Private Thomas Wood was chosen following an internal competition, and appears to have been the best operator of the electric calculating machine to be found in Japan at the time. He worked in the finance disbursing section of General MacArthur’s headquarters. Kiyoshi Matsuzaki, his opponent, was a clerk in the communications ministry; a famously skilled operator of the soroban, he was known to the American sponsors as ‘The Hands’. Both contestants were aged twenty-two. Nearly 3,000 GIs packed the venue, and many of them placed bets (on Wood, presumably).

In fact, Matsuzaki won four of the five events, and overall eight of the twelve rounds of problems. He completed the addition problems in around 1 minute 15 seconds each time, and the others in little more. In one case, Wood trailed more than thirty seconds behind him: thirty seconds that must have felt more than a little excruciating for Wood and his fans, cheering no longer.

The outcome seems to have been a real shock to the event’s American sponsors and reporters, though a close look would have told them that such head-to-heads had been organised before, with victory typically going to the soroban. Stars and Stripes called it a step backwards for ‘the machine age’; the Nippon Times (controlled de facto by the occupying force) stated that ‘civilization’ had ‘tottered’. Back in the USA, reports in Time magazine and elsewhere took a less affronted tone, but a note of surprise was still quite evident. Among writers on the soroban in Japan and elsewhere the contest remained a matter of comment well into the 1950s, and it remains part of the folklore of counting and calculation.

The fact is that a soroban expert can add and subtract numbers about as fast as they can be read out. For multiplication and division, informed comment in the 1950s reckoned that a first-class soroban worker was on a par with an electrical machine for problems with about ten or twelve digits: more and the machine would normally win, fewer and the human would.

When they were not being deployed in exams, competitions and public contests, the suanpan and soroban remained a common piece of equipment in shops and businesses throughout Japan, Korea and China. Compared with paper methods, they were quick; compared with the electrical machines that were now gaining attention, they were cheap and required neither regular maintenance nor an electricity supply. Their portability always remained a boon. Well into the 1980s, the bead-based devices could be called ubiquitous in East Asia. A degree of decline followed, as pocket calculators and electric cash registers displaced them as the most convenient tools for everyday calculation. But suanpan/soroban arithmetic remained an important part of mathematics education, and well into the new century a high proportion of adults reported using their methods at least some of the time. After-school training aimed at competitions retained its popularity too.

An intriguing feature of this culture was that the physical device itself proved to be, strictly speaking, unnecessary for the most skilled users. Regular users of the suanpan and soroban had long found they tended to interiorise the operations, to the point that they could do at least some calculations equally well on an imaginary device. By the second half of the twentieth century the so-called ‘mental abacus’ (anzan, blind calculation) had become a regular feature of training programmes, and it featured in the soroban examinations in Japan and, later, their equivalents in China. As long as the number of digits involved in the calculations remains manageable, it seems to be possible for almost anyone to acquire a rapidity in mental calculation by this means that would be astonishing in other contexts. An observer in the 1980s described a class at Dongyuan:

Children go there after school and usually sit on long benches in large rooms filled with students. A typical mental abacus calculation exercise begins when the teacher, standing at the front of the room, raises his hand, whereupon the room falls silent in anticipation. The teacher then reads aloud a list of 20 three-digit numbers as fast as he can, so fast in fact that the numbers are almost unintelligible. The children are silent, and the room is tense with concentration. After the last number is read, every hand in the room shoots up, and the teacher calls on one child to report the sum. Usually the child’s answer is correct.

Because there is no need to move actual beads, the speeds attainable are even greater than with a physical device, and the development of ‘mental abacus’ training has allowed its users to dominate competitive events like the Mental Calculation World Cup. An expert can add ten ten-digit numbers in under fifteen seconds.

The phenomenon caught the attention of cognitive science researchers, and has been regularly studied since the 1980s. The mental abacus provides a way to count, and to represent and manipulate numbers, that is both entirely mental and independent of language, written symbols or even physical gestures. Experts can hold a conversation, or tap out a rhythmic beat, while calculating. There is evidence that proficiency with the suanpan, the soroban, or their mental equivalents, results in a person using different parts of the brain to perform numerical tasks compared with someone who has learned to calculate using words or symbols: motor centres rather than language ones. It is a helpful reminder that words and symbols are not essential to counting, even of the most sophisticated kind.

