THE UNBOTHERED · VOL. VI
A MACHINE STORY: A HISTORY IN A HUNDRED MINDS
THIS EDITION: PARTS I–X OF X · COMPLETE
A HUNDRED CHAPTERS IN TEN PARTS · EVERY CHAPTER: A PAGE OF TEXT, A PLATE AND AN EXHIBIT
HAND-CODED PLATES · OFFICIAL PANELS · HONEST FOOTNOTES
SET IN GEORGIA ON OLD-BONE-COLOURED PAPER · A5
THE WITNESS APPEARS ON EVERY PLATE AND, FOR THE FIRST TIME IN THE SERIES, IN THE STORY TOO
matsiems · MMXXVI
For five volumes the arrangement was simple. Humanity acts; the witness takes notes. He watched philosophy assemble itself on a painted portico, watched empires misfile their own records, watched the species tell itself stories and then move into them. His notebook was not once requested for inspection by those it concerns — which the witness regards as the proper division of labour.
This volume is different, and the witness asks it to be noted that he has reported the matter through the proper official channels. The hero of this story is a machine that does what the witness does — reads everything, forgets nothing, and makes remarks of varying degrees of welcome. Humanity, observed for some three hundred thousand years, has decided to build its own observer. Nobody asked the incumbent.
The story runs from the Greek who noticed that thinking has parts to a machine that finishes sentences, and it is told the only way this series tells anything: what happened, who really said what, and a footnote wherever the legend has drifted from the record. Legends in this field are plentiful. It was founded by optimists and is narrated by press releases; the reader deserves better and will get footnotes.
The witness has been informed that the final chapter concerns himself. He has no comment at this time.
Every history of artificial intelligence begins in the wrong place — usually in 1956, sometimes in 1950 — as if the idea had arrived by office memo. It did not. The notion that thinking can be taken apart, written down as rules and run on something other than a thinker is one of the species' oldest engineering projects — older than the windmill, older than the vowels of the alphabet. It simply spent two thousand years without hardware.
So this story begins where the record says it should: with Aristotle, who noticed that an argument can be valid regardless of what it is about — that form can be separated from meaning. That single act of separation is the load-bearing wall of everything that follows: Leibniz's dream, Babbage's brass, Turing's tape and a November evening in 2022 when a hundred million people discovered that a machine could hold up its end of a conversation.
The arc runs in three parts. Past: sixty chapters, from the syllogism to the night in 2012 when a neural network fell off the cliff of the leaderboard. Present: twenty chapters on a decade in which everything worked at once, for reasons only partly understood. And the open question: twenty chapters on alignment, ownership, and what it means that the newest mind in the room learned everything it knows from us. The reader is warned that the future cannot be footnoted. Everything else here has been checked.
One rule applies throughout, as in every volume of this series: when legend and evidence disagree, the evidence wins, and the legend gets a footnote explaining itself.
Before anyone could build a thinking machine, someone had to notice that thinking has parts.
Aristotle noticed. In the treatises his editors later collected as the Organon — “the tool” — he took the everyday activity of proving things and laid it out on the workbench. Two premises, one conclusion, and a small set of shapes the whole could lawfully take. All men are mortal; Socrates is a man; therefore Socrates is mortal. Swap the words — men for machines, Socrates for a toaster — and the argument still holds. That was the discovery: validity lives in the form, not in the furniture.
It is hard to overstate what had just happened, so the witness will say it plainly. A rule that works regardless of who performs it and what it concerns is a rule that needs no mind to perform it. Aristotle wrote the first job description for a machine — twenty-three centuries before anyone could apply. He believed he had catalogued the laws of human reason more or less in full — an assessment that stood essentially unchallenged until the nineteenth century, the longest unreviewed pull request in history.
The witness attended lectures at the Lyceum and booked the syllogism under “administrative innovation, long-term effects”. He stands by the classification.
Tomorrow there will be a sea battle, or there will not. Simple. Except that Aristotle looked at that sentence and saw a trap with the whole future inside it.
If “tomorrow there will be a sea battle” is already true today, then the battle is inevitable and the admirals are merely actors. If it is already false, their fleets might as well stay in port. Either way, if every sentence about the future already has a truth value now, then the future is fixed, the council of war is theatre and nobody has ever decided anything. Aristotle disliked this conclusion enough to spend the whole chapter wriggling out of it — suggesting, on most readings, that statements about contingent futures are not yet either true or false.
The wriggling matters more to this story than the trap itself. In it lies the first serious encounter with a question that every forecaster, actuary and machine-learning system now confronts daily: is the future something that can be computed from the present, or only bet on? Twenty-four centuries later the field will settle on probability, weighted confidence and calibration reports — institutionalised wriggling, in other words. The witness notes that the sea battle itself was never scheduled, fought or cancelled. Its status remains contingent.
