Click Multimedia PI.HTM
Click Multimedia PI.HTML
Dar Click para Web Page
Remarks by Bill Gates, Chairman and Chief Software Architect,
Microsoft Corporation Carnegie Mellon University, Pittsburgh,
Pennsylvania February 25, 2004
BILL GATES: Thank you. Well, good morning. It*s good to
know that computer science students are willing to get up this early
in the morning. (Laughter.) That wasn*t my practice when I was an
undergraduate.
Well, I want to talk today about some of the incredible things
that will be happening in the next decade as hardware and software
move forward, we*ll clearly realize some of the dreams that people
have had for many decades, turning the personal computer into far
more than it is today and connecting it up in new ways so that it*s
a natural tool that is really the change agent that*s going to make
business far more effective, and education far more effective, and
just plain old having fun and communicating, a lot better than it is
today.
I*m very pleased to be here at Carnegie Mellon. This institution
has made phenomenal contributions in engineering and computer
science. People like U.S. News and World Report have recognized this
institution as probably the premier institution in those areas. And
certainly Microsoft has benefited phenomenally from the relationship
we*ve had with CMU. The joint research that*s gone on and some of
the great people who have joined Microsoft, including people like
Rick Rashid who runs our research group, Anoop Gupta, who*s running
our communications work, Kai-Fu Lee, who runs our speech effort,
Gordon Bell, one of our top researchers, and many, many other people
who are continuing their relationship with Carnegie Mellon as they
help Microsoft define new software.
As I*m talking about how great it is to be in computer science,
and get not only undergraduate and graduate degrees, I*m a little
bit embarrassed because, of course, I myself am a dropout -- kind of
a paradox, but certainly what Microsoft looks for are people who
have been through the undergraduate and graduate programs, and these
are the people who are going to have an incredible impact on
society.
Computing has gone through a big transition over these last 30
years. When I was young and using that computer at age 13, computers
were big and expensive and only the tools of governments and large
corporations. In fact, people had a very negative reaction to the
fact that the computer would have data about them, and probably
print out bills that didn*t make sense, and some people talked about
putting staples into those punch cards that would come in the mail
so that maybe the big machine would be defeated by that little
staple.
The very first personal computers were quite modest. The Altair
computer, the kit that became available in 1975 that got me to found
Microsoft and leave school, was powered by an 8008 microprocessor,
and that was less than a millionth as powerful as a $300 personal
computer is today. And it really couldn*t do much at all. One of the
big breakthroughs that Paul Allen and I made was coming up with
software that would flash the lights and make it appear that there
was some intelligence inside the machine. (Laughter.)
One of the great discoveries was that the bus was so noisy that
if you put a radio nearby we could actually make sound because of
the electrical leakage, and if you programmed it the right way you
could actually generate a little bit of music with that very, very
limited machine.
That was in an era of eight-bit personal computers: the TRS-80,
Commodore 64, Apple II. You only find those in museums today, but
they did have one thing in common, which was that they all ran
Microsoft BASIC; that is, they came with a piece of software that we
created to let you sit down and program those machines, write games,
write business applications. We started to get disks connected to
these systems, we started to get a decent amount of graphics
capability, but we were really limited, the 64K limitation of those
systems held us back.
And so actually the arrival in 1981 of the IBM PC was a
milestone. This was a machine that Microsoft helped design. That*s
where MS-DOS first showed up. And it ushered in an increase in
volume, and really started a virtuous cycle that had been part of
the vision of the company. And that cycle was that as we got more
software, computers would become relevant to more people, and that
volume would lead to more scale of making the parts, which would
bring the prices down and therefore draw more people in.
And in fact, in 1981, that cycle started so that now, in a
country like the United States, we think of the personal computer as
a tool that a very high percentage of people have access to.
Two-thirds of homes have access to personal computers, and it*s a
tool that in many business settings we simply take for granted.
Now, there was a business model change there to say that all of
the hardware should be compatible. That hadn*t been the case before;
that is, machines from each company -- IBM, Univac, NCR, Digital
Equipment -- all of them ran completely different software. And so
there was really a focus on hardware advances and no ability to
share software across machines or moving forward.
