1
00:00:00,000 --> 00:00:03,780
[APPLAUSE]

2
00:00:03,780 --> 00:00:05,460
Sorry for the delay
in getting started,

3
00:00:05,460 --> 00:00:06,960
we were having
trouble with Windows.

4
00:00:06,960 --> 00:00:12,980


5
00:00:12,980 --> 00:00:17,660
Bill Gates is always
a hard act to follow,

6
00:00:17,660 --> 00:00:21,290
but who better than the
author of the book What

7
00:00:21,290 --> 00:00:25,095
Will Be, our own Professor
Michael Dertouzos, who

8
00:00:25,095 --> 00:00:29,700
has decided today to focus his
remarks on the next 35 years

9
00:00:29,700 --> 00:00:32,110
for the entire world?

10
00:00:32,110 --> 00:00:32,610
Michael.

11
00:00:32,610 --> 00:00:42,810


12
00:00:42,810 --> 00:00:46,080
Thank you, Bob.

13
00:00:46,080 --> 00:00:49,770
Ladies and gentlemen,
there is no reason

14
00:00:49,770 --> 00:00:54,100
we can't talk about the future
while having a little fun.

15
00:00:54,100 --> 00:00:56,670
So, let me recollect
something for you.

16
00:00:56,670 --> 00:01:00,210
Last year, a few of
my colleagues and I

17
00:01:00,210 --> 00:01:03,030
were heading to Taiwan.

18
00:01:03,030 --> 00:01:06,480
The flight, as you know, is
very long, and on the airplane

19
00:01:06,480 --> 00:01:10,020
I was trying to
make something work.

20
00:01:10,020 --> 00:01:14,280
Now, that something was a
laptop, a brand new laptop,

21
00:01:14,280 --> 00:01:16,050
the latest--

22
00:01:16,050 --> 00:01:18,900
it was running Windows--

23
00:01:18,900 --> 00:01:23,580
and a little card on which
you download your software,

24
00:01:23,580 --> 00:01:26,290
you download your addresses,
you download your calendar,

25
00:01:26,290 --> 00:01:28,680
and then you take the
card in your pocket.

26
00:01:28,680 --> 00:01:31,680
I had been at it
for three hours,

27
00:01:31,680 --> 00:01:34,500
and my problem was that when
the card software was happy,

28
00:01:34,500 --> 00:01:39,520
the operating system wasn't, and
vice versa, as you well know.

29
00:01:39,520 --> 00:01:44,550
So, we're on the plane, so I
turn to one of my colleagues

30
00:01:44,550 --> 00:01:47,960
and I say, Tim, would
you please help me?

31
00:01:47,960 --> 00:01:50,330
Well, Tim Berners-Lee
is a very gracious man.

32
00:01:50,330 --> 00:01:52,340
So he said, of course.

33
00:01:52,340 --> 00:01:54,240
He said, I think I can help you.

34
00:01:54,240 --> 00:01:56,060
So he picked up the machine.

35
00:01:56,060 --> 00:01:58,732
An hour and a half later,
he hands it back to me,

36
00:01:58,732 --> 00:02:01,190
still very polite, still very
British, and says, I'm sorry,

37
00:02:01,190 --> 00:02:03,870
I can't make this thing work.

38
00:02:03,870 --> 00:02:08,690
So I turned to my friend and
associate director Ron Rivest,

39
00:02:08,690 --> 00:02:11,540
and I say, the inventor
of the web couldn't do it.

40
00:02:11,540 --> 00:02:15,290
How about you, inventor of RSA?

41
00:02:15,290 --> 00:02:19,310
And Ron, with his characteristic
wisdom, looked at me

42
00:02:19,310 --> 00:02:22,640
and said, no Mike, thanks.

43
00:02:22,640 --> 00:02:25,730
And then, the youngest
member of our delegation,

44
00:02:25,730 --> 00:02:29,270
whose name I shall not mention
to protect the innocent,

45
00:02:29,270 --> 00:02:30,050
jumped up.

46
00:02:30,050 --> 00:02:35,150
And he says to me, you guys
are a bunch of old fogies.

47
00:02:35,150 --> 00:02:38,870
Let me show you.

48
00:02:38,870 --> 00:02:40,490
Good LCS attitude.

49
00:02:40,490 --> 00:02:42,830
Two hours later,
he hands it to me

50
00:02:42,830 --> 00:02:47,170
with expletives
surrounding this.

51
00:02:47,170 --> 00:02:51,220
I went to my seat and I
did what every expert does.

