Advent of Computing - Episode 183 - A Digital Gap?

Episode Date: May 31, 2026

I've been browsing old compur surveys and trying to build up a comprehensive data set. What I've found is a little surprising: between late 1945 and 1949 only 10 new computers entered service. Once we get to the 50s that number explodes. What's going on here? What caused the gap between the first digital machines and the explosion of computers in the 50s? In this episode I try to answer that question by finding out just what was going on during this digital gap. Like Advent of Computing? Then check out the after show! Adjunct of Computing is now LIVE: YouTube Spotify Apple Podcasts

At my somewhat advanced age, or what often feels like it, it seems only yesterday, but it was actually 1937 that I liberated some relays from a scrap pile at Bell Telephone Laboratories, where I then worked, and took them home to start what I thought of as a play project. I had observed the similarity between the circuit paths through relays and the binary notation for numbers and had an idea I wanted to work out. George Stibbitts, 1967, via Datamation magazine. If you read enough about the earliest digital pioneers, you'll find a few commonalities.

One huge one is that none of them intended to build a computer. That word, at least in the way we use it today, is an anachronism. Someone in that period, if you said computer, well, they would have either thought of an analog computer or a clerical worker with a desktop calculator. Stibbitts, sitting in his kitchen table in 1937, had no idea what an electronic digital computer was. Those machines didn't exist. If you had a time machine and went back to his kitchen and told him, well, you're building an electronic digital computer, he probably wouldn't understand exactly what you meant.

The same goes for other pioneers. In 1938, John At Nassoff was doing roughly the same thing as Stibbets. He wanted a machine to solve linear equations and wasn't having any luck with what was already out there. Over a series of attempts and at least one hard-drinking night that may have involved drunk driving, At Nassoff came to the conclusion that he would have to make a brand new digital computing machine. And those were his words, a digital computing machine. you know, as opposed to the analog ones and the human ones that he knew about. Even when these early pioneers used the term computer, it carried a very different connotation.

There are flashes of the future here, sure. At Nasov captures the rough outline of digital computation with wild accuracy. Stibitz has some piercing insight into human interfaces. but these machines weren't anything like what we're familiar with. There wasn't any one point when the computer was invented. We get jumps and starts along the way. No one quite knew where they were headed. And that makes these early days of the digital computer a truly wild west.

Welcome back to advent of computing. I'm your host, Sean Hass, and this is episode 188,000. A Digital Gap? Before we get started, I have a few plugs to make today. First, have you started listening to Adjunct of Computing yet? It's the official after show for Advent of Computing, and it's a good time. If you like the show, then you're going to really like Adjunct. I'll have some links in the description, and you can just find it wherever you're listening

to Advent of Computing right now. The other plug is, I've finally committed. I'm going back to VCF West this August. I already have tentative plans for my talk this year. I think I'm going to do a roundup of all the small and obscure Unix-like things I've been talking about for a while. There's been some new information that's come to light, and I think doing a full overview of all of them together will be a really neat narrative. There's some ideas I have around that that I hope will be enjoyable. Anyway, V-CF West is happening August 1st and 2nd this year in Mountain View, California at the Computer History Museum.

V-CF is always a great event. I always have a fabulous time at West, and it's really fun to see all of the listeners that show up. So, if you're a listener, I know you are if you're hearing this, and you want to come hang out, see me get a little freaked out about weird Unix-like things and meet a lot of other retro enthusiasts. I highly recommend you come down in August. With that, let's get on with the rest of the show. Today, I got a bit of a weird one, a call it a philosophical episode.

I want to solve a problem. But first, I have to actually explain what that problem is. If you're a longtime listener, you may notice how I like to put things into context. I think that's kind of the role of advent of computing. Part of that is talking about scale. I'll often point out how big, the field of computing is at a given time. If we're talking about programming, then it really

makes sense to put that in context of, say, how many programmers exist. The conversation about what makes a programming language important or good is totally different if we're talking about 1945, 1955, or 1995. There's a problem there, though. You see, it's really tricky to say how large the workforce is. So I tend to use a proxy. I'll usually use a rough estimate of the number of computers that existed at that point in time.

Now, that may sound like it would be a nice concrete number, but well, it's not, not really. Early on, there's this wiggle room of what defines a computer, what exactly counts towards the full number. There's also the problem of incomplete information.

