>> Yuval Peres: Okay. Good afternoon. We're truly delighted to have George Dyson here. He will tell us about history of the computer and von Neumann, in other words how the digital universe got its spots. >> George Dyson: So, yes, very nice to be here, thanks to Bayla, who's -- and I first I think met your -- our founders in I think it was in 1984. That's my sister. So when I met Bill he was just a millionaire. [laughter]. In fact, at that time Microsoft was third. The largest software company was Tandy Electronics, Radio Shack, and second I think was Ashton-Tate or one of those database companies, and Microsoft was number three. So things have changed. So I'm going to tell you this story of how this world we're -- we live and work in began, starting really with focusing mainly on two people, on John von Neumann, the curious child who didn't have checklist curiosity squashed by -- you know, by teachers who told him he was -- should be paying attention to other things, let him do math. And Alan Turing, who came a little bit later. But in a way, von Neumann realized in a practical sense Turing's very abstract ideas. So the two -- really neither, neither of these sort of bodies of work would have existed without the other. I mean, Turing's theory wouldn't of come to fruition without von Neumann and von Neumann's computer wouldn't have come to fruition without some of Turing's idea. So that's the machine that von Neumann built. And in these metal cylinders is the actual memory of the machine. In each can is a 32 by 32 array of charged -- electrically charged spots, and there's 40 of those cans. So it's a 32 by 32 by 40 bit deep matrix of memory. Those are the actual charged spots which on a face of a fluorescent cathode ray tube. So now we display memory that exists somewhere else. But that was a reversal of how it started and memory actually was the charge on the face of the cathode ray tube, and then that reversed itself. And so you actually are seeing -- that's an actual 35 millimeter photograph of the dawn of the digital universe from 1952. And that -- so the universe is that existed then was five kilobytes, 40 of those 1,000 bit arrays. And that was all there was. So there you see a schematic of the machine with the memory along this side. These are the arithmetic registers. The digit resolver which when it wasn't working they called it the digit dissolver. So it was a V-40 engine. It was a 40 cylinder engine built by engineers. Really worked very much like a real engine does. So if we go back exactly 60 years, I pulled out the progress report for March 1951, so exactly 60 years ago. They're reporting to their sponsors, which was the army and the navy at that time. Tests were started of reading and writing at specific locations in the memory. So that I just getting this memory debugged and working exactly 60 years ago. So it's sort of software in a way is really 60 years old. And on this project, very ambitious, they had, you know, six engineers, two mechanical, one mathematician. Just one. Three wiremen, four lab technicians, two machinists, one secretary. And the secretary still exists. Her name is Acraval [phonetic]. And so quite remarkable. They got so much original work done it was very a very small team of people. And they had very benevolent sponsors. Here in 1949, they're actually switching sponsors. They're starting to get a little too much bureaucracy from the army, so they're switching to the AEC, the Atomic Energy Commission and saying the army contract provides for general supervision by the ballistic research lab who need sort of progress reports, whereas the AEC provides to supervision by von Neumann. And I think that's really the key -- one of the key things to why this project could be so successful. That really von Neumann, nobody interfered with his decisions questioned what he wanted to do, simply got done. And that didn't slow things down. So he had this amazing ability to be trusted and believed in both by the mathematicians and by the bureaucrats and the funders who -- so he was sort of the intermediary between these two groups, very much like Peter Lee who just walked in. I was just saying how they switched their funding from the army to the AEC because the -- we'll go back, the AEC provides for supervision by von Neumann. No paperwork, just whatever he wants to do. So these were not all von Neumann's ideas. They were clearly other people's ideas, but he was the personal who put them together as Willis Ware, who was the second engineer to be hired said he -- you know, he was in the right place at the right idea with the right connections with the right idea, independent of whose ideas they really were, which will sort of be argued about forever but was not argued about at the time. At the time, every -- Turing's group and the von Neumann group, they all got along really quite well. The sort of arguments about who was first came much later. And who was really first if you go way back, this goes -- this is -- this is Francis Bacon in 1623, who is sort of articulating for the first time how you could encode all communication with binary code. Anything that can be expressed in language can also be expressed as -- in binary symbols. And then Bishop John Wilkins, who founded the Royal Society, he articulated that further how you can encrypt things with binary code. And then Thomas Hobbes took it even further yet to really sort of develop what we would call recursive functions, that any -- anything, no matter how complicated, it can be expressed with logic can be expressed -- encoded simply as binary arithmetic as -- that multiplication and division can be reduced to addition and subtraction. Everything can be reduced just to these two operations. All you need is an adding machine. You can do everything else. And then Leibniz made that much more rigorous and more explicit in -- this is 1679, a manuscript on binary coding. And what's interesting in this manuscript is he talks about -- for the first time he talks about building a digital machine, a computer that's not -- already by this time they had computers that worked with wheels and cranks. But he talked about building a machine which was binary tokens with black and white marbles bells that run down tracks, and he saw that if you had parallel tracks and you could shift them with gates, you could then perform all the functions of arithmetic with that simple operation. And that's exactly how all computation is done today except instead of gravity and marbles bells it's a voltage gradient and shift register. So essentially he invented the shift register, which is what von Neumann's group had to do was reinvent that, but to work at much higher speed with electron. So that's the shift register they came up with, which now we sort of take by granted as sort of given by God in some way, it's not going to change. But at a certain point the idea of the shift register had to take form. And other people had other soared of threads of ideas. This is Lewis Fry Richardson sort of answering the artificial intelligence question that a mind having a will but capable of only two ideas. So this is an indeterminant switch. This is sort of a quantumly indeterminant -- you can't say what state this circuit will go in. It has a -- it has a mind of its own. And if you put together enough of those, who knows what happens. So Richardson then worked on weather -- the problem of numerical weather prediction and developed these methods for