52
00:02:51,220 --> 00:02:54,580
I tried the various
wizards and lizards

53
00:02:54,580 --> 00:02:56,870
that were coming up at me.

54
00:02:56,870 --> 00:03:00,430
And I kept typing, at
random, things in the fields,

55
00:03:00,430 --> 00:03:03,400
until finally, three
hours later, it worked.

56
00:03:03,400 --> 00:03:05,550
Total elapsed time, nine hours.

57
00:03:05,550 --> 00:03:08,550
[? Hold the ?] battery, Mike!

58
00:03:08,550 --> 00:03:09,533
[LAUGHTER]

59
00:03:09,533 --> 00:03:11,200
Good point for those
who make batteries.

60
00:03:11,200 --> 00:03:13,420
Now, ladies and gentlemen,
why do I bring this up?

61
00:03:13,420 --> 00:03:15,820
Our field is young,
so I don't bring it up

62
00:03:15,820 --> 00:03:17,020
to complain so much.

63
00:03:17,020 --> 00:03:18,580
This is to be expected.

64
00:03:18,580 --> 00:03:21,820
But we are at a junction.

65
00:03:21,820 --> 00:03:24,940
What I experienced
with that example

66
00:03:24,940 --> 00:03:28,750
is typical of what we have
done for the last 40 years

67
00:03:28,750 --> 00:03:30,340
with our field.

68
00:03:30,340 --> 00:03:33,760
We have focused on
designing for machines.

69
00:03:33,760 --> 00:03:34,900
We put little envelopes.

70
00:03:34,900 --> 00:03:36,400
We call them
operating systems, we

71
00:03:36,400 --> 00:03:39,100
call them browsers, we
call them network software.

72
00:03:39,100 --> 00:03:43,630
And then we throw all
this stuff at the user.

73
00:03:43,630 --> 00:03:47,060
Now, the image that
this evokes to me

74
00:03:47,060 --> 00:03:51,950
is a user, you, sitting
in your car, a car that's

75
00:03:51,950 --> 00:03:53,450
very early vintage.

76
00:03:53,450 --> 00:03:56,090
And you've got rings
on every finger.

77
00:03:56,090 --> 00:03:58,970
And on every ring,
a wire is attached.

78
00:03:58,970 --> 00:04:01,400
And you've got a wire
for fuel mixture.

79
00:04:01,400 --> 00:04:05,170
You've got another wire for
advance of the ignition.

80
00:04:05,170 --> 00:04:07,670
You've got another one for the
valve clearances, the brakes,

81
00:04:07,670 --> 00:04:08,540
and this and that.

82
00:04:08,540 --> 00:04:10,910
And you're sitting there
with all these wires

83
00:04:10,910 --> 00:04:14,750
and we say to you, go
from Boston to New York.

84
00:04:14,750 --> 00:04:18,060


85
00:04:18,060 --> 00:04:22,040
Ladies and gentlemen,
this must stop.

86
00:04:22,040 --> 00:04:23,480
40 years is enough.

87
00:04:23,480 --> 00:04:30,550


88
00:04:30,550 --> 00:04:35,530
This brings me squarely into
my vision for the 21st century.

89
00:04:35,530 --> 00:04:40,050
Stated in simple human
terms, it is simply this.

90
00:04:40,050 --> 00:04:47,030
People should be able to
do more by doing less.

91
00:04:47,030 --> 00:04:50,360
Now, when I say these
simple words, which

92
00:04:50,360 --> 00:04:54,680
have been characteristic of
every revolution up to now,

93
00:04:54,680 --> 00:04:58,160
I mean three things.

94
00:04:58,160 --> 00:05:02,040
First, we should bring the
technologies of information

95
00:05:02,040 --> 00:05:03,810
into our lives.

96
00:05:03,810 --> 00:05:06,480
Now, when I read the papers
and the magazines and I hear

97
00:05:06,480 --> 00:05:10,860
the pundits talk
about cyberspace,

98
00:05:10,860 --> 00:05:12,610
where we're going to
go, this other space.

99
00:05:12,610 --> 00:05:15,520
We're going to leave our
lives and go into cyberspace.

100
00:05:15,520 --> 00:05:18,355
We're going to put on glasses,
gigabyte [? infected ?] suits.

101
00:05:18,355 --> 00:05:21,640


102
00:05:21,640 --> 00:05:23,920
And we're going to live
in an exciting world.

103
00:05:23,920 --> 00:05:27,280
Well, you know, when the
Industrial Revolution came,

104
00:05:27,280 --> 00:05:29,575
we didn't go into motor space.

