Click Multimedia PI.HTM
Click Multimedia PI.HTML
Dar Click para Web Page
Remarks by Bill Gates, Chairman and Chief Software Architect,
Microsoft Corporation Massachusetts Institute of Technology,
Cambridge, Massachusetts February 26, 2004
BILL GATES: Thank you. Well, it*s great to be here at MIT.
MIT has invented so many things and played such a key role in the
advance in computer science I won*t have time to name them all. But
Microsoft has been very privileged to be associated with MIT and
helping collaborate on a number of projects here.
Microsoft also has some great employees who came from MIT, most
notably people like Gordon Bell, who*s some still very creative
after making so many contributions, and also Butler Lampson, who*s
actually back here teaching courses, as well as continuing to help
Microsoft.
The main thing I want to present today is how I see the next 10
years of computer science as doing some really amazing things,
solving some problems that have been out there for many decades and
transforming most things of interest, transforming the way people
work in businesses, the way they deal with information and meet and
communicate, transforming the way that people at home find
information, the way that they create things and connect up with
other people, and transform even the sciences, bringing the
methodologies of machine learning and modeling and rich data mining
into all of the hard sciences and allowing those sciences to move to
the next level.
It*s a paradox to me that computer science today is poised to do
all these amazing things, and yet in some ways people*s expectation
and even the excitement level about computer science is not as high
today as it was, say, five years ago when we were in the midst of
what we can now look back and say was the Internet bubble.
Many of the challenges, the problems, the things that were hard
to deliver on during that period are exactly the kinds of things
that a combination of great academic work and great commercial
working coming together has solved and will make sure that the
productivity advances that we see in the economy will be
dramatically higher, more than double what we received in the 1990s.
And that*s very profound for innovation, employment, many things on
a global basis.
The computer industry started by making very big, expensive
machines and, in fact, when I was young people thought of computers
as kind of these daunting things in the back room that would often
print out bills that would be wrong and you could never get
corrected. People talked about taking the punch card in your billing
envelope and putting staples in it or mutilating it to somehow
defeat the big machines.
And I was lucky enough to be young at a time when that changed.
My friend Paul Allen saw the very first microprocessor, the 8008,
not very capable but better than a PDP8 was, and he challenged me,
he said, "Hey, Bill, could you write software for this, could you do
a "BASIC, and that really got us going, saying that we wanted to be
in on the ground floor of building the kind of software that
computing would need as it moved from that back room onto the
desktop, as it became a tool of empowerment, a tool for creativity
and communications.
Now, it started in a very humble way. The kit computer that Paul
saw on the cover of Popular Electronics magazine, a freezing cold
day in Harvard Square, and brought back to me and said, "You*ve got
to drop out; it*s happening without "us, -- (laughter) -- that
machine was laughable. It at most had 8k of memory, had no
peripherals, so the software we wrote would flash lights and do
funny things. It was a major discovery that because of the noisiness
of the electronics in it that we could actually put a radio nearby
and if we did certain instruction patterns cause predictable music
to come out of that radio. And it was kind of limited what we could
do, but that was the beginning and the kind of excitement around
that, thinking where could this go, that got our industry off to a
great start.
There was a generation of machines that came after that,
so-called 9-bit machines, TRS-80, Commodore 64, Apple II, all of
which included Microsoft BASIC in as their fundamental software. It
was the equivalent of not only the language but the operating
system, everything, and you could type in BASIC statements to do
graphics and games and business software. And we even got disks
connected up to these things, we moved away from cassette tapes and
paper tapes to these ridiculous eight-inch disks that hardly held
any information at all, but constant improvements.
A big milestone in 1981 was the entry of IBM with the PC into
this business, and that had been a joint project with Microsoft
where we*d convinced them to use the 8086, created the DOS operating
system, very limited system but very appropriate for that machine,
and we kicked off an era that was fairly different, because starting
with that machine, we had a vision that we wanted all the machines
to be compatible, that is to use a software layer to make it so that
whenever you wrote an application it would run on machines from IBM
or HP or Digital Equipment or any of the other computers at the
time. And that hadn*t been the case and, in fact, that prevented the
virtuous cycle that we wanted to have happened, it prevented it from
getting going. And that cycle was that as more people saw
applications that were meaningful to them, they would buy machines,
the more machines that were bought, the more volume of components
would happen, the lower the cost those components would be and the
more people would invest in buying applications so it would make
sense to really build the software industry.
