|
The Physicist
As director of Microsoft's Advanced Technology Group, Nathan Myhrvold oversees 650 serfs and spends $150 million a year to keep Bill on top.
By Stewart Brand
He's only 35 years old, but Nathan Myhrvold has enough to say about
computing to last him at least another 35. With a PhD in physics from Princeton
University, he taught himself programming, read Donald Knuth, and founded a
software company with grad-school cronies. He holds the most prestigious of
jobs in a company built on prestige: after joining Microsoft in 1986, Myhrvold
became chief "technologist" by 1989; the head honcho overseeing 650 serfs in
research and advanced product development. "I work directly for Bill, and
research works directly for me," he says. And this year he's one of six
executives promoted to Microsoft's Bill-centered "office of the president."
Meanwhile, Myhrvold remains a nerd among nerds who reads science magazines and
dreams of deploying the insights of physical science to master that still
unkempt discipline of economics.
Stewart Brand met up with Myhrvold during the TED conference this past spring
to hear him hold forth on everything from software efficiency to sci-fi.
Wired
: What has come to be called Moore's Law was first proposed
by Gordon Moore in 1965. He said that the number of components on a microchip
had doubled every year since 1959, and that the trend would continue until
1975. These days, Moore's Law is treated as a general statement that computers
get drastically better every year - faster, cheaper, smaller - and that this
will occur indefinitely. I take it that you see Moore's Law as something quite
fundamental?
Myhrvold
: The way Moore's Law occurs in computing is really
unprecedented in other walks of life. If the Boeing 747 obeyed Moore's Law, it
would travel a million miles an hour, it would be shrunken down in size, and a
trip to New York would cost about five dollars. Those enormous changes just
aren't part of our everyday experience.
The whole hardware industry has experienced the phenomenon in which every time
computers get cheaper, they appeal to a new set of users; every time they get
more powerful, old customers upgrade. But it turns out that like hardware,
software also has to undergo something like Moore's Law. I did a study of a
variety of Microsoft products: I counted the number of lines of code for
successive releases. Basic had 4,000 lines of code in 1975. Currently, it has
perhaps half a million. Microsoft Word was at 27,000 lines of code in the first
version. It's now about 2 million. So, we have increased the size and
complexity of software even faster than Moore's Law. In fact, this is why there
is a market for faster processors - software people have always consumed new
capability as fast or faster than the chip people could make it available.
Wired
: How do you measure software efficiency?
Myhrvold
: If you write your own software, which is generally true for
mainframes or supercomputers, for every dollar that you spend on software
development, you get about a buck's worth of software. It's a 1 to 1 ratio.
Now, if you look at minicomputers, you will find that the people in those
industries typically have 1,000 to 10,000 customers. They can afford to spend
considerably more on development and still maintain a decent return on
investment. When you spend US$10,000 to buy a minicomputer software package,
the company probably spent $100,000 to $1 million developing it. So, if you're
buying from that market, for every dollar you spend, you get somewhere between
$10 and $1,000 worth of software. Then you come to the PC industry, where for
every $100 you spend, you get a piece of software that cost someone $100
million to develop.
A dollar buys a million dollars' worth of software.
Call it Myhrvold's Law. As the software becomes more complex, doesn't the
likelihood of problems tend to increase? And when you always have to have new
software, won't it always have new problems?
We have gone from 27,000 lines of code to 2 million lines of code for the same
money. If I say I've got two versions of Word - that old one from 1982 that's
perfect, with zero defects; or the new one that's got all this cool new stuff,
but there might be a few bugs in it - people always want the new one. But I
wouldn't want them to operate a plane I was on with software that happened to
be the latest greatest release!
Wired
: What ever happened to the old aesthetic that fewer lines
of code must be better? Has it disappeared because of Moore's Law?
Myhrvold
: No. You absolutely want that, at any point in time. There was
a hilarious phase during our relationship with IBM when we ran into a lot of
difficulty because the key metric for programmer productivity at IBM was the
number of lines of code produced. Our people would go in and reduce the number
of lines of code, generating negative productivity in IBM's eyes. Bill used to
call this "the race to build the world's heaviest airplane."