105
00:05:29,575 --> 00:05:33,010


106
00:05:33,010 --> 00:05:36,020
The motors came into our lives.

107
00:05:36,020 --> 00:05:37,540
We turned them
into refrigerators

108
00:05:37,540 --> 00:05:40,960
and we call them
refrigerators, not motors.

109
00:05:40,960 --> 00:05:43,540
We turned them into
cars, and call them cars.

110
00:05:43,540 --> 00:05:46,300
And they absorb
themselves, the motors,

111
00:05:46,300 --> 00:05:49,090
into the functions and
tasks that ancient humans

112
00:05:49,090 --> 00:05:54,080
have been doing, except
now with new tools.

113
00:05:54,080 --> 00:05:57,370
So, the first thing is to bring
the technology into our lives.

114
00:05:57,370 --> 00:06:00,170
And that's exactly what
I think will happen,

115
00:06:00,170 --> 00:06:04,100
and what I wish will happen,
with the Information Age.

116
00:06:04,100 --> 00:06:06,850
They will come into our
lives, the new technologies,

117
00:06:06,850 --> 00:06:10,490
and they will serve
our ancient needs.

118
00:06:10,490 --> 00:06:13,270
The second thing
that I mean when

119
00:06:13,270 --> 00:06:19,510
I say doing more by doing less
is increased human productivity

120
00:06:19,510 --> 00:06:21,700
and increased ease of use.

121
00:06:21,700 --> 00:06:24,350


122
00:06:24,350 --> 00:06:28,470
With respect to increased
human productivity

123
00:06:28,470 --> 00:06:32,415
first, imagine that
I'm holding a handheld.

124
00:06:32,415 --> 00:06:36,240


125
00:06:36,240 --> 00:06:39,690
Take us to Athens this weekend.

126
00:06:39,690 --> 00:06:44,040
Later, you'll see a
system that does this.

127
00:06:44,040 --> 00:06:46,840
It took me three
seconds to say that.

128
00:06:46,840 --> 00:06:49,090
The computer knows that
"us" is two people,

129
00:06:49,090 --> 00:06:50,710
but we like to travel
business class.

130
00:06:50,710 --> 00:06:53,830
It knows that we can this
Friday, Saturday, and Sunday.

131
00:06:53,830 --> 00:06:56,050
It connects with Eaasy Sabre.

132
00:06:56,050 --> 00:06:57,520
It starts doing the exchanges.

133
00:06:57,520 --> 00:06:58,490
It tries for Friday.

134
00:06:58,490 --> 00:06:59,740
It doesn't find an aisle seat.

135
00:06:59,740 --> 00:07:00,670
We like aisle seats.

136
00:07:00,670 --> 00:07:01,830
It goes to Saturday.

137
00:07:01,830 --> 00:07:03,830
It finds two seats, but
something else is wrong.

138
00:07:03,830 --> 00:07:06,610
It finally books on Sunday.

139
00:07:06,610 --> 00:07:10,120
Total elapsed time in
actual experiments for this

140
00:07:10,120 --> 00:07:14,650
is a minimum of three minutes,
maximum of 13 minutes.

141
00:07:14,650 --> 00:07:19,400
I will take five minutes
as an acceptable average.

142
00:07:19,400 --> 00:07:24,450
Five minutes is 300 seconds.

143
00:07:24,450 --> 00:07:24,950
Nod.

144
00:07:24,950 --> 00:07:27,920


145
00:07:27,920 --> 00:07:29,900
You went to Harvard for a while.

146
00:07:29,900 --> 00:07:34,700
Divided by the three
seconds, the 300 seconds

147
00:07:34,700 --> 00:07:38,390
gives us a ratio of 100
productivity leverage.

148
00:07:38,390 --> 00:07:42,685
But in business
terms, that's 10,000%.

149
00:07:42,685 --> 00:07:47,730
We have achieved a 10,000%
increase in human productivity.

150
00:07:47,730 --> 00:07:50,400
Now, this is actual.

151
00:07:50,400 --> 00:07:52,650
Maybe I do it once a day and
you say it doesn't count.

152
00:07:52,650 --> 00:07:54,980
But if I was the travel agent
for a large organization,

153
00:07:54,980 --> 00:07:56,930
it would count.

154
00:07:56,930 --> 00:08:01,070
Now, I don't expect these kind
of gains across the board,

155
00:08:01,070 --> 00:08:07,760
but I do expect across the
board in the next century

156
00:08:07,760 --> 00:08:13,580
about a 300% increase in
human productivity, primarily

157
00:08:13,580 --> 00:08:16,130
in the office sector.