And what Microsoft did was say, okay, we will be neutral to these
hardware companies and make those systems absolutely compatible. And
that was key, because volume is the secret of the software industry.
If you want to be able to sell software for very, very low prices,
you*ve got to sell tens of millions of copies, and that came into
play. It was phenomenal to see that really come together.
The next big transition point was moving away from the character
mode interface to a graphical interface. And I*m sure probably you
appreciate today that that was a very controversial thing. The
complexity of the code, the number of cycles it took just to put
text up on the screen; we bet our company on that and we had the
pioneering work at Xerox, we had the work that Apple was doing in
parallel where we were doing applications for Mac as well as
building Windows. And it took, well, at least six or seven years
before that became broadly accepted. That led to a whole new
generation of applications, a whole new layer of richness, pretty
phenomenal advance.
And then the next big milestone was the arrival of the Internet.
In a sense, it was interesting, people had been talking about e-mail
and connectivity and information sharing, but it was those standards
that came out of the university environment that really defined the
seed that it all got built around.
And, of course, that became by the late *90s a fairly crazy
period, the idea that things were going to change overnight and
startups that had no profit models would run the world. Today we can
look back on it and say that it was fairly extreme in some ways, and
yet the increased levels of investment and attention were
phenomenal. That gold rush atmosphere, although it had its excesses,
really drove things forward.
Now we*re stepping back and saying as an industry we*ve got some
big challenges. The ability to really do e-commerce, the
infrastructure is not there, whether it*s the protocols, the data
schemas, the security elements, the simple development tools, and we
have to get those in place for some of those dreams to actually be
realized.
So as we move forward, it*s really software that*s going to be
the key element. And yet software is boosted by these rapid advances
in the hardware performance.
At the microprocessor level, Moore*s Law that was sort of a
prediction that every 18 to 24 months we*d have double the
transistors on those processors, that has held true for these last
25 years and will hold true for the next 10 to 15 years.
Now, actually translating those transistors into true computer
performance is a very tough problem. You*d think by this time we*d
be able to execute things in parallel and sort of have the sum of
the performance add up on behalf of a wide range of applications,
but that*s really not the case. That*s a very tough problem, and one
that with these new architectures we*ll have to really solve.
Storage systems, whether it*s RAM or a magnetic disk, has
improved faster than the processor. And so although 10 years ago it
was very tough to even store the documents you would create on your
local disk, today you can type your entire lifetime and not come
anywhere near that capacity. In fact, by the end of the decade
you*ll be able to store basically all the movies you*ve ever
watched, the photos you*ve ever taken, the documents you*ve created,
sort of your life*s history will fit on that disk and it*s up to
software to make it accessible, navigable, fun to use, private,
appropriate so that that will be a real benefit.
The screen is changing. Getting bigger screens with high
resolution, portable screens, it really changes how we think about
computing. Today there is a big divide between the world of paper
and the world of the screen. And most people, if you get a very long
document, at some point will go and read off of paper. That really
needs to be changed, because the ability to navigate and annotate
and share and be up to date is far superior in that digital
realm.
You know, there*s no reason why that weekly magazine or daily
newspaper shouldn*t have this superior experience up on the screen.
And yet the usability and the layout, the feel, the vision, these
are all things that we need to make sure we catch up and do as well
as we provide the advantages of that digital environment.
Part of this is having a mobile screen, a screen you can hold in
your hand. When we did our reading research, it turns out that
anything that*s in a fixed position, you get fatigue, you don*t get
the immersion in the content because it*s very tiring to look at
that fixed viewpoint. So when you*re reading off of paper or a book,
you*re shifting that around subconsciously so that you*re not
distracted by that fatigue. And so getting things like the Tablet PC
and these new screen form factors are very important.
The size of the screen area you*ll be dealing with will be much
larger, and there are a lot of user interface techniques to be
invented there.
The graphics capabilities of these systems will be phenomenal.