dividing the entire globe into cells and then solving the differential equations against adjacent cells but saw how to do that in a numerical way by finite differences. And that again was the thread that was picked up by von Neumann, who finally could realize what Richardson wanted to do when they had the computer to do it with. That's Alan Turing, who came up with this very abstract paper which I'm going to -- all of you know, I don't have to introduce, with the idea of a computing machine just as a mathematical to prove his logical point. So in 1936, this had no basis in practicality. But very quickly it did. And here is saying how there's intuition and there's ingenuity in that you can always replace, you know, a lot of intuition you can replace a little bit of ingenuity if you just do this -- increment this step by step eventually can you -- if you assume that ingenuity is available in up limited supply, which is what you have now, really, unlimited machine cycles, you know, can you replace intuition. And that's one of these open questions. So that became also abstract who cared about Turing but during the war that became very important because with a little bit of ingenuity you can encrypt a message and then it takes a lot of ingenuity to unencrypt it and he worked on that building the Colossus, which was arguably the first real computer and hardware that had high speed internal memory coupled to an external tape. Essentially it was a Turing Machine in all but a few little details. >>: [inaudible]. >> George Dyson: The ->>: [inaudible]. >> George Dyson: That's one of Turing's classified papers debugging the war on encryption and that's a wheel of one of the enigma encryption machines so. This is -- the marvelous thing is next year is Turing's centenary, and they're digging a lot of this stuff out of the archives where it hasn't been seen, so a lot of this stuff is going to be published next year. So this was Colossus was built in secret and unfortunately was kept secret. So that's really what put the Americans so ahead of the British in computing was that the British kept their wartime computing secret whereas the Americans made there's public. So the first stored program in the computer that actually ran was in England in Manchester, by Williams and Kilburn. It was this machine that was actually a copy of von Neumann's machine put it got built even quicker. Sort of the British pulled ahead. So this was the first one to get running. Very small though. But they were very explicit that it was a universal machine, it was capable of Turing universal computation, even though it only had 32 word memory of 31 digits each. But they did -- actually their first problem was produced in prime. They did a lot with it. That's Turing a little later. But then Turing became interested in this question of -- which still interests everyone, the question of can machines think. And that also had a lot of early, early work done on these problems. And Turing made the statement that still true, that being digital is more of interest than being electronic, that the key thing was being digital, not whether you did it with electronics or relays or anything else, it's just electronics was a whole lot faster. But if you were electronic and digital, then you really would pull ahead, and that's what von Neumann did, which was called a mechanical brain at that time. Yes? Yes? >>: Did [inaudible] mention that von Neumann knew about Turing's paper? >> George Dyson: Yes. He -- von Neumann was very clear and told the engineers -- or as far as you can believe, told the engineers who worked for him to read this paper, this is what they were trying to do. And so this is von Neumann's own statement at the time, let the whole outside world consist of a long paper tape. So he really was viewing it as a Turing Machine. Now, we don't have that explicitly in his words at the time but certainly later he talked about that, and the people who worked with him did. And then I always trying to find evidence, I went and looked. And this is the copy of Turing's paper that's in the Fuld Hall Library where all the other volumes of the London Mathematical Society are perfectly bound and pristine and untouched, but this one volume is completely unbound from being referenced so many times. So that's good evidence that the ->>: [inaudible]. >> George Dyson: Computing guys were looking at it. There you can see. So, you know, they're reading it in detail. Or somebody did. So Julian Bigelow says when he went to work for von Neumann the first thing von Neumann said was go get this 1937 volume and read this paper. So I think that answers. So that's von Neumann at 12 years old doing mathematics with his cousin Lily. So he had and extremely supportive family and extremely supportive school system and culture around him. He's there -- they've gone to visit the a artillery position during the World War I. And if you look at where's John -- Johnny, he's sitting on the cannon. So he had a -- just -- he was comfortable with weapons from the very beginning. And that really I think became the theme of his life, just bigger and bigger, more powerful guns and bombs. And he was part of this amazing class, the high school he went to and the other two high schools in Budapest produced all these amazing scholars sort of in one very short period of time. And the dinners every evening were full of intellectual discussion. There was no separation between children and grownups in terms of intellectual discussion. I think that counts for a lot. He gets his first position, assistant professor in Berlin. He starts publishing just a amazingly productive series of papers, I mean dozens of papers, any one of which were important but collectively in just a few years he worked on the foundations of set theory and proof theory which sort of all these threads come together in the way he looked at computing later and then went into the foundations of physics, quantum mechanics, a book that's still in print 80 years later, the same book. Now, probability theory which we were talking about earlier, which sort of becomes game theory, theory of games. And then which they -- very prophetically applies to economics in this model of general economic equilibrium, which is still the base -- I think the last 11 Nobel Prizes in for work in economics all have their origins in this paper in some way, showing that the stable equilibrium is an expanding economy where you have two -- where you have a saddle point between two convex sets. And it's -- it explains -- the problem is you don't know, a bubble also has some of these properties. So it's the difference between a stable equilibrium and something that's a bubble. And we still -- we need to better understand that difference. We could benefit a lot from it. And so he lecturing here in physics. He comes to America. In 1930 they want to get him to Princeton but the gate keepers at the university are very conservative and they're very nervous about hiring a Hungarian full-time, so they make a time, they hire two Hungarians at half time. They hire von Neumann and Eugene Wigner. They give them each a half time position that they can split. And that doesn't arouse the -- you know, the review committee or something who would have said no. So Wigner and von Neumann come at the same time in 19 -- and changed their names. Von Neumann becomes Johnny and Wigner becomes Eugene. And then he gets invited to -- Louis Bamberger, who was really sort