158
00:08:16,130 --> 00:08:18,590
Today, as you look
at your browsers,

159
00:08:18,590 --> 00:08:20,540
as you look at
your email, you're

160
00:08:20,540 --> 00:08:22,250
sitting there
exercising your eyeballs

161
00:08:22,250 --> 00:08:27,020
and your brains just as hard
as our predecessors exercised

162
00:08:27,020 --> 00:08:29,150
their shovels in
the Industrial Era,

163
00:08:29,150 --> 00:08:32,750
except you're using
these for your shovels.

164
00:08:32,750 --> 00:08:34,010
And shoveling you are.

165
00:08:34,010 --> 00:08:38,070


166
00:08:38,070 --> 00:08:40,919
But it isn't painful
because the shovels

167
00:08:40,919 --> 00:08:42,840
are stamped "high tech."

168
00:08:42,840 --> 00:08:45,990


169
00:08:45,990 --> 00:08:49,380
All right, let's
take up ease of use.

170
00:08:49,380 --> 00:08:53,700
Every time I hear the
phrase "ease of use"

171
00:08:53,700 --> 00:08:56,020
from a vendor of
hardware or software,

172
00:08:56,020 --> 00:08:57,640
I become physically violent.

173
00:08:57,640 --> 00:09:00,810


174
00:09:00,810 --> 00:09:03,450
The invariable
image that pops up

175
00:09:03,450 --> 00:09:07,710
is of their operating a
chimpanzee around dressed

176
00:09:07,710 --> 00:09:14,140
in a green suit pretending
to be a surgeon.

177
00:09:14,140 --> 00:09:15,640
I mean, the thing
isn't easy to use.

178
00:09:15,640 --> 00:09:18,790
The fact that they put
floating animals and collars

179
00:09:18,790 --> 00:09:23,060
and ergonomics doesn't
make it easy to use.

180
00:09:23,060 --> 00:09:25,090
What we want is something
that's truly easy

181
00:09:25,090 --> 00:09:27,790
to use, regardless of whether
it's using graphics or text

182
00:09:27,790 --> 00:09:29,110
or whatever.

183
00:09:29,110 --> 00:09:35,220
Let me pick an example,
browsers and operating systems.

184
00:09:35,220 --> 00:09:40,110
Those two worlds, one
catering to local information,

185
00:09:40,110 --> 00:09:42,220
the computer with
its computer system,

186
00:09:42,220 --> 00:09:45,210
the other one catering to
long-distance information

187
00:09:45,210 --> 00:09:49,070
across the browser, have
different commands, quite

188
00:09:49,070 --> 00:09:51,300
different commands.

189
00:09:51,300 --> 00:09:54,950
Now, that doesn't
lead to ease of use.

190
00:09:54,950 --> 00:09:58,930
The image I get from that
is that of an automobile

191
00:09:58,930 --> 00:10:01,765
again, which, when
you're driving it

192
00:10:01,765 --> 00:10:05,320
in your neighborhood,
local, when you turn right,

193
00:10:05,320 --> 00:10:06,940
it turns right.

194
00:10:06,940 --> 00:10:09,220
But when you're driving
it on country roads, when

195
00:10:09,220 --> 00:10:13,520
you turn the wheel right,
it applies the brakes.

196
00:10:13,520 --> 00:10:17,720
It sounds silly here,
but it sounds OK

197
00:10:17,720 --> 00:10:20,660
when you highlight an icon
in the operating system

198
00:10:20,660 --> 00:10:24,090
and take a hotlink on the
browser with the same command.

199
00:10:24,090 --> 00:10:27,690
So we've got to revise
this kind of stuff.

200
00:10:27,690 --> 00:10:30,950
Now, ladies and gentlemen,
this is the second thing

201
00:10:30,950 --> 00:10:34,280
that I mean when I
say that we should be

202
00:10:34,280 --> 00:10:37,550
able to do more by doing less.

203
00:10:37,550 --> 00:10:42,968
The third thing that I mean
is that when we say "we,"

204
00:10:42,968 --> 00:10:46,400
"we" should mean the
people of the world.

205
00:10:46,400 --> 00:10:49,550
Now, we feel pretty
smug today because we,

206
00:10:49,550 --> 00:10:52,610
the people of the
industrialized wealthy world,

207
00:10:52,610 --> 00:10:55,050
are largely connected.

208
00:10:55,050 --> 00:10:58,400
But if you stop and look, there
are 600 million telephones

209
00:10:58,400 --> 00:11:02,360
only, which is 10% of the 6
billion living on this Earth.