I haven't been able to find a good, comprehensive list of every computer ever made with years and dates and everything. I know. It's tragic. I've been trying to get a little tighter in this regard. I want to have a good understanding of how large the field of computing is during these early years, when the field's still small enough that you can get a good handle on the whole thing. So I've been reading old reports and surveys on the industry. In 1953, the Office of Naval research, put out a survey on every computer they could find information on. It's not complete, but it's a good starting point. The Department of Commerce did a very similar survey in 1955, and even later, we get some

surveys from places like the NSA that can be used to fill in a few of the blanks, but again, none of these are comprehensive. As I've been reading and compiling notes, I've noticed a few things. First is that the number of machines spikes dramatically around about 1952. Between 52 and 53, we see around double the number of types of computers in circulation. We go from around 40 to just over 80. I'm going off that odd metric, number of types of machines, because it's also very, very hard to get counts for mass-produced computers in specific years.

I don't have a table of all the sales of IBM 701s by year and month, for instance. But by just looking at a number of kinds of machines out there, that is a good enough indication of how large the field is. Now, the second thing I've noticed is a possible gap. My data seems to indicate that between 1946 and 1949, only 10 new computers entered service. I've alluded to it before and I kind of understood that it took a long time to develop a new computer back in this period. But I didn't realize the implication.

Very few computers were completed in that period. At least that's what the raw data looks like. But data only gives you a rough idea. So this week, I'm hitting the books. I want to figure out a few things here. First of all, is this gap real? Was there a lag between the first handful of computers and actual new machines proliferating? Or is this just an issue of my data being incomplete?

If the gap is real, I want to understand why it existed. Is this the result of secrecy around early machines, or was development really just that slow? By the end, I hope to have a better understanding of just what happened directly after the advent of the first computers. I think it's only fair to start with what is going on in this period digitally. The entire show this time will basically be restricted to late 1945 to 1949. I'm going to basically say we start when Enniak runs the first Monte Carlo algorithm,

and we end when Bynac is delivered, even if it may or may not have been working. As soon as you really dip into 1950, we start to see recognizable things like Whirlwind. I don't want to go there. We've talked about computers we recognize. I want to get to the weirdness in this gray zone. So what do we get in this four-year slice? Well, like I said, it's not much you're likely to recognize. The most well-known that fits in the middle of this pack is the Manchester Baby.

That's widely claimed to be the first operational stored program computer. It was an experimental machine made at the University of Manchester, and it started running in 1948. The only other big player in this period is EdSack. But if you're listening to a podcast that reviews sourcing on 1940s computers, you probably know about EdSack already, so I'm going to skip that one. It's more of a blip in the middle of this field. What we have left are outliers.

We get a few of Bell's relay computers, some of IBM's machines that were almost programmable, and then, oh, then we get the real oddities. One is actually Cizirac, which I've been told I'm pronouncing incorrectly, but I'm sorry. I'm not changing. Cisirac is the first computer built in Australia. We've talked about that at length and about how different that machine is. You can go see the episode of my archive for the full rundown. going further a field gets, well, it continues to get dicey. One of the very early machines was ARC, the automatic relay computer.

However, it's not exactly clear when it was completed and if it was ever fully working itself. Arc gets developed into the Apex C machines. That's A-P-E-X-C. Those spin up in the early 50s. I'm going to focus us in on one of these obscure machines in particular to try and get an idea for what the period is like. The victim of choice is ICCE, the Imperial College computing engine. Ever heard of it?

I doubt it. I'd be surprised. I hadn't until now. ICCE is completed in 1949, so it's just within our time window, or at least part of it is running in 49. This is going to be a very illuminating example of why things are tough in this period. The story of ICCE starts in the Math Department of the Imperial College of Science and Technology in London, England, as opposed to London, Texas. The project was started in 1947 by one Dr. Tony Brooker. Our most detailed history of ICCE comes to us from an introduction that Brooker wrote to a re-release of one of the,

original papers on the computer. How's that for a, for an incidental paper trail? Things start innocently enough. Brooker and another professor, Gordon Barnard, were talking about how frustrating it was to use desktop calculators of the day and how automation, oh, that'd be pretty slick. To quote, George felt that although we hadn't got the resources to join the pioneers, some useful progress could be made if ordinary desktop calculators could be linked together in some way, so that results produced by one calculator could be automatically input to another. That actually rang a bit of a bell for me, specifically a Bell Labs. Around 1940, some researchers of Bell Labs were trying to tackle a very similar problem,

but from a different context. They wanted to automate complex mathematics. The main issue was actually transferring around numbers and storing information. So if someone at Bell proposed, they just link two mechanical calculators together. The idea was rejected at Bell for a pretty funny reason. At the time, there was only one type of calculator. They were all mechanical. It actually proved very, very difficult to transfer numbers from one machine to another via mechanical linkage. You have to use clutches and motors.