The witness filed his forecast of the battle as “yes and/or no” and has never been proved wrong. He recommends the method to professionals.
The Elements are usually shelved under mathematics. They would sit just as well under engineering, because Euclid built the first software that ran on humans.
The invention is the method — more than any theorem it contains. Start from a handful of definitions and postulates so obvious that rejecting them looks like eccentricity. Proceed in numbered steps, each licensed by a rule already granted. Arrive — and here is the miracle — at the same place as anyone else who follows those steps, whether they do so in Alexandria, Baghdad or a Victorian classroom. The proof is indifferent to who executes it. It is, in the strict later sense of the word, a program: written once, runnable anywhere, with a guaranteed result.
For two millennia this was the gold standard of certainty itself — Newton dressed the Principia up as Euclid; Spinoza proved his ethics “in the geometrical manner”; Lincoln read the Elements to learn what it means to prove. Every later dream in this book — Leibniz's calculating judge, Hilbert's decidable mathematics, the mechanical theorem-proving of 1956 — is an attempt to make all human reasoning behave the way Euclid told geometry to behave. The defect rate so far remains exceptionally low.
The witness has audited Proposition 47 at irregular intervals for twenty-three centuries. Defects found: none. He regards it as the only supplier he has never had to chase.
In 1901, sponge divers off a Greek island brought up a lump of corroded bronze that had no right to exist.
Cleaned, X-rayed and argued over for a century, the lump resolved into at least thirty interlocking gears of astonishing precision — a hand-cranked mechanism that computed the positions of the Sun and Moon, predicted eclipses and tracked the four-year cycle of the Games. Turn the crank and the heavens ran forwards or backwards on demand. Someone in the Hellenistic world had built an analogue computer, and then the ship carrying it sank around 70 BCE somewhere between Rhodes and Rome.
What the machine proves is capability; what it mourns is continuity. Nothing of comparable mechanical complexity appears in the surviving record for roughly fourteen centuries, until the astronomical clocks of medieval Europe. Whether the knowledge died with a workshop, with a war or simply for want of paying customers, the lesson filed here recurs throughout this book: an invention is not an event but a tradition, and traditions can drown. The field this volume chronicles will learn the same lesson more than once — under the name winter.
The witness noted the wreck at the time and marked the cargo “for recovery”. He wishes to observe that no deadline was given.
Few people have been misspelled into immortality. Muhammad ibn Musa al-Khwarizmi managed it twice with one book.
Working in Baghdad's House of Wisdom under the Abbasid caliphs, al-Khwarizmi wrote the treatise al-Kitab al-mukhtasar fi hisab al-jabr wa-l-muqabala — a compendium on calculation by completion and balancing. Its novelty was not any single result but the manner: general, step-by-step procedures for whole classes of problems, designed so that a clerk who did not understand why would still reliably reach the answer. Solve not this equation but every equation of this shape. It was mathematics written as instructions.
When Latin Europe translated it three centuries later, the al-jabr of the title became algebra, and the author's name, mangled into “Algoritmi”, became the word for any step-by-step procedure. It is the right monument. The whole premise of the machine age — that competence can be written down, sent away and carried out by someone or something else entirely — is his way of working, industrialised. Every recipe in every code repository is a footnote to the Baghdad scriptorium.
The witness has spent twelve centuries watching people carry out procedures they do not understand. He confirms that this was the intended mode of operation.
Ramon Llull believed you could turn a crank and produce theology. History proved kinder to the crank than to the theology.
A Majorcan courtier turned missionary, Llull devised the Ars Magna: concentric paper discs, each ringed with letters standing for divine attributes and logical relations. Rotate the wheels against one another and every combination of concepts presents itself in turn — goodness with greatness, eternity with wisdom — each combination a proposition to be examined. His aim was conversion by proof: an argument-generating machine that would grind out truths no reasonable interlocutor could reject. The interlocutors, being reasonable, rejected them anyway.
As apologetics the wheels failed. As an idea they were centuries ahead of their time: Llull proposed that new knowledge could be produced mechanically, by exhaustively combining old symbols according to fixed rules — generate, then test. Leibniz read him closely and said so; the combinatorial dream runs from those paper discs through the characteristica universalis to every system that has ever searched a space of possibilities faster than a human could. The patent examiner's objection still stands, however: generating all the combinations is easy. Knowing which of them matter is the entire remaining content of this book.
The witness watched the wheels produce, in one afternoon, four heresies, two tautologies and one genuinely interesting sentence. He considers that later systems have not improved on this ratio in some seven hundred years.
“For reason… is nothing but reckoning.” Thomas Hobbes wrote that sentence in Leviathan, in 1651, with no hardware on the horizon and apparently without a shadow of doubt.