Wired
: So, we now have 25 years of Moore's Law and 25 years of
Myhrvold's Law. Will we have 25 more years of both?
Myhrvold
: On the hardware side, I'm pretty confident there'll be another
20 years at least, which is another factor of a million. A factor of a million
reduces a year into 30 seconds. Twenty years from now, a computer will do in 30
seconds what one of today's computers would take a year to do. So, for
particularly big computational problems there's no point in starting. You
should wait, and then do it all in 30 seconds 20 years from now! That is the
hardware side. The growth of software is certain, because it's only limited by
human imagination.
Wired
: Well, what kinds of problems do you see requiring that
much computational power?
Myhrvold
: Everything will. At every point in the history of this
technology, people have argued that we don't need that kind of power. And every
single time, they have been dead wrong. There's no feedback mechanism yet for
people to stop asking that question, despite the fact that it has such a
miserable track record.
When you double something every year, you're ask-ing the culture to handle a
violent, recurrent change in the rate of change. Instead of an extra something
every year, you get two, four, six, eight, sixteen, thirty-two times as much.
Is that always a good thing?
I tend to be an optimist about this. It's a requirement for the job. This is
about enabling people to communicate, enabling people to do stuff. There was an
old TV special from the 1960s I saw not long ago. They had a young Walter
Cronkite, with dark hair, interviewing people who were terrified that
computerization was going to steal their jobs.
Wired
: I remember that. The secretaries union at Stanford fought
against the university buying word processors because they would be put out of
work.
Myhrvold
: Of course, it didn't really happen that way. I'm not saying
you can't find someone who is affected negatively, but the net positive so far
has been very big. I would make an evolutionary argument that says the thing
driving this is a positive feedback cycle. We like what we get, we want more,
which spurs people to create more. If there is a lack of positive feedback, the
thing just slows down.
Wired
: You started as a scientist, a physicist. What disciplines
do you most identify with now?
Myhrvold
: I noticed to my surprise and delight that, carried over from
past years, my accountant still had "physicist" down as my occupation on my tax
forms. But I also think of myself as a programmer.
Wired
: So astrophysics, then computers. Do you see it as a
challenge to blend what you think would be interesting in science with what is
possible with computers?
Myhrvold
: I think we're in a phase where some of the theoretical
underpinnings of our society are changing. We're in the process of
understanding how society will be impacted by the phenomena of widespread
information and ubiquitous communication. How will business be changed by that?
Can we create such a thing as an effective, scientific version of economics?
I don't mean to be rude to economists, but economics has not been something
that can be reduced to engineering. Economists are at the stage of weather
forecasters: they come up with a new explanation every year. Yet, some of the
real basic stuff can be done by looking at the mathematics of emerging behavior
and evolution.
Wired
: Who are you paying attention to mostly?
Myhrvold
: The coolest single thing is Tom Ray's stuff. (See Wired
3.02, page 126.) It demonstrates the robustness of the evolutionary process.
Ray has created a system that replicates many key features of biological life,
at least in terms of evolution. That this is possible is not so surprising. But
it was done so easily and with such a straightforward approach - his initial
stab at defining an instruction set actually worked the first time out.
Wired
: Where does this lead? Do you eventually get artificial
life models of the economy that provide valid guidelines for business, or what?
Myhrvold
: Ray and others have proposed that people will create software
husbandry. They'll go off somewhere and breed programs. Almost eight years ago,
I and a couple of people at Microsoft wrote a program like that - it uses some
genetic things for finding short code sequences. Windows 2.0 and 3.1, NT, and
almost all Microsoft applications products have shipped with pieces of code
created by that system.
Wired
: For example?
Myhrvold
: One is doing a binary coded decimal conversion.
You have numbers encoded in binary and you want to change them to ASCII. You've
got to shift a bunch of bits around and divide by 10 and then do all kinds of
stuff. This BCD program found a way of doing that much more efficiently. A
human programmer comes at a problem with a mind-set that causes him or her to
solve it one way; in fact, there's a large space of other solutions - and
evolution can find them.
Wired
: Does your Advanced Technology Group in Microsoft do basic
research, or is it more product oriented?
