Advent of Computing - Episode 186 - MADDIDA

Episode Date: July 19, 2026

In 1950 Northrup unveiled their MAgnetic Drum DIgital Differential Analyser. The machine is a wild piece of technology; a rendering of an analog computer in digital encoding. But how did we get here? How did a project to guide a missile lead to a mass produced computer? And what does it have to do with suspiciously dry martinis? Selected Sources: https://sci-hub.st/10.1109/MAHC.2003.1179869 - 2003 paper on MADDIDA Like Advent of Computing? Then check out the after show! Adjunct of Computing is now LIVE: YouTube Spotify Apple Podcasts  

Welcome, Mad Ida. A 31-year-old physicist, Floyd Steele, has invented a new and improved mechanical brain. He calls it Mad Ida. These are the call letters of the phrase used to identify and summarize the functions of this mechanical marvel, magnetic drum, digital differential analyzer. Mad Ida says it's sire, will operate a complete factory without you. human aid. It will take information from one machine, interpret that information, and then run that other machine. Matt Ida makes human beings obsolete. Matt Ida, Mr. and madame, you're welcome.

Here's the key to the whole darn globe. You are the answer to all the prayers arising these days from people too sorely beset by their troubles, real, imaginary, and expectant. to think clearly and act sanely. They need new brains, any brains, any good substitute for brains. Mad Ida, you are the fulfillment of this wonderful machine age. Please sit down, stick around forever, and permit us dumb parasites to fade into the obscurity our approaching obsolescence is consigning us to. Just take over everything and run it.

You cannot possibly do any worse than those you are replacing. New Orleans States, Friday, March 31st, 1950. Welcome back to advent of computing. I'm your host, Sean Hass, and this is episode 186, Mad Ida. And what can I say? I'm finally back to the 1940s. Before I get too sucked in, I do have some. plugs at the top. One, if you haven't already started listening to Adjunct of Computing, the

official advent of computing after show, well, perhaps you should. If you like advent of computing, which I'd be surprised if you're listening to this and you don't like advent of computing, then you'd probably enjoy the after show. I have a co-host Joe, and it's a more casual discussion of the topic that we covered on the most recent main show. You can find it anywhere. you listen to podcasts or via the links in the show's description today. Also, August 1st and 2nd, I'm going to be down at Vcf West. That's Vintage Computer Fest. It happens at the Computer History Museum and Mountain View. I'd be giving a talk about early Unix like clones. I think it's going to be a fun time, but that also means that I timed it right. So next episode will be a live recording of my presentation

at VCF West. So if you can't make it out, there will be the live show for you to listen to. That said, I hope you do make it out. Vcfs are very fun events to go to, and it's always a nice time for me to meet listeners. Now, as I said, I've been sucked back in. Today we're looking at a machine I was initially going to touch on back in my Digital Gap episode, but decided against. There's just too much here that I wanted to learn about. It's more than just a five or ten minute segment in a larger episode. It's not an illustrative example. It's a story on its own. What's the story? Well, it is Mad Ida, the magnetic drum digital differential analyzer. And yes, it is pronounced Mad Ida, not Madida. I only know that because of a Northrop aircraft,

brochure that specifically says it rhymes with mad Ida, as in an angry gal. What can I say? This machine kind of nerd sniped me, and I think it's easy to explain why. Differential analyzers are analog machines used for solving differential equations. Usually this means mechanical analog, as in gears and pulleys and that sort of thing. These are a very old kind of pre-digital computer, and they saw a lot of use up to an even after the, say it with me, advent of computing. So what is the word digital doing here? How can you have a differential analyzer that's digital?

And, wait, it used magnetic drum also? Well, classically speaking, analog computers don't really do memory. That's a pretty digital construct, so we're looking at some kind of weird hybrid thing, almost a chimera. That's just from the name alone. The historical details make Mad Ida all the more interesting. It wasn't a one-off machine. It was mass-produced. Well, kind of.

There were six Mad Idas produced. For the time, that is, well, it's like mass production. So this isn't a one-off oddity. This is an actual production machine. Also, an interesting little wrinkle here. Mad Ida isn't general purpose. It's a special purpose computer. But as we've discussed before, that category is very broad.

What exactly was Mad Ida? How useful was it and how was it hybridized between analog and digital? In 1953, the Office of Naval Research published my favorite survey of digital computers. What can I say? I don't get out enough. Luckily for us, this computer survey used a pretty loose definition of the term computer. So this report has all kinds of handy calculating machines included. The entry on Matt Ida is noteworthy because it actually ruins the report's fastidious formatting. Each computer in the survey is reduced to a nice single-page description.

That includes things like how many vacuum tubes and diodes are used, how long addition and register transfers take, and how much memory is available. Half of the data is actually rendered as a table. So it's a very well-formatted survey. The entry on Matt Ida doesn't use the table. Someone just typed over it with a few key details, since addition speed isn't really applicable. In place of well-formatted performance figures,

the table says, quote, 44 integrations are processed in one drum rotation. Again, this is typed over gridlines on a table. I kind of love that even in 1953, some nerds at the O&R are like, oh, this is a computer, but it kind of doesn't fit in. That should tell you how different this machine is. In the 1950s, the definition of a computer was much more flexible, and even under those definitions, mad-eyed is kind of weird. To get to the heart of the Mad Ida matter, we need to look at how and why it was developed.