It doesn't work out well. You have to have an electronic calculator to make this easy. The path eventually led to Bell's complex number calculator, or the Bell Model 1, see Episode 174 for the Sorted Tale. Anyway, Brooker is on a very similar path. The key difference is that in 1947, technology is different. Two key technologies now exist. the computer and the sequence-controlled calculator.

These are basically cousins. The only difference is that a computer is turning complete. A sequence-controlled calculator isn't quite there. It's more like a calculator that you can give a list of commands to. Brooker and his colleagues had access to some of the early papers on computing, so they saw what was coming. They also watched as other researchers in England took stabs at making their own machines. It didn't take long for Berker to want to give it a try himself.

He initially planned to do something similar to a sequence-controlled calculator. To start that, you need to build an electronic calculator. That, on its own, is a huge task. Papers like the draft report on EDVAC were floating around, but that didn't tell you how to wire up logic elements to do math. It gave you vague ideas about how to make a computer and how a computer would function. So Brooker was really on his own.

It was also viewed as a very costly and perhaps wasteful project. Brooker settled on using relays for his machine for two reasons. He knew how relays worked, and with post-war supply chains, it would be impossible to get vacuum tubes. So he went around town looking for hundreds of relays. The best source ended up being a surplus shop. The total bill came to 120 pounds, due of some inflation and conversion, that's around $5,600. Brooker had to basically pull teeth to get the math department at his college to pay for that.

I think the fight for funding alone illustrates a crucial point. In 1947, it was not clear to everyone that electronic digital computers were the future. Heck, it wasn't even clear that the future was digital. or that anyone should invest in that. If the math department had a crystal ball, which, if they're being stingy about some relays, they couldn't afford a crystal ball, but let's assume they were gifted it as a legacy.

If they had a crystal ball and could see the future, then yeah, they'd understand that this was a good investment, that what Brooker was making was very promising. But there were few that could see that future. Crystal balls are very expensive. We see the same pattern in the states. The U.S. government is, oddly enough, the visionary force here, at least for finances. Due to federal investment in computing during World War II, the feds were excited to get more computers.

That led to funding for projects like whirlwind, hurricane, and cyclone. It leads to UNIVAC and a pile of other semi-commerical. machines, but I'm getting ahead of myself. Over the next few years, Brooker built his electronic calculator. In 1949, it was working. But this wasn't quite a full computer. What he had produced was the ALU, the arithmetic logic unit. That's the circuit that actually carries out commands. It does computing. It worked, but getting there was a struggle, everything had to be done from scratch. There was no book to go off of. Brooker did, however, get one singular break to quote, and by a great stroke of luck, we discovered the blueprint of a circuit for a relay adder in the

drawer of a desk in the department library, end quote. That just show you how early and how challenging this was. Everyone was working from the ground up. Prior art was basically scraps of paper, and a small amount of public information, plus, if you were really, really astute, some early work on binary logic formulation. Information about ENIAC was disclosed in 46, but that did not include circuit diagrams. And besides, even if you had ENIAC circuit diagrams, those bear little resemblance to a working programmable computer. The hard part for me is, does this ALU meet the criteria of a computer? Does this go on the list as a computer operating in 1949?

That is a very tricky question. An ALU on its own is not turning complete. So therefore, it's not a computer, at least not by the book. There's no way for it to carry out a series of instructions, let alone, decide what to do next. It just can't work. But zoom out for a second. Forget the definition and look at the larger context. Iniak is considered one of the first computers. Its date of first operation is pegged to 1945. That's when the first Monte Carlo simulation is carried out on it, just a few months after the bombings of Hiroshima and Nagasaki. It's very technically

turning complete. But it's not a stored program computer. It doesn't have memory that can store instructions. Everything is wired by hand. In that sense, it doesn't really have conditional execution because it's not really executing anything. In 1947, INIAC is expanded to allow for some stored program capability. So, do we say that INIAC wasn't a computer in 1945? Oh, but it was in 1947? There's no satisfying answer here. At least, I have yet to find an answer that satisfies me. Researchers are working towards computers in this period, but at what point does a very complicated circuit become a very simple computer? This is one of the reasons that I approach my discussion of firsts with a lot

of caveats in context. Timing around the early development of computers, especially unique machines, is fuzzy at best. So, by 1949, the ICCE. is almost a computer. The process to get there from nothing up to an ALU took two years. Once we get to an ALU, Brooker actually steps back and another researcher takes up the baton. That researcher, Ken Tusher, finishes turning ICCE into a fully stored program computer. It was fully described in the fall of 1952. The final step took three more years, so in total,