The kinds of things we see that are only done on a non-real time
basis, we*ll be able to do on a real time basis. And so the PCs and
videogames, the successor to today*s PCs, the successor to Xbox will
provide high-definition, realistic scene generation. And the
question of whether those user interfaces now can move into things
like browsing to find a book or browsing to find a file or being in
a social environment with a set of friends talking about things, do
those 3D interfaces come in, I think that*s very right that it*s
time that we found a way to apply that, because the quality of
graphics is so unbelievable.
In connecting these systems, of course, network speeds go up even
faster than storage capacity, and the wireless element of this has
come in to make a huge difference. Ultra wideband wireless will let
us disaggregate the PC, so that the screen, the storage, the
computation are all thought of as different pieces and you can
dynamically assemble those pieces on behalf of the user, control
multiple screens, use different screens at different times.
And wireless is going to take what is today the most expensive
part of computing -- broadband expense -- that really makes it
prohibitive to get this out pervasively in developing countries,
particularly in rural areas, and solve that problem.
So advances in wi-fi with direction antennas, peer-to-peer
networks that self organize using mesh techniques, so-called Wi-Max,
802.16, that*s going to come in and provide more distance, that
should bring the broadband costs down the same way that the cost of
the hardware and software have advanced.
Now, all these devices are going to have to work together. The
single system centric point of view is very much obsolete. It*s
tough to make systems work together today: moving files between PCs,
upgrading a PC, synching your schedule or your mail onto the
different devices. So this holistic approach of learning what the
user is interested in and applying across those devices, there*s
definitely work to be done in that area of ubiquitous computing.
We see that devices from wall-sized screens to the huge
desk-sized monitor to the tablet to the pocket device, which is the
evolution of the phone, all the way down to the wrist-sized device
will need to work together.
The watch I*m wearing here is actually a so-called SPOT watch
that has a microprocessor in it. This is an ARM microprocessor. It*s
running at 10 times the speed of the original IBM PC. It*s got 10
times the memory of the original IBM PC. In fact, if you take this
watch onto a U.S. nuclear submarine, there*s more processing power
in here than is on the submarine, because they use older systems
that they chose a long time ago.
So the power is quite phenomenal, and we can deliver in a secure
way, code down to this watch using a byte code CLR approach. And so
whether it*s watching baseball updates or stock updates or any
information you care about, you just go to your PC and choose what
you*re interested in -- news, sports, weather, your calendar,
messages -- and those messages are sent to the device so that it*s
working on your behalf, showing information that you care about.
It actually uses a wireless network that piggybacks FM
transmission, the so-called FM sideband, and so we didn*t have to
build out a new network infrastructure; we just inject the data into
the transmission of the FM radio stations and then anybody who*s in
range gets connected up.
So the hardware will not be holding us back. The hardware will
give us the speed, the storage, the connections, the screens, the
graphics to let the software work on behalf of the user.
It*s interesting to think what scenarios can we advance, what
kind of change can we bring forward.
The economy is largely made up of people dealing with
information. If you have somebody who*s designing a new product or
providing customer service or working with a customer on a
purchasing situation, their ability to do their job is based on
navigating to information, knowing what*s going on, communicating
with people at a distance.
And even the modest advance we made in the late *90s at giving
people better tools, electronic mail, browsers, a little bit of
electronic exchange, that caused a big change in the economy.
Advancing productivity is the magic thing; it creates jobs, it makes
the products and goods available far better. And yet the
productivity we gave in that time period is a very small percentage
of what we*ll be able to give in the next decade.
As software tackles these very tough problems, we can make those
information workers far better. The information they have today is
limited, the way they visualize that information is limited, the way
that meetings are organized and followed up on, the way that you
record the notes and the video, the way you draw in somebody who*s
not in that location; the way we think about workflow, what is it
that people should get done and the boundary between software
systems and people. Just take a simple incident where you get an
invoice that you disagree with; today, the inefficiency of pointing
out what*s wrong, trying to get their software system and your
software system to deal with the phone calls and the e-mails that
are an ad hoc way of coming up with a solution to that problem, it*s
10 times more complex than it should be and it doesn*t lead to the
record that would let you monitor those things and understand those
things.
