1

advertisement
1
>> Amy Draves: Good afternoon. My name is Amy Draves, and I'm here to
introduce George Dyson, who is joining us as part of the Microsoft Research
visiting speaker series.
George is here today to discuss his book, Turing's Cathedral, The Origins of
the Digital Universe. Over 60 years ago, a group of eccentric geniuses
gathered to create a universal machine. An idea that had been put forth by
Alan Turing. The computer they built came to be the digital universe we know
today.
George Dyson is a science historian as well as a boat designer and builder.
is the author of Baidarka, Project Orion, and Darwin Among the Machines.
Please join me in giving him a very warm welcome.
He
>> George Dyson: Thank you. It's great to be here once again. Can we dim the
lights? If there is a way. And trying to remember how many times I've been
here, but for different subjects and different things. And I'm here on the end
of a -- was here in Seattle three weeks ago and have been all over the place.
I think the high point -- or the low point of my trip was leaving New York City
a couple of days ago on the train to Princeton and I had -- had to be in
Princeton at 2:00 for a radio interview from Boston, from NPR, and I missed the
train. And I called the publicist and she says, sorry, you got to do the
interview from the train on my cell phone. And if you'll see what the story's
about, but this is about the [indiscernible] project to get this very fast,
first five kilobyte computer running, primarily to solve the feasibility
question of hydrogen bomb.
And I'm on the train, doing this interview on my cell phone, and that's what
the interviewer starts asking me about, this whole, why did they need the
computer to solve this hydrogen bomb problem? And I start trying to explain.
I'm in between the cars, and then, you know, I notice everybody in the car is
looking at me, like they probably called the New Jersey transit police. And I
totally lost it. It was really -- otherwise, everything has gone well.
So I think I first came here to Microsoft in 1982 with my sister, Ester.
Anybody here know her? In the old day, 80 percent of the people in the room
would have known her. But she was a good friend of Bill's in the old days and
used to stay at his house, swim in his pool, in a way helped him get started.
2
And but she doesn't drive so whenever she came -- and I lived, at that time in
Vancouver. So whenever she came here I would be her driver so I got to see the
inside of what was going on here. I think you had 80 people then. And
everyone always says, you know, because my father, Freeman Dyson, how hard it
must have been to be Freeman's boy and, you know, live in that shadow. It was
much harder to be Ester's little brother. That's Ester at school.
Because she was so good,
later I would come along
And there we are playing
and I'm 10. And now she
and she was a year and a half older. So a year half
and just always be a disappointment to her teachers.
a scene from Casablanca. So I think there she's 12,
starts writing a newsletter.
This is like a blog. 1967. Her first newsletter. She's just going to write
what she -- I think she has 40 subscribers, or followers. That's Ester's. So
that's April of 1967. Just a few months later, she goes off to Harvard at age
15. I'm working on the Crimson, the Harvard newspaper that all those famous
people once worked on. And that's where she met -- that's really what put a
lot of what you have here as you know came from Harvard.
Then she got hired by Forbes and then got bored with Forbes and kind of moved
over to the other side, to the stock analysis side. So now it's 1981 and she's
writing, they asked her to report on this new thing called software. You know,
is it going to be worth investing in for Oppenheimer securities. What is
software? Software turns the computer from an engine, raw computer power, into
a machine. A computer system that can accomplish something useful. So she
sees that.
Then she takes over from Ben Rosen, who started Compaq, and ran this
newsletter. It started out being just about the semiconductor industry, about
what chips were hot and stuff. But turned into sort of a newsletter. The
first newsletter, really, about the personal computer industry.
So now they send her to visit Microsoft, October 1982. A visit with Microsoft,
a nice little software company grows up. So you see Microsoft's really moving
up. It's now fourth behind Tandy, VisiCorp and MicroPro. It's got $11 million
revenue for the year. And it's pulled ahead of digital research and
Ashton-Tate, who the year before were still ahead.
And then she gives her analysis that the company has just released its first
3
applications package. Starting to think about strategy and business plans.
This is sort of the beginning of the end. And Microsoft has more than just
money. It has about 180 people, including 90 programmers. That was last week.
How many do you have now? Anybody know? Many thousands.
And it's no longer going to be a language and systems software playground for
creative programmers. We suspect that the company might move on to direct
end-user sales. Hasn't made this jump yet. There's no way Microsoft can
become the success it looks to be without abandoning some of its charm and
becoming more structured, more organized, and more market-oriented. It's
always a little sad to see a company grow up, but the result can be fairly
terrific if the process is handled well. So she sort of called the whole
thing.
And the rest is history. Those were the people, the big players at the time.
This was Ester, she started doing a conference, and had a marvelous ability to
sort of bring, it was like a high school dance with the guys like Bill on one
side of the room and the Wall Street bankers on the other, and Ester got the
bankers to invest in the company. So there's Dan Fylstra, does anybody
remember him? He wrote visicalc, which was the first -- or he was the
co-author with some dispute of visicalc, was really the first spreadsheet. And
Bill and Gary Kildall. Anybody remember Gary Kildall? Yeah, Gary Kildall
wrote DR-DOS; which in a way became MS-DOS. And John Sculley, who was, thumbs
down for John Sculley. John Sculley was brought in to kind of, you know,
handle Steve Jobs, who was a little out of, you know, a little outlandish at
that time. This was the year -- in fact, this year, they were still really
buddies. John and Steve were working together and then, of course, they had a
falling out and Steve left. So here we are, all these years later.
I was a -- at that time, I was living in -- I wasn't living in a tree house
anywhere, but I had lift in a tree house up in Canada. I was living on the
ground but watching all this through reading Ester's newsletters. And it drove
me to write this book with the long-term evolution of computing. I went to
Ester's conferences and was very disappointed. I thought I was going to meet
all these people who really thought about the fundamental importance of
computing in a serious way, and all they did was argue about, you know, whether
we should have five-inch floppy disks or three-inch floppy disks or which file
structure would win or, you know, what kind of -- whether to do debase 3 or
debase 4. So my revenge was sort to have write this deep, in a way too deep
serious book about sort of the history of computing going back to the 17th
4
century.
