Advent of Computing - Episode 35 - Analog Computing and the Automatic Totalisator

Episode Date: July 26, 2020

A lot of the technology we associate with the modern day started on anachronistic machines. I'm not talking about mainframes, I'm talking older. Today we are looking at George Julius's Automatic Totalisator, an analog computer used to manage betting at horse tracks around the world. These were massively complex machines, some networked over 200 input terminals, and they did it all mechanically. Like the show? Then why not head over and support me on Patreon. Perks include early access to future episodes, and stickers: https://www.patreon.com/adventofcomputing Important Dates: 1913: Premier Tote installed in Auckland

Nearly every episode I research, I run into surprises, and that's one of the reasons I like producing this podcast so much. The more I look into the history of computing, the stranger and more convoluted the story gets. And there's something that I really like about that. I try to find more obscure or less well-covered material for episodes when I can, because I want to share that feeling of amazement and excitement with all of you listening. And today, I have something that fits that description to a T. Now, we're

all pretty familiar with the idea of distributed systems, even if you may not know them by that name. The internet is a perfect example, but there's smaller networks, like IoT setups or even something as mundane as a central heating and cooling system. However it's packaged up, they all work off a very similar principle. You have some central system that many multiple nodes connect up to. Take an isolated server on the internet, for instance. It can have hundreds of thousands of people connected up to it over the web, all simultaneously. Each time someone connects, it has to handle a request and send back data, and it has to keep track of who's requesting what. The central server has to

do a lot of tasks, but anyone logging in doesn't have to do all that much at all. While networks may be the easiest place to observe distributed systems today, the concept is not exclusive to networks. In fact, it predates networking very considerably. This isn't some kind of gotcha where I'm going to talk about theory or writings. I'm talking about actual working machines. As far back as 1913, this idea, something that feels really darn modern, was being used for gambling at racetracks. Today, we're going to be talking about George Julius and the Automatic Totalizer, a series of analog computers that mechanized betting. But this isn't all fun and games.

Julius' machines did things that wouldn't be replicated for decades to come, and they all were made using gears, pulleys, and shafts. Welcome back to Advent of Computing. I'm your host, Sean Haas, and this is episode 35, The Automatic Totalizator. Today, we're traveling to a part of computer history that I've been avoiding for some time now. We're going to take a look at some analog computers. Part of why I tend to avoid analog systems is because, well, I just don't understand them very well. Since 1945, computers have been all digital, and that's where I'm most comfortable. If you've worked on modern computers enough, then you can kind of get a feel for how early

mainframes must have felt. But analog systems, they work on a whole other principle. It's kind of like looking at a whole other evolutionary line that died out. And I think I'm not alone in this. The coverage of analog computers in general is somewhat lacking. And I think that's partly because the specifics of these systems are pretty enigmatic to people. Despite being brushed aside, analog computers are important to the story of computing in general. It fleshes out a lot of early history, and it informs a lot of decisions

that digital pioneers would make. A lot of the first digital systems were made as reactions to analog computers, so understanding this era gives us more insight into the broader story. This episode, we're going to be looking at one of the earlier and more obscure analog systems, the automatic totalizator. I first read about this type of computer in a paper called Forgotten Machines, The Need for a New Master Narrative, written by Doran Swade. The crux of Swade's argument is that a lot of people get the history of computing plain wrong. There are

machines tucked away that fit the general description of a computer, but they rarely show up in histories of the topic. It's less of a forgotten history and more of a misplaced one. Automatic totes are one of the examples that Swade brings to the table. Basically, a totalizator, often called a tote or totalizer, is a machine that's used for managing bedding at horse tracks. They were around for quite some time before automation caught up to them, and by the 1910s, a transformation began to take place. Writings from the period use different words than we'd use today, but essentially, machines that networked together hundreds of bedding booths

all connected back to a central processor would start to be built in that decade. So yeah, part of the misplaced history of computers involves networking. To make things even more bizarre, these early computers were totally mechanical. So let's dive into the Julius Totalizator, a machine nearly forgotten by the 21st century. Along our twisting path, we're going to see how a machine like this came to be, how it was even possible, and dive into how analog computers worked in general. Now I'm going to try to keep things light, but we are going to need to get into some math, so hang on to your seats. But before we get too far into computing's history,

let's talk a little bit about horse racing and betting. Now, I'll be the first to admit an utter lack of knowledge in this area. I don't really bet on anything, so I'm speaking here in a purely academic sense. And I gotta say, horse racing is a lot more complicated than I initially expected. The race itself is simple. First, you get a bunch of horses and a bunch of jockeys, usually 5 to 10 of each. You set them on a track, either straight or oval. Then, at a signal, they all run from one end of the track to the other. Whoever gets to the finish line first is the winner.