So-called business intelligence where you try and see where are
your quality challenges, where are your sales problems and where are
your sales opportunities is very, very poor today.
Even the basic process of buying and selling, finding a seller
that you don*t know, checking their reputation, checking your
transaction, we don*t have the XML standards and the protocols to
solve that, and yet every one of those things is being put
together.
The idea of visually looking at workflow processes, so that if
you*re customizing software, you don*t go in and write thousands of
lines of code, you simply look at that visual process and make those
changes, today companies that are 99 percent identical end up with
vastly different application software that they have a hard time
upgrading and connecting up to others, because the way we*re
expressing these business process differences are way, way too low
level. We*re writing too much code, the code is not abstract the way
that it should be.
In communications the inefficiencies are something we can all
relate to. We have multiple phone numbers that our ability to
control when somebody gets a hold of us or puts mail in our Inbox is
very limited today, our time is being wasted and even things that we
should be notified of, it*s hard to say, "let me know if this
schedule "changes or "this account is "overdrawn or if this person
is unhappy about something. You should have exactly the information
you care about and not be forced to deal with things that are not
important to you.
An extreme example of this is spam. Spam is a major problem
wasting lots of people*s time. I have to say sometimes I actually
get a kick out of the spam that shows up in my inbox. Here*s one
that I got and it looks like if I get out of debt I*ll get to meet
new interesting people, so that could be a lot of fun. (Laughter.)
Then I got one that offers to solve my problems of not having a
university degree, that I thought that would be pretty good, 24
hours a day I could call those people up. And then finally there was
one that really spoke directly to me. (Laughter.) Being able to get
legal help for just pennies a day would be a significant
productivity advance for me. So it*s interesting what comes in there
and yet clearly spam is something we*ve got to get rid of so that
people aren*t wasting time and so they don*t miss e-mails that are
of importance.
And there is a lot of technology we can bring to this in terms of
what we pass through, authenticating the mail, that spam is actually
even a security problem because the original mail protocol, SMTP,
doesn*t authenticate who it*s from and so you can pretend to be
something that you*re not and cause people to do all sorts of things
they shouldn*t do because of that. And it*s also a spreading
function for malicious code, so making sure that we change that is
important.
We also have to take all these variety of communication
techniques and bring them together. We*ve got blogs, weekies, we*ve
got instant messaging. We*ve got still the PBX systems and the voice
calls, both wireless and wired. And being essentially able to create
a software agent that as people are trying to get a hold of you or
trying to schedule things with you, understanding your sense of
priorities and the context that you*re in, what device you*re with,
what the cost of interruption in terms of your present activity is,
that*s got to come. And there*s a lot of very sophisticated software
that can work on your behalf to do those things.
If we just think about the consumer, there*s clearly a revolution
taking place in how we consume music, how we deal with photos. And
it won*t just be photos; it will be movies and photos and audio
annotation, simple presentations you make around those things.
Dealing with cherished memories and making them accessible, that*s a
scenario that every personal computer basically ought to just come
with that capability and as you take pictures they should
automatically be filed, they should be backed up in the network in a
way that makes sure you can never lose them and yet where you retain
total control over who has access to those things; a lot that needs
to be done to take advantage of this revolution towards digital
media.
Now, coming up with interfaces that make it natural to deal with
those thousands and thousands of things is fairly difficult. One
thing we*re seeing in terms of devices is that people want their
media with them wherever they go. So inside the house you*ll have
high-speed wireless networks and the PC will be able to project onto
any screen or send the audio out to any speaker, and so you don*t
have to try and have it in many different places.
The device I*m holding is called a Portable Media Center and what
this is, it*s a 40-gig disk that you can just connect up either
wirelessly or through USB and download movies, download photos,
songs, and so it*s an evolution beyond just the pure audio player
because there*s a nice color LCD here. So whether it*s kids in the
back of the van or sitting on the airplane, your video is
available.