There essentially was no software industry before the PC. There
were about 20 different companies and the highest award in the
industry was one that you got for selling a thousand copies of a
piece of software; it was called the ITP award. And I filled out the
application after we*d sold 2 million copies of BASIC, I sent it in
and they said, "Well, we*d love to give you this award but there
must be some scientific notation error here, you*ve got too many
"zeroes. And I said, "No, I*m not kidding, we sold 2 million
"copies. And they said, "Well, geez, how come we haven*t heard of
you?" And we said, "Well, I don*t know where you*ve been, but we do
have this very high-volume, low-price model, so we*re not like on
the New York stock exchange or anything but we have sold two million
"copies."
And so personal computing gained in momentum, it gained in
excitement. One of the great issues was to move from character-mode
interface to graphics interface. It may seem strange today, everyone
takes that for granted, but the cycles required, the difficulty of
writing these programs made a lot of people say, "Hey, we don*t need
this, that*s just really too many icons, too many fonts, let*s stick
to the serious stuff, we love these mono space characters up there
on the "screen. And there were about six or seven years where taking
some of the pioneering work from Xerox, Apple, and Microsoft really
pushed forward with this idea of graphics interface and that became
Microsoft Windows, it was integrated in the operating system.
And then we got to the next frontier. That frontier was
connecting all the machines together. And year after year we*d
always say e-mail is coming, connectivity is coming, online services
are coming, and, in fact, it didn*t happen and didn*t happen, and
then all of a sudden, coming out of the university environment, the
standards of the Internet exploded, along with the decrease in cost
of optic fiber and the increase of the speed of the network and a
critical mass was achieved. And that was a period where things
definitely went a little crazy but the gold rush atmosphere actually
accelerated the investments people were making and raised the
awareness that people had of what a revolution this was.
Now, some of the things that are hard about e-commerce or
workflow or modeling or making sure these systems are
ultra-reliable, ultra-secure, some of those were revealed as
shortcomings that needed software breakthroughs, needed software
advances. And so as we look forward, it*s kind of a bias I have, but
the thing that*s really going to make a difference is software. It
is the new generations of software that let us interact in natural
ways, that connect these devices up in new ways and problems that I
see being solved in the near future.
Now, the hardware guys, I have to give them credit. They*ve
always provided a more powerful platform for us to exercise our
software creativity against. And that*s why Paul Allen and I could
say back in 1975, personal computers will be mainstream. The slogan
we had was, "A computer on every desk and in every home," and in
some small part in some countries we*ve come a long ways towards
that. It*s not yet the machine that we envisioned in terms of the
ease of use or breadth of things that can be done, but it*s
certainly a good rough draft that*s on its way.
The hardware people have given us Moore*s Law that predicts a
doubling in chip performance every 18 to 24 months. That*s held true
these last 25 years. And something like the next 10 to 15 years it*s
very likely to continue to hold true.
Now, that increase in transistors, there are some very
interesting software techniques related to parallelization that are
needed to take transistor count and map it into performance. It*s
not automatic that just because you have twice as many transistors
that you get that performance. And so finally some of these issues
of automatic parallelization and understanding the algorithms that
let you do that, we*re making progress on those.
The storage people do an even better job than the chip people.
Their doubling rate is something like 12 to 18 months. And this is
very important, because when storage was expensive, the idea that
you could deal with photos and videos and audio annotation and
replicate information around so it would be immediately available,
even if the system is not connected up to the network, that just
wasn*t possible. People didn*t think in those terms.
And, in fact, storage is so available now we have to be creative
in thinking about what we*re going to do with it. We*re getting lots
and lots of that. In fact, a good example of how cheap storage is,
is that we have this device that comes out this fall called the
Portable Media Center. It*s a 40-gig disk, beautiful LCD display,
and you can just connect it to a PC that*s recording TV shows or has
your movies or whatever is there and it automatically downloads the
movies, videos, photos onto this device that you can, of course,
carry around and use anywhere you want.
And these devices will come out fairly inexpensive in the $400
range, and the price will just come down and down and down because
this is the magic of that hardware innovation. Eventually we*ll just
take it for granted that kids who want to watch movies or people who
want to watch shows have this available, and so it won*t just be
portable music players but devices that deal with the video as
well.