The second arrival gets second place and so on down the line. I don't understand how that can get complicated, but oh boy, humans have found a way to make that very complicated. The crux of our discussion comes down to a system called parimutuel betting. Essentially, anyone gambling on the race will place some amount of money on a given outcome. Let's just say it's who comes in first place to make the whole thing easy. The amount bet on each horse is tallied, then the total bets are placed into a pool. At the end of the race, anyone who bet on the winning horse wins a payout. Now, to calculate that earning, you start

with the pool of all total bets. Some of that is skimmed off for taxes and fees to the racetrack and such. The remaining pool is then split amongst the winners based on the amount of money they paid in initially. For each $1 bet, you get some X dollars out. So how much you can actually make off winning is determined by how much money is bet on the dollars out. So how much you can actually make off winning is determined by how much money is bet on the winning outcome compared to how much is bet overall. That ratio of possible payout for any given winner is called the odds. If for every $1 you put in, you could win $5, that makes for 5 to 1 odds. When you place your bet, those odds are usually displayed, but in the lead up to the race, those numbers are constantly in flux, so odds have to be constantly updated and recalculated.

Now, I really don't like the naming convention here since it's not the actual chance of an outcome, it's just the payout, but that's how it is. I've been using the example of betting on a given horse getting first place. But that's just the tip of the iceberg. Depending on local ordinance and laws, there will be a lot more options to choose from. You could bet on first and second place, or bet that a horse will be in the top three of the race. You could bet on an exact order for the top three winning horses, or that a horse will

win this race but lose the next race, the list can get pretty long and can vary a lot depending on the country you're betting in. From the record-keeping standpoint, this gets really complicated really quickly. Each outcome the gamblers bet on will need to have its bet tracked, they'll need to have odds calculated and updated, and then once the race is over, the final payout will need to be calculated and passed out. If you have, say, 10 different types of bets, that's 10 different calculations that are constantly being ran and need to be updated. And since gamblers need to see odds prior to betting, that makes each calculation pretty

time-sensitive. But that's just the on-paper struggles of parimutuel betting. The people managing betting and keeping the books were, after all, just humans with pens and paper. As we all know, humans are kind of fallible. It's pretty easy to fudge around some numbers and skim a little money off the top. By the same token, it's possible to coerce a human operator. The ideal solution would be to take humans from that side of the equation entirely, but that turns out to be a pretty tall order. With advances in technology, some mechanization made its way onto racetracks. Access to analog calculators made computing odds a little bit

easier. Some tracks adopted semi-automated tote systems. These allowed an operator to increment counters as bets replaced, automatically displaying the running count. And they helped a little bit when the final payout was calculated. But most of this development was happening around the turn of the 20th century, so there were some hard limits on how complex these systems could be. To fully automate gambling would take a much more sophisticated machine. Now, with that rough explanation of horse betting out of the way, I think it's about time to introduce George Julius.

He was born in England in 1873, but his family moved to Australia just after he turned 10. By 1890, Julius had earned a degree in mechanical engineering and was well on his way to a career in the sciences. In the coming years, he worked as a railway engineer and eventually picked up some contract work for a logging firm. As with a lot of folk in the sciences, his home and work life tended to mix quite a bit. After hours, Julius would spend time tinkering on his own pet projects. As a child, George had been fascinated with clocks, spending hours disassembling and repairing their delicate mechanisms. Now as an adult, he found himself drawn back into clockwork mechanisms,

but not necessarily for timekeeping. He had a big idea brewing, one that would slowly start to materialize. That was to create a voting machine. In Julius' words, quote, A friend in the West conceived the idea of getting me to make a machine to register votes, and so to expedite elections. I invented one that aroused some interest, and it was submitted to the Commonwealth government. It had been urged, however, to adopt the hair-spins system. I then produced another machine which would register votes only when an elector had voted formally, which automatically

recorded preferences, informed the returning officer if the vote was informal, and gave the result without any human intervention, being purely mechanical." The concept here is relatively simple. For any democratic system, voting is really important. It's kind of the core issue of it. There has to be some way for every eligible citizen to submit a vote. You need some measures in place to prevent votes from being tampered with. And ideally, you need a way to quickly tally all votes. Having slow returns, or worse, having inaccurate returns, can lead to disaster. At the time Julius was tackling the idea, most voting was done with paper ballots that were then counted by hand. That arrangement is viable for relatively small populations. But as the number of voters grow, then you can run into some problems.