And this whole idea of digital video recording, that you get to
control what you watch and when you watch, is very addictive. I
mean, there*s no doubt most video will be consumed on some sort of
time delayed basis as things move forward. And as we get devices
like this, this will come out this fall from a variety of
manufacturers but with a very consistent user interface, driven by
our software, these will come out at $300 or $400 and the price will
simply come down from there. And so we really do need to solve the
issues of making it easy to get the information there and navigating
that information.
I just want to quickly show you a few prototypes that have been
done in Microsoft Research that suggest the kind of visualization
that*s possible as we*re dealing with lots and lots of media.
This is a movie application here. We can see there*s a movie at
the center there. So what happens is, it*s taking the movie and
putting it at the center and then finding things that are related
like the director and the various actors. So if we go over here, we
can see all the movies by Ridley Scott, we can just select one of
those and that goes to the center. And then I see it goes into the
database and it calls up all the things that you might want to pivot
on related to that. And then over here it has all the different
movies in the genre.
Of course, you*d like to have this annotated by professional
reviews that you trust, friends who have made comments on it, things
that when you might have seen the movie, what you thought about it.
But this type of visual interface starts to deal with the variety of
information that you*re going to have.
One other prototype I*d like to show you -- look at the photo
realm in particular, or actually, photos and movie clips, and this
is called Media Frame. And this is the idea that when you have all
these different photos it*s a little bit tricky to organize them. In
fact, there*s one person at Microsoft Research who actually carries
around a little camera, sort of a lapel camera, and it automatically
during the course of their day will take photos sort of noticing
there*s a scene change or that people are talking loudly or
somebody*s laughing and so over the course of the day there might be
a hundred photos taken and then those, of course, are automatically
sent to a computer system, but kind of taking those, selecting the
ones that are interesting, we need a lot of help from software to do
that.
So here we can see a lot of photos and I can just go look at any
one of these. Actually, this one I*ll click on is a movie clip
that*s stored there, and so just quite a variety of different
things.
Now, I can put keywords on all of these things, so I can say,
okay, which of these are related to Thanksgiving? When they come in
from the camera, they are time marked because most cameras now will
have time information, and more and more they*ll have GPS
information. So the idea of seeing a map, seeing the timeline
definitely makes sense.
Here you can see all the things tagged about Thanksgiving. I
might want to say, okay, which of these, if I pick one of these, I
might want to say what*s similar to that photo and I can relax the
condition of how similar things are and various things get selected
because of that.
I can say, okay, which of these photos have a face in them and so
this is recognition software that goes in, and to some high degree
of accuracy, is able to see where the faces are. I can say, okay,
which ones are indoors, which ones are outdoors and software is
helping out.
In fact, this whole database as it came in, the software
automatically tried to figure out the orientation and proposed the
ones that should be rotated so that it was in there in a very strong
way.
If we go back and select all, we can say, okay, what would it be
like if we see this in a 3D fashion? So here we*re organized by
date. I can take that X axis and go to finer grain groupings of
these things. I can take a set of photos, select them and say, okay,
those all relate to a certain event, so I can apply the keyword
there, so then I get that very rich navigation.
And obviously this is a database, but it*s a very different
interface than sitting with a typical forms entry, select an item at
a time type interface, and we think this is the kind of thing that
you*ll need to be fully empowered to deal with all your rich
personal media.
Well, let*s talk now about R&D. The software field, the
economics of it are fascinating. All the costs are up front and once
the software is created, the actual marginal costs of making copies
is almost zero. Certainly there are support costs and upgrade costs,
but it*s a very different equation than many parts of the
economy.
Microsoft is showing its optimism about these software advances
by, as your president said, driving our budget upward to about 6.8
billion a year that we spend. And that is substantially the largest
technology R&D budget in the world. So that either means we*re
crazy or there*s going to be a lot of software breakthroughs coming
out of that.
A very important part of that is the Microsoft Research function.
That got started now about 12 years ago when Nathan Myhrvold at
Microsoft recruited Rick Rashid, who was a professor here, to come
and be the founder of that activity. And it started fairly small, it
started on things like linguistics and graphics and now has expanded
to the point where we cover a lot of different areas, and, in fact,
I*d say the vast majority of the areas that we have counterparts
here at Carnegie Mellon who are working on very similar problems.