I think that should answer just about all of our questions about the machine. I started down this rabbit hole with two sources. I say start, and that will matter in a moment. One of these was a textbook, Oman's Analog Computing. That and Paul Saruzzi's Beyond the Limits, Flight enters the Computer Age. Now, as always, these are the ones. are not primary sources. These are books compiled after the fact.

In the case of Saruzi, he's writing in the 1980s. So there are limited sources out there, at least limited source availability. I'm going to be tempering things with contemporary data in a lot of spots. Again, this is where I started. I mainly want to name these sources because if you go looking for information on Matt Ida, this is what you'll find initially. All right, so what's the actual story? The origins of Matt Ida go back to an earlier project called Snark.

This was one of the first cruise missiles. If you haven't seen a cruise missile before, go find a photo. It's kind of like a mix between a small plane and a rocket with wings. Development of Snark started soon after the Second World War. It was part of a larger project to develop new delivery. methods for these new-fangled atomic weapons. Snark's development was contracted out to Northrop. The idea is it would serve as something like a predecessor to an intercontinental ballistic

missile. Just an intercontinental winged missile, I guess. One of the wild parts about cruise missiles is that they're just tiny airplanes. Snark was powered by a jet engine and cruise and used J-toe rockets for takeoff. In fact, the jet engine that ends up being used by Snark is the same one used in the Boeing 707, so it may be better to think of these weapons as self-guided planes than missiles. Northrop, at the time, didn't know much about computing, but we can forgive them. In fact, no one knew much about computing in 1946.

It's not just that the market for computers is small, it's that there really is no market. There are no experts, but there were a few folk who knew more about these computers than the average bear. Northrop ends up contacting Eckert Moutchley Computer Corporation. They signed to have EMCC create an experimental machine for them. That machine would be used to go.

guide a snark missile. Now, why would Northrop jump to a computer? This is actually an interesting question, and answering it involves some math. Well, a little bit of math. So welcome to Physics 101. It's relatively easy to describe the motion of an object mathematically. There is one complication, though. Equations of motion are described using differentials. Velocity is the derivative of location with respect to time, the change in location over the change in time. Acceleration is the derivative of velocity. This gets hairy when you want to go backwards.

If you know where you want to go and know your current location, how do you need to move? To figure that out, you have to solve equations of motion. That means you have to take the reverse of a derivative. You have to integrate, which isn't the hardest thing to do, but it is computationally complex. ENIAC had been designed to do exactly this. It was originally built to compute firing tables for artillery.

If you were given information about an artillery piece and its ammunition and know where you are and where you want to hit, then how do you need to aim your artillery? To figure that out, artillery teams would consult books of tables. Those tables were solutions to these. equations of motion. They were usually made by hand, which was slow and error-prone. Iniac was designed to spit out those tables, to solve a bunch of equations of motion, or, in other words, to do a bunch of integration. Inniac had been a state secret during the war, but in February of 1946, the war department

disclosed the existence of this new electronic brain. The press release made it clear that INIAC was a huge leap in technology. It was faster than any calculator ever built before. It was more flexible and could plain do more than any analog machine. Quote, it is capable of solving many technical and scientific problems so complex and difficult that all previous methods of solution were considered impractical, end quote. It was really the new hotness. and it could solve the exact kinds of problems that Northrop was facing with their snark. At least so the story goes. Saruzi gives us a little more detail.

Supposedly, Northrop wanted a miniatrised computer to put inside snark missiles. They contracted EMCC to create a machine as proof of concept for the project. That may have been an ambitious goal, to quote Siruzi. In retrospect, it is hard to believe, that the builders of Inniak, which filled a room and consumed 174 kilowatts of power, were able to convince Northrop that they could provide a version of the same machines small enough to fly on the snark, end quote. Now, I think Saruzi is a little harsh here.

No one really knew the capabilities or restrictions of computers yet, or at least very few did. ENIAC was disclosed as this huge machine, but it wouldn't necessarily have been cleared in Northrop that miniaturization would take decades. I don't think it's crazy to start in on the program with a big computer and just see where it goes. In 1949, EMCC delivered their first test article, Bynac. And yes, that's right, this is where Bynak came from. It was planned to be used as a ground test machine to examine the feasibility of using a computer in Snark. Also, a note here, if you read Saruzi's chapter on Matt Ida, he spreads a myth about Bynac. He claims that Bynac never worked after it was delivered to Northrop.

We now know that's not correct. Northrop would publish a number of papers off the back of calculations done on Bynac. My suspicion is that those papers weren't widely accessible until more recently. So the general story here goes that Northrop engineers were disappointed in Bynak, so they started work on their own computer. That machine was Matt Ida. I say that's the general story, but that story is not correct. I want to be clear here, I'm not trying to be mean to Paul Saruzi.

In 1983, the story I just outlined would have been correct, but times have changed. In 2003, a group of Northrop engineers who worked on Mad Ida published an article in ACM. It's titled West Coast Contributions to the Development of the General Purpose Computer. The primary author is Donald Eckdal, and it's all about Mad Ida. And it wouldn't have been available when these earlier histories, were written. This is, just as a side note, one of the reasons it's really important to work from primary sourcing or close to primary sourcing. When historians synthesize information, and this includes me. I do this too. We're limited by our own biases, what we have access to,

and also the lens that we're looking at history through. When you look at primary data, a lot of that is either not relevant or not there in the same way. It gives you a better picture. There's one of those cases where, as we've reached the information age and more information has come to light and been written, we can just tell the story better. Anyway, going back to the Ectaw paper, it kind of rewrites the whole story.

