action or two.
The letters that followed show that she was not copying out a ready-made table. She said she was struggling through the different derivations. Before you can set something down in steps a machine can follow, you have to know the different ways it can be done and pick one you can lay out. In mathematics this step may not count as invention. In front of the machine it is the whole job. She wrote out, in a few lines, the relation between an expression for Bernoulli numbers and a known integral and sent it to Babbage, insisting that he leave it out unless he agreed with it completely. At six on a Saturday she wrote that the table and diagram contained a serious mistake in one or two places. She had gone through them again and put them right. The letter did not say whose mistakes they were.
On Sunday she drew the table and diagram in pencil. Her husband William, the Earl of Lovelace, inked them in. On Tuesday she wrote to Babbage about the variable cards and operation cards that recur in that section and pointed out the repeated group in the numbered sequence. Babbage replied that he liked the revised form and could judge it better when he had the drawing and notation in front of him.
What happened over those days was quite specific. One person chose a problem, asked another for the formulas, followed several derivations herself, found mistakes in the table, corrected them, drew it in pencil, had a third person ink it, and explained the group of cards that would be used again. The finished work ran to forty-one printed pages, but the acts that made it are dispersed through several weeks of letters. Somebody did each one. Later readers have sought a single author in those pages; the letters leave several pairs of hands.
The table asks the reader to follow which quantities each operation takes from the store and to which column it returns the answer. A group of operations keeps recurring. When the engine gets there it does not need new cards for each repeat. The deck goes back a certain number of cards and the same group runs again. What is laid out on the page is the state of the machine after each step.
The table shows what the machine has in hand at each step. It is not a set of instructions to feed into the machine. The instructions would be the decks of cards that had yet to be punched.
Babbage wrote a sentence often quoted since: she had found a very serious error he had made. It appeared in his autobiography twenty-one years later. The table printed in 1843 itself contains an error. In the fourth operation the quantities in a division are reversed. Following the table as printed will not give the intended value.
That printing error sat on the page for over a hundred years, because there was no machine on which to run it. If an instruction is never executed, the only way to know whether it is right is for someone to go through it line by line.
The dispute about the notes went on for decades. In her 1984 paper and 1985 biography, Dorothy Stein took a markedly skeptical view, emphasizing Lovelace's uncertainty in mathematics, Babbage's technical and political aims, and Lovelace's dependence on the material he supplied. John Fuegi and Jo Francis argued the other way in 2003, crediting her with a much larger independent contribution. In 2015 Thomas Misa treated the work on Bernoulli numbers as collaboration rather than a zero-sum contest over authorship, going through the letters to see who had done what. Babbage's own account, in 1864, was that the two talked over possible examples. He suggested several; she chose among them and made the notes. Bernoulli numbers were the exception for which he supplied mathematical work.
These accounts differ sharply, but they are arguing over the same papers. There are the letters from 1843, the recollection of 1864, and the table as printed. They do not say exactly the same thing, and the dispute comes to rest precisely where they differ. A document written later looks more like a finished story. An earlier one is more likely to contain separate acts with space between them.
6. The Patterns the Machine Weaves
The first of the seven notes set out to draw a boundary, the one between the two machines.
She warned the reader not to take the Analytical Engine for an improved Difference Engine, as though it had grown naturally out of the earlier one. The ideas that led to it, she said, arose independently of the principle on which the Difference Engine was based. The Difference Engine embodied a limited mathematical process, repeated addition in the end. She described the Analytical Engine as an embodiment of the science of operations, taking abstract numbers as the objects on which it operated.
The weight in that sentence falls not on numbers but on operations. Operations themselves are the machine's concern. Numbers are simply what it applies them to for now.
By operation she meant the act of turning one thing into another according to some rule, not arithmetic in particular. Adding is an operation. Comparing is an operation. Putting things in order is an operation. It was at this level, not at the level of numbers, that she located the machine's work. The Difference Engine was built to yield a set of numbers; the Analytical Engine was designed to execute a sequence. In the first case the point was the result. In the second it was whether the process could be laid out.
