You鈥檙e in the throes of trebling the size of your research arm. Is there a
big change in Microsoft now, in its approach to research, to become a founder of
big ideas?
Microsoft Research started five years ago, and in the intervening time, code
from Research has been in every major product.
Now that鈥檚 even against a backdrop where in the five-year horizon, most of
our major research projects have not come to fruition. So one day, our
speech-recognition work will be ready for prime time. We think that will be
within five years. Five years ago, we thought it would be five years, I鈥檒l be
honest about that鈥攊t doesn鈥檛 matter if it鈥檚 25 years. Our customers speak,
computers don鈥檛. On the day that you鈥檙e able to hold conversations with a
machine, it鈥檚 going to enable a whole new class of applications.
Advertisement
But in terms of Microsoft鈥檚 breakthroughs, you could include graphical user
interface applications, because the first graphical
spreadsheet鈥擡xcel鈥攚as Microsoft鈥檚. You might look at Microsoft
Windows and Windows NT which started in our group. Windows NT was part of our
strategy to expand our operating system franchise from little machines up to the
most powerful computers in the world鈥攇iant multiprocessors. It must have
been seven years ago we started that project, but in the last 12 months it made
over $1 billion. People project that it will be a couple of billion
dollars in the next year鈥攊t鈥檚 got an incredible growth rate. So there鈥檚
billions and billions created by an idea which was as radical in its time as the
research things we鈥檙e doing now.
So the perception that Microsoft is not an innovative company is wrong?
It鈥檚 completely wrong. The average software company is a single product
company. In fact, all of them except Microsoft are single product companies.
They find one great technical insight. That single product becomes between 60
and 100 per cent of its revenues, between 60 and 150 per cent of its profits,
because often other parts of the company lose money, and they never innovate
beyond that. So these companies lose their way technically.
In contrast, Microsoft has continually reinvented itself. It started off as
the Basic company. Programming languages were Microsoft鈥檚 sole income. Turns out
that today we鈥檙e still the leading supplier of programming languages, but you
don鈥檛 think of that because we鈥檝e added all these other things. As the use of
computers broadened from being purely for programmers, we developed graphical
user interface applications. We鈥檙e the world number one for graphical
applications.
We developed operating systems: we鈥檙e the leading operating systems supplier.
First that was character-mode operating systems鈥擬S-DOS. Then, we made the
transition to the next generation鈥攖o the graphical interface. Then we made
the transition from there to a fundamentally restructured, written-from-scratch
operating system鈥擶indows NT.
As multimedia capabilities came along we鈥檝e added multimedia products. We鈥檙e
the world number-one encyclopaedia company, with Encarta. We鈥檝e added games, and
entertainment and we鈥檝e become one of the leading Internet
companies鈥擨nternet and Intranet.
In a sense though, you鈥檝e followed in the footsteps of Netscape. They
pioneered the area
That isn鈥檛 true historically. The Internet is a thirty-year-old thing. And
for almost all of that time, most people considered it to be completely
insignificant. From a technology perspective鈥攖he technical protocols that
make up the Internet鈥攁re all very old. So the Internet isn鈥檛 a technology
phenomenon. It is a social phenomenon.
You can鈥檛 put your finger on the breakthrough that made it happen. But
suddenly you got a critical mass of people exchanging information. And I mean
critical mass in a technical sense. Metcalfe鈥檚 law says if you have n
people communicating, the value of that communication increases as the square of
n. Suddenly you get this enormous social phenomenon.
Now Netscape was certainly a company that commercialised it. But the World
Wide Web was invented at CERN, the first graphical browser was at Berkeley.
Founders of Netscape at a supercomputer research lab did come up with something
and that research lab should be congratulated. And Netscape certainly is to be
commended for being very adroit at commercialising this external technology. But
they didn鈥檛 invent the Internet.
Where is the focus of Microsoft Research today?
Progress goes in fits and spurts. We have the greatest collection of computer
graphics experts in the world at Microsoft now. And they鈥檙e doing tremendous
work.