And one of your people here, Charles Shimoni, who for a while was chief
architect, who's Hungarian and actually was the person who brought Ester to
Microsoft so when we first came here, we went to a Hungarian restaurant in
Bellevue somewhere with Charles. He's a huge fan, as all Hungarians are, of
Johnny Von Neumann, the person whose three-dimensional holographic bust is in
Budapest on his 100th birthday.
So Charles was very taken with what I wrote in Darwin Among Machines about
Johnny Von Neumann, and ten years ago, twisted the arms of the people at the
Institute For Advanced Study in Princeton to allow me to go look for relics of
this project that Von Neumann had done. And that year was very successful. So
I not only -- there was not only lots of stuff in basement of the Institute For
Advanced Study, I found other basements that had stuff in it.
Sort of invariably, the best stuff is in a filing cabinet like this next to the
water heater. And the sort of dark, interesting period to me in sort of the
origins of the digital universe is from like about 1937 or 1938, before World
War II, until, you know, like 1957, after Sputnik, everything was very well
recorded. But there's sort of a period there we really, very interesting stuff
happened we don't know exactly what happened.
And in the bottom drawer of that filing cabinet was all the correspondence back
and forth between Johnny Von Neumann and his second wife, Klara, who he married
in 1938, and she came to America. But they were always -- he's in Princeton,
she's working, doing the programming for the [indiscernible] at the ballistic
research laboratory. They were always writing back and forth.
They were writing back and forth about all the problems in their marriage and
the problems in the programming of this early, these early computational
problems. So it was a wealth of material, all in the original envelope, dated,
postmarked from Los Alamos to Klara in Washington, D.C. In fact, some of this
was when they were out here in Seattle.
And that gave, if you read this book, that's what makes it into a kind of a
living book and a story, because there's that firsthand, like having the
emails. The what people are really thinking day to day, as all this exciting
stuff was happening before people could really even see what was happening,
because they were in the middle of it.
5
So I work at a former tavern up in Bellingham, and my kayak workshop, so it's a
beautiful place to sort of lay out those, you know, chapters that became a book
and see what's missing. I took ten years to write this book, which is almost
incomprehensible how long it is. And to put it in perspective, the group
working with John Von Neumann, they conceived, designed, built, programmed,
debugged and solved the important problems, finished the project in less time
than it took me to write about it.
But finally, it became a manuscript. My subtitle was A Creation Myth to sort
of remind people that this was not necessarily the truth, but it was the story
of how this digital universe came into being. And the publishers couldn't have
creation in the title. So it's, now it's the Origins of the Digital Universe.
And if you're a writer, you live in fear, there are sort of three things you
live in fear of from your publisher. First thing is the letters of rejection
telling you that, you know, it's a great idea, or it's a wonderful manuscript,
but we can't publish it. It's not for us.
And if they accept your book, then several years later, you get the letter that
I got lots of, saying, where is the manuscript? You're late, we're going to
cancel your contract. And if you make it past that hurdle and you actually
write the book, then another two or three years later, you get a letter, an
email now that says, we've come up with a cover design and we think you will
like it.
And it's almost invariably not the case. Covers are -- it's incomprehensible
how they come up with covers that so misrepresent, you know, the book you've
worked on all these years. In this case, Pantheon, this cover's brilliant. I
had no -- didn't question a thing about it from the moment I saw it. Just fell
in love with this idea of the cover that is an actual example of, you know, how
we made that switch from analog to digital at the beginning with punch cards
and then electronics and how it puts Alan Turing, the sort of meaning behind
the of Turing's Cathedral is like this whole structure is like a cathedral that
you can't tell really who built it, but somehow the fundamental mathematical
ideas of the Alan Turing are at the foundation of everything. So putting him
on the inside cover just works really well.
The upper graph is the dedication to Leibniz in the 17th century who made the
statement, who built digital computers and actually invented but did not build
6
the digital computer that ran with black and white marbles running down tracks
that were shifted, effectively exactly the same as a shift register in a micro
processor, but with marbles and gravity instead of pulses of electrons and a
voltage gradient.
But he reminded us it was not made for those who sell oil or sardines. It's
not all about ad revenue and -- although maybe it is, but there's something
else going on that's of fundamental importance.
But as some of you were sitting here the last few minutes, the book has 80 some
pictures. I'm just going to run through them, which sort of gives you the
whole story with a little more color than just the black and white in the book.
So the first picture is simply a view. This is the digital universe in 1953.
It's not a picture -- it's not a mapping of bits somewhere else, as what you're
seeing now is actually, you know, is map from the bits that are in solid state
in my laptop through various stages. But at the time, the bits were actually
stored as spots of electric charge on the face of a phosphorescent oscilloscope
tube. So those bits actually were the memory of the computer. And when you
watched this computer work, you could actually see the bits changing state and
moving around. So fundamentally different. And then it reversed so that then
we used cathode ray tubes to display memory that came from somewhere else. So
that's February, 1953. Machine's been running for about six months.
At that point, it's easy to remember there were 53 kilobytes of high speed,
fully random access memory on the entire planet. You know, which is, what, the
memory of an email today with no attachments. And a lot was done. And this
particular machine was five kilobytes.
This piece of paper came from another basement. Jillian Bigelow was hired as
chief engineer by Johnny Von Neumann, and somebody threw this piece of paper
out and he saved it. At the top, it says orders, let a word 40 BD, which is an
abbreviation for binary digit, be two orders. Each order equals C of A,
command, and then it gives the address coordinates and an address. That to me
is a fundamentally statement. As far as I can tell, that's the origin of the
command line. The fact that they would have a command and an address, and then
that sort of was cast in stone and that's how we do all our basic programming
still. And they hadn't invented the word bit yet.
And there's an entry in the journal also from March, 1953, stop at 18, 8.
18,
7
8 is the coordinates of a memory location where the computations stopped. So
the computation it stopped was being done by Nils Barricelli, a Norwegian
Italian viral geneticist who was running not simulations of evolution, but real
evolution within this five kilobyte universe. He was inoculating it with
random strings of bits and allowing them to cross-breed and self-replicate and
do things like that.