Things shake out to be pretty similar to the information problem, but with a little bit of a twist. As population explodes, a government will have to hire more and more clerks to count ballots. Eventually, a point will be reached where, no matter how many people are pouring over ballots, you simply won't be able to get a full count before the next election cycle. There's just way too much information for humans to handle alone. But the twist comes from all those humans. Each time a ballot is counted, well, that introduces one chance to make an error. Even worse, people being how we are, each ballot introduces a chance for forgery. With enough coordination, it's possible to cheat an election. And if there's

so many ballots that counting them all is a challenge, then, well, it can become very easy to hide forgery in plain sight. One solution to this problem was our old friend the punch card. Starting in the 1890s, punch card-based voting machines would pop up in the United States. These worked by punching votes onto a paper ballot and then running a stack of ballots through a tabulating machine. Votes could be counted up pretty quickly and very accurately, and some of these early machines even had fail-safes built in to detect tampering. Punch cards were already being used to modernize census taking,

so extending it to voting made a lot of sense. While these machines were an impressive leap forward, it was only a partial measure, really. Humans were still involved with the process. A savvy operator could simply swap out a stack of ballots before feeding them into a tabulator. Another issue came down to speed. Generally speaking, voting will take place at multiple polling places. Just look at elections in the US. They're often spread to multiple vote stations in each city. If you only have a handful of tabulators, then a lot of time will be spent ferrying around stacks of punched cards. Even if you have a tabulator installed in each location, well, you still need to combine all those totals.

These types of systems are a step up, but it can be taken further. I can't find out if Julius was aware of US punch card systems with the state of communication in the 1890s. I think it's likely he hadn't, but the sourcing's not that great, so we can't say for sure. But whatever the case, Julius took a much different approach to his voting machine. The basic idea was to have multiple voting booths all tied to a central system. That central component would count up each vote, and once booths closed, an accurate total would be reported. It sounds simple to us today, but how Julius went about designing his system is anything but. Now, not a whole lot is

known about Julius's early voting machine. The only source I can find about it comes from the man himself, and he's a little bit scant on details. We can't really examine the machine either, since, as Julius recollects, quote, the government had changed its mind again and wanted another system. I was so discouraged that I broke up the machine in disgust and abandoned further End quote. So, in a fit of frustration at some bureaucrats, he broke apart the only example of his voting machine. To me, that paints a picture of a bit of a mad scientist. And, well, he may have had the right to be a little bit mad. Inferring from his later work, the voting machine would have been a pretty large and complex device, but all this

genius that went into it was lost on the Commonwealth government. Julius was done with voting, but the machine he built would stick around in his head, even though it may have been smashed to pieces. The details on how he took his next jump are also pretty thin. Julius provides about one sentence to account for it, so if you'll allow me some creative license, I like to imagine it something like this. George is standing in his workshop, in his hand a large wrench. The wreckage of his voting machine is strewn all over the floor. Just as he finishes dismantling and crushing the last gearbox, a friend walks in and shouts something to the effect of, man, you gotta calm down. How about we go to the races? Now, I've already touched on the technical side of horse racing, at least as far as gambling goes.

I've already touched on the technical side of horse racing, at least as far as gambling goes. But there's another cultural side. Bob Duran, a professor at the University of Auckland, gave a great explanation of why horse racing was a big deal in a talk at the Computer History Museum. To boil that down considerably, it all came to time and place. At the turn of the 20th century, before World War I and the Depression, the middle class was on the rise. A lot more people were finding themselves with disposable income. If you lived in the big city, there were myriad outlets for this extra cash. But not really so in the boonies. When Julius broke apart his voting machine, he was living in New Zealand, which even today is pretty far out of the way.

There wasn't a whole lot to do, but there were racetracks. Attending races and gambling on races became an accessible pastime for a good number of residents. I can only imagine that it took a little cajoling from his friend, but eventually Julius made his way over to a track to try and relax. And as with a lot of more technical-minded individuals, Julius wasn't that interested in the leisure aspect. Instead of enjoying a day at the races, his attention was drawn to the mechanisms being used to manage bets. By this point, most likely sometime around 1909 or 1910, the racetrack had automated away some of the annoyances of parimutuel betting.