Our customers, from four year-olds on, are trained by movies and TV to expect
the best in graphics. Why shouldn鈥檛 they be able to play with that? Why
shouldn鈥檛 video files be as common as text files? In the past couple of years,
we鈥檝e seen still images become as common as text files. And as memory increases
and chips get faster, we鈥檙e going to keep crossing these new thresholds.
Software is like a gas. It expands to fit the container it鈥檚 in. Every time
computers get incrementally more capable, we add incremental features and old
users will upgrade. Every time computing becomes cheaper, it suddenly becomes
accessible to a whole new set of folk. And every time it goes over an absolute
threshold of performance, you can get some brand new feature you never had
before. Graphical computing was ludicrous when computers were substantially
weaker. Suddenly, it became compelling. We could do desktop publishing. We could
do a zillion things.
In the next year, video will arrive. Video has the potential of
revolutionising the computer experience in every bit as much as still
images.
Will we all be downloading films from the Worldwide Web?
Yes. And you鈥檒l be creating films and animations. A memo that you might have
handwritten a few years ago, tomorrow will be a video presentation. It鈥檒l
include computer graphics from these new Intel processor machines that will go
under people鈥檚 Christmas trees this year鈥攖he MMX, Pentium 2 machines.
They鈥檙e more powerful than the workstations used to make Jurassic Park just a
few years ago.
And that鈥檚 just one element of this. Then there鈥檚 the Net. At the moment, for
example, the Net is fairly retro in that many of the things we take for granted
in other aspects of computing don鈥檛 exist yet on the Net. You can鈥檛 get decent
fonts, for example. But over the next few years we have an opportunity to pump
more and more technology into the Net. It鈥檚 already revolutionising the
communications world. It鈥檚 threatening the tariffs of the telephone
companies鈥攊t may make voice phone calls nearly free and give us incredible
video bandwidth as people take fibre to our homes or to the kerb.
Every way you look at it, there鈥檚 more opportunity in our industry today than
there was when Bill Gates started the company. And to capitalise on that
opportunity, you need to have both strong technology people and people who
understand the intersection between technology and business.
How much of your research is focused on creating technology for immediate
application?
One of the great mistakes that people make is directing research to be too
applied. It ceases to be research. It becomes development and often loses the
spark that makes it worthwhile.
There are things in our products that are examples of long-term research
reaching fruition. Take the grammar checker in Office 97, which checks your
grammar while you type, and is by any measure vastly better than any other
grammar checker. It is based on a parsing technology that was developed by our
natural language research group over five years. From that base, within a couple
of years we鈥檒l have it for a dozen major languages. It鈥檚 a framework on which
you鈥檒l see dramatic innovations in natural language processing at every
step.
That鈥檚 an example of benefit from a long-range project. But other benefits
have come incidentally as side effects of other research projects that we鈥檙e
doing. In fact, there鈥檚 sort of a scale-invariant thing here鈥攁nalogous to
scale invariance in fractals. Research results exist on all scales. Every 10
years you get your 10-year result. But it turns out that the same projects are
generating lots of little results in the meantime鈥攊nterim things that if
you鈥檙e clever you can grab and attach to something else.
Why have you decided to treble your research budget? Is it because you think
the research group is making money so you want more of it, or is there something
going on in computing that says now is the time to increase your research
budget?
It鈥檚 some combination of those. Mostly it鈥檚 because we think we鈥檝e been very
successful at it. When we first started Research, there were two real
challenges. The first was 鈥渃ould we create a world-class research group?鈥� And
there were plenty of people who said we couldn鈥檛. We weren鈥檛 known for research.
We were known as a company that worked on those iddy-biddy computers back then,
which weren鈥檛 cool. People wanted to work on minicomputers or supercomputers or
some such. Other people said your culture won鈥檛 be right to create a great
research group.
The second challenge was 鈥渃ould we transfer technology from research into
products?鈥� And, again, people said you won鈥檛 be able to do that. They said that
it had been 鈥減roven鈥� that research can鈥檛 affect products鈥攂y Xerox PARC,
Bell Labs and IBM Research. Even though they did good work, somebody else
commercialised their inventions. I argued that both of those were dead wrong: we
could create something and we would have an effect on products. And six years
later, we did.