of the Bill Gates or Jeff Bezos of the time decided to fund this institute, really sort of Microsoft Research, a pure learning and research. And von Neumann got one of the first positions. And the motto of the institute -- how many of you have been there? There's a lot of cross-fertilization between the IAS and Microsoft Research thanks I think a lot to Jennifer Chase who, you know, encouraged that. The motto was a usefulness of useless knowledge, that if you just let people work on whatever they want you'll be surprised at how useful it might be. And the computer of course is a perfect example of that. It just worked out great. So not that it's going to be useless but that you can't predict the usefulness. What could be wiser than to give people who can think the leisure in which to do it? Some people will do something, but some people will -- if one out of 100 people comes up with something great, it's worth it. So von Neumann gets the fourth point after Veblen or Einstein, Veblen, Alexander, then von Neumann. He actually -- he took Herman Weyl's position who Weyl declined so they offered it to von Neumann. They build this beautiful headers which is a paradise in the middle of nowhere in New Jersey where Einstein came and made his home. And if you go in the front door, there's Veblen who was first, Einstein was second and von Neumann next to the tea room. And then odd second floor above von Neumann is Godel. So they also bring Godel in 1933. Who doesn't stay. He goes back to Vienna but then finally comes back permanently. So Godel there becoming very close friends with Einstein. And Godel I think was important to this whole story because there he credits von Neumann. And now you see -- and these two had reasonably small amount of money but they applied it incredibly wisely to sort of maximize who they could bring out of the collapse of Europe with the budget they had which was sort of in the order of $10,000 a year in stipends. So they get Erdos for $750. And the people who you say, you know, why was Erdos always sort of begging for food, that's why. He had -- I mean that was $750 for the year. And Godel they get -- they bring for $200 a month. So also very, very little. But very effective. If you could offer these people a position at the institute they -then they would have a foot hold in America and they could go find jobs somewhere else. And that's what happened. And now von Neumann is doing applied game theory will otherwise known as gambling. So here he's at the casino on the park in New York City and he's not good -- he's good at game theory but not good at gambling. So he's lost all his money and applied for credit and the casino is worried enough they write to the institute asking if his credit is good, because he's overspent his limit. And then it was talking about Peter earlier how we need to, you know, we need to think of good -- the key is thinking of the right question. And here von Neumann is just, you know, sort of talking blue sky with Ulam, and the economy is going bad. It's 1939. He's saying there must be something that explains this besides just the sheer stupidity of the guys on Wall Street. And that is actually a good question. And that's what led to, you know, there must be an explanation of what happens which makes no use of the tact that stock broker's are stupid. And that was really the basis of theory of games and economic behavior, which one way of looking at it is it's a way of showing how can you build a reliable economy out of entirely unreliable parts. And that's another theme that comes back in computing, how you get a system as reliable as an economy that where all the pieces are sort of unpredictable and unreliable. And then as if he wasn't doing enough already, von Neumann steps into the nuclear weapons business. This is a letter that is -- was unknown to historians I think until I found it. The letter we all know about is from Szilard and Einstein to Roosevelt. This was a letter to the Rokefeller Foundation from von Neumann and Veblen saying that they heard good solid rumors that the Heisenberg and the Germans were starting to working on nuclear weapons and what the Rockefeller Foundation should do -- the best thing that could be done would be to not start the Manhattan Project but again scientists out of Europe so they will be in a safe place when the war is declared. And that's what happened. It was actually Rockefeller Foundation money through the institute that brought the Niels and Harold Bohr, Wolfgang Pauli and those other physicists just happened to be in America when the Manhattan Project got started. It was on the basis of that letter. And von Neumann was helpful is that he had a profoundly good understanding of high explosives because he's -- I didn't mention at the beginning was his father didn't want him to count on a career in mathematics, thought it wouldn't pay enough, so he should also get -- so he actually got his first PhD in chemical engineering. So he understood high explosives in a way that most physicists don't. And if you're trying to build a nuclear weapon it really is not so much a nuclear problem, it's a high explosive problem. How do you actually compress the plutonium or uranium to a high enough density to explode in and that's -- so von Neumann solved that implosion problem that really was key to making that. This was the Nagasaki weapon which was -- could not have been done without this theory of reflected shock waves which is something where the mathematics came from von Neumann. So that 16th July, 1945 they set off the first nuclear explosion. It was von Neumann, Richard Feynman who was doing -- running the punch card computing machines at the time and Stan Ulam. So people who had worked before the war only in pure logic and abstract things and suddenly they're allowed to work with real hardware and machine shops and high explosives. It was very, very seductive. Some people went back to pure math and some didn't. So von Neumann asked what are you working on, something much more important than bombs. He's going to start working on computers. Actually he was working on both. So they needed -- this was irresistible at the -- before the bombs were used on Japan, there was a lot of question about whether they should be set up underwater and cause a -- deliver the energy by wave energy or above ground. It was like after the war they couldn't resist actually setting a bomb off underwater. That was done at Bikini in the South Pacific and that's really the start of the Cold War. I think after the war, we could have just said, okay, we're never going to use nuclear weapons again, we're putting them away. But it was very provocative to go out and do those tests. The Russians really had no -- you know, it had to be responded by tests on the other side and then the whole thing escalated. So von Neumann, there he is with Werner von Braun. You just -- you can tell those generals just love him. I mean he was the guy who made -- evidence time they came to von Neumann he would have some idea for some fantastically successful, you know, atomic bombs, computers, intercontinental ballistic missiles. I mean whatever it was. So he was highly favored. That's the ENIAC, which was built by the Americans during the war. That's Herman Goldstine who then became the administrative leader of the institute project. So even 1946, even the maintenance manual for the ENIAC is still a restricted document. We're keeping that stuff secret at the