Computers and Automation: Stored Instructions and Industrial Control

Older machines stored a rule. They were not this

The Antikythera mechanism is a geared instrument for astronomical cycles. Tony Freeth and his colleagues reconstructed that gearing in Nature in 2006. The engineering page already said what that paper is and what it is not. Calling the wreckage an ancient computer is a metaphor. Metaphors are how a straight line gets smuggled into a history that did not have one.

Jacquard’s head, from 1804–1805, stores a pattern for which warp threads rise. Bouchon, Falcon, and Vaucanson came first. The weaver stayed. The textile page is the place for that mechanism. Charles Babbage, later, planned an Analytical Engine that would take operations and numbers from cards, with a store for the numbers and a mill for the arithmetic. Doron Swade’s account is the one to read when you want the drawings without the myth of a finished Victorian computer. The engine was not built in his lifetime. A plan is not a machine that ran the payroll.

Herman Hollerith’s equipment for the 1890 United States census is a different real thing. A clerk punched a card. An electrical brush sensed the hole. Counters tallied the people. It is data processing, electromechanical, built for a count. It is not a stored program, and it is not a machine-tool control. The cards rhyme with Jacquard’s cards. Rhyme is not descent. Thomas Haigh and Paul Ceruzzi’s history of modern computing is useful here because it will not let you start the story only when the tubes light up, and it will not let you pretend the census tabulator was already a laptop.

Lovelace wrote notes on an engine that was not built

In 1843 Ada Lovelace published a translation of Luigi Menabrea’s account of the Analytical Engine and added notes longer than the account. Note G walks through a calculation of Bernoulli numbers as a sequence of operations the engine would perform. She also wrote, carefully, about a machine that might manipulate symbols other than quantities. That is a remarkable set of notes about a machine that did not exist in metal.

She is often called the first programmer. Historians still argue how much of the method was hers, how much was Babbage’s, and what “programmer” can mean when the mill was never finished and the collaboration was real. The Computer History Museum’s page on her notes is honest enough to show you the fight instead of crowning her and leaving. I will not crown her, and I will not erase her. The notes are in print. The engine is not in a museum as a working original. Both of those sentences can stand.

Relays could calculate. They were slow, and they counted

After Hollerith, punched-card shops spread into accounting and statistics. The machine sensed a hole and added a count. A room of people punched, sorted, and filed. That is industrial information work. It is not yet a computer you reprogram for a new kind of problem by loading a new routine into memory.

Relay machines went further. An electrical switch can stand for a bit, and a chain of them can step through an arithmetic operation. Konrad Zuse’s Z3, in Berlin in 1941, was an electromechanical, programmable machine. It was destroyed in the bombing. It belongs in this story so the “first” does not quietly become an Anglo-American word. It does not belong as the winner. It was not electronic, and it did not survive to become a product line.

Wartime money built several machines. They were not the same machine

Anyone who names the first electronic computer has already chosen a definition and hoped you would not notice. Colossus, built in Britain under Tommy Flowers for Bletchley Park, was running in the middle of the war as an electronic digital machine for one cryptanalytic job: the Lorenz cipher, not the Enigma machine. The Enigma work was a different device, the bombe, and it was electromechanical. Colossus was secret for decades. It was not a general-purpose computer you could point at a factory problem, and it was not a stored-program machine in the later sense.

In the United States the Army wanted firing tables faster than a room of people with desk calculators could finish them. ENIAC was built for that at the University of Pennsylvania’s Moore School, under contract. John Mauchly and J. Presper Eckert led the design. A large crew of engineers, technicians, and draftsmen built it. It was finished in 1945 and shown to the public in February 1946. It filled a room. It used vacuum tubes as switches. You changed its job by replugging cables and resetting switches. That is programming. It is not the programming of a machine that keeps its program in the same memory as its numbers.

Six women who had been computing those tables by hand — Kathleen McNulty, Jean Jennings Bartik, Betty Snyder Holberton, Marlyn Wescoff Meltzer, Frances Bilas Spence, and Ruth Lichterman Teitelbaum — were among the people who turned mathematical problems into ENIAC setups. Later public stories of the machine left them out for a long time. “Programmer” is a word the institution learned later. The work was real: they made the machine carry out the calculation. They were not the whole staff, and they were not a footnote.

The Atanasoff–Berry machine, at Iowa State before and during the war, was an electronic digital device aimed at systems of equations. It was not a general programmable computer. A 1973 patent case took the crown off ENIAC’s patent. A court ruling about who may own an idea is not the same thing as a historian’s “first.” Raúl Rojas and Ulf Hashagen’s collection, The First Computers, is useful exactly because the plural is the point.

The program moved into memory. That was a practice, not a lightning bolt

ENIAC did not start life as a stored-program computer. In 1948 the Moore School crew changed its control so that a form of program could be read from function tables instead of being entirely replugged. Thomas Haigh, Mark Priestley, and Crispin Rope tell that conversion in ENIAC in Action, and in a 2014 paper they take apart the textbook notion of one “stored-program concept” that arrived whole. Read them before you repeat the eureka.