The screen is another place where innovation is critical for us.
If we think about how we can move reading from paper to the screen
so that we get the rich searching, updating, annotating, sharing
that the digital world allows, that requires screens with very high
resolution. It requires screens that we*re comfortable holding in
our lap and just sitting there paging through the information. It
requires a thin, light device, long battery life; big challenges but
certainly what*s gone on with LCDs and other screen technologies
says that in the future we can assume a 30-inch LCD on a knowledge
worker*s desktop or three 22-inch displays, which is the
configuration I*m using right now and lets me work with information
in a much better way. And the same way that a newspaper gives you a
big, wide field of vision, this does that and there are certainly
some advances in window management use of the screen area that have
to take place as we get to extremely high DPI on big screen areas
that come with it.
We believe that reading will move onto the digital platform, that
the superiority of the cost structure, all these things argue for
that as we get devices that are based on this new screen
technology.
A nice milestone in that is the arrival of the Tablet PC. That
got kicked off about a year ago. It*s based on the miniaturization
of hardware, the ink software, ink-recognition technology that we*ve
been working on for over 10 years is now bootstrapping in terms of
the quality of the hardware, learning from that software, doing that
better and better and making that mainstream. And so all the
portable devices will become Tablet devices, and they really will be
like Tablets, which they*re a little bit heavier than a Tablet
today.
The graphics processors, there the improvements are you can get
higher transistor counts there because of the number of duplicated
components and actually if we look to the future of CPU
architecture, we can see that more predicted by what*s happened in
the JPU level, because they*ve thought about parallelization. Now,
they do it in a domain specific way that we need to open up, but
it*s really the blending of those that is the next stage there.
A big element, of course, is wireless. As we get things like
ultra-wideband wireless, which is hundreds of megabits, the idea
that you connect a computer to the screen will become obsolete. The
computer will find the screen that*s nearby and take advantage of
it. The idea that the computer and the storage have to be associated
with each other, there*s no reason for that. You can carry your
storage with you and whatever PC is around, in a very secure way
your storage can be made available to it and you can be given
guarantees that that information isn*t left on that machine after
you log out from that machine. So we*ll see the dis-aggregation of
the PC that way.
We*ll see the arrival of rich new peripherals. Digital cameras
are now the most popular way to take photos and that*s happening in
the motion video space. Well, those devices will have ultra wideband
and so they*ll mark not only the time but the location of that
information and deliver that to your storage system.
Your storage system will be a combination of data stored in the
cloud and sort of a far-sight ocean store type way where you don*t
have to worry about whether you*ve backed it up, because there are
many copies that are stored in encrypted ways that mean that only
you have control of that information. Or you*ll have the storage
that you carry with you physically that will give you total control
over it, and making those two things work well together is very
important.
Now, in the area of wireless, one of the tough challenges has
been the cost of broadband. When you would think about what*s
expensive, getting a PC, say, into rural India, the hardware is $300
or $400, the software is less than $50; it*s that broadband cost, a
monthly cost of paying again and again and getting that
infrastructure out there that*s really the prohibitive factor.
And we believe through some software techniques around mesh
networks and some advances in the wireless hardware, particularly
the Wi-Max-type approaches, bringing in not just omni-directional
approaches but directional antennas as well, that we will get the
kind of connectivity that can make sure that connecting everyone on
the planet becomes very feasible.
So you*ll have a range of devices, wall-sized screen devices,
desktop Tablets, pocket-sized. We even believe in a wrist-sized
device. In fact, I*m wearing my SPOT watch, which just came out in
the last month, and this -- I don*t know if you*ve seen it -- it
lets you see sports activities, stock prices, your schedule, you get
messages on it and a lot of different things that are being
transmitted to this watch.
It was actually when I was at MIT over 10 years ago that I first
saw a demonstration of FM sideband data networking and this watch is
based on that approach. Of course, the modulation techniques are
several generations later.
There*s a microprocessor in here that we paid National
Semiconductor to create that*s based on the ARM architecture. This
microprocessor just on my wrist has 10 times the power of the
original IBM PC. It*s got 10 times the memory of the original IBM
PC. This thing is powerful. And, of course, the battery life is on
the order of many, many days because these things are low power.