time. >>: [inaudible]. >> George Dyson: It -- ENIAC could -- it was later made into a universal computer. It wasn't as first conceived, but then -- and this is a very complicated part of history how the von Neumann group started writing code for their own machine that they just couldn't get build and von Neumann had the idea that, well, we could adapt the ENIAC and actually run these codes on the ENIAC by putting -- by changing the function tables that were supposed to be storing ballistic coefficients and so on as storage registers to store instructions. So they kind of turned the ENIAC into a stored program computer. And von Neumann's wife Klari helped with that. So von Neumann was already thinking about the next machine which they were going to call the EDVAC. So this is June, 1945. So here he's set of reinventing live notes and showing how to build a true universal machine. In a report that was published very few copies but had tremendous influence. So you see there now what we call the von Neumann architecture, the central arithmetic unit, the central control, which is CPU, the internal high speed memory and then the input/output and then the low speed memory. And them most importantly the instructions which govern this operation must be given to the device in absolutely exhaustive detail. So we all know that, the software has to be bug free. And he already -- 1945, he's already consulting for IBM. Which was not known to the other people on his project to some, you know, irritation later it tunneled out that von Neumann was actually consulting for IBM all along. And that's one reason sort of IBM got such a head start on a lot of the other companies. But at that -- oddly enough, at that time IBM had no interest in computing. They were interested in building the input/output equipment, the punched card stuff. They weren't interested in -- this central computing didn't is there them yet. The company that was interested was RCA, who then abandoned computing. Probably one of, you know, the worst business decisions in all time. The head of research at a RCA, they also had a very large, called a David Sarnoff lab, or later called the David Sarnoff labs, Princeton Research Labs which was very much like this, it sort of did pure unapplied research and applied research all under the same roof. That was Vladimir Zworykin, Russian, and he and his protegee, Jan Rajchman, they had this vision for a digital vacuum tube called the Selectron. That's a Selectron which stored 4,004 kilobits in a single vacuum tube that was fully randomly accessible. So it really was an integrated circuit on a vacuum tube, not a chip. And they had a very hard time getting it practically realized. But they started to. So there's Jan with one of those. It was a big vacuum tube, but it would have cautious that would have -- so von Neumann believed that had this thing was going to be ready in about a year when he started his project. He was going to have a sort of plugin -- just like plugging in a chip today, you plug in your little memory modules and you would suddenly have however big you wanted your computer, just plug in more of those tubes. So the origins of von Neumann's project was actually at RCA. The first meeting was in Zworykin's office at RCA. You can see there John Tukey is part of the group, who's sort of representing the Bell Labs, Claude Shannon side, and then the bulk of the people are actually from RCA. Not -- nobody from the IAS except von Neumann and Goldstine from the army. So in that first meeting they issue these sort of three commandments that the heart is a central clock, it has a central clock cycle. The design is modular, so you can plug things in and out if they don't work. And words coding the orders are handled in the memory just like numbers. So that's the origins of the stored program, that the data and the instructions are intermingled. And here we have one of the more amazing things I found in Julian Bigelow's basement boxes of stuff of after he died. This is a crumpled up piece of paper that had been uncrumpled, and it -- with no date but it says -- it doesn't say bid, it says BD for binary digit. So it would have been early 1946. Let a word 40 binary digits be two orders, each order is command and then address. And as far as I know, that's the first reference to command line that you -- actually they gave the address first and then the command. And then at that address you may have another command because you -- when you go to an address you get 40 bits, so there's two more instructions and two more addresses. And there -- this is after the war. The institute is absolutely full. They not only had all the professors and visitors, they had the entire league of nations economic group from Geneva who were orphaned by the war and got invited to Princeton were on the top floor and squeeze everybody else in. So the only room they had to start this computer project was this room next to Godel had an office and he had an officer for his secretary, but he didn't want a secretary because he didn't -- he didn't produce much. He felt nervous having it. So he had an empty secretary's office, and they moved in there. So that office 211. So it's ironic that the genesis of this sort of modern computing architecture actually is right next to Godel. And there's similarities because what -- really what Godel did with his idea of Godel numbering concepts with arithmetic numbers and manipulating the numbers that then manipulates the concept is essentially what you're doing in this method of computing, that you're giving -you're assigning these words, numerical addresses and then manipulating the addresses. And it very much was sort of Godel's ideas brought to life. And that -- so Arthur bushes, Herman Goldstine and von Neumann wrote that sort of this founding document in that closet next to Godel. It's their total budget from November to April, you know, four months is $4,705. And now the next month you see the first $4 for electrical work. So that's sort of the beginning of the end. There's moving from theory to actually building something. And the and/or gate. So they're doing all this with -- you know, trying to get this binary behavior out of these very analog vacuum tubes. Told number of tubes an machines, 3,474. Most of them were little miniature 6J6 tubes that were found to be more reliable than more expensive tubes just because they made so many of them. Diagram for putting them together. So it was a total shock at the Institute for Advanced Study to have people -- you know, real engineers, machine shops, soldering guns. That's Julian Bigelow, who was the chief engineer, Goldstine, Oppenheimer, who didn't do very much except sort of approve it. And so then other people came from other -- all over the world to sort of build the machine and build duplicates of it. Gerald Estrin went to the Weizmann Institute in Israel and built their first computer. The women who did the coding. Now they're getting the unpleasant letter from the director -- they're trying to figure out where to put these people. The only place is next to the men's room in the basement is all they get. And even then the historians complain about engineers moving into the basement. >>: [inaudible]. >> George Dyson: Huh? >>: The Weizmann [inaudible] decide to move from some portion of the work from there to Weizmann [inaudible]. >> George Dyson: Yeah. Well, Gerald Estrin came to the institute and worked for a couple years and then everything