Northrop wasn't floundering and looking for a computer. In fact, the announcement of Inniac pushed them to start their own internal computer group. That group, led by Floyd George Steele, wasn't really planning to put a computer in a missile, at least not at first. Their initial plan for Snark was to create something like an automatic firing table. A tape of solutions to equations of motion would be created on the ground, either by computer or by hand.

That would be loaded into the missile. The missile would then use its sensors to look up a course correction on that internal tape storage. Bynac plays more of a bit part in the story. Sarkeesian, co-author of the Mad Ida paper and Boots on the Ground Engineer, was himself pretty confused by the whole Bynack thing in general. He didn't really see a use for the computer, to quote, perhaps they had other reasons for asking for this. it may have been a way for the government to sponsor development, which apparently was what was going on, end quote.

Steele also explains in an oral history that the Bynak team was kind of separate from the team that led to Mad Ida. This was a separate initiative going on inside Northrop. They didn't get mad about Bynac and decide to make their own computer. To put this in a more academic sense, there's actually no causal link between Bynak's supposed failure and Mad Ida's creation. Mad Ida was already under development before Bynac was ever contracted for. The only connection is John Mouchley. During the saga, Mouchley became a regular guest at Northrop's offices, so much so that he would

give impromptu lectures to the new. computer group. In fact, the Bynac debacle directly detracted from development of Matt Ida. Some members of the computer group were rotated through tours of EMCC's factory in Pennsylvania. Those tours involved training ahead of Bynac's delivery. That took up months of time that could have been spent on Matt Ida. But I get ahead of myself. The bottom line is that Matt Ida didn't come as a response to Bynac. It was well underway. It was a direct response to Computer Mania in 1946 and the very real constraints of the Snark Missile Program. The true first step towards Matt Ida was a smaller device called, I know it's, it should be

pronounced Dita, D-I-D-A, a digital differential analyzer, but I kind of want to call it DIDA. This is how we're going to get to the root of the weird analog digital debacle. The solve comes down to the people at Northrop's Computing Group, namely Floyd Steel, reading a little bit too much. In order to solve equations of motions, you have to integrate. At this point, it doesn't really matter if those equations are solved on the ground and stored inside Snark, or if they're solved by some in-flight device. The equations still need to get solved. There was an existing device that could do this,

the ball and disk integrator. In my mind, this is the magical heart of early analog computers. It's like those crystals that they put in lightsabers that make them work. There's this whole class of machines called differential analyzers, which, despite the name, are used for integration and can actually be used to solve any number of equations? They're called differential analyzers because you use integration to analyze differential equations. I guess it's like calling Advil a pain pill since it fixes pain, but whatever.

Differential analyzers had existed for quite a while by the time 1946 rolls around. Steele had specifically been reading up on Vannevere Bush's analyzer, which was used for computing things like firing tables. Yeah, Enniak was basically designed to put these machines out of a job. The ball and disc is cursed in a number of ways. It's composed of a metal disc, a metal ball attached to a sliding rod, and a metal cylinder. The ball rides on top of the disc and presses up against the cylinder. This way, when the disc moves, the ball works.

the ball rotates, which in turn rotates the cylinder. The ball can be slid so that it rides the disc at different locations. If the ball is in the middle of the disc, for instance, there'll be no movement on the cylinder. But if it's at the edge of the disc, the cylinder will spin quite a bit. Many differential equations are in terms of time and some location. Call it X. Think of DX, DT, if you're familiar with more trad. math. For that kind of equation, the disk would be used to represent time and the location

of the ball would encode X, your position. You'd spin the disc to model the interval of time you're trying to solve for. So if you want to integrate over three seconds, you might end up spinning the disc three times. You move the ball back and forth to model movement during that time span. You can find videos of these devices in action. They make a lot more sense when you can actually see them. Their operation has to do with gear ratios and how motion transfers across the disc. There are a few crucial concepts here. The first is repetition. A ball and disk computes an integral by repeated rotations. If you want to integrate something over a larger time period, a larger delta T, then you usually have to spin the disc a lot more. Alternatively, if you want to

make something very accurate, then you can scale things in such a way that you can just spin the disc a whole lot and have each revolution represent a smaller change in time. The point is repetition is what really matters for accuracy and really even for how big your numbers get. The second is a fact that this is a continuous device. The numbers you deal with aren't discrete. They don't come in steps. Instead, they can represent any number.

You don't have something like rounding here. That can get messy because the accuracy of calculations are dictated by how well the device is made. If there's an imperfection in the disk, your results will be off. That error can be very hard to track down because it, well, it'll just be an error in how much some rod or gear moved during calculations.

Notice how this is a very different mode of operation than a digital computer. It's not just a slight difference either. Analog machines operate on their own principles. Digital just doesn't enter into the equation. Maybe you're seeing why the concept of a digital differential analyzer is so odd to me. These are two incompatible forms of thought, right? The DIDA is basically a recreation of a ball and disk integrator, except it represents numbers in binary.