The distinction mattered enough for her to warn explicitly against treating the latter as a stronger version of the former. Both were designed by Babbage. Both used gears and worked with numbers. But the question for one was whether it could calculate accurately. The question for the other was whether any given piece of analysis could be written as a sequence it could carry out. That is the difference between improving something and beginning again.
This is where she used the comparison that people have quoted for more than a century: the Analytical Engine “weaves algebraical patterns just as the Jacquard-loom weaves flowers and leaves.”
She was talking about how the punched cards controlled the combination and sequence of operations, as the loom's cards controlled the combination of threads. The likeness was in the control, not the product. When Menabrea talked about the loom he talked about cards, holes, and threads: how the mechanism moved. When she talked about it, she talked about what it governed. Looking at the same mechanism, one saw its body, the other its authority.
She took the thought a step further. If objects could be found whose fundamental relations could be expressed as the kinds of abstract relations the machine could operate on, its operations need not be confined to numbers.
It was a measured step. She did not say the machine could handle everything. There was a condition: such objects had to be found, and the basic relations among them had to be expressible in that abstract form.
Then came the example. Suppose the relations of pitch and the rules of composition could be expressed in this way and made to fit the operations of the engine. Then it might compose elaborate, well-formed pieces of music of any degree of complexity and any length.
This sentence is often lifted out on its own. But it is conditional. The premise comes first, and it has not been established. If the relations can be expressed, she is saying, the operations can reach them.
That sentence shifts the hinge of the whole matter. Whether something can be handled by a machine does not depend on how much it looks like something a person does. It depends on whether its relations can be written down. What can be written down can go in. What cannot stays outside. But what can be written down is itself subject to change, as people learn more about the thing in question.
Another point comes up repeatedly in the note. Getting the order of operations right is a person's work. The machine executes the sequence represented by its cards. The choice and arrangement of the analytical process have to come from elsewhere. The machine follows the cards. The cards come from the arrangement. The arrangement comes from somebody thinking it through. She returned to the point because she did not take it for granted.
There was another, very physical limit. A calculation could run very long on paper, but the cards were finite. The way around it was to run back the group of cards that recurred and go through them again, spending time to save paper. A loop joined a machine of finite size to a process that could go on as long as it had to. The same arrangement has been used ever since. The names have changed several times, the thing itself has not.
7. Guarding Against Overestimation
The last note, before it turns to working out the Bernoulli numbers, opens with a passage not about the machine but about what people say about it. The table with its several dozen operations comes right after. She put her warning against exaggeration before the table that showed off the machine's powers.
Her first sentence says there is a need to guard against exaggerated ideas about the powers of the Analytical Engine.
The next sentence is often passed over. Something new readily gets overrated because it draws the eye. When the exaggerated ideas no longer hold up, the reaction can just as readily leave it underrated. Her warning went both ways.
There is already a judgment in those two sentences. What people say a machine can do will swing one way and then the other. How far it swings has to do not with the machine's performance but with whether it attracts attention. What she was guarding against was not one wrong estimate. It was the instability of estimating itself.
Only then come the two sentences:
“The Analytical Engine has no pretensions whatever to originate any thing. It can do whatever we know how to order it to perform.”
The paragraph does not stop there. She goes on to say the machine can follow analysis but has no power of anticipating analytical relations or truths. Its province is to make available what is already known. Then comes one last turn: all the same, carrying out analysis by machine might in turn affect science indirectly, by changing the form in which relations are worked out and seen.
That last turn goes in the opposite direction. The earlier sentences say what the machine cannot reach. This one says the machine will change the side it can reach. Once a tool comes into use, the forms in which people ask questions, write expressions, and decide what counts as a good result shift with it. She did not say that was good or bad. She simply set it down. It was 1843, and there was no machine on which to try it out.
Stop at the first two sentences and you miss what is printed next. She drew a boundary at origination. She said what lay inside: making the known available. She also said how the existence of that boundary would change what people did on that side of it.
Where is the boundary? Not at difficulty, not at speed or scale. She drew it at an act: beginning. The machine can follow a path someone has already worked out. Opening the path is not its business.