ICCE took five years to construct. That's from just a vague idea to a fully turning complete computer. An interesting point is that by 1949, the field had already begun to change. Brooker wasn't totally isolated. He knew of other projects, but wasn't exactly drawing deep influence from them. Too sure, at this point, was able to be in dialogue with the rest of the field from Brooker, quote, I can still recall his acerbic comment on one distinguished pioneer. Of course, Mr. M never aimed to build the world's best computer, but rather the world's first computer. Kin deliberately sought the perfect architecture for the limited source at his disposal.

The high cost of storage explains why Kin placed great importance on using every bit in an instruction bite. Some would say that M was a pragmatist. Ken was an idealist, end quote. We can only assume that the M is one John Mouchley of Inniak fame. It's kind of funny to imagine a British dude bent out of shape about John Mouchley, but there's something important here. Tushar knows enough about Mowchley's work to critique it.

He doesn't just know other computers exist. He's an active dialogue with their design. By the time we reached 1950, things are plain different. Some machines already exist. Papers have been published and people are talking. There is, after all, something to actually talk about. In 1953, a book called Faster Than Thought is published. It's one of the first books about computers to hit print

and probably the first book on computer history. It has a pile of chapters about all the computers in operation in England at the time, including one co-written by Tushar about I-C-C-E. You see, that's dialogue, not just someone scrounging for a circuit diagram in a desk drawer after they saw a single article about a computing machine. Anyway, the forward to faster than thought illustrates this point perfectly, to quote, some four years ago finding myself under the necessity of acquiring certain limited knowledge of how electronic digital computers functioned, I turned to the patent literature as the only comprehensive digest of

information on the subject. It continues, I found the mental struggle a severe one, end quote. As someone else who's read many of these early patents, correct, this effect is reproducible. Four years previous would have been 1949, right as ICCE's math circuits were completed. So let me throw down hypothesis number one for this gap. It just took that long to make a computer. We see the gap because folk start more digital projects directly after World War II. But it took a while for things to spin up. Because there weren't many functioning machines yet, there wasn't enough material for researchers

to work from. So everyone has to start from scratch. That ends up taking maybe four to six years before you have a programmable computer. I think I can get behind this one. In 1945, there were no stored programmed computers that were turning complete. It wasn't just a matter of making a machine like ENIAC that was barely programmable. Researchers all jumped at simple stored program computers and sequence-controlled calculators. That, plus the early state of the art, introduced a bit of a lull while projects were completed. We end up with a bunch of computers in the early 50s

because that's just how long it took these post-war projects to wrap. We end up with more computers in the next few years because many of these early problems were solved. There was a pattern to work from. You didn't have to consult patents or scrounge for parts, so the snowball rolled. But that's only one angle. What else was going on in this period?

There are a number of machines that are cousins to the electronic digital computer. I've already mentioned one, the sequence controlled calculator. IBM produced quite a few of these SCCs in this gap. There's another part of the family tree to consider, the electronic analog computer. The history of electronic analog machines parallels the development of their digital cousins. We also have some fuzziness around the actual birthday for this type of machine. The first definite electronic analog computer shows up at the end of the Second World War. But there were mechanical analog computers that had electronic components since way back.

As early as the 1930s, we get differential analyzers that used electronic motors. but let's not go there. That's an argument best saved for another venue. A key difference between the analog and the digital was programmability. In 1945, it's still not clear that programmability is all it's cracked up to be. The church-turning thesis is already out in the open. That boils down to the whole compatibility thing that I always harp on. If a computer is turning complete, then it can be used to do anything that any other

turning complete computer can do. Programmability opens up a huge world of possibilities. In the earliest epoch, the size of that world was not very clear. There are actually a number of arguments made against programmability. They all boil down to, essentially,

why bother? At that point, there were many single-purpose machines already in active use, like, for instance, the tape comparators that we talked about last episode. Those machines were basically computers just not programmable. Punchcard tabulators also fall into this category. They're a totally special purpose and can be configured to do just about anything you want, but they don't have that programmability piece. They don't have the

superpower lurking inside. Electronic analog computers fit into the same mold. It's hard to talk about analog machines because they had a very wide range of capabilities. But in general, they didn't take steps. Everything was done at once, so it wasn't really possible for them to make choices about what to execute next. There wasn't a concept of next or a concept of execution, really. So they weren't turning complete. They weren't even programmable. Rather, they were configurable, usually via a plugboard, but there were exceptions. Again, very wide field. Go read hybrid computing by corn and corn if you want to fully enter the psychosis. I don't recommend it unless you're super interested.