he learned he moved -- he then went to Weizmann and built their first machine sort of as a direct copy of this machine. And then the actual -- the old drum, there was a one kilobyte drum, and it actually went to Israel. The drum actually was shipped. And here the director is complaining that the computer people are coming up and taking too much sugar in their tea and so very much you can see all the sort of behavior starting. And everything was published in reports that as Jack Good says not only did the reports explain the design, how it was designed, they explained why it was designed the way it was designed. So that was really von Neumann's doing that he -- he declared that everything would be published openly. And of course that was very much to the benefit of IBM because there were no patents in the way when they wanted to commercialize this. So if you look at who was checking these out from the library, it's RCA, IBM, National Cash Register company. There are persons requested interim progress reports, IBM corporation. If you look at the original order codes, down here this is really where IBM was coming in because they were counting on doing all the input/output equipment. So actually a hard coded order codes were designed -- you have -you really have that sort of affect of oh, we don't have the drivers for your other printer, but we've got the IBM driver. So IBM had this inside track. The machine itself was very small. I mean that's not much bigger there on the screen -- about the same size as life size. You see here in the registers, address, instructions, address, instructions. So 40 bits wide, 10 bit address, 10 bit instruction, 10 bit address, 10 bit instruction. Down here the memory tube, which included a video amplifier because the signal had to be amplified, the faint little distinction between a charged spot and a not charged not, that signal was amplified 30,000 times. So any noise would destroy it. So you had to do that processing right in the memory tube itself. You couldn't take it over a wire, you would introduce too much noise. And then the discriminator that discriminated whether it was a zero or a one and the difference between a zero and one is this very fuzzy distinction. So the memory was always going haywire. And because you had one bit in each tube, all 40 tubes had to be working or nothing worked. It wasn't like a car that, you know, could run on seven cylinders. So that's James Pomeroy with one of these tubes which was just an off the shelf war surplus oscilloscope tube. That was again von Neumann's genius was to say we're going to do this only with available war surplus parts, we're not going to invent anything new, we're just going to invent new ways of putting it together. So what they did was took -- this is really analog technology and then figured out how to digitally switch it. Because the problem with memory, we think that later somebody invented core memory and that was a revelation. But these guys knew about core memory. The problem isn't the memory element, the problem is the switch. How do you have a thousand position switch? So the clever thing was this was that the switching was done by the analog deflection of the cathode ray electron beam. So the spots were the memory and then the switch didn't have any moving parts except electrons and that's sort of how they could get it to work and work so fast. The access time was 24 microseconds. So in 24 microseconds, you could go to any one of these spots and a get all 40 of them in parallel and get the word out. So it all became dependent on these very complex clock cycles where while the thing wagon being refreshed you went back in to any point. Each tube had its own little log book documenting its problems. Here he's -- memory is not working so he's taking a tube out and he's hitting it with a hammer, which would dislodge a piece of dust or something. They did bitmap character -- as far as I know the first bitmap characters. That was their high speed wire drive for input/output. Bicycle wheels. And that's an oscillogram of a word. That's a 40 bit word, which shows you sort of how fuzzy things were. And you're trying to use that to, you know, design nuclear weapons and do really hard problems where you need accuracy. So von Neumann was fascinated with this in a theoretical away. How you get reliable computation out of unreliable parts. He worked on that in the abstract and then his guys -- his engineers had to deal with this every day. So it was always the question machine error or human, where the people making coding errors or was the machine making machine errors? No use, went home before closing down in disgust. And midnight oil. And I'll just run through this. Really the reason -- probably the reason I uncovered this and I'm here is I was back in Princeton about 12 years ago and discovered this material in cardboard boxes in the basement of the institute because as a child I had been fascinated by the fact that they had built this computer, demonstration for Dr. Von Neumann. And Jennifer Chase happened to be there, also visiting. And I told her about it and she said oh, you got to tell Charles Schimony [phonetic] and Nathan Myhrvold who started this operation there at a trustees meeting. And she grabbed them out of the meeting and said George is going to show you these log books. And we went down and looked at these log books and that's when -- up till then the institute had sort of denied that this -- they didn't have this material in archives or anything, they just really -even people from the Smithsonian were told that no documents exist, and it was really Charles and Nathan who convinced the directors they need to do something with this material, and they invited me. So it's really how this stuff has finally seen the light of day. We can all -- we missed the -- this is 6:48 in the morning. So they worked all night. Stage 10, stop. Big trouble. There was a mixture of code problems. This is the problem with the -- this is really a disk crash. The drum surface has become scored. And they're utilize arguing, you know, was it your problem or my problem. So it's a arithmetic problem, not input, output. False start, machine or human, maniac, which was the -- the acronym for the machine was mathematical and numerical integrator and computer or maniac. Maniac lost its memory, maniac regained its memory. They say see Pandora for details. I think Pandora was a box where they kept notes sort of to the next shift, you know, we're having problem on number 32, stuff they didn't want to put in the log book. But they wanted -- AHA. Code error, machine not guilty. And all these hexadecimal codes are quick sums. They would quick sum the entire memory. Because every quality was done twice. You couldn't just do something once, you had to duplicate it and then see that it came out the same. Code error, machine not guilty. Found trouble in code, I hope. Dammit, I could be just as stubborn as this thing. The hell with it. [laughter] master control off, way off. Now a real problem is this is New Jersey where it's very hot. IBM machine putting a tar like substance on cars and tar is tar from roof. It was so hot in the machine room that tar was melting out of the seams in the roof. Now a mouse has climbed into blower behind regular greater rack, set blower to vibrating. Result, no more mouse and a hell of a racket. And you can tell this would have been a book calculation because it was 4:50 a.m. on May, 1953. And