Now, I know, this is kind of a cop-out. The details are, well, they're a little complicated, and they only mostly understand how it works. So, the first DIDA is built at Northrop, and it used 18-bit registers to store numbers. It was composed of three components. a register called Y, another called R, and a transfer device called T. An integral is defined, at least graphically, as the area under a curve.

So you can calculate it by taking the sum of a lot of small changes in variables. Those little changes are called deltas. In an analog integrator, the deltas are represented by spin and they're continuous. So you get very smooth lines. But in a DIDA, things are different. We have discrete numbers now because we're binary. So we have to take discrete deltas. Here I'm working off of a progress report on Matt Ida,

but it uses kind of confusing notation. They talk in terms of delta Y, delta X, and delta Z, as in the Dita takes in information about changes in X and Y and gives a result that's related to change in Z. It's a little too general for me, so I'm going to use slightly different naming conventions. I'm going to assume we're using a data to calculate distance. That makes things more straightforward and more, I think, applicable. It also gives us a concrete problem, so there's that.

If we want to get distance, then we need to integrate velocity. In other words, distance is the integral of DYDT, change in position over change in time. The Y register takes in Delta Y. So you send in binary numbers that represent changes in position. Crucially, Dida used pulsed binary. That means instead of a wide binary number coming in 20 wires at a time, it uses a single wire. Each pulse increments the register.

T took in pulses too. Those pulses would represent delta T, a change in time. But these aren't numbers, they're triggers. When a pulse hits T, it would add the current value of Y into the R register. It's basically saying, take in all the change in position that's happened in the last time period, and then add that to my running total, then start counting up changes again. Then we get to the final part, which I'm too stupid to fully understand.

For some reason, you don't read the value of the R register. You read how often it overflows. These are digital registers, so once you reach a certain value, the register wraps back to zero. If you had a 4-bit register, you'd hit 1-1-1-1, add one more, and you're back at nothing. A dida had a wire that would pulse when that overflow happened.

You read the rate of that overflow pulse to get the actual result. That would be your distance. That also means the size of the registers matters. Specifically, it's related to the coefficient of integration. If none of that registered, ha, with you, then that's fine. What's important to know is that the dita is an analog. It used discrete values and it used binary encoding. But it isn't digital in the way we know it.

A more familiar digital solution would be, well, like a stored program computer, where you'd read values out of registers. I'm not familiar with techniques that read the rate of overflow as an output. So, while DIDA does use binary numbers, numbers, it doesn't function like something we're more used to. Dita is much closer to the ball and disc integrator. In fact, the first Mad Ida progress report has a diagram of a ball and disc directly above a diagram of how a Dita works.

This is meant to be a digital adaptation of analog technology. In that sense, we can think of the whole overflow part as equivalent to the spinning disk. Both give us a cycle. That's used to drive output. The machine here fits in a weird spot, right? But it's in good company. The data is specifically meant as a digital recreation of an analog technology. So is ENIAC, at least in large part. When ENIAC is designed, it takes a lot of really important cues from earlier differential analyzers. It's not as directly related as DIDA, but it fills a similar role. So we're in the same ballpark as really, really early computers.

Looked at in another way, DIDA is a transitional technology. It's between two worlds, right? Northrop is using newer technology and ideas to replace an older technology that they know works. But they aren't making a whole new device. They aren't taking a radical approach. This is more of an evolutionary step. They're adapting new ideas

to directly replace older ones. The first experimental didas were made from vacuum tubes. Specifically they used flip-flop-based registers. And again, I mean register in the most technical sense.

These are small 18-bit memories that count pulses and roll over. Each dida used a lot of tubes. Ectol explains that a single register used 36 tubes, so a single dita would need at least 72 tubes, plus more for inputs, outputs, and the transfer device. That's pretty complex. And to do useful math, to actually make a circuit that could solve an equation, you would need multiple didas wired together. All this is to say, that the concept of Adida proved useful, but it didn't prove very small. You couldn't put a handful of Idas into a missile and expect it to fly.

So where did Northrop go from here? I'm just going to read an extended passage from the Ectal paper verbatim. He quotes Sarkeesian discussing one aspect of the Snark program, and I think it's particularly illuminating. Quote, there were wire recorders, so one of the first things I was supposed to do was pick out digital signals from a wire recorder. The specs were very difficult to meet for that time. The wire recorder was supposed to record digital information and then play it back in a precise manner.

Sarkeesian continues. Upon asking for background information, I suddenly discovered that there was not any on tape recorders, either at Northrop or even in the literature, except for some very basic things about the early history of wire recording on magnetic recording. This launched me, more or less, into research of what was then the state of the art and how we could make use of it, end quote. We've arrived at the classic bug bear. The 1940s is an era where we're laying out the most simple,

basics of computing. That includes logic and programming and comp-sci theory, sure, but it also includes a whole lot of frustration about magnetic tape specifically. What I want to highlight here is how truly difficult it was to construct digital memory in this period. Sarkesian ends up grinding through early German magnetophone patents. It's also very very very very. very likely he ran into Poulson's patent for the first wire recorder, which was actually a drum wrapped in wire, but I'm sure that's not important at all. Or at least, I'm sure there isn't a source that could prove that's important. It wasn't a simple choice to just say that Snark would read data from a prepared tape. When you hear that today, it sounds like a neat

solution. A computer can't fit inside Snark, but we can put a magnetic tape in it. Tape drives are small and cheap, and they can be cram-packed with data. So that's simple. It just works, right? But that was actually almost impossible to do in 1946. No one mass-produced digital tape drives. Heck, no one mass-produced tape that was really suitable for storing digital information. Thus, Northrop was left starting the project from first principles. Sarkeesian made headway, and that led to an interesting development. That development? Well, would you believe it was a magnetic drum? This phase of the project started in 1948, well into the development of magnetic tape and the data.