We can download arbitrary programs to this device, so as we get
new ideas about sports presentation, information presentation, as
people have neat things they want to do, it can come down in an
automatic way.
Today, the watch is in receive-only mode, but we actually have
the capability to send data as well in a local area, and so you can
find people with common interests, a lot of applications that the
glanceable information platform will be particularly appropriate for
working with the other devices.
We have to think in terms of the scenarios. The photo scenario
that you use all the different devices, the scheduling scenario
using all those devices, and it*s way too hard today to get those
things to work together.
One of the big places that software advances will change things
is change things in the way business is done. The information
visibility that a typical information worker has is extremely low.
They*re used to it in a way so they don*t know to complain, but
their ability, say somebody gives them a sales printout. They look
at these numbers and they must think, "Wow, that one*s really big.
Wow, what did we do? That one is kind of small. Geez, are we in
"trouble. Well, their ability to just dive into that data and see it
by time period, product, cost structure, they don*t have it. It*s
not there. The schematization and model approach to bring that down
to every employee to just naturally expect that they can see those
things and understand those things, that*s not there. The world of
business intelligence hasn*t delivered on that.
The XML foundation that*s advanced so fantastically over these
last six years is the foundation to make that happen, to build XML
into the spreadsheet, to build a knowledge of business processes so
you can visually see what*s the state of this activity.
Businesses today do all these custom modifications to the
application programs they run, the enterprise applications. And
that*s very strange because the differences between those
businesses, you ought to be able to express it in some other way
than code. Code is complex. When people update the applications, you
don*t know how to combine that new code with the other code because
it*s not orthogonal. There shouldn*t be code in that process; there
should just be visual business processes that you*re connecting up
to and explaining how this business is different than this business.
How does the order process work, how does the collection process
work, how does the analysis process work and that*s the kind of
thing we*re really on the verge of, because XML gets our semantic
level up and lets us finally address making this information really
available.
If we look at meetings, meetings are a source of a lot of
inefficiency, as any information worker will tell you. Things that
they didn*t need to be there for, meetings they had to fly in to
that they would have preferred to be able to do at a distance,
things that didn*t get followed up on, things that somebody who
wasn*t there you wanted to explain to them that you couldn*t just
link in and see the transcript or see the video of what went on
there.
Well, storage is almost free and cameras and software to scan and
understand this stuff will be almost free, and so we can take the
meeting and have that be something that we bring a lot of efficiency
to. If you make meetings in general 10 percent more efficient,
that*s tens of billions of dollars of extra productivity every year,
and that can be used just as cost savings, it can be used to make
better decisions, to drive quality into processes and it will do
every one of those things.
Even the basic process of buying and selling hasn*t been made as
efficient as it should be. Can you find all the sellers of a
particular type of product? Can you check their reputation? Can you,
if you engage in a transaction, see the state of that transaction in
a very rich way? If your computer is talking to their computer and
their computer is somehow malicious, are you protected from that
kind of behavior?
If the software is talking to the other software, what about the
workers? Say that there*s a delivery that*s defective, how do you
coordinate the negotiation on e-mail in an ad hoc way with these
back-end systems so they can understand things and check the state
of things? These things are incredibly inefficient today, so basic
workflow is not built-in.
E-commerce has not happened. E-commerce only really happens where
every seller can find every buyer, every buyer can find every
seller, independent of location or previous knowledge of each other,
and that rich transaction is done in a pure digital way.
In communications, what we*ve got today is kind of a hodgepodge
of different things. The latest thing is blogging. That comes after
instant messaging, which comes after e-mail. You*ve got your
wireless phone and your wired phone. Lots of times you*re
interrupted, the phone rings when you don*t want it, things come
into your Inbox that you don*t want and your time is a scarce
resource. And so these activities are wasting your time, causing a
lack of productivity, even in some cases you have enough spam that
you filter out or don*t have time to read e-mail that would have
been of value.
Now, for me spam is this awful thing, but sometimes when I look
at the spam I get I have to just step back and laugh about them.
I*ve got a few examples here. This is one of my first ones.
(Laughter, applause.) And it*s clear once I get out of debt I*m
going to be meeting a lot of nice people who are going to be
friendly to me. (Laughter.)
The next one looks like it might be more targeted. (Laughter.)