And the deal was that the bomb people from Los Alamos got the machine. They
had first rights to the machine. They would go home at midnight or 2:00 a.m.
and then Barricelli came in and ran -- that's the first night he starts letting
these creatures loose inside the memory. And then the engineers come back in
the morning and it's over, back to the thermonuclear hydro dynamic codes. And
it's this beautiful transition between these two problems they were working on
at exactly the same time. One whose solution could end up destroying all life
on earth and one whose solution could end up possibly creating an entirely new
form of life.
Alan Turing at age 5. And Johnny Von Neumann at age 7. They grew up, Von
Neumann in Budapest and Alan Turing in England in a very similar time. But
between World War One and World War II.
So what we know Turing best for is his paper published in 1936, written when he
was 23 years old in 1935 that we remember as the proof of universal
computation. That if you build one machine that do not, you know, just a
number of very simple operations, then it can do anything any other digital
computer can do if you give it the proper software to do so.
And great debate has raged among modern sort of historians over how much credit
did Von Neumann give to Turing. Did he read Turing's paper or not. Were
Turing's ideas important to Von Neumann. And I thought I just would go look.
So I went to Johnny Von Neumann's library at the institute, and now it's behind
one of these compact shelving where you have to turn the cranks to get the
vessels apart so you pretty much know no one's been there recently. And all
the volumes of London [indiscernible] mathematical society are there, and
they're all immaculate and perfect and everything else, except the volume with
Turing's paper in it. It's completely just absolutely unbound from having been
read so many times.
And that's what the engineers told me.
The surviving engineers I spoke with
8
said yes, when we showed up in 1946, Johnny said, you know, read this paper.
This is what we're trying to do. We're going to build one of these machines.
And such machines existed. Very difficult to get people to not say, oh, George
is talking about the first computer. I'm not talking about the first computer.
There were lots of, in fact, if you read the book, there was 11 computers that
were ahead of this one that are mentioned.
But until then, the memory was accessible only at the speed of sound. It was
delay line memory and things like that. So suddenly now, Von Neumann figures
out a way or his engineers do and we start getting memory that's accessible to
speed of light and that was a big transition. And the memory became
two-dimensional. The Alan Turing model, as powerful as it is, it's a
one-dimensional model where you have this not infinite, but unbounded string of
tape and that's where the memory is.
And then Von Neumann had the sort of practical implementation of that, which is
two dimensional, the two dimensional address matrix that still governs
everything today. Everything we do has a -- you know, you a have a 64 bit
computer. That's a 64 bit matrix.
So in England at that time, looks to us as if computing just sort of suddenly
came out of thin air after World War II. But it was incubating during the war
but in secret. On the British side, they were trying to break -- I mean, in
fact, it was their lives depended on breaking the codes. The communication
between the German high command and the U-boat fleet. If you've read
Cryptonomicon by Stevenson, just a fantastic portrayal of what that was like.
England's survival depended on breaking those codes. Alan Turing had this
realize idea in 1936. Only two requests for reprints of that paper came in
when he published it. Suddenly, it had a real implication, because what they
did with this machine called the colossus, of which they made 11 copies, it was
a, effectively a universal Turing machine. And by changing the state of the
bits in those vacuum tubes, they could imitate what they suspected might be the
state of the German machine that was doing the encryption and then they could
run through a whole lot of those permutations in a short period of time and if
all went well, break the codes within enough time to be useful. And they did.
And probably very good argument could be made that England won the war because
-- our side won the war because of breaking those codes.
9
That's Alan Turing in 1946 getting on a bus to a long distance -- a running
race. He was a long distance runner. And Johnny Von Neumann in 1952 with that
machine built in Princeton. So the memory is in those canisters with a 32 by
32 bit array in each can. There's 40 cans. So there's -- they're running 40
bit words in parallel, and one bit of each word is in the separate tubes. All
the tubes have to work perfectly, or it doesn't work.
Why did it happen in Princeton? Because Princeton is in the middle of New
Jersey. That's the Delaware estuary on the top where Philadelphia is and the
Rareton estuary leading to New York. And there was an Indian foot path between
the two. Which started being used by the people going back and forth between
the growing cities of Philadelphia and New York. A settler built a tavern in
the middle, Henley Greenland, so you could stop for a beer or change your
horses.
And that tavern grew into the town of Princeton, and then became Princeton
university. But the other side of the trail, there was still wilderness and a
group of Quakers led by William Penn, they settled there for the opposite
reason. They were trying to get as far away from the corrupting influences of
the city and they built a quaker meeting house, and that land ultimately became
the Institute For Advanced Study. And you still have these sort of opposing
cultures in Princeton, where the university is drinking a lot of beer and the
institute is drinking a lot of tea.
And they built this fantastic building in 1939. It was just the right thing at
the right time, because things were going terribly bad in Europe, and the
Bambergers, who funded this institute, were very strong on saving, rescuing all
these European scholars. Oswald Veblen sort of masterminded this. He brought,
some of you may know Paul Erdos, the great mathematician. They got him out of
Hungary for $750. They brought Stan Ulam out of Poland for $300. So all they
needed was some token appointment and they could get around the Visa rules
because it was an exemption for lectureships and save people from Europe. And
they saved a huge number of people.
Veblen also hired Norbert Weiner, the sort of founder of cybernetics. He had a
real eye for -- Norbert Weiner was writing articles for the encyclopedia
Americana. Couldn't get a teaching job. So Veblen just had an eye for
rescuing people.
The place was run by Abraham Flexner.
And his -- he sort of was the, you know,
10
the mastermind who convinced the Bambergers to do this crazy thing and his idea
was the usefulness of useless knowledge. That if you just let people work on
whatever they want to work on, something good will come of it. It's sort of
that Google gives people 20 percent. He gave them 100 straight off. You
either got 100 percent for one year or you got that 100 percent for life.
There were two classes of membership at this place. Bamberger is who funded
it.
The first mathematics faculty at the beginning. You've got Einstein, Marston
Morse, Oswald Veblen, Johnny Von Neumann and Oscar Morganstern, who founded -they wrote the theory of games and economic behavior together. Einstein and
Kurt Godel. Godel, the great logician, who I think didn't get enough credit in
the origins of digital computing. In this book, I try to give him a little
more credit that he deserves. His office, he worked closely with Von Neumann.