Herman Hollerith and Kawaguchi Ichitaro: Counting machines

What about the electrical machine that Private Wood operated, and that was so soundly beaten in 1946? Where does it fit into the story of counting, in East Asia or elsewhere? To find out, it is necessary to take a step back to the years around 1900, when modern anxieties were coming to dominate the very ancient practice of census taking.

Japan – probably Tokyo – in 1905. An operator sits at a prototype machine. Made of dark polished wood, it is the size of a large upright piano. The operator has a flat desk to work on, and faces a rack of forty dials, each one like a clock face, with two hands.

She inserts punched cards into the machine one after another, taking them from a dedicated card-sorter that stands to her right. She places a card in position, and turns a handle. The machine makes electrical connections through the holes in the card; the pattern of holes determines which circuits close and which do not. That in turn determines which of the forty dials advances a step: perhaps one, perhaps several. Each dial has markings up to one hundred: with two hands, each can count to ten thousand. At the end of the working day – or the set of cards – the numbers can be read off and written down, and the dials reset to zero.

Kawaguchi’s Electric Tabulation Machine.

IPSJ Computer Museum; collection of the Statistical Museum.

The machine was designed by Kawaguchi Ichitaro, an engineer working in telegraph signalling technology, to assist with the tabulation of census data. The request for such a device came from the Japanese cabinet via the ministry of communications and transportation. His prototype was complete in 1905, and was used to process some of the data from the previous year’s demographic survey.

The creation of this counting machine was one of the outcomes of a thirty-year agitation for more and better statistics about the Japanese population and economy, led by individuals from both government and business. A system of land maps and family registration, very broadly similar to the old Chinese taxation surveys, already existed, but after the Meiji restoration in 1868 the government increasingly perceived a need to gather more information about the country’s economic activity, to help plan its growth and industrialisation. By the time a new national census law was promulgated in 1902, it had become clear that the volume of data to be gathered would benefit from mechanical processing: hence the request to Ichitaro to build his tabulating machine.

Those familiar with the history of computing and calculating machines will have a sense of déjà vu about Ichitaro’s machine. It bore a marked resemblance, both in its appearance and in the way it worked, to the machines designed in the USA over the previous two decades by Herman Hollerith, and that were spreading rapidly throughout the world during the decades around 1900.

In the USA, carrying out a regular population census had been an obligation ever since 1787, when the Constitution specified that the apportioning both of members of the House of Representatives, and of direct taxation, should be based on the populations of the various states in the Union. It specified that a census – an ‘enumeration’ of the population – should be taken for this purpose within three years after the first meeting of Congress, and at least every ten years thereafter. The first was duly carried out in 1790.

Over the following century, the details of the census and some of its assumptions changed drastically. The population of the United States grew from under four million to around sixty million, and both government and people came to demand more and more information about that population. From the initial handful of common-sense questions about names and numbers, the work of the census enumerators steadily grew in complexity, with new questions added about agriculture, industry, births and deaths, schools, libraries, churches and property. The 1880 census asked twenty-four separate questions, produced hundreds of tons of raw forms (‘schedules’), and resulted in the publication of twenty-two large volumes of data. Officials worried whether the situation was sustainable. A census is as much about reducing data to manageable form as about collecting it in the first place: any set of data can be processed given enough time, but there were limits to the delays that could be tolerated and the money that could be appropriated to fund the work.

‘Counting’ the census results, up to this time, had meant making tally marks on sheets of paper. Specifically, sheets were prepared covered with a grid of small squares, and in each square up to five pencil marks were made by a census employee who was counting, say, the number of census schedules that reported their subject as practising a particular occupation. A completed line of five squares contained twenty-five tallies; a complete block of four lines contained one hundred. Making the marks by inspecting the original census schedules in this way was repetitive, boring and susceptible to error, and converting the completed sheets of tallies into Arabic numerals, written on another sheet of paper, was also eminently liable to be done wrongly by tired, hurried or stressed human beings. Furthermore, if several kinds of data had to be extracted – age, occupation and marital status, say – the original census schedules had to be handled several times, creating a bottleneck in the whole process.

To avoid the expected disaster at the 1890 census, its director arranged a competition to test possible mechanical solutions to the problem of large-scale, repetitive counting. Mechanical aids had already been used over the previous twenty years, including a tally sheet that was automatically moved on rollers, and a system in which the raw records were initially transcribed onto more easily handled separate cards or papers. At the trials in 1889, systems were shown involving paper slips and coloured inks, or cardboard ‘chips’ of different colours. But the competition was easily won by a system using punched cards to transcribe the census data, invented by a 29-year-old son of German immigrants, the Columbia graduate Herman Hollerith.