These early machines were called totalizers, or totalizers with a Z, or totalizators, or toteboards, or just totes. The sources I've read seem to disagree greatly on the canonical name or kind of shift back and forth between them, which makes researching the topic a little bit annoying. to shift back and forth between them, which makes researching the topic a little bit annoying. But no matter what you call them, there's no dressing up the fact that early totes were very primitive. It may actually be a little bit grandiose to even call these totes machines at all. They were still mostly human-powered. Each racetrack tended to have its own setup, so details varied, but the example that Julius came into contact with on that fateful day was a so-called jam tin tote. At the track was a series of betting booths, where gamblers bought tickets from a small army of clerks.

When a ticket was issued, a clerk grabbed a small metal ball and dropped it into a tube that corresponded to the horse that had been bet on. The overall pool and number of bets for each horses was totaled by simply counting the number of balls going through each tube. Odds were then manually calculated by a human. Above the betting windows was a large odds board. It displayed all the grand totals, and it had to be updated manually. From the outside, it looked like a complex machine, but really it was just a building full of frantic people and maybe a couple desktop calculators.

Julius quickly realized that he'd seen this type of problem before, and recently. Placing bets was really just the same as casting a vote. The only major difference between parimutuel betting and election comes down to a matter of a little bit of extra math. betting, and election comes down to a matter of a little bit of extra math. More specifically, these early marble totes bear a striking resemblance to how some Greek city-states ran elections in antiquity, just swap out the tubes and balls for pots and stones. The way Julius saw it, his failed voting machine could be rebuilt and remixed, so to speak, into a tote machine. And thus, he set about designing an ungodly complex device. What he ended up with was a rudimentary analog computer,

a type of device that not many today are familiar with. The overall scope of work for a tote can be broken down into a few pretty simple steps. You need to increment a counter every time a bet is placed. You need some way to select every time a bet is placed. You need some way to select which counter is being incremented, but also bump up the total counter at the same time. You need a way to divide each of those counters by the total to get the odds for each racehorse. And you need to be able to handle inputs from many multiple stations. The only difference from a voting machine turns out to be the addition of division.

I want to go over each of these and give a little bit of an explanation because here the devil is really in the details. The first of Julius's totalizers would be completed in 1913 and installed at the Ellerslie racecourse in Auckland, New Zealand. This machine, initially called the Premier Totalizator, gives us a good overview of what we're looking at. The racetrack had 30 betting stations. Each of these were outfitted with a data entry machine that fed information back to a central building that held the brains of the operation. The main building was packed with a mechanical computer for processing incoming data. It had a large tote board for displaying odds to gamblers also. On its first

day in operation, around 83,000 bets were placed using the machine. So, just keep in mind that nothing here is on a small scale. Julius wasn't the first person to create a mechanical computer. It wasn't a common occurrence, but there had been a history of these types of machines already. The earliest example is probably Charles Babbage's Difference Engine. This was a mechanical computer designed by Babbage in the early years of the 19th century, with the first working model completed in 1822. In the intervening years, a number of similar devices had been created, so there was at least some kind of precedence. What exactly were these devices like, and what were they really even

capable of? First and foremost, these are specialized computers. That is, if we choose to call them computers at all. I think we should, but some people will argue the point. You didn't program these kinds of machines. They were built to do one task. In Julius's case, that was parimutuel gambling, or in Babbage's, it was solving polynomial equations. In operation, you gave the machines inputs and you got outputs. But what it actually did was always the same every time you ran it. You could think of the machine itself as sort of a program. This is one of the major ways that these systems differ from what we're used to using today. They were very complex and very powerful number-crunching machines,

but they didn't have the same kind of flexibility of later computers. Once built, it was very, very hard to change what it did. You basically had to make a new machine. Another distinction that probably won't matter for most people but I find interesting is that these were not logic-based computers. The smallest unit of operation for these early analog systems all came down to mathematical functions, whereas later computers are built up from much more simple logic operations. It's a little bit of a fine distinction, but it shakes out to make analog computers

only good at math. Anything else, and you get into some pretty involved and pretty weird territory. Numbers are also handled in a totally different way. Digital wasn't even really in the vocabulary yet. Well, unless you were a very specific kind of math nerd at this time. Instead, numbers were represented using analog means. The details would vary a little, but the core concept remained constant. Julius opted to encode numbers based on the rotation of a shaft. It was set up such that every 36 degrees represented a number from 0 to 9. To add 1, a gear on the shaft was engaged, causing a 36 degree rotation. To represent larger numbers, you need more shafts, one for each digit. Then, everything's set up so that once a shaft rolls