Now, this doesn't mean that analog computers were viewed as weaker or less useful machines. Remember, programmability wasn't a proven solution. It was an option that was new and exciting, but not everyone was convinced. Case in point, the U.S. Navy. I haven't seen a definitive name for this codename scheme, so I'm just going to keep calling them the foul weather projects. Starting in 1946, the Office of Naval Research pumped money into computing. That's how we get Project Whirlwind, which leads to MIT's Whirlwind One computer.

That's one of the most influential machines ever built for a number of reasons. We also get Project Hurricane, which leads to Ravak, which exists. That's fine. We also also get Project Cyclone and Project Typhoon, both which lead to analog computers. The same forces that jumpstart the digital computer revolution are also interested in analog computing. That should tell you something about the importance of analog computers in the 1940s. Now, the path to Project Cyclone is far from clear. The foul weather projects had, well, let's call them varied success.

Project Whirlwind went very, very well, so it's very well documented. Project Hurricane didn't go so hot, so there were like five papers I can find on it. Project Typhoon occurred. Next paragraph. Cyclone is a bit of an odd one out here because it was successful. and it lasted a long time, but it's not super clear what its original intent was. Cyclone ran into the late 1960s and funded a pile of related research. The main issue is I don't have a single document to point to that describes the contract for Project Cyclone.

If only the government could be that transparent. Anyway, Project Cyclone started off as an agreement between the O&R and R, and Reeves Instrument Corporation. According to a 1953 paper published by Reeves, quote, the primary function of Project Cyclone is the development and operation of a guided missile simulator and the establishment and operation of a simulation laboratory. Problems in aerodynamics, engine control, aircraft stability, dynamics, and navigation have also been studied with the aid of the computing facilities of the simulation laboratory, end quote.

That sounds crystal clear, right? But there's a catch, a twist, if you will, and this is why I really wish I had the original contract. During Project Cyclone, Reeves develops this machine called React, the Reeves Electronic Analog Computer. But they first attempted to develop a computer called ReeVAC, which would have been a programmable computer based directly off the design of EDVAC. In fact, Reevec was going to be part of a lineage. It was designed by Samuel Lubkin,

who had previously worked at the Moore School. You know, the home of INniac and EDVAC. The Reeves computer was going to be an improvement on the draft design of EDVAC. Reeves went public with the project in 1948. By May of 48, the digital computer idea was scrapped. Lubkin left Reeves and went to work for the National Bureau of Standards. While there, he worked on the design and construction of SAC, another very early digital computer. After the split, Reeves went on to construct their analog machine, React, which they would produce in quite a good number of machines. Note the timing. Project Cyclone started in 1946. That's when the O&R and Reeves entered into a contract to do something. Again, I don't have the contract, and that makes me sad. But let's just trust what Reeves

was saying in the 1950s, that the O&R wanted machines for work around guided missiles and simulation. That mandate doesn't say anything about what kind of machines the Navy wanted. That means Reeves could plot their own path. All the evidence points to Reeves attempting a digital solution, dropping it and then going with an analog computer instead. Was this seen as a step backwards? Well, in 1948, I don't think it would have been. Electronic analog computers were just as new as digital computers. They are both new and promising technologies.

Reeves wasn't pivoting to make an easier machine. They weren't under-delivering. We know that for a fact. Project Cyclone continued for decades. The Navy didn't dump the contract when Reeves came back with React instead of revak. A similar decision was made during Project Whirlwind. The mandate of that project was to create an aircraft simulator.

The contract didn't say the Navy wanted a computer. It stated a problem to be solved. The MIT team's first swing was an analog computer. That ran into some issues, so they switched to a digital computer. If nothing else, this should show how both analog and digital computers could solve the same issues. They were both capable. It was a matter of picking the right tool for the job. Reeves chose analog, MIT chose digital.