The core of the idea was to smash these two technologies together. The vacuum tube registers used in DIDA were too large and complex. All they really needed to do was store little data. And magnetic storage was looking viable. So why not replace the registers with something more magnetic? But how exactly did Northrop land on drum memory as the solution? This is a more important question than you might first assume. ERA would complete some of the first magnetic drum memories

in 1947. So one could posit that it's possible someone at Northrop caught wind of the device. But the story is more complicated than just that. Steele had laid down designs for Mad Ida, or what would become Mad Ida, before he knew about magnetic drum. In an oral history interview, Steele describes this new machine as loop-based, as in it operated on recurring loops of data. The first choice was to use an acoustic line of some sort, either mercury or something more exotic.

But again, we're working from first principles. Everything takes a lot of time and effort. Everything has to be made by scratch. That led to a strange compromise. To quote, When these didn't materialize very rapidly, meaning delay lines, we decided that perhaps we could simulate them with just a loop of tape, like a tape belt, so that we could get the effect of recirculating line and get the computer going. We started originally to lay out a loop of tape, and this led to the idea of pasting it around the rim of a wheel.

Then, because of the crack and the tape, it seemed best to just spray it on. This led effectively to the disc. He continues. At the time, Aiken was doing drums that had bearings on both ends, and I was trying to do a simple disc, just a single loop. The drum ended up as a simulator, with one end of the drum simulating for the loop, actually.

End quote. This paints a wild picture, but one that I do find very self-consistent. The drum isn't the key here. It's recirculation. It's looping. Remember how a ball and disc integrator work. It all comes down to the disc spinning in a cycle. Dida also works on a cycle thanks to the fact that its registers overflow and reset to zero.

The crucial fact here is the cycle. Steele wants to work with acoustic delay lines because they can be used cyclically, just like a ball and disc integrator. When delay lines took too long, Steel tried using tape loops. Notice that he chose tape loops as a substitute because they were cyclic. It should be able to work just like a delay line or, I mean, maybe a spinning disc. And then he made the jump to drum. Note again, there is an attempt here to straight up paste magnetic tape to the face of a drum,

or the edge of a disc, as the case may be, but really, isn't a disc just a very short drum? What the Northrop team eventually landed on was a spray-on magnetic coating. That can make it sound like a sophisticated device, but I wouldn't go quite that far. Ectall explains that Northrop didn't make the coating, at least they didn't make it from scratch. They dissolved the coating from pre-purchased. magnetic tapes in a solvent, I'd assume acetone, and then sprayed the mixture onto drums. That basically took the place of adhering tape to a drum, but with more steps. Magnetic drum memory was cutting-edge technology, but it was produced in a pretty ham-fisted way.

Matt Ida would also work on the same principle of cycles. It replaced the pair of vacuum tube of the DIDA with a pair of two magnetic tracks on the drum. The transfer mechanism was now automatically tied to the drum itself. When the drum rotated, that triggered a transfer. Each full rotation would do the whole add rollover reset output cycle. This is, in very simple terms, not at all how a digital computer works. There's just no line to be drawn. here between Matt Ida and any other programmable machine.

Just think of the relationship to memory for a second. On a normal drum-based computer, the rotation cycle isn't really connected to execution. It's more like an inconvenience that you have to work around. It dictated how fast certain bits could be accessed, and it dictated certain timings in the computer. But on Matt Ida, the drum cycle is everything. It is the very execution cycle of the machine. So then, how do you even program such a device?

Well, it doesn't take machine code, that's for sure. Matt Ida was programmed with patch cables, ENIAC style. You wire up the output of one integrator to the inputs of another integrator. You build up a network of integration devices. That may sound pretty special purpose at first, but it's actually pretty useful. The obvious application is, of course, differential equations. You can also use integrals to represent just about any mathematical expression, or at least many useful expressions.

I'm not going to say any because I'm not actually a math guy. A 1950 brochure from Northrop shows how you can do everything from simple addition to exponents to trig functions using networks of integrators. In operation, you first wire up all these integrators, then you have to input initial values. That's basically how you handle the arguments to your quote-unquote program. Matt Ida does have a front panel, but it doesn't look like anything that you'd expect from this period. The most important operations you handle via the front panel, panel are putting those initial values into the integrators. So you have to have a way to select

integrators, right? Matt Ida has 44 of them by the time it's in production. You could use a dial or some rotary switches to select the integrator, but that's too pedestrian. Instead, you punch in the integrator using push buttons. The control panel has two vertical rows of numbered switches. To select an integrator, you punch in its number. Now, I'm not entirely sure how addressing worked here. Matt Ida has 44 integrators. We know this because Northrop says it everywhere. The integrator select your button, well, the buttons range from 2 to 7 in one column and 0 to 7 in another.