The idea of mobile and wireless computing, the ubiquitous thing,
they*re doing of lot of interesting things there. The language area,
speech recognition linguists, actually our whole group there, the
majority of them are people who had some of their academic
experience here at CMU. CMU was one of very few places that believed
in that and did before it was fashionable. Then it was fashionable,
now it*s not fashionable again, but it will be fashionable in the
next five years.
Speech recognition is one of those things where as we*ve gotten
into it, our appreciation for how good human recognition is just
goes up and up. If we take a fairly isolated case of it and say
let*s take words where there*s no context, that is randomly selected
words in an environment where there*s no background noise at all,
the perfect audio environment. The difference between computer
recognition and human recognition is not very dramatic. But as you
give the human the advantage of context and as you put noise into
the environment, there*s more and more of a divergence of the human
ability versus the computer ability.
And so although even at the basic acoustic level there are still
advances, it*s really that noise elimination and understanding of
context are the things that are going to close the gap in the most
dramatic way.
We believe that speech over the next several years will be ready
for primetime. In fact, it was kind of a fun thing in China just
last year where XD Huang, the head of this group, was there saying
how great it was and somebody in the audience challenged him to say
that they could type faster than he could enter things completely
accurate with text. And, in fact, they had this runoff and speech
won by a huge amount.
Now, in the case of Chinese, that*s a little bit easier than it
is in English because the keyboard isn*t nearly as good when your
alphabet has literally thousands of characters, but it*s a milestone
and we are starting to see that bootstrap of acceptance in China and
Japan, which you*d expect. But as we talk about ubiquitous
computing, the keyboard will be less and less available and speech
has got to come in.
It turns out that ink recognition is actually an easier problem
than speech recognition. Part of the reason is that you not only
have the processing of the software, in the case of handwriting it*s
done at a very conscious level. So if you enter something in a
Tablet PC and there*s about 400,000 of those out today, and we*re
gathering the database of what people put in and what the mistakes
are and using our learning machine technology to get better and
better at this. In fact, there will be a major release this summer
that will raise the accuracy very, very dramatically.
But in any case, you can see that if you do an E where the loop
is really small, sometimes it will think it*s a C. And so you can
even subconsciously collaborate with us and say when you*re going to
make Es in the future you*ll actually make them sort of unnaturally
large and the system just gets better.
Speech doesn*t work like that. You have no conscious model. And
so it*s very upsetting and it appears to be completely random when
the system gets something wrong. And, of course, the thing it picks,
because it doesn*t have context, is beyond irritating because
there*s no logic that the choice it*s made should show up within
that corpus of material. And so it*s these contextual things that
are one of the important advances.
In fact, by the summer release of the ink recognizer, one thing
we do is we just look through your documents and your e-mail
addresses and your URLs and we bring our dictionary to understand
what you personally are likely to be dealing with, and that alone
makes a huge difference in terms of how well that system works.
This morning I had a chance to meet with the faculty and talk
about some of the big bets you*ve made and I was very excited to
hear that one of the big bets you*re making relates to Trustworthy
Computing. This is the most important issue in software today. In
some ways it*s not that sexy because if you get it right, it*s just
something that you don*t run into. People don*t call you up and say,
"Hey, I didn*t have a virus today, thank you." (Laughter.) But in
order to realize the potential of these systems for e-commerce and
collaboration and communication, all the things that essentially
society implicitly is starting to rely on these networks for very
important information, we have to have a level of reliability,
security, and isolation that we just don*t have today.
And this is leading to so many fascinating problems in computer
science, whether it*s the languages themselves and the kind of
explicit typing systems and constraints, making those very easy for
people to express, finally taking something that was slightly in
vogue in computer science when I left school, which was this idea of
proving programs and actually saying can this program ever have a
memory fault. We*re actually, based on some work that was done here
at CMU, some work at MSR, now taking very large bodies of code like
Windows device drivers, and able to say, hey, this thing has a
problem, because it can actually leak memory. And the way the
proving works is it actually shows you the exact path and the
example of how that*s allowed. And so these very rich tools are
important.