His office was above Von Neumann. And in his incompleteness proof of 1931, if
you look at it carefully, it's very much like a computer program that he
encodes logical statements with a numerical code and then gives them numerical
addresses, manipulates the numerical addresses with arithmetic that then
manipulates the underlying logical statements in a very profound and important
way. And I think those ideas translated directly to what, sort of how Von
Neumann set up this machine.
Von Neumann, at age 11, with his cousin Lily, doing math. And in 1915, at an
Austria Hungarian artillery position, and he's the kid sitting on the barrel of
the gun. He was very comfortable with the military. At a wedding of his
cousin in Budapest. And Budapest, between World War One and World War Two,
Budapest was just phenomenally a rich place. Rich in literature, art,
mathematics, and tremendous parties. And it changed the world when those
people left Hungary. A big part of the, what we think of as Hollywood really
was driven by Hungarian immigrants.
Don't know how many people know Bailey Bulabash who comes here. He comes to
MSR every year for two or three months, and his -- these letters of Klara's
that so much of the book is based upon are half in Hungarian and half in
English. They drift back and forth. And Gabby Bulabash, his wife, translated
all of those letters into English. Fantastic job.
Von Neumann's identity card from the university of Berlin that he, where he
resigned in protest against the Nazis in 1935. He came to America in 1930.
Princeton university -- Veblen was trying very hard to get Von Neumann, but the
11
gate keepers at Princeton at that time, there were only 23 Jewish students at
Princeton and it was almost impossible to hire a Jewish faculty member.
So Veblen had a marvelous way of getting around the rules. He had this
brilliant idea he would hire half of two Hungarian Jewish professors and that
didn't break the rules. So he brought Von Neumann and Eugene Jager, each a
halftime position. It was ten times what they could make in Europe. They both
said yes and came to Princeton and both stayed, greatly to our benefit.
Princeton in the 1930s. The guy who's had so much to drink he's lying on the
floor is Percy Robertson, who is teaching quantum physics to Alan Turing at the
time. Alan Turing came to Princeton in 1936 and was a, essentially getting his
Ph.D. under Alonzo Church there. So everybody's Hungarian there except Wagner
standing in the middle from Madison, Wisconsin, and Percy Robertson, who was
from Hoquiam, Washington. How he got to that position from Hoquiam, I don't
know. Von Neumann and Dick Feynman standing at Los Alamos in 1949. The photo
is by Nick Metropolis, who brought us the Metropolis algorithm.
Feynman, it's my favorite picture of Feynman. He's playing James Dean or
Marlon Brando. So these were peer mathematical physicists who worked
completely in the abstract and suddenly during the war they get to work with
high explosives and machine shops and things. They weren't going to go back.
They were not going to go back to just academic work. And the question was
what was next.
Obviously, more weapons. Feynman didn't want to work on weapons. But computer
was the next -- was going to be the next big thing. That's the Trinity, the
first implosion/explosion. So the British had built this series of colossus
machines. In America, we built the ENIAC, the electronic numerical integrator
and computer, built at the university of Pennsylvania at the Moore School.
Incredibly sophisticated machine that we ignore, we tend to forget how advanced
really was. Effectively, it was a multiple core processor. It had 20 parallel
processors that ran, did its computing at a very different way that's more
similar to the way we do it now in some ways.
But when Von Neumann walked in there and saw this thing, I think they
immediately, in a great debate over whose ideas they were, but he saw how you
could transform this machine into unless and actually run these more sequential
codes, which is what he wanted to do. So he wrote a paper, again under much
dispute, because he's the sole author on the title page. The record on the
12
EDVAC, which is sort of how you can build a much more sophisticated, fully
stored program machine. So all the elements of this modern computer were the
central processor and memory and all that are in that -- the architecture's in
that document. And how you do an adder, basic things like that.
The person in the middle here is Vladimir Zworykin, who also, I think,
strangely ignored who was director of research laboratories at RCA at that
time. And he was very strong -- he had brought RCA into television, been very
successful. They were doing well. He wanted to go into computing in a big
way, and then RCA said no. But the first meetings of Von Neumann's project
were all held in Vladimir Zworykin's office. You look down the list you the
fifth person is John Tukey, the statistician who worked with Claude Shannon.
And he's the guy who finally got up in one of these meetings somewhere in early
1946 and said, binary digit, come on, let's just call it a bit. Got tired of
saying binary digit. Then we had bits.
And in my view, how do these amazing things happen? You know, they
whole project with really between one dozen and two dozen people in
years with under a million dollars. And how does that happen? And
people helped. So I'm interested not just in the name brand people
get left on the project, but everybody from the bottom up.
did this
about six
all the
whose names
So important person there was Bernetta Miller, who was the administrative
assistant. She's the person who, you know, wrote those contracts with the
government, made sure the checks were cleared, and that kind of thing. She was
also the fifth woman in the United States to get a pilot's license. And here
she is in 1912. She's demonstrating the Bleriot monoplane to the U.S. Army who
at that time only flew airplanes with two wings. They were not going to buy,
you know, contract with a defense are contractor for a plane with only one
wing, because if you shoot out one wing, it's going to crash.
So they decided that the way to convince the Army this plane was really safe
was to get a woman to fly it. So she had the job of flying this thing for the
Army. And then her eyesight began to -- they wouldn't let her fly in World War
One as a combat pilot so she volunteered on the ground, for bringing supplies
under fire. And then her eyesight deteriorated and she ended up being the
administrative assistant on this computing project.
Emanuel Levies was hired at 16 to work on the ENIAC, just as a high school
student and then came to Princeton with Von Neumann and Herman Goldstein and is
13
still alive and well in Los Angeles. Came to my talk two weeks ago at the
computer history museum in Mountainview and they didn't want to put her on
stage, because it was being filmed for KQED. But at question time, she came on
stage and absolutely stole the show because she could, you know, microphone in
her hand, tell stories about Von Neumann and all these other great people
because she was there.
We're losing the last eyewitnesses, which is what I'm trying to do, get the
story down, because I think it's an important one. Were going to be living in
this digital universe for as far as we can see, and we ought to know how it
really started.
Norman Thompson was doing meteorological codes trying to predict the weather.