you can go to the other notes and see they were actually doing a thermal nuclear calculation where they would run all night. And then another engineer has written here lies Marston Morse. And that's a -that's a dig as Morston Morse who was one of the pure mathematicians to objected to the computer being there. So high speed, which was -- top high speed was about 16 kilocycles. The low speed was about two kilocycles. But the machine, they could run the machine -it was asynchronous, so they could run at any speed. You could -- while you were debugging you could run one instruction at a time and just -- and then you could actually see the bits in the tubes and figure out what was going wrong. I have now duplicated both results. How will I know which is right, assuming one result is correct? This is now the third different output, I know when I'm licked. So, you know, it's like having three watches and a they're all different. And now later, now it's 1958. Von Neumann has died in 1957, they're about to close down. All troubles were code troubles. So it really switched -- they got the hardware running reliably, now all the problems were in the software, which is sort of the state we're in today. What's the use? Good night. And then over -they weren't supposed to say that they were working on bombs, but over to the bomb guys. And that was the result. So this huge explosion is a result of these calculations in this little artificial universe. So Oppenheimer, who was so criticized for slowing down the hydrogen bomb project actually was supporting it through this machine but couldn't talk about it. So now this is the day that we bombed Nagasaki Stan and Nick can now act openly, so now this veil of secrecy is lifted. That's Stan Frankel and Nick Metropolis who also Nick was another I think great example. He ran computing research at Los Alamos for a lifetime and a place that did fantastic amounts of pure and applied research, you know, and he was just had a gift for supporting the right things and encouraging real science. So they invent Monte Carlo for statistical methods in neutron diffusion. These are actually codes that are being run by hand. They don't have the machine running yet but they're doing the flow diagrams and the algorithms and they just have Klari von Neumann and Francois Sulan [phonetic], people like that just sat in a room and did this on paper. This is Klari and Johnny's notes on -- so this is where they turned the ENIAC into a stored program computer. This new method is based on a vocabulary, a set of orders which is conveyed to the machine on two levels, the background coding and the problem coding and that's really the ore gyps of sort of the operating system and the application that you set up the background coding and then you run the problem, and it's all -- you're potentially creating a virtual machine and then running on that. And there's Klari who just happened to be the right person in the right place at the right time to do these codes for her husband. And they had phenomenal energy. I mean so between 1946 and 1955 they crossed the United States 28 times in that kind of car, no seatbelts, no air conditioning, no FM radio, just leaded gas you know in a two lane road. And but gave people time to think. I think for a lot of von Neumann's best ideas came from these long road trips, Klari's security clearance which also just took very little paperwork to get cleared. She's explaining how she started doing this coding. She became John I's experimental rabbit and just found it fun. She had worked during the war when John was in England probably consulting with Turing, so we don't know, but she worked -- she got a job for the population research center in Princeton. So she worked on these problems of population theory which turned out to be exactly the problems of nuclear weapons. Because what you're doing in these big Monte Carlos, you're seeing how many -what percentage of the population dies or fishings, which is reproducing or emigrates which is leaving the zone in which you're doing the simulation. So she sort of was perfectly prepared for this when it happened. And did most of these codes. So here -- and the codes are very small so. In that envelope was a code for one have the thermal nuclear problems in 1950. And then all this marvelous correspondence going back and forth which Gabi [phonetic] translated -- revert to Hungarian when things got interesting. As a historian, it's terribly frustrating. He's talking about something and then suddenly it would turn to Hungarian, you knew something interesting was going to be said, and Gabi translated all this. And here is the answer to the question. The prevailing view is that the women who did this coding were just doing arithmetic and they didn't understand the physics and clearly that's not true. She's talking about changing the parameters of the tamper and these computations ran times for five or six weeks, and you had to adjust -- you sort of watched what was happening. So the event itself would be less than a microsecond, but it would take six weeks to calculate. And things had to be changed as it moved along. And then the find answer was a one bit answer. So it wasn't like today where you have a whole lot of input/output. You have almost no input/output until the very end. You're just recycling all these numbers in doing this -- these Monte Carlos. And then von Neumann, you know here he's hinting at -- you know, they were working on bombs. They were working on the most destructive thing ever conceived on the face of the earth. And here he says tell her factor four, which was in favor of the at this time the debate was whether you could actually produce a thermal nuclear explosion. Factor four is a gift of God or of the other party. So he's making some reference to this might not be entirely a good thing. And so to balance that, like a lot of physicists, I think Solard [phonetic] did the same thing, so he turned to biology later in his life. And then unfortunately he died too early but became fascinated with evolution, questions of genetics or he had sort of foreseen the structure of DNA in a way in a lot of his theoretical papers on self reproduction. So he brought this reasonably crazy Italian mathematical biologist Barricelli to the suit to work on these problems of experimental evolution within this artificial universe. So Barricelli comes in 1953, you know, can we create numbers that sort of have a life of their own and start evolving and doing things we wouldn't expect. It was a possibility of an evolution similar to that, a living organism is taking place in an artificially created universe. That's the first reference I find to really viewing this as an artificial universe to itself. So Barricelli had amazing rapport with the machine. It always most of the time worked for him. He thought directly in binary. He didn't even use arithmetic really. He just did this all in straight code and then would watch these patterns evolve and deal with these -- here the belts are burning up and machine will not duplicate Barricelli off. They let him have the machine even after the bomb guys left like at midnight or two a.m., they would let Barricelli come in and have the machine -- if it kept running he could have it until dawn. So a lot of his work was done late at night. Barricelli claims machine is wrong, code is right. So he made this universe cyclical so that it -- you know, it was only five kilobytes so he would