Security is a very deep set of issues that will continue to be a
top issue for some time to come. There will be progress. The ability
to keep software up to date, the ability to firewall up systems,
already we*ve made a lot of progress on those things. So the actual
level of these problems will get reduced but in order to get where
we need to go, there is certainly a good decade*s worth of work
there.
Some other tough problems: The most interesting problem of all in
computer science, the one that*s always been the most appealing and
the toughest, is artificial intelligence. CMU even going back to the
*60s was a pioneer of looking at those problems and thinking about
what progress could be made.
And progress has been very modest. Maybe someone in this audience
will be the one to make the breakthrough that we need here.
I happen to be an AI optimist myself. I think these are very much
solvable problems. I think that some of the modeling techniques
around Bayesian networks and other learning technologies really are
starting to mature.
The only product in the market that uses AI today is this thing
that goes around and vacuums the floor and tries not to get lost, so
that*s the level we*re at today, just down there on the rug trying
to keep it clean. (Laughter.)
A next generation of using these technologies will include things
like computer games. Computer games are a great domain to try things
out in. Often when you play against the computer it becomes
repetitive and boring, and the computer doesn*t know how to get
better or how to be at the right level.
Well, by using what we call Xbox Live! where we get to monitor
gaming experiences and seeing how people are playing and what
strategies they have, we can use machine learning technology to
create really wonderful opponents that try out these different
strategies and have those available. And so if you want to train and
get really good or if you just want it to work at a basic level, we
have a reasonably rich AI engine in there that*s learning on an
ongoing basis.
And, in fact, these AI engines can be applied to so many
different problems, low level problems like being smart about memory
management, high level problems about looking at systems and seeing
are they behaving in a natural way. Machine translation, this is
another thing that we*re extremely optimistic about that there*s
been enough progress that it will be practical to have extremely
high quality translation within a domain commercially available
literally in the next several years. So the breakthroughs, the
opportunities for the breakthroughs, are pretty phenomenal.
One thing that has always been a theme here at CMU is getting the
various disciplines to work together. And that*s particularly
appropriate now because many of the hard sciences are relying more
and more on data. Just take astronomy. Jim Gray, who*s one of our
researchers; he*s a great computer scientist and he*s a hobbyist in
astronomy. Well, astronomy used to be that you*d be there at that
telescope at three in the morning and some supernova would blow up
and you*d write a paper about it and that was a big contribution.
But today, the amount of scanning information at various wavelengths
and various time periods is a monster database, and yet it*s not
really a database because it*s stored in different locations and no
one has characterized the nature of the data in each of those
things.
And so the problem that Jim took on with some collaborators in
that field was saying this science is going to be data driven, so
let*s create a national virtual observatory where all of these
observation things are available and so you can have a query like
have any quasars been seen in a galaxy like this in this time period
and see what the correlations are and test out theories and have
deep understanding of the information that*s been observed.
Now, astronomy was a good place to start with this, but we see
this in even tough areas like biology. The amount of data, genomic
data, proteomic data, trial data is phenomenal, and yet that isn*t
gathered together, and so it*s really a computer science problem to
take that data, understand that data. The characteristics of
biological systems makes people from computer science are very
important there. And so I*d just pick those as two examples of areas
where it*s people with a computer science background who will make
the breakthroughs. It*s the rich modeling techniques that come out
of our discipline that will be, I believe, a primary
contribution.
Now, computer science has a lot of challenges. One of them is
that the Internet and the PC have become such basic important tools
that we want everyone to have access. And this is often talked about
as the digital divide. Now, of course, the most direct way to help
with this problem is to make sure that the price of software and PCs
and broadband networking continues to come down. That drives up
availability. But there*s even special outreach efforts that all of
us should consider and hopefully get personally involved in.