I think there's some encoding going on here, but we don't have an actual manual. We just have a brochure and a spec sheet, so I can't tell you for certain. Once you selected the integrator, you could inspect, clear, or fill it. That's actually pretty similar to how other early machines work, or how one would operate a computer trainer. You select an address in memory, and you can mess around with it. However, there's a big catch here. Matt Ida did not have a way to display numbers. That sounds, to be frank, a little crazy, right?

You were expected to use an external oscilloscope to read off values, as in something that displays a waveform. The front panel had jacks for hooking up one of these scopes. Those jacks would output the raw waveform from the integrator track you selected. So you don't even get a number per se. you just get a train of pulses. I think this is fed directly from the magnetic pickups, or at least pretty close to them,

probably from an amplifier connected to a pickup. Since an integrator was composed of multiple tracks on the drum, you can also access each track independently. Matt Ida had separate push buttons and dials for choosing to view contents on a certain track or clear or fill that track. As a result, we actually end up with quite a lot of buttons on the front panel for such a small machine, but no outputs. Again, external oscilloscope. Crazy to me.

Filling in data is equally wild. In the very middle of the front panel are two buttons. Their labels are one and zero. And that's right. Matt Ida has a straight up binary keyboard. I've seen many jokes about binary keyboards. They've even seen ones that folk have made, you know, as a bit. But I've never seen a serious binary keyboard.

Echdal even mentions specifically how fiddly this was to use. Some members of the computer team become de facto Matt Ida operators because they were better at running the binary keyboard. Now, we can easily dunk on this interface choice. So get it all out, point and laugh a little. All right, y'all good? Because we need to actually consider a few things here. First of all, this feels so clunky because it's supremely primitive.

The binary keyboard would be sending pulses directly onto tracks. There's no binary to decimal conversion anywhere. outputs would be difficult to construct, so why not just use a scope? Matt Ida is a lot closer to an analog computer than a digital computer, and it's a lot closer to lab equipment than a complete project. It's not that while to use an oscilloscope as an output in a lab setting. All electronics labs will have some scopes nearby. Matt Ida was also supposed to be plugged into something else. I mean, first off, duh, it did come out of a project to build a guidance computer for a missile.

But I mean this in another way. Mad Ida is a special purpose machine. It would make sense for its output to be fed to some other device, and its input to come from something other than a binary keyboard. Many other analog computers functioned this way in this period. And again, I'd argue that Matt Ida is a lot closer to that mold than to a digital one. There's also the point of programmability here. This, I think, is one of the strongest pieces of evidence that Matt Ida isn't really a digital computer.

It's just an analog computer with a binary coat of paint. By 1946, the EDVAC report has already leaked. Inniak was the last serious plugboard-based computer. Once the EDVAC report drops, everyone is racing to construct stored program computers. At least, broadly speaking, Bell was kind of doing their own stored program thing in this period, and there is still a large swath of computers that don't conform to what we would think of as store program, binary electronic digital computers. But the point is, big machines aren't really plug-based once the EDVAC report comes out.

Computers after 1946 all have some relationship to stored program. The concept of representing programs becomes key. And central to that is the idea of steps, of a cycle. I mean this in a very specific way. The machine internally has some cycle of fetch instruction, decode that instruction, execute it, store some data, or something similar. Programs are then composed of instructions that tell the computer what to do during that cycle. Mad Ida doesn't fit on that lineage at all. It doesn't have a cycle that fits into this ideological mold.

Its cycle is, integrate, integrate, integrate. Its program is, where do I integrate? Where do I integrate? There are no programmatic steps. It's a network of discrete components that all do the same thing repeatedly. Programming Mad Ida, if I can even use that word, would be almost identical to programming an analog computer. In fact, if we're talking about electronic analog computers,

then things are even closer. Electronic Associates sold a pile of electronic analog computers in this rough period. Those machines had electronic integrators. To input a program, you would wire up those integrators connecting inputs to outputs.

You would put in initial conditions and then let it rip. The only visible external difference between an EA machine and Matt Ida would have been that Matt Ida had to use digital inputs and digital outputs. But you could probably port an EA program to run on Mad Ida just fine. The first Mad Ida unit was completed in 1949.

From there, things got strange. It's time for me to back up a little bit and explain why Snark has kind of fallen out of the picture. It has to do with, well, some office politicking. From Echdal, quote, We trusted Steele to represent our viewpoints to management. We believe that he really tried to demonstrate the importance of our developments in digital computers and the impact they might have on the problem of internal guidance of the SNARC missile. Ectal continues.

We later found out that Steele had, in fact, instead communicated to both the project management and to Jack Northrop the far more radical ideas that the new techniques associated with digital computers were so powerful that the company should scrap its entire guided missile program. End quote. The computer group was getting high on their own supply, or at least steel may have been. There was a point where it looked like Matt Ida could power Snark internally, at least a fleeting moment. But steel wasn't sending that up the chain of command. Instead, he was going up to Jack Northrop himself and taking him.

telling him they needed to become a computer company, that they should scrap this whole missile thing, doing analog computation, doing traditional guidance, garbage, we should focus on a digital future. That's either very brave or very foolhardy. This would lead to some, well, let's say some odd outcomes. Steele was pushing very hard for a digital future at Northrop, but that wasn't necessarily a future everyone saw. At least, not everyone believed in the mad Ida future. So he played some politics here.