Bell Labs and IBM are well known for blue-sky research. They have people who
are paid just to sit around and think鈥攏ot about products. Are you
following that model? Or are you doing something fundamentally different?
Prior to us starting our research group, we didn鈥檛 do research. We did a
bunch of innovative advanced development stuff. When we started Windows NT, that
was considered a very advanced development project. But it was a development
project. We set out to develop a product, not to say 鈥渃ould you do it?鈥� We
actually thought it was quite possible. There鈥檚 a real difference between
advanced development and research in that sense. So the notion of blue-sky
research was quite new to us five years ago.
To say that we鈥檙e just like Bell Labs or IBM is wrong. I鈥檇 say we鈥檙e
not鈥攊n a couple of ways. First, the management of our company is
technically oriented. And that has been enormously valuable in getting our
research applied in products. In all too many places, there鈥檚 sort of an ivory
tower that reports up through nontechnical management, and the product
development is completely separate in the corporate hierarchy so there鈥檚 really
no relationship. In addition to running Microsoft Research, I have a role in
deciding the strategy of the company鈥攁nd I report to Bill who runs the
place. He鈥檚 been instrumental in making that happen and in building the right
kind of culture in the research groups.
Some of the places you鈥檝e mentioned, not only are they blue sky, but they
have a perverse 鈥渁nti-applied鈥� notion, where people are proud of not applying
their research. And I think it鈥檚 quite possible to do blue-sky work without
that.
A funny thing happened when we first recruited for Research. The only people
we could really interest in working at Microsoft were people who wanted to see
their stuff affect the lives of millions of people. And if you really couldn鈥檛
care less about that, what proposition would I have for you? You鈥檙e a successful
scientist, you鈥檝e got a career some place, why should you switch?
But if you were someone who said, yeah, I love doing this but it pisses me
off that no one ever picks it up, we had a unique proposition. If you do
something cool, we can put it in the hands of a hundred million people. That鈥檚
appealing to a certain set of people. Hire a few people like that and they hire
a few more people like that and it creates a culture where they work on stuff
that鈥檚 ten or twenty years off from being applied, yet they know that if they
solve it, we鈥檙e going to put it on every desktop or in every laptop in every
school or home or business in the world.
So the culture at our research group is that it鈥檚 very cool to see your stuff
applied. It isn鈥檛 an ivory-towerish place. That鈥檚 even true for our new theory
group.
A theory group?
Over the past few months, we鈥檝e been hiring a bunch of mathematicians, in
fact some of the world鈥檚 best mathematicians. And, again, there are some
mathematicians who think it鈥檚 fascinating that not only can they prove a theorem
but that that theorem could actually result in changing the way world
computes.
What we鈥檙e talking about now is far in the future?
Yes. And as part of our expansion, going from 200 researchers to 600, we have
room to do a few more cool things. We鈥檙e willing to support and create
interdisciplinary groups that couldn鈥檛 exist otherwise. A number of years ago,
we started a group in Bayesian inference.
You have this fundamental problem of how to understand uncertainty in
computing. Certainty is great, you have zeros and ones. But reasoning under
uncertainty is difficult. This Bayesian inference idea says that you should
represent uncertainty with probabilities. You use Bayes鈥� theorem as your
fundamental tool. And we鈥檝e assembled probably the best group of people in the
world to do that.
And the world in general has underinvested in this approach versus other
approaches. It turns out that as one aspect of this we developed the inference
engine that鈥檚 in Office 97 for the help system. So there鈥檚 been a direct
spin-off. But mostly they鈥檙e working on far more advanced, far more esoteric
problems.
Over time there鈥檚 a variety of other areas that we鈥檙e going to be working in
that are much more speculative. I鈥檝e been very interested in artificial life and
evolutionary programming. When we find the right sort of people to support, and
the right sort of synergy then we鈥檒l go ahead.