And this is not one that any of you need worry about, since I hope
you won*t drop out.
And finally there was one that really related to a serious cost
problem I*ve got. (Laughter.) The shareholders really want me to dig
into this one and understand what*s going on there.
So it*s a serious problem but it*s amazing the things that are
out there.
Letting people send billions of pieces of mail very, very cheaply
devalues the time of the person on the other end. And this is a very
solvable problem. We need mail that comes from people we communicate
with regularly to be authenticatable, and we announced on Tuesday a
way of doing that, leveraging off the DNS that we think can be
applied and uses a standard literally within months. Mail that comes
in from a stranger, some type of proof is necessary. If the filter
thinks that looks like spam, then you need some type of proof to
distinguish it from the other e-mail. And there are several forms of
proof that will be used and they all work in parallel; any one of
them is kind of an "or" condition.
Proof that*s computational, where you solve one of these problems
that asymmetric in the opposite way that cryptographic functions
are, that is it*s asymmetric in the sense that checking the answer
is easy but actually doing the computation is hard, and so for
somebody sending a modest amount of mail it will just happen in
background, they won*t notice it, but if you were sending millions
of e-mails it would be a significant computation cost to do that, so
you screen out.
Human interactive proof where you bounce back and make somebody
solve something that software alone can*t do is another approach, or
if there*s connection into a payment system, making somebody put a
little bit of money at risk, not that gets charged to them and this
is only e-mail to strangers, but that it*s at risk so that if it
really is junk, the person who receives it and spends time reading
it, at least they get the benefit that whatever the threshold they
set for their time, their Inbox rate, that gets credited to them,
but if it*s their long lost brother or somebody saying their house
is on fire, hopefully they won*t debit that, they*ll say, okay, that
was just at risk and I*m glad that person connected up to me.
In the home environment, when we think about media and memory,
there is so much that can be done and yet we*re going to have to
deal with a lot of volume. All the music you like, all the movies
that you*re interested in, all those photos that you take, it*s kind
of amazing.
We have a researcher at Microsoft Research who wears as she goes
around in the day something that*s just a camera, and it notices
when there*s a big scene change or when there are people laughing or
anything loud or something, and it takes photos. And so at the end
of the day that researcher has over a hundred photos that might be
interesting to put in her journal and save and even annotate with
some voice or something. But software has got to help select which
one of those things are interesting and to navigate amongst those
things. And so there is a lot that has to be done on that.
We want to put users in control in the home. This idea of
watching TV shows only on a schedule, slowly but surely that*s going
away. People who use that, whether it*s built into what we call the
Media Center PC or a TiVo or the satellite receiver, get very
addicted to it. It*s kind of like e-mail where you know it*s not
perfect, but you don*t want to give it up; you just want it to get
better.
And so as we think about that, we think about what kind of
interfaces would deliver on that. In fact, I*ve got just real
quickly a couple prototypes from Microsoft Research I wanted to give
you just a sense of this. In fact, these were both done by an MIT
graduate.
The first one here is pretty straightforward. Let*s say you*re
looking at movies, you*re looking at "Blade Runner" here and what it
shows is, okay, the director is Ridley Scott. Well, then I can go
over here and see other movies directed by Ridley Scott and I can
just select one of those, "Alien," that*s brought to the center and,
of course, then all the things related to that come out and I can
see, okay, these actors and see the different things they were in
and see if one of those might be interesting and just pivot through
these sets of movies in a simple digital way.
Of course, this will be annotated with the reviews that you
trust, comments from friends, if you*ve seen the movie what you
thought about it, and so very navigable to get around the movies of
interest.
The other one I wanted to show has to do with photos. In photos
we*re dealing with lots and lots of photos, literally, if you take
your lifetime, tens of thousands of photos that you and your friends
are sharing and you*d like to be able to get back to in a rich
way.
And so here we see them as miniatures. I can just hover over
these things, lots of photos. We*ve even mixed in video clips as
well. Here is Gordon Bell at the computer museum. Because we don*t
think the boundary between stills and motion will hold up. In fact,
these audio comments that we call Photo Stories bring a lot more
emotional connection to that experience.