over from 9 to 0, the next shaft is incremented by 1. It's really similar to those old flip clocks. As the smaller value digit rolls over, the next larger digit rotates 36 degrees. If you chain together, say, four of these shaft adders, then you can handle numbers as large as 9,999. This type of adding machine is called, rather fittingly, a shaft adder. In general, that's how an analog machine would operate. Need to add four to five, just rotate the proper shaft four times, then rotate it another five. 4 to 5, just rotate the proper shaft 4 times, then rotate it another 5. When done, you should see a 9 somewhere. That works for a desktop calculator, but Julius had to take a bit of a departure from the norm. The specific problem he faced was in parallel inputs. In other words,

how do you get an accurate count if someone adds 4 and 5 to the counter, but they add those numbers at the exact same time. In simple calculators, this just couldn't happen. You're forced to input each number in serial, one at a time. A single person can't really jam their fingers fast enough to get simultaneous inputs. But Julius, well, he wasn't designing machines for just one person. In practice, many operators would have to share the system. Each window where bets were placed would be sending back information to a shared calculator. He had to find a way to deal with simultaneous inputs. He had to make a system that could be shared. Now, shared infrastructure is very common today. I've been hammering on this point pretty hard,

but the internet is one easy example. Even something as mundane as a telephone uses shared infrastructure. You don't have a switchboard in your home, but you can pick up a phone and connect up to a shared switchboard. In the realm of digital electronics, this is really simple to deal with. Using specialized encoding, you can send many multiple signals over the same wire, and then on the other end you just have to have a program to decode them. But analog, and especially mechanical analog, that gets really, really complicated. Julius's tote handles this using a differential. Now, if you own a car, you should at least have a passing familiarity with that word.

In something like a car, a differential is a component that controls how fast each wheel spins. It makes it so that the average rotational speed of the two wheels remains constant. The reason for this is to make turns work out better, since the inner wheel can slow down while the outer wheel speeds up. But in general, a differential can be built to handle more than just two inputs. You can make it as big as you want. The differential and Julius's totes were a good deal more complex than what you'd find in a normal car. His early designs could handle around 30 independent inputs, while he claimed that a larger one could handle up to a thousand or more.

In essence, it worked as a parallel-to-serial converter, safely taking in each separate input and then passing it up the chain to the central calculator, and without ever affecting the other inputs. Everything was physically connected with gears, so this type of device was essential, since otherwise a twist on one shaft at one input could crank another shaft that you didn't want to move. But here's the crazy part if a huge differential isn't crazy enough. So how were all these bedding stations connected up to the tote machine? I mean, we couldn't use something like TCP IP quite yet.

At any good-sized track, at least any track large enough to afford a mechanical tote, bedding booths were pretty well dispersed. Not everyone could get to the central tote, so there had to be some booths on all sides of the track. Julius's earliest machines were fully mechanical, so he couldn't use wires and electrical signals to send data from remote booths into the processor. Instead, a series of pulleys, wires, and tubes were used to mechanically connect everything together. So when a bet was placed, it was transmitted back by the pull of a wire.

This may be one of the earliest systems that remotely transmitted data from recording nodes to a central processor full stop. And I guess this takes us pretty naturally to the betting booths themselves. And another complicated device in its own right. At first, Julius' machines were only able to deal with bets for winning horses. This wasn't a limitation of the technology so much, just more a limitation of space. Each type of bet needed its own nearly isolated machine. You had to have a different belt to transmit the betting data, a different adder, a different section of the display board, even a different differential. It was possible,

but just came down to a big duplication of work. But while you couldn't bet on a horse placing in the top three, or any complicated bet type, you could gamble away your money at a number of different denominations. When placing a bet, you don't really go up and say, hello, I'd like $101 bets on horse number three. You'd probably just place a single $100 bet. In the parimutuel system, a $100 bet is really the same as $101 bets, but it's easier for people to think in terms of the actual value. But that raises an issue. terms of the actual value. But that raises an issue. How do you handle inputs that can range from $1 up to, say, $1,000? How Julius got around this is central to the design of the bedding booths. A clerk working one of these booths wasn't expected to take a belt, calculate the correct degrees of rotation, and then precisely advance it. Rather, Julius devised a somewhat simple machine to do the job. Now,

simple might not be the right term. On the outside, Julius's ticket machine looks very simple, deceptively so. The input is a swinging arm that rotates over a set of labeled holes. Each position is labeled with a horse number and ticket cost. An operator simply swings the arm to the correct location and depresses it, which transmits the bet back to the tote board and then spits out a ticket for the customer. On the inside is where things get much less simple. The ticket machine is actually a very fancy clutch connected up to a set of pulleys and wires, one pulley for each horse. When the arm is depressed, the clutch engages and spins the proper pulley the proper amount of time. For the smallest denomination bet, like a $1 bet,