He went back to his boss from Seal. I asked Northrop if he would get a consultant to appraise the machines so we could go on that basis. I was suggesting that he undertake the manufacture. I didn't want to do it myself, but I thought it might be a good idea. End quote. Jack knew a few people that had some ideas about computers. and Steele was hoping that if the good word came from outside the house, maybe Northrop would see the digital future for what it was.

So, they did find a consultant, one John von Neumann, or, as his friend's column, Johnny. How this all went down is hilarious. It's accounted in Eckdahl's paper as well as Steele's oral history. I'm using the two to stitch together the full story here. The mad-eyed a team got an appointment to go show Von Neumann their new computer. An issue here is the whole go part. Northrop was based in California, and von Neumann was out in Jersey. At this point, no computer had been successfully moved very far.

Bynak was the closest. It made a cross-country trip from Pennsylvania out to California, But that machine had taken months of troubleshooting after the move. It came in pieces had to be reassembled, and even then, didn't really want to work for a while. So taking a computer cross-country for a quick demo, that wasn't a thing. That was no small task. So how do you go about moving a computer in a way that's never been done before successfully? Why?

You use air freight. I'm not even joking. and the mad-eyed a crew put the thing in a box and shipped it. To be fair, steel had a custom crate built for the machine, so it was well-handled. But still, they're shipping one of the first computers in the world by airmail. And it's after they had direct experience with moving a computer at all, not ending very well.

There were some hiccups even before shipment was done, though. Even the destination proved a problem. Plan A was to send the machine to the IAS, the Institute for Advanced Study, where Von Neumann worked, and then do the demo there. But von Neumann called him up saying he didn't like that idea. You see, Johnny was doing this as a favor, so he didn't want to do that favor at his work office. That didn't seem ethical. But he offered up his basement as a venue. That idea also didn't work out.

out from Steele, quote, we were ready to ship it when he called again and said that his wife didn't like the idea at all, and she wouldn't let us near the place. We said, well, we'll find some place around there, and we sit it to the hotel, end quote. It's chaos, complete and utter chaos, but hey, you kind of have to accept these things in business. For the Mad Ida crew, Steele and Echdal among them, flew out to meet the machine at the Princeton Inn in Princeton, New Jersey. But even once they arrived, the chaos continued.

They weren't set up in a convention center or anything fancy. They were going to be running Mad Ida in a hotel room. The room's power outlet wouldn't really work for that. So what is a poor engineer to do? This is where the steel and ectaw stories diverge on a very funny detail. According to Steele, a worker at the hotel said they couldn't use the room's outlet, that it would trip a circuit. But offered them a giant extension cable. They then strung the cable out the room's window, down the side of the building and into the hotel's basement. As such, they had to sleep with the window propped open during a cold New England winter. The poor Californians then spent a night freezing in their beds.

Echdahl's story is slightly different. Steele claims the power issue came down to fuse, as that Mad Ida would have drawn too much power and thus tripped a few somewhere in the Princeton Inn. Eckdahl claims it came down to phasing. Mad Ida needed a three-phase power input, which is common in labs and workshops, but not in hotel rooms.

But the crew lucked out, because, according to Ectal, the Princeton Inn was across the street from an electric company. He quotes Reed, another engineer on the product. He quotes Reed, another engineer on the project, as remembering the saga like this. Quote, before we went to bed, we decided to see if we could get a cable strung across the street. That was done. A cable was run from the power company to the window on the fourth floor of the Princeton Inn, end quote.

Either way, the lads from California end up, exposed to the cold winter air, since their window has to be propped open for a power cord. Electricity aside, after some tinkering, the crew got Mad Ida to work. That may make it the first time a computer was shipped cross-country and brought back to life without a major refit. At least, that's a claim that the Mad Ida team makes. I'm inclined to believe them until I run into an earlier source. Reed explains the few days in great detail. I'm only going to hit the highlights and point out some interesting points along the way.

The next morning, the crew rolled out to Von Neumann's office at the IAS to talk about Matt Ida. Johnny had already read up on the documentation, so this was less a tour of the machine and more discussion. What's interesting is that the machine itself was still over at the Princeton Inn. They talked about its construction, operation, and how it was programmed, but this was all theoretical, physically removed from the device itself. The next step was to socialize. Quote, before von Neumann came over to see the mad Ida in our hotel room, we were invited over to his house where we had one of his very dry martinis.

He was famous for his martinis, which were probably pure gin, end quote. This is a bit of color I've never heard before, and I kind of love. This may mean that gin is actually an important contributing factor to the early days of computing. Turing once suggested that gin could be used in place of mercury in a delay line memory. So there may be a connection here. If you want to truly understand the earliest era of computing, perhaps you have to make a strong martini. And the recipe here is simple, right? pour a jigger a gin, and then look in the direction of a bottle of dry vermouth.

Stir with ice, strain into a glass, garnish with an olive. After the lead-up, it was time for the main event, at least after they slept off the quote-unquote martinis. The next morning, von Neumann actually came to see Mad Ida in person. The machine was programmed to calculate the vessel function, and it plugged away, slowly spinning, as von Neumann checked its outputs and asked questions. Overall, he was impressed by the device. What I want to point out here is how odd the whole interaction seems, at least to me, right?

In my head, if you're showing off a computer, it would make sense to start with the computer. But we're working backwards. Part of that must come down to practicality. Folks spent more time talking about computers in this period than they did actually using computers. So it probably would have been strange to go in the other order. Would have been odd to start by looking at the computer and then talk about it.