One of the most exciting ones that Microsoft and my foundation
did together was going around to the librarians in the United States
and saying, "Would you like to have a PC in the library?" And we
were worried about it, because would they find it a negative thing
and if the kids came in to use it, would they use it for appropriate
things, could we keep those systems up and running in a very
effective way, even if people were kind of hacking around. And, in
fact, that project, which took place over a five-year period, is now
complete and that is that 18,000 libraries that did not have
computers in there now have a total of 50,000 computers. And these
things are monitored for reliability. There are training programs.
And the phenomenal thing is the librarians love it. More people
come in, the rate of using books has actually gone up as people are
coming in to use this tool, so it hasn*t been a substitution effect,
and it*s really reinforced this idea that the library plays a role
in equity, that any kid who doesn*t have a computer at home, if they
can reach a library can get in and use these state-of-the-art
things.
Now, if we look on a global basis and look at education, it*s
just scratching the surface, so there is a lot to be done there.
As we think globally, there is a new dimension coming to global
competition. People are just starting to get used to the fact that
the advances in the Internet and software are making it so that when
we think about jobs being done anywhere in the world, that it*s not
just manufacturing. Global shipping and logistics systems meant that
manufacturing jobs were moving around the globe even 20 years ago.
Now any job, including doing architecture work, answering phone
calls, not just software development, through the use of technology
can be done wherever you find the most qualified people.
And this is creating opportunity. This is going to create better
products and services, it*s going to tap into the incredible human
resources and energy in countries around the world, but it will be
an interesting challenge in terms of the United States renewing its
edge, saying, okay, research and a commitment to long-term tough
problems has gotten us and kept us in the lead and are we going to
maintain that even as countries like China and India are coming and
becoming major players, major contributors. I think it*s very much a
win-win situation and some of the concern and humility will get us
to folks who can do the right thing.
After all, it was during the 1980s when U.S. computer science was
doing very good work that led to what happened in the *90s, where
there was a humility, and in that case it was about Japan, that was
Japan a superior industrial model, did their sort of centralized
research where they had this fourth generation project. There was a
question of whether that going to make some big breakthrough versus
the sort of diverse approach, using U.S. universities, and I think
we very wisely didn*t follow exactly what they did, we strengthened
the things that we did well and what came out of that was a
phenomenal contribution, both at a national level and at a global
level.
And so there*s going to be a lot of good debate here about what
this means to the U.S. and a need both at the university level and
commercial level to up the ante in terms of the research that gets
done.
Well, one limiting factor in all these breakthroughs is having
great people to do the work. And when I talk about the work, it*s
important to know I*m not just talking about writing code; I*m
talking about people who understand the customers, the business.
There is such a range of activities and yet to do those jobs well,
really a background in computer science and understanding in
computer science is pretty important.
It is interesting the challenge that we face in terms of getting
a diverse set of people to go into these fields, and again I would
congratulate Carnegie Mellon for very early on recognizing this
problem and trying different things, making people aware of the
problem and making some progress, not solving the problem but being
a model for some of the things that can be done well.
One of the things I*ve done personally through my foundation is
create what*s called the Millennium Scholarship Program, and that*s
for minorities to be able to go in both at the undergraduate level
and graduate level and be in the sciences and create a role model
who will hopefully bootstrap and get these numbers up in a very big
way.
If you think of what are the exciting places to work, I*ll admit
I*m very biased, but if you go, say, into law -- and my dad*s a
lawyer -- things aren*t going to change that dramatically. If they
change, it will be because of technology and the issues around
technology. If you go into investment banking, it*s not going to
change that much.
There are only two fields that you can make argument I*d say that
they are going to improve the world in a very dramatic way, and
those are the things around engineering and computer science, both
hardware and software, and the things around biology and medicine,
and with a lot of very interesting things at the intersection. The
biology and medicine guys will move forward in large part because
they can take the data, the visualization, the tools that we*ll
create out of these things.
And so I think this should be the most attractive field, and I*m
very excited to be part of it. Even though I*ve gotten to
participate in a lot of milestones, I think the milestones of most
interest are the ones to come in the next 10 years. And I know a lot
of you here are going to make huge contributions to that, and I look
forward to seeing what you*ll do.
Thank you. (Applause.) <!--START RIGHT NAVBAR--></TD>