the clutch only spins 36 degrees. A larger bet rotates more, so a $10 bet would spin one full rotation. To the central calculator, a $10 bet just looks like 10 $1 bets. All knowledge of what a ticket actually means is held in these little booths. Now we come to the last stop on a whirlwind tour of an early tote machine. And that's the actual tote board itself. In other words, what someone sees at the races. We've already seen how a lot of engineering behind the Totalizator is a mix of general innovations with more specific mechanical touches. Its input method is a perfect example. Julius had to solve the problem of parallel inputs to a normally serial device while dealing with the shortcomings of a mechanical

machine. The idea of sharing central infrastructure and building out a system to transform parallel data into a single stream of information, well, that's pretty generally applicable. But the actual implementation of the idea, an epicycle-geared differential, that's a lot more specific to Julius' application. The display board is another perfect example of this type of thinking. You see, Julius actually created a buffered display with built-in memory, but he did it using gears and springs.

But not all is as it seems. Working in the realm of digital, we're used to operations being pretty much instantaneous. There is some delay, but everything operates so fast that a human can't really tell the difference. The only real limit is how fast we can zip electrons around on a chip. Mechanical devices operate on a whole different timescale. Gears have this annoying little thing called inertia. This means that they take time to start moving, and they take time to stop moving. The amount of time is dependent on a handful of factors, but one of the big ones is

how much the gear actually weighs. A bigger gear will resist change in speed much more than a smaller gear. Julius's calculating unit used relatively small gearing, except for one component. The display board, the large sign that showed up-to-date odds to gamblers, that had to be pretty big. The digits displayed had to be readable from anywhere on the racetrack. With odds changing every time a bet was placed, the calculating unit chugged away pretty quickly. So much so that large shafts and gears, they couldn't actually keep up with that speed. It took a good amount of time to get the numbered shafts and display boards spinning.

And by the same token, it took a bit of time to get them to stop. Overshooting your goal wouldn't be all that hard to do. How Julius got around this is a pretty cool invention in its own right. Instead of driving the display directly, the output from central calculators were connected to the display via a spring. As the shaft from the calculator rotated, the spring was coiled up, and that stored each degree of rotation as energy. Then, the display gears were powered by the uncoiling of that spring. With this system, the calculator was able to rotate its input as fast as it needed to, while the larger display could unfurl at its own pace. The parallel here should be pretty clear. What Julius built was an

analog memory element. Remember that in analog town, every number is represented as a rotation on a shaft. So Julius is really just saving over a number but onto a spring. You can crank that spring up to whatever value you want, within reason that is, you can't break a spring, and then you release it to read off the value. He even calls this part of the tote machine a, quote, storage spring. So, in 1930, we're already seeing a very clear take on early computer memory. That's the basic, well, a little more than basic, but the general overview of how

Julius's automatic tote functioned. To sum everything up, he created a distributed computer with edge nodes that all reported to a central processing node. It had a mix of shared and separate infrastructure, mechanical networking, and even a very crude form of computer memory. So, what did this thing look like in practice? Was this clockwork tote actually ever used? I already mentioned that the Premier Totalizator was installed at the Ellerslie Racecourse in Auckland in around 1913. It had a central calculator, 30 betting stations, and one really big display board. So, how exactly did George Julius convince the racetrack to bet on an untested and radically new technology? Well, we don't actually have a solid source that explains the details.

From my research, it seems that there just aren't many primary sources surrounding this side of the business. Even George's own recollections go straight from working on a small-scale model to installing the premier tote. But we do have enough details that I think we can speculate. One factor came down to speed of operation. Laws surrounding gambling were pretty tight, especially so for horse racing. The codes at the time stated that the race couldn't start until all odds and total calculations for bets were finalized. Remember that parimutuel betting is pretty math intensive, so running

odds by hand could take some time. Given enough bets, a race ran the chance of being delayed. That was a frustrating outcome for everyone involved. But the premier tote could circumvent the problem completely. William Mackey, a worker at the Ellerslie track who would later go on to run its racing club, put it this way. The machine did not reduce the staff required to operate the totalizator. In fact, the reverse was the case, but it did enable the betting transactions to be recorded accurately to show immediately the investments being made

on various horses, and the machine could now be balanced in two minutes instead of the 10 or more it required under the former manual system. End quote. That's a massive reduction in time. And there is another byproduct of this improvement. That's throughput. This is something that you see in any computing system. As you can process a job faster, you start being able to just do more jobs in general. In the more specific case, being able to take