And so when it*s just like this it*s hard to find exactly what
you want. And so people will tag these things with keywords. Here
are things that relate to Thanksgiving. We can do software analysis
and so if we want the photos with faces, we just select those. If we
want the photos that are indoors, the software can select those. If
we want the outdoor photos, we can select those. If we want to see
photos that are similar, let*s select this bridge photo and say, OK,
I can relax the constraint and say what*s similar to that. OK,
that*s a lot like it, that*s a little bit more like it and I can
select groups of things to be used.
This software automatically when it brought the photos in helped
orient the photo by being able to recognize the cases where things
were kind of, at least to the software looked like they might be
misoriented.
We can also start to use a 3-D way of looking at these things, to
group these things, and what that means is now this by timeline so I
can select this set and tag these, I can change the timeline and get
to finer groups in terms of when they were taken or where they were
taken, and this makes it very easy to just step through these but
also deal with groups that I want to organize and tag in different
ways.
So in a general sense we can say, well, that*s just a database
but we need much better ways of interacting with the database than
just the common query processor. People won*t be writing SQL
statements to navigate through their photos.
Now, this optimism I have about computer science and its impact,
a little bit the proof of how serious we are about that is the
R&D spending that has been increasing at Microsoft. Today, it*s
$6.8 billion a year, kind of an intimidating number, at least to me
since 10 years from now people will say to me whether that was wise
or not, but I*m quite confident that it is. That*s the largest
technology R&D budget that there is. IBM is about 20-percent
less than that, but, of course, it*s not all focused on software,
and then other commercial entities you*d have a big drop down,
particularly if you take the long-term components, the equivalent of
Microsoft Research.
We actually do our research work in three different locations, in
Cambridge and in our headquarters, and in Beijing. We have smaller
groups in some other areas but those are the primary areas.
We*ve had, as I mentioned earlier, a strong collaboration on a
number of things with MIT. For example, the iCampus project, we*re a
key partner in that. I*m thrilled at the things that are coming out
of that. Some involve learning at a distance, some involve the
Tablet PC. The idea that we can make learning better there*s no
doubt and I think that*s a great pioneering project.
Natural interfaces for learning, that*s this magic paper idea.
That*s a fantastic thing.
We*ve gotten involved in things that think. We*ve got a lot of
people on the faculty here that are helping drive our agenda. For
example, Victor Hsu is on the technical advisory board for our group
over in Beijing.
So it*s been a good, strong relationship and the progress being
made in the combination of academia and commercial research labs is
really fascinating. It*s phenomenal and it*s not getting that much
visibility, and yet these advances are extremely relevant to
problems that we have, the problems that are of critical
importance.
Take, for example, security. Of course, MIT has a strong program
on that. Of course, you*ve got Professor Rivest, who just got the
Turing Award, which is a fantastic thing. Security is something that
if we*re going to achieve the potential of these systems has to get
a lot better. And that*s a tough thing, because code reliability
gets into that, how you configure up these systems, how you watch
behavior as part of that, so it*s going to take some
breakthroughs.
Over 25 years ago, when I was leaving Harvard, this idea of
proving program correctness was sort of in vogue. And unfortunately
for many years, although there was some progress, that scale of the
program it could be applied against didn*t get very large, hundreds
of lines of code.
Now, working with universities, people in Microsoft Research are
taking literally things that are a million lines of code and being
able to go in and prove very important things about those programs.
Or if they can*t prove them, they*re able to show the counter
example that says, yes, this can touch memory that it shouldn*t
touch or it can acquire the lock and never release a lock. And so
you see exactly what the pattern is and how to fix that.
Now, all this proving technology is having a wonderful effect on
innovation in programming languages, because we want to take
everything that the programmer knows about the data types and the
constraints and express those in as high a level, as strong a
fashion as we can through contracts. And this idea of contracts very
easily, having languages that are very explicit about those things,
that takes all the theory of language innovation and brings it into
the mainstream and says we really need those capabilities.
Keeping systems up to date, being able to look at a system and
say is this behavior normal, we need this both at the single
computer level and looking at the network, is there a type of
traffic that*s exploded in terms of usage at a time where the
overall traffic is starting to be too heavy for the network to deal
with. There should be automatic tools out there that are doing that.
And, in fact, machine learning techniques that build the model of
typical behavior and then see these things that are unusual will be
used at every level of the system, at the memory management level,
at the API level, at the network level and at the wide networking
monitoring level. We call that behavior blocking and that will be a
critical component in solving those security issues.