But I don't know, it strikes me as funny. At the same time, this whole interaction is basically unprecedented. Up to this point, it wasn't really possible to take a computer over to someone and show it off. Computers were massive. They lived and died in special purpose computer. rooms. They didn't move. Or when they did, it was a great cost both in time and money. Machines didn't travel by airmail. One of the first demos of a computer, perhaps the first, was when George Stibbitts showed the operation of Bell's complex number calculator off to a room of mathematicians

in New York. The actual computer was still in New Jersey. The demo was only possible via teletype. Stibbitts was actually remotely connected to the computer over telephone lines. He had to use a trick to even show off the machine running. During that demo, no one saw the complex number calculator. They just saw its output on a teletype. But Mad Ida is very different. It was small enough and sturdy enough that it could travel. And travel it did. It would be shipped and showed off at Wright Patterson Air Force Base and Rutgers University, both to great acclaim. Truly unprecedented for a computer in this period. But there's still a nagging question here.

Is Matt Ida a computer? Or rather, was it viewed as a computer? Have you ever thought about the power of magic in a historical context? Reality aside, just consider belief. Let's say someone says there are power. powerful warlock, and they can cast all manner of nasty spell over you if you upset them. I think it's fair to say that if you heard that today, you'd probably smile and nod. It would have pretty small impact on how you treated that person.

But what if you actually believed them? What if you were raised in a place in time where folk actually believed in magic? If that same proposed warlock approached you, then you'd have a very different reaction. I admit, I'm stealing this idea from Dan Carlin's excellent hardcore history podcast, but I think it's applicable here. The point Dan makes is that belief in magic, regardless of its efficacy, have profound impacts on how ancient societies interacted. Belief impacts behavior, oftentimes more than reality.

That's a useful lens here because we're so early in the history of computing. Did folk in 1950 think math, IDDA was a computer. If they did, then it stands to reason it would have some impact on the lineage of machines, that it would matter more than a device that we might consider a computer today, but folk in 1950 did not consider a computer. So what was the determination of the time? Luckily, we have John von Neumann's own verdict we can go off of. After the demo, he sent a letter to Jack Northrop with his assessment of Matt Ida. Quote, I think that your magnetic digital differential analyzer is a most remarkable and promising

instrument. The principles involved in its design and its engineering embodiment seem to me very sound. I would view the machine as it stands now as an example of what can be done, rather than as the actual solution for any particular purpose. But I consider that you have established the principles of a whole family of very new and most useful instruments. Von Neumann continues. Your equipment can probably serve as a basis for a worthwhile, all-purpose computer of the intermediate type, but I would not consider this as the most important implication. In quote.

Thon Neumann goes on to suggest that Matt Ida should serve as the basis for future special purpose machines. Yeah, it could be used in a general purpose computer, but Matt Ida, on its own, is not general purpose, and von Neumann doesn't think that's the best route for the technology. That, dear listener, is very interesting indeed. I know I rag on how Von Neumann architecture is a misnamed concept. We get that name because von Neumann wrote the first draft of the Edvac report. That describes a shared memory architecture for a stored program computer.

It gets leaked, and the name sticks because Johnny is the only one on the bottom. line. The design was a team effort, but however you slice it, von Neumann was there at the very start. He was a huge proponent of stored program general purpose computers since the beginning. He had huge influence on their design. For him to say Mad Ida shouldn't be seen as a starting point for a general purpose machine, well, that carries a lot of weight. At the same time, he doesn't say Mad Ida is useless. He doesn't really position a special purpose. computer is somehow less useful than a general purpose computer. In the letter, he describes it as a different tool. I think that lines up very well with what I've seen in this period. All these different

types of computers are coexisting in 1950. Analog, digital, general purpose, special purpose, even strange chimeras like Matt Ida. Matt Ida isn't turning complete. It's not general purpose, so technically it's not what a 21st century nerd would call a computer. But in this period, it is called a computer. It's used as a computer. All this is to say that the past is a strange place, and Matt Ida is a very strange machine indeed. All right, that does it for a dive into Mad Ida.

What I've left out is the spread of this machine. Over the early 50s, a total of six Mad Ida's were produced, and sold. I left this out for a reason. When Matt Ida switched over from development to productization, a different team took the reins. Part of that team eventually leaves Northrop and starts their own computer company. I want some more space to tell that story another time, so we're ending here, where Matt Ida's proven its usefulness. Matt Ida gives us another glimpse into a transitionary period. When computers, as we know them, haven't fully formed. It's a machine that's much closer to an analog computer than a digital computer. But just saying that doesn't tell the

full story. There were a number of machines that fall into this broad category, chimeras, not quite analog, but not digital in a way that we would commonly understand. Inniak also fits this mold. I think looking at these old machines is important because it stretches our definition of what a computer is. It forces us to actually examine why computers are important and what it actually means to be a computer. Truly to understand computing, I think we have to understand these very early days where the concept itself is still in flux. Thanks for listening to Advin of Computing. I'll be back in two weeks with a live episode from Vcf West. Until then, go to adjutant ofcom. You'll find links to the Patreon.

you'll be able to donate to the show. You can find anything you've ever wanted and go listen to Adjunctive Computing. Until then, and until next time, have a great rest of your day.