bets faster meant that you could take more bets. In 1916, the West Australian ran an article titled The Totalizator, a machine for Perth. Julius was already expanding beyond a single installation. A very similar tote was currently in the works at a racetrack in Perth, Australia. The article is mainly about the new installation, but it has a few jewels of information about the Auckland machine. Some few months ago, a machine practically the same as this was installed at Auckland, and from the first race in which it was used, it was hailed with delight, by press and public. In fact, a prominent Sydney sportsman who had only recently

returned from one of several trips to New Zealand said that the new machine had revolutionized wagering in the northern city. He stated that he found the old antipathy of the totalizer almost entirely eliminated, and expressed the opinion that nothing more satisfactory could be desired by anybody. Mr. Joss Brennan, president of the WA Trotting Association, yesterday received a telegram from Auckland, stating that during the four days meeting of the Auckland Jockey Club just concluded, the sum of £200,225.410 passed through the totalisator, to Mr. Brin's telegram, it worked splendidly throughout. End quote. The new automatic tote was fast and accurate, which meant that it could rake in a lot of money. With a little inflation and conversion math, the total mentioned for the four-day meet at Auckland translates to just a

little bit over $25 million. That's a whole lot of money, even if the racetrack only took a small percentage off the top. And I think that speaks to why racetracks ended up using a machine that the government wasn't able to. An election comes once a year, so it would be hard to justify the expense of a fleet of massively complex machines. Whereas, you can have races nearly every day of the week. With each race bringing in millions of dollars, a tote will just pay for itself very quickly. And in that capacity, the automatic tote excelled. So much so that in 1970, Julius formed Automatic Totalizators Limited, or ATL for short. From there, his clockwork computer took off.

By 1920, there were already 10 installations, with the largest networking 150 booths. But the premier totalizator wasn't perfect. I think Julius would have been the first to agree with that. These first few installations were very primitive. In a lot of ways, Auckland and Perth were getting a rough draft. They worked, but there was a lot of room for improvement. Space was one of the big issues. One of the overarching tides over time has been the slow march towards smaller computers.

As we rewind in time, we run into larger and larger mainframes. The Premiere Tote is a prime example of this. You can find pictures of this beast pretty easily, I'll post some around after this episode comes out, and you can see what I'm talking about for yourself. It took up an entire two-story building, with components stretching from the floorboards up past the rafters. The machines were so large that there's actually stairs and ladders set up to access different components. As far as weight goes, I haven't seen any figures, but I'd estimate in the range of tens of tons, maybe a lot more. The entire machine is built from metal, and going off images of preserved components, it looks like it's

mostly a combination of brass and steel. That is not going to be lightweight in any respect. Even the networking cables, if you can even call them that, took up a lot of space. The actual wires used to transfer bets from windows to calculator was made from piano wire. Each window had one wire per horse, so 30 windows and a max of 10 horses at the Auckland track gets you about 100 tensioned piano wires. Some installations ended up having miles of this wire inside them. The channels from bedding station to calculator and the final destination where the cables all connected up had to be laid out very carefully and with plenty of space. If lines got tangled, cut, or bent, the entire machine would go down. The other wild and

very dangerous part of the Premier Tote was its power source. Remember, Julius's device used no electricity. Instead, it was initially powered by falling weights. And this part of the Tote actually worked a lot like the clocks that George tinkered with as a child. The tote had a series of drive gears that provided the power for the thing to actually run. Slung over each gear is a chain, and attached to the end of that chain is a large weight. As the weight descends, the gear turns, thus providing power. Like I mentioned, this is basically the same system that powered old mechanical wall clocks, just on a much grander scale. And it works, at least in

general. But it doesn't work indefinitely. Once the weights reach the floor, you stop getting any power out of them. Julius' solution to that was to have one of the machine's operators run over, grab the large weight, and then scamper up a ladder to raise it back up. We're looking at a human-powered machine, so in addition to normal staff, a track had to make sure that they had a few pretty tough and burly people on their team. But for all its faults, the premier tote took off in New Zealand and Australia. Sure, it took about the same amount of people to operate one of these new machines, but they were fast, they were accurate, and they were massively reliable. And in the coming years, Julius would vastly improve on his early designs.