Did an amazingly good job if you get a five-day forecast, it's -- it really is
his codes, just running with a whole lot more processing and a lot better input
data.
RCA, under -- I think we -- we lost our sound? They were building what was
called -- they called the Selectron, and it's like the dinosaur with feathers.
It was a 4096 bit digital matrix. Fully digitally switched inside a vacuum
tube. So it's like the missing link between the analog and the digital. The
problem is they didn't get it debugged and working in time. By the time they
got it working, sort of Von Neumann had moved ahead with something else.
But in a way, this is, the architecture of this machine they built was designed
for these tube, and that's in a lot of ways why the architecture we end up with
was so well-suited when we did get silicon memory and silicon chips. It was
already there to plug in. The architecture had been expecting that from the
beginning. So that's the ancestor of your USB stick. Fully solid state
memory.
Oh and they built one machine that ran the [indiscernible] in Los Angeles was
built with selectrons, and they had 100,000 hours mean time between failure on
those tubes. It really worked well once they got it running. Now, the only -they were only 256 bit tubes. They had to scale back down. So they were sort
of too ambitious to begin with.
So the Von Neumann group was desperate. There was real pressure from their
sponsors to -- this hydrogen bomb question was imminent and they couldn't wait.
They went to Manchester, England, and they borrowed the memory tubes that
14
Turing's group in Manchester had developed, what we call the Williams tube,
which stores the memory as it spots a charge on the face of those just standard
orser plus oscilloscope tubes. And a very high gain, 30,000 gain amplifier at
the face of tube.
And they aren't calling it ones and zeroes. They're calling it a dot and a
dash. But how you distinguish between a -- you know, they had 0.7 -- three
quarters of a millionths of a second to make that distinction, whether it was a
zero or a one.
The machine itself was a work of, I think, engineering genius. And that's part
of the story I'm trying to tell is how that came to be. So Julian Bigelow was
a real engineer. I remember going to his house as a kid. He had an airplane
engine dismantled in his living room. He worked on real engines. And he built
this -- the computer really was built as a V-40 engine with 20 cylinders of
memory on each side, overhead intake and exhaust valves that are memory
registers and all done with what we, I think, would now call positive
interlock, with no bit moves without going into an intermediate register that
signals back, okay, I've got all the bits. You can clear. It doesn't -- so
you don't lose bits in transit. And that's one reason why, you know, again why
we can have these computers that run billions of cycles a second and don't lose
bits.
So it wasn't that big. It was two feet wide, eight feet long, six feet high.
The remains of it are still in the Smithsonian. Julian Bigelow, Herman
Goldstein, Robert Oppenheimer, who we remember as being so opposed to the
hydrogen bomb but was actually the director of this institute in supporting
this project and Johnny Von Neumann and all the engineers who went, as this
project was successful, they disbursed officers of other labs to RCA, IBM,
Stanford research, you know, all the places that built copies of these
machines, including about 11 different countries overseas.
Some of the women who did the actual programming, there was no place for
anybody to live in New Jersey after the war, so Bigelow went up to upstate New
York to an iron mine that was closed down after the war and cut these buildings
in half and brought them down to Princeton, against a lot of protests, and
that's where the computer people lived. That's the ancestor of your hard disk.
It's two bicycle wheels running a loop of magnetic recording wire, and they
were able to get 90,000 bits per second input/output, just using that very
crude machine. So that's a 40 bit word with index bits at the start and end of
15
the word.
So all of this was done with vacuum tubes, very -- that's a shift register so
you can physically see how there's that sort of intermediate register that
stores the bits before they are shifted back down to the next register.
A lot of the work was done by high school kids. All the wiring, everything had
to be done multiple times. That's the wiring diagram for one of the shift
registers. Jim Palmer with the pipe, Julian Bigelow in the middle, Herman
Goldstein and Willis Ware, who then came to Rand and was a very important part
in the beginnings of the internet at Rand. That's sort of another story.
They're trying to run what we now -- what they called Monte Carlo, probably, if
you look at what algorithm has solved the most real problems in the real world
is probably the Monte car low algorithm. And that didn't come out of thin air.
It sort of came out of the mind of Stan Ulam, who we rescued from Poland for
$300. And then he was also working deeply on the hydrogen bomb problem at Los
Alamos. And then got a near fatal case of encephalitis and was put in the
hospital and the doctors told him to stop thinking. Which was very difficult.
And his answer is okay, I'll stop thinking. He started playing solitaire.
I'll just mindlessly play solitaire. He kept playing solitaire, but he
couldn't stop thinking. And while playing all these games of solitaire, he
started wondering, well, how could you calculate what the outcome of the game
of sol /TAEURGS and then he realized that the game itself was a statistical
model of the process you were trying to solve and that would be a better way to
solve it.
He had this blinding insight and that became Monte Carlo, that you sort of run
this, using random numbers in the computer, you follow this statistical chain
of process, which was very -- which solved these sort of neutron diffusion
problems which had been very resistant to any other kind of mathematics and
solved the problem. And Klara Von Neumann who came in 1938. That's her 1939
French driver's license. And Johnny needed somebody to start writing the code.
There was a lot of housekeeping. They had nothing that we would term an
assembly language or anything like that. It was just all done in raw, absolute
addressing, raw binary.
Somebody had to do all that housekeeping and he trained her to do that. Most
of the early Monte Carlo codes were all coded by her. That was his, I thought
16
a Cadillac, but friend corrected me a couple of days ago, it's actually a
LaSalle. Bought a new car every year. Loved to drive fast. That's on their
honeymoon to key west.
Stan Ulam with Francois, who was another great source. She died just this last
year. Kept journals and could tell me what people were thinking at the time.
You know, why did they really want to build this terrible, horrible weapon?
She had lost her entire family in the holocaust and except one brother, and
Stan had also lost everyone except one brother. So they had their reasons to
want to make sure that these, if there were going to be such weapon, they would
be on our side.
Nick Metropolis, his badge photo. Playing the first game of chess against the
maniac at Los Alamos. They're playing on a six by six board with no bishops to
make the board simpler. You can see the input/output is still by five hole,
paper tape. Johnny Von Neumann at -- on a trip to the Grand Canyon in the
1940s. And all the horses are facing this way, and his -- or mules, and his
mule is facing the other way. That's Klara there.