Another set of areas that I think are making wonderful progress
are getting a more natural interface between users and the computer.
Victor Hsu here has been a big advocate of that, building some
wonderful systems that take speech all the way up into particular
domains and lets people interact with those. And making it so that
it*s not a huge technical exercise to build one of those systems,
that you just have a general runtime for that, I think is something
that will be solved in the years ahead.
The progress in speech recognition is very good. If you take an
isolated, simplistic case where there*s no context and no noise and
a perfect microphone, three great simplifications, the difference
between a human and a computer is not very drastic. It*s as we relax
those constraints and bring in context, crummy microphones and noise
that then the divergence between the computer and the human is quite
substantial.
And, of course, human users of these things are very demanding.
Because speech doesn*t operate at a conscious level, as it makes
mistakes you just get irritated and talk louder. And, of course,
it*s been trained, it*s learned you*re speaking in normal tones, so
it just gets worse and worse and so it just degenerates while you*re
yelling at the system.
Now, ink is not quite the same. It*s a little easier because you
process ink at a conscious level. And so although it*s irritating
when it doesn*t work, you can look and say, well, hmm, could I have
recognized that, is that E looped so closely that it looks like a C,
and, in fact, as we monitor people using our handwriting system, the
plasticity is partly in our subsystem but a lot of it is in the
user, that the user consciously or subconsciously is actually
writing more explicitly the features that have caused the problem in
the past. So that*s partly why ink is coming into the mainstream,
say, a few years in terms of general input, the equivalent of
dictation a little bit faster.
The place where we*re seeing our speech work really catch on is a
combination of people where the keyboard is unattractive for them
for any reason, including repetitive stress injury, or people in
China, Japan or Korea where the keyboard is relatively less
effective as an input technique. We can already beat the fastest
typist of Chinese with a Chinese speech recognition system. And so
that*s a milestone along the way that is pretty exciting.
General artificial intelligence, this is the Holy Grail, and when
I was talking to the faculty today I was impressed that MIT has kept
its commitment to this area throughout all the years. It is one
where some very interesting approaches, statistical approaches,
Bayesian approaches all are now starting to be used in different
fashions.
The actual product on the market that apparently is a spin-off
related to a professor here is the thing that goes around and
vacuums the rug, and that*s pretty low level, better than nothing
but we want to move up in terms of the things that go on.
For gaming, one place we*re using our machine-learning technology
a lot is we on Xbox Live can watch player behavior in different
strategies and the machine can learn. And so if you want to pick an
opponent, typically if you play the computer historically it*s an
algorithm that isn*t that much fun to play because eventually you
see that as being very predictable. We can take all the play styles
that we*re seeing across this network and create any sort of level
of difficulty or different fashion of play and make it as
interesting and as varied as playing with human opponents, including
letting you win every once in a while, which on videogames for me
that is pretty tough. You pick these things up and they are geared
to, well, to you and not to people who haven*t used them nearly as
much. And so we*ll make these things appeal and even if you start to
beat the system, boy, we*ll crank that up to a level that will keep
it challenging.
So AI is going to be applied in a lot of different ways, modeling
things in the other sciences, helping with dynamic behavior in
systems, very, very important.
All the natural interface techniques -- vision, speech -- we*ll
come to take those for granted in a very strong way.
Now, the boundary between computer science and the other sciences
historically was a fairly hard boundary, and that is breaking down.
One great example of that is the research we have of Jim Gray, who
looked at astronomy and said, boy, there*s a lot of data there and a
lot of the advances come in proposing something that you can either
validate by looking at that data or invalidate. And so we really
need to get all these databases connected together.
And the semantics are very high level. It*s again not just a
relational problem. But collaborating with a lot of astronomers who
know the domain, he*s been hooking up those databases and now
navigating through this logically connected database is a very
important tool in astronomy. And many of the sciences are going to
where those rich data collections are necessary for everyone to have
access to in a high level way.
Biology, of course, is perhaps one of the most challenging
because of the breadth and the differences in the data, but even
there this is starting to happen and I had a great discussion with
some of the faculty who are pushing off in that direction and really
seeing that the boundary of computer science and biology are very
much coming together and we need people who understand both of those
things to make advances in solving diseases. And I*m very optimistic
about how fast that will move forward.