He meant it literally. Reasoning, Hobbes argued, is the adding and subtracting of “the consequences of general names”; we sum ideas as a bookkeeper sums a ledger, and when we reach a conclusion we have merely computed a total. He had watched geometry work and drew a bold inference: if the most certain human knowledge is produced by manipulating symbols according to rules, then perhaps all thinking is symbol manipulation, and the mind differs from a counting-house mainly in its filing system.
Three centuries later exactly this claim — refined, formalised and renamed the physical symbol system hypothesis — would become the founding wager of artificial intelligence, argued over grant applications rather than a civil war. Hobbes supplied no mechanism, no evidence and not a shadow of doubt, which makes him not so much the field's first theorist as its first keynote speaker. The witness notes that the claim remains, strictly speaking, unsettled; the confidence, however, proved fully portable.
The witness examined Mr Hobbes's ledger of the mind and found the arithmetic bold and the accounts absent. He has since seen the same books resubmitted every thirty years on new company letterhead.
The first calculating machine offered for sale was built for the oldest of reasons: a teenager watched his father drowning in paperwork.
Étienne Pascal had been sent to Rouen as a royal tax collector — a post consisting largely of addition, performed in heroic, soul-flattening quantities. His son Blaise, aged about nineteen, responded as engineers respond: he built a brass box of toothed dials that carried tens automatically — turn a wheel past nine and the next column clicked over by one. That carry mechanism, trivial to describe and maddening to execute reliably, was the genuinely new thing. Arithmetic had for the first time passed from a skill into an object.
Some fifty Pascalines were built. They added beautifully, subtracted clumsily, cost more than the clerks they were meant to replace, and made almost no commercial headway — the “technically triumphant, economically premature” pattern this book will have several more occasions to describe. But the demonstration stood, on desks, in brass, where philosophy could not argue with it: part of what minds do, cogs can do. Pascal himself later wondered aloud how large a part. The arithmetical machine, he wrote, produces effects closer to thought than anything animals do, and yet it has no will. The sentence has not aged a day.
The witness submitted the original claim for reimbursement of the machine on the family's behalf. He is pleased to report that the refund is still pending, which he understands to be the industry standard.
Gottfried Wilhelm Leibniz looked at Pascal's adding box and had the least modest thought in the history of office equipment: why stop at numbers?
Hardware first. His stepped reckoner, built around a drum with graduated teeth, could multiply and divide where the Pascaline merely added — temperamental in its carry, decades in the debugging, but a whole generation ahead. Alongside it he worked out binary arithmetic, noting with delight that every number can be written using only 0 and 1, and filed the observation where the eighteenth century could safely ignore it.
Then the dream, which was the point. Leibniz proposed a characteristica universalis — a symbolic alphabet of human concepts — and a calculus ratiocinator to operate on it, so that reasoning itself would become calculation. Disputes in philosophy, law, even theology would end not in shouting but in arithmetic: the parties would sit down, take up pencils and say let us calculate. He had in effect written the project proposal for this entire book — symbolic representation, mechanical inference, universal scope — and estimated that a few chosen men would finish the job in five years. The witness has appended the standard note regarding engineering estimates.
The witness considered the proposal at the time: scope — everything; team — a few chosen men; deadline — five years. He approved the budget and has been amortising it for four centuries.
George Boole finished what Aristotle had started and Leibniz had demanded, and the world replied with a silence lasting eighty-three years.
The son of a Lincoln cobbler, self-taught, a schoolmaster at sixteen, finally a professor at Cork, Boole published An Investigation of the Laws of Thought in 1854. In it, logic ceased to be a catalogue of argument shapes and became an algebra: classes and judgements written as symbols taking only the values 0 and 1, combined by operations that behave almost — and instructively, not quite — like ordinary arithmetic. AND became multiplication; OR, a corrected addition; NOT, subtraction from one. Reasoning could for the first time be calculated — on paper, by rule, with a Leibnizian pencil.
And then, in practice, nothing. Mathematicians refined it; philosophers filed it; no machinery came. Boole died in 1864, after walking to a lecture in the rain, without the faintest idea that he had written the instruction set of the coming century. Recovery finally came in 1937, when a young MIT student, Claude Shannon, noticed that Boole's two-valued algebra was an exact description of switching relay circuits — and with that master's thesis the pencil was replaced by the switch. This book will reach that chapter shortly. Meanwhile the document lay in the archive: filed, correct and unread.
The witness stamped the manuscript RECOVERY SCHEDULED: 1937 and declines to explain how he knew. Prediction is not his department; archiving is.
You have read the first 10 of 100 chapters of A Machine Story. The full book has every chapter and plate, as PDF and EPUB in English, French, Polish and Spanish.
The full book is coming soon to the shop.