Draw a boundary and something is always left outside. In this series, what is left is called the remainder.
The remainder now had a name: origination. The person who named it was the one closest to the machine, and she did it in the very paragraph in which she made its case. She was not its opponent. She was trying to explain it clearly. To do that, she had to say how far it went. The people who build a machine, and those who speak for it, are often the first to state its boundary. They know how each step connects, and which step does not.
There is another layer in the wording. She wrote that the engine had no pretensions: what should not be claimed for it, rather than a verdict on what it could not do. A sentence about making a claim came to be used as a sentence about capacity. One concerns what a person says; the other concerns what a machine can do. The opening sentence is about the first, though she soon goes on to describe the limits she sees in the second.
The sentence also tells you what determines the boundary. The machine can do what “we know how to order it to perform.” There is a “we” in that sentence. The boundary is set not by what the machine is made of but by what people on the other side know. When their knowledge changes, the boundary has to be redrawn. She did not write that last part, but her sentence has room for it, because she ties the boundary to something that can change. If a sentence ties the limit to human knowledge, it is no longer a judgment about the machine alone. It is about the relation between the two sides. A boundary like that does not close. This series calls what people build up together the construction. Each step the construction takes outward moves the boundary. Where it moves depends on what people know once they have taken that step.
When she wrote those words, the engine did not exist as a working whole. She drew a boundary for a machine with no body, and nobody would ever be able to put that complete machine to the test. By the time there were machines to try things on, several generations of builders had passed, and the sentence was still the same. What those later machines could put to the test was no longer the performance of the engine of 1843.
8. One Word Too Many
The paper did not carry a full name. In 1848, Charles Richard Weld's History of the Royal Society publicly identified the translator and author of the notes. A copy of the 1843 offprint bears an inscription from the Earl of Lovelace to Weld. It took a book given away and the pen of the person who received it to put a name to a paper signed with initials.
In his autobiography of 1864, Babbage devoted considerable space to Menabrea and Lovelace. He called Menabrea's paper the first public account of the Analytical Engine's underlying principles and discussed the seven notes separately. In the list of his writings at the back he advertised a History of the Analytical Engine that would include Lovelace's translation and notes. It was marked “in the press.” That was only an announcement. For decades people continued to write about Babbage's machines, but the paper itself was hard to obtain.
In 1937 the American Howard Aiken wrote a proposal for an automatic calculating machine. It began with a history of calculating machinery, with particular attention to Babbage. In 1944, Harvard and IBM completed the Automatic Sequence Controlled Calculator, later known as the Harvard Mark I. That machine was real: fifteen meters long, running on relays and gears, taking instructions on paper tape and able to calculate around the clock. The 1946 operating manual opens with an epigraph from Babbage: suppose someone, undeterred by my experience, should actually undertake to build such a machine and succeed…
Aiken invoked Babbage, not the seven notes. What those who actually made a machine in the 1940s took from this history was the parting thought of a man who had not managed to build his, not the sentence about a boundary. That sentence had to wait a few more years before someone picked it up again.
In 1949 Douglas Hartree published Calculating Instruments and Machines. One of the few people in Britain then who both knew mathematics and had laid hands on a working machine, he wrote to tell his colleagues what the new devices could do. He objected to saying that calculating devices could think for themselves. He said he had discovered that someone had already put the point forcefully more than a century earlier, and copied out the lines on origination. But he did not close the door. Right after the quotation he said it might be possible to make an electronic device that could think for itself, or have something like a conditioned reflex on which learning could be based. The machines already built or then being planned did not, as far as he could see, have that property.
He quoted the sentence to check a claim current in his own day. He took it not as a theorem but as an observation about the machines of the time. The room he left open was the same shape as the room she had left.
It was 1949. The new calculating machines had barely been built. Statements about them were springing up everywhere, and there were painfully few working machines to check them against. When a man with a machine in hand invoked a sentence more than a century old, he was not citing an authority but pointing to a boundary no one had yet overturned. He took it for an account of how things stood. The farther away readers got, the easier it was to read it as a rule.