So let me hit you with Hypothesis 2. We see a gap in the development of new digital computers in this period because there was an alternative technology. Electronic analog computers were still showing considerable promise. It was not clear that the exact digital formula we know and love was correct. So, therefore, the energy going into computing was spread out among a few different options. I like this hypothesis because it refutes something I dislike very heartily. I never liked the idea that digital computers just burst onto the scene and took over.

That just doesn't sit well with me. It doesn't make sense. It must have taken some time. There must have been a window where digital computers were fighting it out against some other thing. Well, maybe that window was the late 1940s, and that other thing was the electronic analog computer. I can further complicate matters. I keep mentioning some special purpose machines. What exactly is the deal with them? There are some really funny examples from further back in the day, the totalisator board used for horse bedding and other neat single-purpose machines, but that's too far back.

There were also some early golf simulators, but again, that's 1930s technology. In fact, we just covered the perfect example last episode. Goldberg and Demon were both special-purpose cryptography machines. They were nearly as complicated as programmable computers. They just weren't programmable. There are a number of cool NSA reports about the history of computing devices at that agency. What's crucial to note is that those reports group single-purpose machines like Goldberg and Demon with general-purpose computers.

One report recounts the history of both devices in the same paper. Today we put a pretty rigid delineation between what's a computer and what's not a computer. Either it's turn-complete or it's not. That's not the case in the earliest epochs. The line is more blurred. I mean, these NSA reports are from the 1950s and the line is still somewhat blurred. Electronic analog computers, single-purpose machines, and general-purpose stored program computers are all computing devices.

They're all machines that serve. the same purpose. It's only as time goes on and as stored program machines prove their usefulness, that we realize that's the path forward, that that's the lineage we should invest in. But in the 1940s, and even into the early 50s, that is not clear. There's another angle to this possible gap. Have you considered just how long it took to make a computer from scratch? I know I said this before, but But stick with me here. It took years and years. Univac took five years to develop. So even if a lab were able to start in on a computer as soon as possible, say, 1946, it's unlikely that they'd have

a computer ready by the end of the decade. That's part of what we're seeing. I'm increasingly sure of it. There are a lot of projects that start during this gap, but don't make it to completion. That leads, I think, pretty naturally to the next line of inquiry. What if someone was working on a computer prior to 1946? I'm pegging 1946 as the year here because that's when the EDVAC report is leaked, and it's when disclosure about ENIAC is made public. That's when folk outside of a handful of labs would suddenly know that these digital computer things existed, and, hey, maybe programmability is something interesting to look into.

Many first-generation machines are inspired or based off the EDVAC report, after all. Then, who knew about computers prior to disclosure? Well, there was Conrad Zeus, but he was doing his own thing. Actually, I think in 1946, he was still hiding inside a farmhouse. There were the researchers at the Moore School. There was the lab at Harvard, a handful of researchers at IBM, and you could probably count some folk at Bletchley Park. The Moore School was tied up with EDVAC and then brain drained, a pile of researchers left to join the private sector or other digital projects during the gap years. IBM didn't introduce anything turning complete until 1950 or so.

Bletchley Park I'm no expert on, so I refuse to comment something something pilot ace, I believe. That leaves Harvard, which actually used their head start. In 1947, they completed the Mark 2 computer. That's neat and all, but I want to take a slightly deeper cut here.

There's a dark horse in the race, that is, Bell Laboratories. Bell had been developing computing machines since way back. Their first, the complex number calculator, was completed in 1939. So they were very much in the field. In 1946, Bell Labs turned on their new Model 5 computer. If we want to talk about Head Starts, then Bell is our best example. They were working in the field before the word binary existed, or at least was in common use. A number of relay computers were developed. developed at Bell during this period. George Stibbitts, the force behind these machines,

called the Model 5 the first step towards a modern computer. But this was a step on an existing path. As early as 1937, Stibbitts have been thinking about digital computers. His first complete machine was the complex number calculator, which gets renamed the Model 1 after another machine is constructed. The management angle here is a little complex. The series of machines follow Stibbitts, not necessarily Bell. To quote via a series of articles in datamation, the very ill wind of World War II blew me some good in this area,

where I was loaned by Bell Laboratories to the National Defense Research Committee, NDRC, end quote. It's while working at NDRC that Stibbitts starts his second computer, the Model 2. This was a really based machine like the previous model. It was also a somewhat special purpose machine. It was designed to solve fire control problems, and it wasn't exactly programmable, per se, at least not in any way we'd recognize.