Myhrvold
: We have a group called Microsoft Research that operates at the
same level of exploration as a university or some of the industrial research
labs like Bell Labs or Xerox PARC. We've got around 100 people doing a lot of
forward-thinking stuff. We spend quite a bit of money in the areas of speech
understanding, natural-language understanding, and various ways of reasoning
under uncertainty, new paradigms for programming - stuff we may not productize
for five years, or ever.
Wired
: How much does research drive Microsoft?
Myhrvold
: Well, we're in this technologically driven world, and we're
all surfing this wave of technology. There are two ways you can go with that.
You can put a guy who doesn't understand technology in charge of your company.
He's going to be up on that surfboard while a bunch of guys on the beach shout,
"Go to the left! Go to the right!
Over to the side!" It's a bad way to manage a technology company, in our view.
One of the reasons we think we're going to be successful - and god knows, maybe
we won't be - is that the guy who's on our surfboard, Bill, understands that
wave of technology and grew up with it. Bill controls and manages the evolution
of technology across the company. I work directly for Bill and the research
group works directly for me - so, fundamentally, when you have a new thought in
research, you have to convince two people, me and Bill.
Wired
: As a manager of researchers, do you have any rules of
thumb about what projects might be promising to pursue? By "rules of thumb" I
mean the kind that I used as an editor - you publish something only if it is
new, true, important, and well written. The difference between research and
advanced development is the difference between doing something that no one
thinks is possible, but you do, though you don't know when. That's research.
Doing something that's possible - people agree that it is, but no one's ever
done it - that's development. When we evaluate new research proposals, some of
it's based on how cool they are.
Myhrvold
: What makes a project cool?
It's intellectually interesting; it does something that really would be an
advance.
Wired
: Every company has some mottos or slogans that everyone in
the company knows about. What's one of yours?
Myhrvold
: Programmability is a drug. I'm an addict. Everyone is. All
successful programs are programming environments of one sense or another.
You're using the fantastic capability of computers to be customized for a
specific purpose, and you are programming them, whether you know it or not. The
magic of programmability is something that most parts of society, most parts of
our culture, haven't touched yet. And when they do, they're going to get just
as damn hooked as I or anybody else is. The magic of having the world's most
pliable, flexible machine whose characteristics can utterly change just by
sliding in a disc - that is intrinsically powerful.
Wired
: That's an empowering, if busy, future. What else?
Myhrvold
: Positive-feedback-cycle economics. This phenomenon is
essential to understanding how you move into the phase where you have
million-to-one leverage rather than one-to-one leverage.
Wired
: What do you read? Books? Magazines?
Myhrvold
: I read a ton of science things. Science, Science News, New
Scientist, Scientific American, Discover. If it's about science, I get it. I
don't read newspapers.
Wired
: Is science fiction something that's of interest or use to
you?
Myhrvold
: I've read science fiction novels off and on for years. But
I've always thought it should be more useful to what I do than it is. We've had
science fiction authors like Greg Bear and David Brin come over to Microsoft
and give talks; to date that's been less successful than I would have liked. I
think the real thing is smart, free-thinking individuals - of which science
fiction writers may be an example, but they don't have any particular lock on
it.
Wired
: Is there anything all of your peers believe that you
don't, anything you ramble on about so people mutter, "Oh god, there's Nathan
going on about X again"?
Myhrvold
: I'm very interested in economics. We're getting to a point
where there has got to be a way for us to understand the dynamic processes in
economics at the same level we understand dynamic processes in computing and
the physical sciences. And perhaps biology.
Wired
: This is a physicist talking. A biologist would never say
that.
Myhrvold
: Today's biologist wouldn't. Before Newton, physicists wouldn't
have either.
Wired: There is a fundamental difference between
systems that build on their own unique histories, like life, and systems that
don't, like fire or the behavior of atoms. Economics is more like biology than
like physics.
Myhrvold
: Well, then, I guess I'm disagreeing with a peer!
Copyright
© 1993-2002 The Condé Nast Publications Inc. All rights reserved.
Copyright © 1994-2002 Wired Digital, Inc. All rights reserved.