Punched cards or punched paper strips had long been used to control mechanical looms and player pianos, or to store information about the holders of railway tickets. Hollerith’s innovation, prompted at least in part by a suggestion from his fellow census employee John Shaw Billings, was to use a punched card to store information about each individual recorded in the census. Crucially, he realised that information in that form could be sorted and counted automatically by a mechanical or electrical device. By the time of the 1889 trials for the federal census, his system had already been patented and successfully used to tabulate health statistics for Baltimore, New York and New Jersey, and for the US Army.

Starting from Monday 2 June 1890, nearly 47,000 census enumerators went door-to-door across the USA, gathering data. There were eight million square kilometres to cover. One printed form was used for each family, with twenty-six questions to answer about each person, and five extra questions about the household. Additional forms were used for veterans, for agriculture, manufacturers, the disabled, homeless, convicts … a total of nearly seven hundred different pieces of information were collected. There were penalties for refusing to answer. Special agents, meanwhile, were separately collecting information about property, mortgages and a huge range of industries and products: from agricultural implements to hosiery, from brick yards to silk goods.

Each enumerator sent daily reports of work completed to a supervisor, as well as sending in the completed forms themselves. After checking, the forms went to Washington:

The blanks which had been filled up were laid one upon the other on a piece of straw board. Each pile contained the schedules of a single enumerator. On top of all was placed an empty portfolio, to whose center was pasted the label with the enumerator’s name and the designation of his district upon it. The bundle was then corded together and a number of such bundles, representing from 13 to 15 enumeration districts, were placed together in a box which they exactly fitted. The box, 27 inches long and about 18 inches in its other dimensions, properly closed and sealed, was sent in this shape to the Washington office. One hundred such boxes were received daily, and several trucks were kept busy transferring them.

At the Washington census offices, a total of something like twenty million forms were eventually received, weighing around two hundred tons. The data on the forms was transferred to punched cards, using a device much like a typewriter keyboard rigged to a hole punch. One card was prepared for each person surveyed: thus about sixty million cards, each about the size of a dollar bill, with an average of perhaps twenty holes per card recording different pieces of information according to a precisely defined scheme of hole positions. With practice, a person could punch a thousand cards a day, and it was reported that by the end of the process the clerks had handled so many cards they could ‘read’ the information from them just by looking at the holes.

After six months the cards were ready to be counted. A journalist reported visiting ‘a very tidy and airy machineshop … where nice-looking girls in cool white dresses are at work at the long rows of counting machines’ that ‘remind one of upright pianos’. One of these clerks would place a card flat on a hard rubber plate and pull a handle to bring down a frame on top of it. A grid pattern of electrical points in the frame met a similar grid of mercury contacts in the plate, making contact in just those positions where there was a hole in the card. Circuits were thus completed, a bell rang, and a selection of the machine’s forty dials would each advance by one. A quick-fingered clerk could process 7,000 cards in a day, and there were up to a hundred clerks.

A Hollerith tabulator in use in 1890.

Scientific American, August 30, 1890, cover. Public domain.

A preliminary count of the whole population was completed by 12 December, just over a month after the last census returns came in. A total of 62,629,250 people had been counted. Subsequent work provided detailed breakdowns of this number by state, city and town, and similarly broke down the count of people by their place of birth, age, occupation, parentage, literacy and marital status. The electrical circuits could count a complicated pattern of holes just as easily as a simple one, and they could count several such patterns simultaneously: up to forty in principle, as many as there were dials on each Hollerith machine. Furthermore, when the circuits closed they not only rang a bell and advanced certain counters; they also performed the final task of opening the lid of one of two dozen boxes to the side of the machine. The clerk took the card out of the machine, put it in the open box and closed the lid. Thus, as one set of counts progressed, the cards were automatically being sorted ready for the next set of counts.