run it, it was like an ant crawling around inside the wine bottler or something, he wouldn't see where the edge was. And then sampled it every hundred generations, he took a slice of the memory and then would output it and print it out. And he claimed to see all the sort of biological behavior. A lot of people thought he was nuts. But I think in a way he was right. These are essentially one dimension -- one dimensional cellular automata that reproduced. And he found that the real driver of revolution was cross-breeding, just like we found in biology. It's not some mutation is going to produce some fabulously more successful organism, it's just by cross-breeding between different organisms, you will have more fertile offspring. And here he sees a virus that invasion technique is based on its ability to make use of the game strategy program of the host itself. So that's really I think the first virus I know of. All done -- all the output was done on punched cards. So again, thank goodness this was saved. There's a whole pile of these im -- sort of fossilized imprints of these universus. Then he's realizing he really needed diversity. He started three universities in exchanging things, running for more generations. But they just -they're not going to become living organisms, they're just numbers. So they need to have -- in biology we have a genotype and a phenotype that you have to let them express this somehow and so he started doing that with taking the genetic sequences of these numerical organisms and using them to code for moves in a game program like a very simplified game of checkers or chess and then he started noticing improvements. So very interesting early stuff. And because he was a biologist he saw all the parallels between this and what was going O I mean his real interest was in explaining how the origins of his genetic code in, you know, at the beginning of life. And he had a theory that I think has a good chance of being right that if you're trying to explain how the genetic code evolved, how did these nucleotides become something as complex as DNA and RNA. His view is that you have these perimeter molecules that form collector societies like social insects and that the nucleotides would go out and they were sort of collecting amino acids and bringing them back to the nest the way ants go out and collect food. And that that became just more and more evolved and sophisticated until you started getting these long self-reproducing strings. And that's a way of sort of making, no, we don't need any hugely improbable step to get from the beginning to what we have now. And in a way that's similar to what happened in computing. You the take a very abstract view of computing you have the order codes who start also collectively in a sort of Barricelli way sort of cooperating to bring back machine cycles and things that are valuable to the order codes and make the order codes become more useful to the outside world so they get reproduced. And so the way that Barricelli imagined was trying to create in the laboratory actually went out and happened in the real world. He died in 1993. And then when I went back there that time I met Charles and a Nathan and got to look in the basement it turned out there was a seventh box. And in the bottom of that box was this box of punched cards, and that was the code for one of these universes of Barricelli there with the instructions from the engineer. Or his instructions to the engineer telling how to run it and then the engineer saying back there must be something about this code that you haven't explained yet. And I think that's the -- still the mystery is how do these purely deterministic formal systems that sort of [inaudible] believed in and you have purely sequential codes you get -- you know, you end up getting -- they're not unpredictable but interesting behavior that's beyond what you you would expect. So von Neumann wanted very much to work on this. His collaborator was Ulam, and they were working on a book that was going to be -- was going to do for self-reproducing codes and sort of what they -- what he and Morganstern had done with theory of games and economic behaviors. This is an outline for one of the chapters. Turing, not Turing. So there's all the things you can do with a Turing machine, the things you cannot do with a Turing machine. And then at the bottom Ulam. And I don't know what that is, but I have a guess. There's Ulam. And this is from a note -- letter that they exchanged in 1952 after the Mike bomb explosion. So the -- they're talking about a unbounded two dimensional digital universe in which you start having Turing complete two dimensional cellular automata that are universal and can sort of organize the random space around then, and they start competing and you start having evolution. And so they were arguing about where this would go. And I think that's what he meant by Ulam explanation mark was if you had this sort of world that he and Ulam were imagining what would happen. And that was this big -- you know, they only had five kilobytes memory and they were thinking about that. And there was a trajectory that von Neumann of course died before he could really do the and see this come to fruition but he -- this was later sort of extracted and published by the office of naval research. So he was the perfect project leader. Bigelow and Goldstine fought constantly about everything. They argued about everything. They didn't get along. They said nasty things about each other, but they were unified by von Neumann to work together because he just had the ability to convince people the goal was more important than fighting about it. Here Bigelow explains that the tidal wave of computational power was about to break and inundate everything in science and things would never be the same. So that's what motivated these people to give up all their patent rights, to give up jobs in industry, you know, to make this thing worker in such a short period of time. And Bigelow worked probably closer than anybody else with von Neumann about actually getting the machine done and knew -- von Neumann didn't write anything -- he didn't publish anything until he was absolutely sure. He didn't publish any half baked stuff. So we don't really know what he was thinking about the future of computing. But there was hints in what Bigelow says. And this was also -- I think profound insight that we sort of ended up doing all this sort of recursive programming because the machines are so fast it's the only way we can feed them enough instructions is to write very iterative programs. But that the machines themselves actually could operate very differently. And I think we're starting to see some of that in the commercial world out here now where a lot of computation is happening that isn't being programmed, it's machines -- I mean it can be completely trivial like Facebook pages updating and so on, but we're -you know, you're not actually programming an individual event. So von Neumann was thinking about this in 1949 about the future and very much the future unfolded, other than he did not envision these incredible reliable small machines and saw all too true that it's easier to write a new code than to understand an old one. And I think we all suffer from that. So he just had all these -- and then the institute, very un-- you know, in a way they were right. I could see both sides. But they decided this project should be -- what should they do? So they