John von Neumann’s “First Draft of a Report on the EDVAC,” circulated in 1945, put a logical design on paper: a store that holds both numbers and instructions, a unit that does the arithmetic, a control that steps through the instructions. The draft carries his name. The conversations it came out of included Eckert, Mauchly, and the rest of the Moore School group. The architecture is nicknamed for him. The nickname is not a sole inventor.

The machine usually credited with the first running of a program held in electronic memory is the small experimental machine at Manchester, in June 1948, Freddie Williams and Tom Kilburn. It computed a trivial thing and proved the memory. It was not a service anyone queued for. Cambridge’s EDSAC, in 1949, is the one that then did regular mathematical work from stored programs. “First stored program” and “first computer you could use” are two sentences. Do not weld them.

The switch got smaller. The plant did not change on the day the switch did

A vacuum tube switches fast and fails like a light bulb. It wants power and a room. The transistor, at Bell Labs in 1947, was the work of John Bardeen, Walter Brattain, and William Shockley — a point-contact device first, not a lone spark from Shockley. The integrated circuit is two claims in two years: Jack Kilby at Texas Instruments in 1958, Robert Noyce at Fairchild in 1959 with the planar silicon version that could be manufactured. I will not pick a father. A circuit on one piece of semiconductor is the fact. The patent fight is not the fact you need.

The microprocessor people mean, when they name one, is Intel’s 4004 in 1971, built for a calculator contract, by a group that includes Ted Hoff, Federico Faggin, Stanley Mazor, and Masatoshi Shima. A processor on a chip made computing cheap enough to hide inside other products. It did not, that year, replace the controller on a transfer line. The PLC page already said machine-level sequencing stayed with controllers built for the floor. A cheaper chip is why a later controller, a later drive, and a later screen could exist. It is not a decree that every relay panel died in 1971.

Numbers reached the machine tool before the personal computer did

Guiding a cutter from a list of numbers is older than the chip. After the war, John Parsons pushed the idea of machining from numerical data. MIT’s Servomechanisms Laboratory, on an Air Force contract, made a milling machine follow those numbers with servos. The demonstrations sit in the early 1950s. That is numerical control: a tape or a card, a special control, a machine that moves where the numbers say. Computer numerical control — a computer inside that control, able to hold and edit the program at the machine — comes later, as the electronics shrink and cheapen. Do not use CNC for a 1952 tape. Do not give the whole thing to one man. Parsons had the production idea. MIT built the control that made a demo. The shops that bought it came after both.

Above the machine, other computers took on planning, scheduling, and the count of what had been made. They did not absorb the PLC. A general-purpose computer is a poor substitute for a box that scans inputs on a known rhythm and fails in a way an electrician can explain. Plants kept both. They still do. Haigh and Ceruzzi are the wide map. They are not a wiring diagram of your cell.

A network moves a number. It does not make the number true

Once controllers and computers could speak, a count, an alarm, or a temperature could show up on a screen in another room. That is useful when the number is the number. It is a new way to be confidently wrong when the sensor drifted, the address was swapped, or the message arrived late and was treated as fresh. The PLC page already said upstairs must not be the only thing between a press and a hand. This page will not take it back.

Connecting a line does not optimize it. Somebody still has to decide what the number is for, and somebody still has to throw out the number when the gauge is lying. A dashboard is not a process.

Where this sits in the history of automation

A plant now can have a PLC, a robot controller, an industrial PC, a vision system, and a database, each doing the job it is bad to steal from the others. The computer’s share is to hold a lot of instructions, to calculate, to remember a history, and to pass a result. The measurement still comes from a sensor. The motion still comes from a drive or a valve. The interlock that must be true every scan still wants a controller designed for that, not a model that is usually right.

What gets called artificial intelligence can spot a pattern in that history: a drift, a likely fault, a forecast. It does not become conscious, and it does not inherit the safety job because it noticed something. A person remains accountable for what the system is allowed to do when it is wrong. I am describing tools a plant can buy. I am not describing every shop, and I am not describing 1946.

The long use, without the straight line

Later systems answered limits these machines actually had. A stored program answered a cable plant you had to replug for a new problem. A transistor answered a tube that burned out and heated the room. A chip answered a cabinet of boards. A network answered a number that lived on one machine. Each one arrived when somebody could build it and somebody would pay. None of them was hiding in a Jacquard card, and none of them replaced the person who has to say the answer is fit to use.

Computers earn their place because they made instructions into data a machine could keep, change, and run at electronic speed, and because industry eventually aimed that speed at machines and at records. They do not earn the claim that they created automation, or that they taught it to think. The instruction does not care what we call it. It only does the next step the program, and the data, still say to do.