Now, this tool that we*ve got, the PC connected to the Internet,
all this great empowerment, it*s such an important thing that we do
have to worry that there are people who are benefiting from this and
people who are not, and people talk about that as the Digital
Divide. It*s something that I think people in computer science
should care a lot about and in various ways contribute to trying to
minimize that difference.
One of the projects that Microsoft got involved with, together
with my foundation, was saying, well, what about libraries, would it
be appropriate to have computers there. And we were a little worried
about this because would the librarian like it, would it be at
expense of the books, would kids come in and just hack up the
machines, would they be doing enriching things as they were using
those computers; it was unclear. But six years ago we kicked it off,
we did pilot projects and over six years in all 50 states in 18,000
libraries, rural, everywhere in the country, we put in over 50,000
computers. And the response of librarians was just phenomenal of
wanting to be trained, wanting to reinforce the role of that
library, and traffic to the library increased, not just to come use
the computer but also the number of books that were being lent out.
And we were able to monitor and support all these things in a very
efficient way that made it work very well.
Throughout the project we learned things. We came up with a
version of the software you can just hit a button to switch from
Spanish to English for a lot of these libraries. We had a button you
can just hit to switch to big print so if you don*t like reading the
font that we typically use, boom, all of a sudden it*s a lot better.
We came up with things to help people with the common scenarios.
And so it*s great to see if you give people those tools they*ll
use them and it really makes a big difference for them.
Getting these tools out to schools, getting them out to all
different countries, a lot of challenges remain there that need to
be addressed.
When we think about computer science, one thing computer science
has done through the Internet, through software has made the world a
smaller place. In fact, people now worry that this is going to
create a new level of global competition. And the answer is it is.
People*s opportunity to have great jobs in the future will be far
more determined by their level of education than by what country
they happen to be in. Historically, your educational level didn*t
matter that much. If you were in a rich country you made a lot of
money, and if you were in a poor country you made very little money.
Now the opportunity for educated people worldwide to help out, to
contribute to products, not just software products but anything you
can imagine, architecture, law, answering the phone, it will be done
where people have those skills.
And as people look at that, they go, wow, what does that mean?
Well, it means the U.S. has to keep its edge in terms of doing the
best work and that means research, it means intellectual property,
it means improving the education system, rededication.
It*s very similar to what happened in the 1980s when there was a
lot of angst about Japan. Japan at the time appeared to have a model
where they would just pick an industry, the car industry, the
computer industry, the consumer electronics industry and boom they
would do it better.
And the great thing that happened in the *80s was there was a lot
of humility, a lot of thinking, well, do we just match what they do
exactly that way or do we just push forward on our strengths, our
approach to things.
During the *80s they did this AI project and it really because of
the way it was done it wasn*t done with a diverse academic approach
that we use here, it really ended up not generating much.
So we rededicated ourselves and it was actually the work done
during that period that led to that productivity increase that
benefited all countries but the U.S. in particular during the 1990s.
I see that same thing repeating itself as we question our unique
role and reinforce what needs to be done.
One challenge that we have in the field, in all science fields
but particularly in computer science is the issue of diversity. To
do the best work we want to draw on everybody*s talent and give
everybody a deep involvement. The variety of jobs, the need for
great people is pretty phenomenal. And the diversity numbers in some
professions like law and medicine have been going pretty strongly in
the favorable direction.
One thing I personally have gotten involved with to try and help
push this forward is a scholarship program that*s called the Gates
Millennium Scholarship. And here at MIT, out of actually a thousand
people who get those scholarships, 60 people are here at MIT, so
it*s a great thing and a real endorsement of MIT that there are more
Gates scholars at this school than at any other school.
These science problems are tough but they*re fun to work on. The
jobs that are involved with them, you can have an impact, you can
work with other people, it*s not just somebody isolated off coding
all night, although if you want to do that, that*s fine, we still
have lots of jobs that are like that.
And so the sense of reward of being involved in changing
business, changing entertainment, changing education, giving tools
to those new sciences, including to help with disease, I think
that*s a phenomenal opportunity. So that*s why I*m more excited
about computer science than ever and I*m very excited to see what
some of you here can do, taking that to the next level.
Thank you. (Applause.)