The Model 2 read instructions from paper tape loops. A program would consist of instructions punched onto paper tape, to make it run repeatedly, the ends of the tape were glued together. That way, when the final instruction was reached, the computer would advance to the end back up to the first instruction. I've been trying to find any deeper information on the Model 2, but I've met with not much luck, so I can't give you an instruction listing. What I can tell you is that it was used like any other early computer.

Problems were programmed, time was scheduled, and the thing ran for hours and hours on end. In that sense, the concept of conditional transfers doesn't really matter. The Model 2 was filling the role of a computer. Oh, and I should add the year here, shouldn't I? The model 2 started clicking away in 1943. That's the same year the ENIAC project started. Stibbitts was way ahead of the curve. As the years went on, new models were constructed.

The model 3 was completed in 1944, that added the notion of subroutines. These were tape loops that could be called up by the main program. It also added these things called tape tables. These were kind of like a read-only memory. For this to make sense, you need to kind of reorient your view of a computer. The Model 2 had a pile of tape readers and punches. Some were dedicated to these tape tables. The tapes themselves were a simple punch-paper tape,

so read-only storage. You can't really rewrite them. Data was broken into blocks. Each block was addressed. That means you could call up a certain chunk of data buy its address. There was a little technical trick that made this all possible. A hunt unit.

That's right. We found another one. The Model 3 could read tape both forwards and backwards. When you gave it an address, it would see what the current tape address was and then decide which direction it needed to scan in. That, yet again, is a level of sophistication you wouldn't expect to see in this period. But there's one crucial thing the Model 3 still doesn't have. memory. It has registers. It has all these paper tapes, but it doesn't have addressable memory.

That means it doesn't fit our more modern definition of a stored program computer. There really isn't a place to store a program. In modern parlance, we might call this a Harvard architecture machine. That's a bit of a retroactive label based off the early computers developed at Harvard. In this architecture, you have separate memory. for data and code. The Model 3 doesn't fit that description exactly. You have registers for read-write data.

You have special tapes for read-only data and separate tapes for program data. It matches the isolation aspect, but I still wouldn't really say that the Model 3 has memory per se. Anyway, the Model 4 comes next. It's similar to the Model 3, but with more features. Not a whole lot changes until 1914.

46. The Model 5, made back at Bell Labs, comes out right in our digital gap, right at the start, and it's much more sophisticated than Stibbitt's earlier machines. I'm a little fuzzy on the earlier Bell machines, but details on the Model 5 are more clear. It is 100% turning complete, so this is a fully-fledged digital computer. The only ifs, ands, or buts come down to memory. The Model 5 still has no form of addressable memory. It relies on the tried and true tape tables, registers, and routine tapes. The details of this machine come to us via two articles in Mathematics of Computation from 1948. Note the year.

Even the formalized descriptions of this computer are still right in the middle of our gap period. That's neat. I love when stuff flies up, right? Anyway, the Model 5 had a few tricks that made its power undeniable. The first are its routine tapes. Earlier Bell machines had these, but the articles on the Model 5 describe them in a way that finally makes sense to me. The Model 5 had five special tape readers that can be referred to in your code. You can tell the computer to go execute the program stored on tape 2, for instance.

Those tapes are called routine tapes, and they're most often glued into loops. The loop trick was crucial in common from Stibbets. Quote, our room was decorated in a somewhat avant-garde style. We kept spare tapes glued into loops hanging on pegs near the computer. Somewhere along the line, we had failed to instruct the computer to reach out and take a new one off the peg when needed, end quote. That's trick one. the routine tapes and their loops. Trick two is the section.

A program could contain section numbers. It was possible to issue a command to jump to a specific section. This is another spot where the hunt mechanism comes into play. Program tapes, that includes a main control program and the routine loops, couldn't be hunted backwards, only forwards. That's why it's such a big deal that routines were kept as loops. You could call up a section of any routine at any time. It just might take a second for the tape to loop around to the right spot.

The final trick is discrimination. I realize that sounds bad now that I say it out loud. Discrimination is what Bell Labs called a conditional transfer. You may be able to see why conditional became the preferred language. The Model 5 has very simple conditions, but that's all you really need. Conditionals plus looped routines and addressable routines made the computer fully turning complete.

All things considered, the Model 5 was a very sophisticated machine for the time. Remember, the time is 1946. There are a scant handful of computers in the world at this point. There's no fully operational stored program computer with addressable memory that exists. The timeline here is truly wild. As I mentioned earlier, 1946 is the year the EdVAC report is leaked.