One of the first big changes was bringing electricity into the picture. As early as 1917, the transition got started. The first big upgrade was replacing the miles of piano wire with miles of electric wiring. Instead of pulling on a long band, ticketing machines now sent an electrical impulse down a wire. On the other end, that signal actuated a solenoid that advanced an input gear by the requisite number of degrees. It's a simple change, but I think it's interesting for one big reason. We're already getting into digital stuff. The actual data going between

bedding terminal and calculator was an electrical signal, an on or an off. In other words, a one or a zero. Once that reaches the calculator, a set of motors and gears turns that digital signal into an analog value on a shaft. Julius didn't call it an analog-to-digital converter, but that's just what it is. Sure, it's a super-limited one-bit converter, if we can even call it that. But there it is in 1917, tucked away at a racetrack. Over the next few decades, ATL installed a total of 74 automatic totalizers, ranging in complexity. The largest machine was built in 1928 for a track in Paris. It networked 273 terminals. The automatic totalizers that Julius designed and built

became the de facto system for racehorse betting. If you wanted speed and reliability, then there was only one place to go. Some installations would remain in constant use for close to 60 years. The technology was amazing, especially for the time. The engineering was superb, and it solved a very niche problem better than anyone else ever could. If these machines were so popular and long-lived, then why have they become so obscure? Let's go back to Doren Suede's main argument. In Forgotten Machines, they argue that computers like Julius's mechanical totalizers, well, they don't fit nicely into existing narratives of computing's history. It has features that are both analog and digital. It does some really modern things, but it uses really antiquated methods.

In other words, it's messy to classify. You can't easily plot these kinds of machines out on some master graph of the development of computers. So a lot of people either don't know about the totalizator, or if they do, they might not see it as relevant to computing's past. Another big reason for this comes down to the march of progress. Even for Julius and ATL, analog computers were just a step. The switch to electric networking was just one example. Over the years, ATL kept up with technology. Information gets a little sparse, but the last mechanical totalizer was

probably installed sometime in the 1950s. By then, computers were starting to appear, as in digital electronic computers. ATL had already started using more and more electronic components in their machines, and pretty soon they'd make the switch to more modern computers. Massive, building-sized clockwork mechanisms just weren't the order of the day. They were replaced with fleets of PDP-11s. Simple electrical impulses were replaced with Ethernet. Julius' mechanical marvels became just that, a marvel, but no longer useful. ATL would continue to play a big role in gambling, but their future, like everyone else's,

became much more digital. It's time to wrap up this episode. Analog computers hold a very strange place in the history of the discipline. It was hugely important early on. A lot of what made computers so revolutionary in the middle of the 20th century was started much earlier on mechanical monsters. But despite their importance, analog computers are often sidelined. George Julius' automatic tote is a prime example of that. He was dealing with problems that computer scientists would still grapple with nearly 50 years later. His intricate networks of bedding terminals and central calculators

are something that just shouldn't be possible in 1913. But they existed, and that's amazing. It wouldn't be until the 1940s that anyone would build a network on such a large scale. Ultimately, the rise of digital computers totally outmoded these early systems. But that doesn't mean that analog computers should be forgotten or misplaced. Just for some context, the Harvard Mark I, often held up as the first electronic digital computer, well, that was only in operation for four or so years, and there was only ever one of that machine. In that time, maybe a few dozen people worked on it directly, maybe a hundred if we're being generous. By today's standards,

it's totally obsolete technology, but countless papers, books, and documentaries have been written about it. By contrast, George Julius and ATL installed a whole heck of a lot of automatic totes. The machines operated for decades, with a rotating cast of employees, clerks, and engineers using them on a day-to-day business. But in my research, there's only actually a few lectures on totes that I can find. There's scant mentions in a handful of papers, and maybe half a dozen websites on the topic, if that. All things considered, that just doesn't seem right to me. Julius's work is breathtaking. It's as simple as that.

It pioneered technology far ahead of its time. The same can be said of many early analog computers. It may be easy to forget them in the large scale of computing history. But to do so, that's at your own peril. Thanks for listening to Advent of Computing. I'll be back in two weeks' time with another piece of the story of the computer. And hey, if you like the show, there's now a few ways you can support it. If you know someone else who's interested in computers past, then why not share the show with them? You can also rate and review the podcast on Apple Podcasts. And if you want to be a super fan, you can now support the show directly through Advent of Computing merch or signing up as a patron on Patreon.

Patrons get early access to episodes, polls for the direction of the show, and other assorted perks. You can find links to everything on my website, adventofcomputing.com. And if you have any comments or suggestions for a future episode, then go ahead and shoot me a tweet. I'm at AdventOfComp on Twitter. And as always, have a great rest of your day.