At the Ulam household. Francois, Klara, who is still alive, Stan Ulam, Johnny.
And these are the -- drawing by George Gamoff that puzzles me because it's
Stalin with the bomb and that's Robert Oppenheimer looking like a saint. And
Stan Ulam and Edward Teller and George Gamoff. And what those three physicists
are doing, I think what they're doing is some -- represents somehow different
approaches to trying to ignite a thermonuclear reaction, but I don't
understand.
Von Neumann and von Braun. They together conceived the Atlas missile program.
And a whole 'nother thread in this sort of story of how we got this numeral
world came from Lewis Frey Richardson and the British meteorologist who, during
World War One, spent the war by hand calculating a difference equation model of
the atmosphere over northern Europe for one day. And the results were totally
wrong. But Von Neumann and Churny, who came in to do these calculations,
believed that the principle was right, just that they needed to reduce the
noise in the equations. And when they did, and could do it much faster, it
started to work.
And Richardson, I think, also answered the artificial intelligence question,
you know, can machines ever be intelligent. He said they can be intelligent
already so here's a model of a machine that's a nondeterministic circuit.
17
Having a will but capable of only two ideas. That's their first trip to run a
meteorological calculation. That's the first day got a 24-hour forecast in
less than 24 hours. They're celebrating. And very quickly, it went faster and
faster.
So if you look at what they did in just five kilobytes, they worked
fundamental problems where the mathematics was similar but the time
very different. In the middle, we have a scale of seconds in time.
worked on nuclear explosions where everything is over in, you know,
billionths of a second and they worked on shock waves, sort of what
the next few seconds.
on five
scales were
So they
in
happens in
And they worked on meteorology, which is sort of in the middle. That's sort of
in range of time scale of hours and days, after a few weeks, months, it becomes
climate. And Nils Barricelli showed up and worked on biological evolution.
What would happen over millions of years. And Martin Schwartz came and ran
stellar evolution codes. So looking at the evolution of stars over billions of
years.
You have 26 orders of magnitude in time. And the good thing about stellar
evolution was you actually had stars up in the sky and you kind of check your
results against the real world evolution of stars.
And if you compare that with things that we are familiar with, the fast, you
know, the smallest unit of time that we can perceive is like the blink of an
eye, third of a second. Or your entire lifetime, 90 years. And exactly in the
middle is kind of, you know, a working day at Microsoft. Eight hours. And I
don't understand why that is. Why are we right in the middle of the
perceivable. Is it because we are in the middle, it's going to be sort of
equally perceivable on both sides or what.
But to me, that's still some puzzle there I don't understand. And what the
computer did, which is what -- what Von Neumann very intentionally set out to
do is to enlarge our perception and did that very successfully. Brought us
into ranges of time that before fast digital computers, we really could not,
you know, calculate what would happen to stars over billions of years and we
couldn't see what would happen in a billionths of a second of a nuclear
explosion. And that just has continued to expand.
If you go into the world of science today, you'll see people who are working
18
down in the fento-second range and, you know, looking now at the evolution of
the universes over even longer periods of time. So Von Neumann was extremely
interested in biology. He and Alan Turing, they both died working on biology.
And to sort of in a way hedge his bets of inventing this terrible weapon that
could destroy all life on earth, they brought in Nils Barricelli, who was a
viral geneticist, ended up spending quite a few later years here at University
of Washington. So there's still people here who remember him. He ate at Ivars
regularly. Took all his students there.
And he worked again, directly -- most of the programmers worked in hexadecimal
notation. So, you know, you have a hexadecimal code for a 40 bit number. He
worked directly in just raw binary and these are outputs of his results. He
would run these simulations until they kicked him off the machine and then he
would save the memory in the core and print it out. Phenomenally interesting
stuff.
That's Alan Turing in 1951, back in England. He's now working on the Ferranti
Mark 1 machine. We remember Turing for this model of deterministic
computation. What we forget is that by the time he got to Princeton, he
already saw the limits of deterministic computation and what he worked on in
Princeton was a model of nondeterministic computation. He called them oracle
machines, machines that followed logical sequences for a certain number of
steps and then made intuitive leaps that were nondeterministic. And he
strongly believed that no machine could ever have any degree of real
intelligence as long as it was deterministic.
And so this machine, he insisted it have a source of electronic noise, sort of
like Lewis Richardson. And I think that's, in many ways, future we're living
in now, where we're building large, very large machines that are essentially
nondeterministic and start to do other, more interesting things.
Julian Bigelow would be a hundred next year. Alan Turing would be 100 this
year. Von Neumann dies in 1956. They turn the machine over to the university,
who can't get it running. They finally bring Julian Bigelow back -- they fire
him but then bring him back to can he please get this thing running. He gets
it running really well with about six people helping. And then they pull the
plug. So that's the last entry in all those thousands of pages of log books.
12:00 midnight, July 15, 1958. Off. It's a real tragedy for Julian, who still
had problems he wanted to work on.
19
Thanks so Charles Shimoni, when I really started digging this stuff out -actually, this is what I showed to Charles, and that's what prompted him to
twist the arms of the trustees and let me spend more time there. In the bottom
of one box in the basement in the corner, covered in greasy teletype manual
stuff was a box of punch cards. That's the actual cards and the source code
for one of these universes that Barricelli was running. With a note that said,
there must be something about this code that you haven't explained yet.
And when I saw that, I thought that's the last sentence in the book. All I
have to do is write, you know, the other 130,000 words. It took eight years.
But this takes us back to what Turing was trying to do at the beginning. You
know, Alan Turing invented this model, digital computation, not to invent the
digital computer. That really wasn't what he was thinking about. He was
concerned with this very abstract problem in pure mathematics called the
[indiscernible] problem, the decision problem of whether and, if I'm speaking
to programmers, I'm going to garble this for you. But the question of whether,
by given a string of logical symbols, is there any systematic way to determine
whether that string is a provable formula or not.
And David Hilbert, who posed that problem, thought the answer would be yes.