That really is the starting point for the proliferation of computer projects. But Bell was so far ahead that they completed their machine in that year. That was only possible because they had been working on digital computers prior to 45. It's also interesting that the Model 5

isn't a von Neumann architecture computer. The EDVAC report popularized that design, a single memory for data and code. Bell's machine, of course, predates that design. That's cool as proof that the EDVAC report really is where that shared memory architecture comes from. This also makes the Model 5 a little hard to understand, at least for modern computer nerds.

When you learn how to program, or really anything about computer science, you learn about the Von Neumann architecture. It's the most popular design out there by far. As a result, when one encounters the Model 5, it may not look like a computer. But it really is. So, Hypothesis 3. For our digital gap years, it takes a long time to build a computer from scratch, right? We've established that.

There were very few labs working on computers before 1946. Therefore, relatively few computers were completed in the latter half of the 40s. It really just comes down to timing. This also helps me, I think, tighten down my numbers a little bit. This gives us a nice, let's call it a back of the napkin, rule of thumb kind of calculation. It must have taken just about five years to construct a new computer. That number lines up very nicely with the spike in running machines we see in the early fifth. Projects inspired in the middle of the 40s come to fruition in about five years.

That's handy to work out this way, because not all machines are well documented, especially when it comes to their development histories. For a recent example, look at Radak. There are only a handful of documents in scattered interviews. That may seem sparse, but that's a pretty good paper trail, actually. The ICCE has almost no paper trail. Bynak is the same. There's just so little information on these machines. As a result, it's very hard to peg down dates for when different projects began and when they were completed.

A bit of a lightning journey this time, but I'm going to close us out with a longer reflection. And yes, I think I'm finally out of my 1940s rut. I'm ready for something different, but I'm glad I stuck around in this period for so long. One of the most fascinating things, to me, is this very early period. Part of this is just how other it feels. Nothing's established yet. That doesn't just mean that machines in the 40s and early 50s are wild and wacky. By examining this period, we watch norms become established. This is when the field of computing forms.

The current state of computing is the result of a long, long series of arguments, attempts, and best guesses. We eventually land on something that we can recognize. Electronic digital computers with random access memory. Those computers use the Von Neumann architecture and are all turning complete. It's a very, very narrow definition. In 1945, only some of the important arguments had been considered. We've taken 10, maybe 15 attempts at making computing machines. And our best guesses were pretty fuzzy.

There are some machines that pop up that align with the current model of what a computer is. Machines like Edzac or Whirlwind are all things considered pretty darn close to modern. Those machines don't start running until the 1950s, really. And even then, the field is still wide open. We get blips of modernity, but not a well-established pattern. The gap between 45 and 49, I think, shows us a wide-open. field. The problem that people are trying to solve is never, how do I build a modern computer? That is a completely anachronistic notion to have. No one in 1945 sat down and thought,

gee, I sure wish I had an IBM PC. The problems that people are facing all have to do with building automatic computing machines, building devices to solve mathematical problems. As we've seen, there are a lot of ways to do that. The electronic digital computer, with all the qualifiers attached, is only one very narrow solution. You could go fully analog, or electronic analog, or hybrid. You could even make a digital computer that's non-programmable that just does a single task. All those types of machines existed in this period. They all competed.

Eventually, over a matter of years and decades, the formulation of a modern computer emerges. The general purpose with stored programs, Bonn-Noyman architecture, electronic, digital, all that. It's a tight definition, but it becomes popular, it becomes dominant. Each step towards that formulation is something like a fight. Programmability proves to be more useful than special purpose machines. A shared memory space proves to be more useful than I-Sycephor. data and code. Digital proves to be more effective than analog. All those fights are completely new during this gap period. That's why it's interesting. Once we get to the 50s and 60s and beyond,

there are different horizons to chart. But in this early epoch, every question is fundamental, every single one. And we see a wide range of answers that succeed, fail, or simply disappear. After a fashion, we do arrive at the design we're all used to, but that outcome wasn't guaranteed. With that said, I'm done and dusted. I'm done with the 40s for a bit. I finally have a better handle on the state of things in that decade, so I can move on, at least for the time being. I know I will one day come back. Thanks so much for listening to Avon of Computing.

I'll be back in two weeks' time with another piece of computing. passed. I think next time it might be about programming finally. I'd like to get back to a cool language episode. Until then, you can find links to everything at adjunct ofcom you can support me on Patreon

and you can listen to the adjunctive computing for more computer history content. As always have a great rest of your day.