had these illustrious review committee review the project that, you know, one consisting of professor von Neumann he sort of had a group of them to himself. What should they do with it, should they fund it and hire full professors, should they shut it down and this was pretty much the verdict, that it was time for von Neumann to move on to something else. And so they decided at that time ->>: [inaudible]. >> George Dyson: Lighthill said that. So by that time von Neumann had gone -he was -- had become a commissioner of the Atomic Energy Commission. So the question was should the institute really establish what you have here, should they have a center of academic scientific computing research? And the answer from the institute sort of board was no. And this is Bigelow arguing that, you know, actually what they wanted to go into. They didn't want to build new or faster computers, IBM was going to do that, they wanted to deal with these fundamental questions of languages, computability, all the stuff that's still sort of wasn't worked on for the next ten years and then sort of caught fire again. So he wanted to do -- you know, biology, astronomy, Earth sciences and that all these things would come to life through computing, which is exactly what happened, but unfortunately it didn't happen there. So if you look at the problems they worked on, it was quite an amazing distribution. They really worked on five main problems. They worked on these nuclear explosions which go down to about 10 to the minus 8 seconds. Von Neumann loved working on shock waves sort of this that period. And then they worked on meteorology, periods of days -- hours to days to weeks to months and even climatology would be years. And then Barricelli worked on biological evolution and then Martin Schwartz ran a long series of code on the evolution of the solar system and stars. So a range of 10 to the 17th to 10 to the minus 8 seconds all in five kilobytes. And then if you look on the other side of sort of things we can comprehend, the shortest is a blink of the eye and the longest is your whole human life. And interestingly enough is right in the middle of that range. And then the exact middle is about the length of a Hollywood film, sort of the most you can keep somebody's undivided attention to tell a story is sort of two hours. So for some reason we happened to be right in the middle of the time scale of the observable universe and that -- whether that's an accident because we're in the middle or just coincidence but what computing did is sort of expand our horizons into these areas that were too fast to perceive or too long to perceive. And we're still -we're still very much doing that. So von Neumann then tragically 1956, he gets cancer, he's now in a wheelchair, Eisenhower is giving him the metal of freedom. And now he's in the hospital, spent the last year in Walter Reed Hospital. And this is a note left by someone who visited him and kept notes, he no longer could write. We talked somewhat randomly. So he's saying he can no longer, you know, hold a book, he has to be read too out loud. But he imagines a device -there really is the personal computer, you can project books, read them, voluminous pointer to be in several colors with a method of erasure, all the things that -- you know, Power Point we just take for granted today, he sort of imagined there. He said such an intervention was difficult but not impossible. The idea is to be able to read and write purebred in consciousness without physical interference. And that's really the world that he brought us is where we can just do all this intellectual stuff without having to worry about the details of keeping track. February eighth, 1957 he dice, is buried in Princeton, the same cemetery as Godel. Then he turns the project over to Princeton University who 12 midnight July 15th, 1958 it's shut down for the last time. That's off 12 midnight Julian H. Bigelow. So he ran the last -- he ran the first code and the last code. Ulam. He died so prematurely, seeing the promised land but hardly entering it. And it really was, it was this dream of Leibniz of a purely digital universe that would encompass having. And that's -- and you guys are taking this wherever it goes next, which we don't have to wait long to see. Thank you. [applause]. >> Yuval Peres: Questions? >>: You mentioned Bacon, Leibniz and others as the forerunners. What about Babich? >> George Dyson: Yeah, Babich was certainly a forerunner. >>: [inaudible]. >> George Dyson: Yes. Using mechanical engineering Babich envisions really the whole thing, programming and sort of reinvents it. And what Babich -- what's odd because Babich -- electricity was accessible to Babich but he just wasn't -he was at the time when he could have been thinking electrically but he just only saw mechanically. So I left out Babich essentially because Babich has been so well covered by historians of Babich. But it's a fascinating story. And Turing new of Babich -- there's a direct linear and between Babich and Turing. >>: Also the future sort of James pronounced judgment on [inaudible] when you were even younger. >> George Dyson: Yes. Well, I may have taken that a little bit out of context. He did say that this work is important, but it doesn't belong at the institute. Which may have been -- may have been true. I don't know. >>: [inaudible]. >> George Dyson: I think it was a great loss. I mean, you see the institute now, they've built new buildings. They have different schools. The biologists get along with the historian's. They could have supported a school of computer science with one machine and theoreticians it would have been fantastic. >>: [inaudible] somewhat ironic that this huge advance date and programs all the same, today is the biggest weakness of [inaudible] gets from anyplace to anyplace. >> George Dyson: You're exactly right. So why do we except these things as absolutely assumptions? That was done at the time because bits were so scarce you had to. But we don't have to now. So why don't we rethink that? I think that's a very important thing to work on. >>: People do. >> George Dyson: They are. Good. >>: [inaudible]. >> George Dyson: Yeah. Yeah. So I'm glad -- I think that's very important to look at these architectures completely fresh because of the -- if you gave those people the technology we have now, they would do it completely differently. I mean ->>: Do you know what [inaudible] Princeton? >> George Dyson: The which? >>: [inaudible] computers? >> George Dyson: No. >>: [inaudible]. >> George Dyson: Using solitons for computation? >>: [inaudible]. >> George Dyson: Yeah. Anyway, that's very existing. I think that's exactly -and I understand you're doing -- you know, you're working on quantum computing here. Even -- I think the application may be completely different than what you expect. But any thinking about different ways of doing things is ->>: You say a few words about Klari's original background. It's pretty impressive that she became such an expert on ->> George Dyson: Yes, Klari was a -- I mean her main achievement in life had been to be national figure skating champion at age 14 and then went off to boarding school in England. I mean it was very non-academic and just loved life. And the fact that she did [inaudible] work and took to it so well was quite extraordinary. And sad that, in a way that it was because of the classification problem it's not really known: >> Yuval Peres: Thank George again. [applause]