And Turing said out of instinct, the answer would be no. The way he got to a
firm no was by inventing this abstract machine, we now call it the universal
Turing machine, that he was able to prove could do anything that any other
machine could do, yet one of the things that this machine that could do
anything that any other machine could do could not do is come up with a
systematic way of determining by looking at a string whether it is a provable
formula or not.
And the implications for the real world are exactly sort of the world you're
all working in today, that you really can't tell -- and again, this is not
being formally correct. You know, it doesn't mean that you can't tell by
looking at a code what the code will do, but it almost translates to that.
That you can't -- you know, you're never going to be able to -- you cannot have
a firewall that keeps out all malevolent code simply by inspecting it. You
have to let the code run and you can never debug your products without, you
know, letting them run and seeing what happens.
To me, that makes the world much more interesting. I think that's why the
digital universe is never, no matter how much trillions of cycles per second we
have, it's always going to become more interesting at a faster rate than we can
20
sort of figure it all out. I think that's -- Turing left us with this very
profound result that's going to be with us forever. And it takes us back to
the world of Leibniz. This is a medal that Leibniz designed for the duke of
Brunswick in 1697, and the explanation of it is really that everything in the
world can be described in digital code and that if we understand digital code,
we can start to understand everything in the world. And that's all we're
really still doing.
The last picture in the book is me at the -- in front of Fuld Hall, where they
started this project. It's 1954. The project is already fundamentally over.
They've done the interesting stuff by the time I showed up. But I grew up in
this place where most of the people working there were doing very boring stuff.
But in this back building, you know, Julian Bigelow was still building this
machine or at least they were, you know, the scrap pieces of it were lying
around. To me as a kid, that was by far the most exciting thing. The hardware
was still really hardware.
And thanks to all the people and the institutions who supported this and let me
into their archives and basements and their sort of family secrets to try to
put this whole story together. And we have a little bit of time for a few
questions. Do we have any questions?
>>: You may not be able to answer this, but I was wondering, when I was
looking at photos that you had of the big machine that, the guy who put the
airplane engine in his living room, he was putting the machine together. In
your book, do you go into any of the reasoning of what he went through maybe.
Not so much the physics, but just the general explanation of his reasoning on,
you know, putting this here and putting that there, and putting it
[indiscernible] juxtaposing and all that? Because I think that's fascinating.
In other words, you know, what was going through their mind, what was their
reasoning on how they put something together. Why they would think that this
was the right way to do it.
>> George Dyson: Yes, I'll repeat the question. The question is -- do I go
into any detail about why the machine was designed physically the way it was.
And yes, I mean, according to some critics, I go into way, way, way too much
detail, because I think it's a very important question. Because every micro
processor you use today is functionally an exact blueprint of that machine.
Not because, necessarily, that machine was so much better. But it just
happened to be the one that was copied. Once you started copying it, people
21
wrote codes for it. Once you started writing codes for it, you had to follow
that design if you wanted to run the codes.
We're stuck with it like we are with the genetic code in biology. But Julian
Bigelow thought very deeply about this, and there's endless arguments at the
beginning. And physically, he was way, way ahead. I mean, the architecture of
that machine is actually three-dimensional, because he was concerned about the
time path between components.
So we now put things flat on a chip just because a chip is so small, it doesn't
really matter. But in this bigger machine, he saw very clearly that you wanted
the structure to actually be three-dimensional so there was a shorter
connection path between the components to get these very, very, very high
speeds that they got.
And then it was beautifully designed, which is hard to imagine to us today.
That computer required 37 different voltages. Micro processor just sort of has
plus or minus five volts or something. So all these different voltage
supplies. Vacuum tube has to have heaters to heat the -- heat it first and
then it has five or six inputs and outputs. So it was amazingly complex. The
wiring was brilliantly done with copper sheets that supplied all the heater
voltages like that.
So it's a sad thing that this computer is not on public display. It's in one
of these storage warehouses at the Smithsonian because it is, I think, a true
work of functional art. And it was, you know, it was built by hand. Another
question here.
>>: Von Neumann, in the [indiscernible] work doesn't mention Turing
universality. But he clearly knew about it. So I wonder, is there any
evidence that he deliberately left it out, or he was thinking of it, or what?
>> George Dyson: Yes, he clearly was familiar with Turing and universality.
The EDVAC report is a very strange document. There's a historian, David Greer,
who is looking into this with great detail with sort of handwriting analysis
and so on, because it was produced under very odd circumstances. And there's a
-- you can take a sinister view that it was, you know, that it was pushed out
there to avoid any possibility of patents on some of these ideas.
>>:
It was rumored that Von Neumann was a consultant for IBM at the time.
22
>> George Dyson:
1945.
>>:
It's not a rumor.
I found a consulting agreement dated May
So it was a bit hard on Eckardt and Mockley.
>> George Dyson: If you read this book, you'll find several smoking guns that
are very quite unpleasant, I mean, in the sense that it was Eckardt and Mockley
found the their own company and they had a good special computer, the Univac,
but they didn't have unlimited government funding. But they had a purchase
contract for three machines. Would have put them in business. And at the last
minute, their security clearance was questioned and they lost the contract.
And that put them into bankruptcy and that's what opened the door for IBM. And
so there's no direct chain of who did what or so on, but Eckardt and Mockley
were very bitter and I think they had good reason to be bitter.
So I admire Von Neumann tremendously, but I also would be the first to say that
there were people who were very annoyed by what happened.
>>:
It was Goldstein who put out the report with only Von Neumann's name?
>> George Dyson: That seems quite clear, yes. Goldstein issued it, and what
those circumstances were, it's very hard to tell. It's a very interesting
report, but it's very vague because at that time, you couldn't really talk
about electronic architecture. It was still -- the ENIAC was not declassified
until February 1946. Report came out sooner. So this is marvelously important
period in history. I think some day it may all get sorted out.
>>: Are there any documented meetings between Turing and Von Neumann in terms
of work they did together?
>> George Dyson: Sure. Oh, yeah, Turing was in Princeton for two years.
That's why it baffles me. Why do people make this big deal about no contact.
But Turing, they were in the same building. They were -- they drank tea in the
same common room. They went to the bathroom in the same toilet.