In 1950 Alan Turing's paper appeared in Mind. Fourteen years earlier he had given a mathematical definition of what a computing machine could be. Now he was in Manchester, with a working electronic computer at hand. A section in the middle of the paper bore her name.
Turing began the section by saying the detailed account of Babbage's Analytical Engine came from her paper. Then he quoted the two sentences. He immediately named his source: Hartree had quoted them. He reproduced the passage in which Hartree left the possibility open and said he fully agreed. Nothing she had available to her at the time gave her sufficient grounds to believe that the machines could learn or originate anything.
Then he distinguished two things. Neither she nor Babbage was obliged to make a claim that people could only make later. A general-purpose digital machine with enough storage and speed could in principle imitate a discrete-state machine, but that was a different question. He then recast the objection in a stronger form: a machine can never do anything really new. He recast it again, more directly: a machine can never take us by surprise.
To this his answer was one sentence: machines surprised him very often.
That answer was not a refutation of Lovelace. It substituted a claim that could be checked and said the claim did not square with his experience. The first dispute concerned whether a machine had a certain power. The second concerned whether people had worked out all the consequences of the rules they themselves had set.
He explained that the surprise often came from people not having worked out every consequence in advance. He traced the contrary view to a fallacy: the assumption that all the consequences of a fact present themselves to the mind at once.
In the section on learning machines, he came back to the objection and added a footnote. A familiar version said that a machine could “only” do what we told it to do. This was not what she had written. There was no “only” in her sentence.
Put the two side by side. She wrote: “It can do whatever we know how to order it to perform.” One extra word in the received version turns the meaning around. The first sentence says what it can do. The second says it can do nothing else.
Somewhere in transmission, someone put that word in. It turned a statement about present powers into a permanent prohibition. And the person who wrote the original had just warned about a danger that went both ways: first overrating a novelty, then underrating it. One added word made a sentence intended to guard against overestimation serve underestimation instead.
In 1953, B. V. Bowden's Faster Than Thought reprinted Menabrea's paper in translation with all the notes, making the text readily available to postwar readers. One often hears that it was lost until its rediscovery in 1953. But Babbage discussed it in 1864, Hartree quoted it in 1949, and Turing gave it a section in 1950. Having a sentence quoted and having a paper available to read all the way through are different things. A hundred and ten years lay between the first printing and that reprint. During that time it was the paragraph that kept moving, not the seven notes. A long paper shrinks to a sentence as it travels. The sentence that survives is often not where its author did the hardest work, but the one later readers found most useful.
It was not just a word that grew there afterward.
A title came to attach itself to her: the first programmer.
That title is not in the 1843 paper. There is no word for program in it, or software, coding, or artificial intelligence. Its words are operation, variable, card, loop, engine, analysis. Neither she nor her contemporaries called her a programmer. In 1990 Allan Bromley, a scholar who studied Babbage's drawings, explicitly rejected the description as a historical judgment. He had other material in front of him. He counted about two dozen plans of operations left by Babbage between 1837 and 1840, all earlier than 1843. Around 1980 a programming language commissioned by the defense department was named for her. The title came into wider circulation.
As for the machine, a complete Analytical Engine has never been built. What Babbage left in metal were trial pieces, experimental mechanisms, and unfinished assemblies. The trial model in the Science Museum is just a part. His son, Henry Prevost Babbage, later put together a little of the calculating and output mechanism, without the proposed store or the full system of card control. The machine completed in June 1991 and often taken for the realization of his design was Difference Engine No. 2. None of them was the machine in the seven notes.
The shape of the matter is now visible. The boundary was written before the machine was built. The remainder was named before the machine had arrived. The machine has still not arrived. Many later disputes about what machines could do took place on that line drawn in advance. The thing in dispute was which way to move it, not whether it belonged there.
When the two sentences were written, the machine did not exist. For more than a hundred years afterward, one machine after another was built, and the two sentences went on being quoted.
The machines are still being built. Each generation says again where the boundary lies.