Journalists who visited the workroom were invariably impressed, if half deafened by the bells (‘… your tympanums all tingle / At the jingle, jangle, jingle / Of the Bells!’). Two dozen volumes of statistics were printed over a period of seven years, and the census of 1890 was unarguably the largest and most detailed view of the American population and economy yet produced (notwithstanding a widespread belief that certain cities, and perhaps the population as a whole, had been under-counted). It was one of the printed bulletins from this census that famously noted that ‘the unsettled area’ of the American continent ‘has been so broken into by isolated bodies of settlement that there can hardly be said to be a frontier line’, prompting a long-running conversation about the meaning of the frontier – its presence, its movement and now its absence – for American history. The census also resulted in an increase in the size of the House of Representatives from 325 to 356 members.

Hollerith’s machine was a genuine mechanical counter – not a machine for doing arithmetic, but one whose sole purpose and capability was to answer the question: how many? No human being had to pay individual attention to the items counted; they could simply be fed into the machine and a count accessed some time later, in the form of a number symbol indicated by the machine. The heart of the counting process – repeated attention plus keeping track – was carried out by the machine itself, raising an interesting question about how or whether a machine can ‘pay attention’ to anything. For the purpose of the census, the items counted were people as represented by punched pieces of manila card, but it was clear that almost anything could be counted by similar means. This was a genuine novelty in the history of counting, and one that, naturally, proved capable of enormous development over the subsequent decades. It is, perhaps, no great surprise that counting machines went on to change the world.

Machines of substantially the original design continued in use for processing the next US census, but from 1910 the Census Bureau ran its own machine shop to manufacture similar devices rather than rent from Hollerith at $1,000 per year for each machine. Hollerith’s designs, meanwhile, were taken up for census work and other statistical tasks in a succession of countries including Austria, Canada, Norway, Russia, France and Italy. By 1912, an admirer could speak, admittedly with some exaggeration, of their adoption by ‘nearly every civilized country’.

Later improvements included a system to record decimal numbers on the cards – using a choice of ten holes for each digit – and machinery to add up numbers so stored. Subtraction, multiplication and the means to represent negative numbers followed in due course, as did a device to feed cards into the machine automatically, doing away with much of the need for dextrous human operators. Special control cards could tell the machines to stop, and printing devices could automatically print the totals so that humans did not need to transcribe them.

Hollerith sold his limited company in 1911, staying on as a consultant; it became part of the Computing-Tabulating-Recording Company, which in the following decade changed its name to the International Business Machines Corporation. For much of the twentieth century, IBM would be one of the world’s largest data-processing companies, supplying equipment to governments and businesses around the world.

There is a disastrous epilogue to the 1890 census itself. In a departure from the practice of the US census up to 1870, no copies were taken of the census returns before they were sent to Washington. One reason was to save labour, and to reduce the burden on local administrations of producing and storing literally tons of paper forms. Another was the belief that the punched cards themselves constituted an adequate copy of the data: one official wrote optimistically that the original schedules ‘might every one of them be burned up, and the Eleventh Census could be taken over again from beginning to end, by means of the little slips of manilla’: that is, the punched cards. In fact, that was not the case, since the names of individuals were not recorded on the cards, only reference numbers denoting the original schedules where they could be found: data was irreversibly lost during the transcription onto punched cards. A third reason was that some felt the original data itself no longer had value once it had been counted, reduced to aggregate statistics and those statistics published: an ill-omened side effect of too-efficient counting practices, perhaps.

By March 1896, an accidental fire had already destroyed portions of the original 1890 census schedules. The bulky remainder were still in the portfolios in which they had been submitted, tied up with twine and piled flat in racks. Limited funds were available for their preservation, and together with 180 tons of punched cards, they were stored in poor conditions in a basement in Washington. On 10 January 1921, a five-alarm fire tore through the building. There were no human casualties, but four hours of fire and the jets from twenty hoses damaged perhaps two-thirds of the 1890 records beyond repair.

The remainder were disposed of as useless in the 1930s, missing out very narrowly on the advent of new attitudes to records management and the preservation of historical documents. The cornerstone of the American National Archives building was laid just a day before the disposal was authorised, and a programme to photostat and bind old census records was under way within the year. It was too late for the 1890 schedules and cards, and with their destruction a permanent, much-lamented gap was created in the records of American genealogy and social history.

What of the Japanese census machine, based on or inspired by Hollerith’s long-lived invention? By February 1904, Japan was at war with Russia, in a conflict that continued for a year and a half, with huge consequences for the government’s priorities and budgetary allocations. The census was postponed indefinitely, and a national population survey of Japan was not taken again until 1920, by which time Ichitaro’s prototype machine was obsolete, perhaps forgotten.