>>:
Von Neumann research --
>> George Dyson: Yes, and Turing, partly because of the war and partly because
he was homosexual and Princeton was not a very welcoming place, whereas
23
Cambridge was. And, I mean, he -- it's interesting. When Von Neumann comes to
America, the day he gets off the boat, he says, I'm in America. I'm Johnny.
You know, he just loved America from the first day. And the day -- Turing was
never really happy in America. And I think it's clear why he went back.
But then he came back during the war. Von Neumann went to England during the
war and Turing came to America during the war and that's sort of completely
black. We don't know what they worked on. But like Jack Goode, who worked
with Turing very closely, said that when Turing came back from his trip to
America and the war he talked about -- because, you know, they were chess
players. He talked about this, what if I had a chess board and you had bags of
gun powder equi-distributed at the intersections on the grid. What algorithm
would you use to decide whether you ignite one bag of gun powder does the whole
thing go off. That's really the criticality problem.
I think Turing had some input into the computational problems they were doing
at Los Alamos and I think Von Neumann, in fact, Von Neumann explicitly says
that he got his ideas about programming from a trip to England. He actually
says that in a, you know, a written report.
>>:
How alien would those early programs look to modern programmers?
>> George Dyson: Not alien at all. And vice versa. That's what's so
interesting that if you brought Turing and Von Neumann back today and took
them, showed them your code, they'd completely understand it. I mean,
nothing's changed. They would be horrified -- I mean, Von Neumann would be
horrified, why did you name this crappy architecture after me. Because they
were just interested in solving -- and Bigelow was heartbroken. I mean, you
know, he saw how to make this design so much better. I mean, he was just
wanting to go on to version two and we stayed on version one for 60 years. I
mean, once it started working -- and what happened was it became so cheap and
so fast. So we sort of tolerated it.
But they were both thinking about very different forms of computation.
>>: Online, someone says I'd be curious to see the computational logic that
you've immersed in as any [indiscernible] designing, piloting your small water
craft or [indiscernible].
>> George Dyson:
So the question is does this thinking about digital design
24
have any relation to designing boats, which is something else I do. No, I
would say no, that those are really different things. And I think you know
that from your work. I mean, that's sort of designing code where you -- things
have to be in perfect logical sequence is very different from, you know, making
pottery or something where you are dealing with form. It's a very different
kind of design. I mean, there might be some elements that clean, elegant code
is similar to a clean, elegant, you know, boat design. But very little.
I mean, so unfortunately not.
Yes?
>>: At the beginning of the lecture, you mentioned your sister.
the book? What are some of her thoughts?
Has she seen
>> George Dyson: Has my sister, Ester, seen the book. Yes, she has seen it.
Has she read it? You'll have to ask her. She keeps asking me, because they
did an audio version, they kept calling me up, how to pronounce Hungarian names
I don't know how to pronounce. So finally, I referred them to Von Neumann's
daughter, who is still alive, to help with the pronunciation.
Ester kept asking can I send her the audio version so that might mean she
hasn't read it yet. But Ester, I was just in New York and we had a great day
together, and she's zooming around in her orbit doing all kinds of great
things. And in a way, it's very satisfying to sort of put these worlds
together.
And I gave a talk in person to our third grade teacher, one of these teachers
who I thought had been so disappointed in me came to the talk and said no, you
were just different than Ester, but we weren't disappointed. Yes?
>>: First from Twitter, on the cover, the response is it's full of epic wind
and full of geek awesome. Good choice.
>> George Dyson: The cover is great. The cover doesn't mean anything, if
you're trying to decode it. The designer very -- which I think was
tremendously to their credit, I gave them one of Barricelli's cards. When they
said they had this punch card idea. And then I got a message back, is it okay
if we move the positions of the holes around. I said sure.
>>: The question, though, is as a technology historian, you said you started
writing this book around eight, ten years ago, right? And that influx at the
25
time we were in a very unlimited CPO, unlimited power mindset in software. Now
we've traded that off for this kind of internet of things, sensor becoming the
nervous system of the cloud and the cloud being this culmination of the big
machine, right.
How do you see that going in terms of your job as being historian of technology
when science fact starts outpacing science fiction? Are we on this technology
elbow? How is that making your job harder in that role?
>> George Dyson: So the question is as a writer about technology, how do you
deal with the rapid change in technology? And, of course, that's why I'm a
historian. If I write about things that happened in 1945, you know, my book
won't be out of date by the time it's finished, which would be the case if you
were trying to write about modern developments.
As to where this is going, I have very strong belief that we actually are going
back to analog computation. And people just won't admit it, because analog is
sort of a dirty word that is out with like polyester tires and 45 RPM records.
But it's a true provable thing that analog computation can do a lot of things
that cannot be done digitally. And if you look or if I look at what's going on
in the computational university, you see analog computation all over the place
and I think it's the new wave of the future.
We kind of keep calling it other things, like web 2.0 or web 3.0. But the
simplest way to describe it is something like Facebook, the complexity is not
in the digital code. The complexity is in the architecture. And it's like you
say, it's in the connections between things so you give every kid in high
school a very simple piece of code, and suddenly then you have a direct analog
computer solving the problem of who is friends with whom that purely digitally
is a very difficult problem to solve.
So these problems, you know, very known problems traveling salesmen problems
that can be solved much better in that way. And the companies that are doing
that, those are the ones that are doing so well. That's what Google is doing.
It's easy to collect all the digital data. It's hard to find the network of
meaning as to what -- so you let people do that. You let them link what they
think you give them 100 results and, you know, they click on one and you know
where the meaning is. And that's where I think is the future and it does
change the way you think about code. Last question.
26
>>: Can you comment on analog as one who has helped people with analog
computers. They have a habit of going unstable. They have artifacts in them
that cause them to go to unstable, having nothing to do with the problem you're
trying to solve. In fact, I helped a friend who was working on his Ph.D.
convert his analog to digital, resolve the problem and then put it back on the
analog computer so it wouldn't go unstable.
And I think the message is, these systems are so complex and they will have
instabilities that just will arise. And won't have anything to do with what
they intentionally design for.
>> George Dyson: Yes, there's really more an answer to a question pointing out
that analog systems, which is part of their power, right, that they have a lot
more power than you see. And you have to live with that and work with that.
Thank you very much.
Download