210
00:11:02,360 --> 00:11:06,170
And in terms of these 600
million telephones, only 15%

211
00:11:06,170 --> 00:11:10,710
are interconnected, about
80 or 90 million people.

212
00:11:10,710 --> 00:11:11,730
Multiply the two.

213
00:11:11,730 --> 00:11:14,860


214
00:11:14,860 --> 00:11:18,710
1.5% to 2% of the
world's population,

215
00:11:18,710 --> 00:11:21,310
ladies and gentlemen,
is interconnected.

216
00:11:21,310 --> 00:11:22,480
That's all.

217
00:11:22,480 --> 00:11:26,260
98% is disconnected.

218
00:11:26,260 --> 00:11:30,280
We cannot hear the millions
of voices from the Indian

219
00:11:30,280 --> 00:11:34,250
subcontinent, from
Africa, and from China,

220
00:11:34,250 --> 00:11:37,270
although increasingly they're
becoming connected there.

221
00:11:37,270 --> 00:11:39,700
All we're getting is
the television feeds

222
00:11:39,700 --> 00:11:41,710
and the government feeds.

223
00:11:41,710 --> 00:11:44,290
And these people do not
have the opportunity

224
00:11:44,290 --> 00:11:46,690
to sell their services
or their products

225
00:11:46,690 --> 00:11:51,310
yet on the world's markets
through the new medium.

226
00:11:51,310 --> 00:11:55,510
Now, I think it's very
important to bring these people

227
00:11:55,510 --> 00:12:01,870
in because left to
its own devices,

228
00:12:01,870 --> 00:12:05,010
the Information Revolution
will increase the gap

229
00:12:05,010 --> 00:12:06,930
between rich and poor people.

230
00:12:06,930 --> 00:12:10,820
The proof is
excruciatingly simple.

231
00:12:10,820 --> 00:12:12,470
The rich get more computers.

232
00:12:12,470 --> 00:12:13,860
They become more productive.

233
00:12:13,860 --> 00:12:15,290
They get richer.

234
00:12:15,290 --> 00:12:18,920
The poor have no computers,
stay where they are, formula

235
00:12:18,920 --> 00:12:21,410
for exponential gap increase.

236
00:12:21,410 --> 00:12:23,240
Everybody says we can
help them to get out

237
00:12:23,240 --> 00:12:24,920
of the problems, et cetera.

238
00:12:24,920 --> 00:12:26,390
Yes, but we've got to help them.

239
00:12:26,390 --> 00:12:29,240
Alone, they cannot get the
machines, the communications,

240
00:12:29,240 --> 00:12:32,270
the training, and the
skills to do that.

241
00:12:32,270 --> 00:12:37,900
It's important that we do so,
not only for compassion's sake

242
00:12:37,900 --> 00:12:42,260
and for altruism, but also
for self-preservation,

243
00:12:42,260 --> 00:12:45,140
because history teaches
that when gaps like this

244
00:12:45,140 --> 00:12:50,240
are allowed to grow,
they result in bloodshed.

245
00:12:50,240 --> 00:12:54,140
So, ladies and gentlemen,
there's so much potential

246
00:12:54,140 --> 00:12:57,020
if we do help the poor,
not only poor countries,

247
00:12:57,020 --> 00:13:00,260
but poor people within
the rich countries.

248
00:13:00,260 --> 00:13:03,060
The kinds of
services and products

249
00:13:03,060 --> 00:13:08,130
they could sell, the
way they could rise,

250
00:13:08,130 --> 00:13:11,820
it's not always helping
from the rich to the poor.

251
00:13:11,820 --> 00:13:13,950
Imagine a Sri Lankan doctor.

252
00:13:13,950 --> 00:13:19,620
A Sri Lankan doctor makes
$20 a day and speaks English.

253
00:13:19,620 --> 00:13:24,990
Imagine that doctor connected
through the world's networks

254
00:13:24,990 --> 00:13:28,920
administering health care
to homeless people in Boston

255
00:13:28,920 --> 00:13:30,870
who go to a kiosk
where there's a nurse,

256
00:13:30,870 --> 00:13:33,150
there's a blood cuff
meter, a monitor.

257
00:13:33,150 --> 00:13:36,480
They can talk to the doctor.

258
00:13:36,480 --> 00:13:38,480
Maybe it's not the best
health care in the world

259
00:13:38,480 --> 00:13:41,780
because the doctor isn't
there, but it's a lot better

260
00:13:41,780 --> 00:13:47,500
than no health care, and it
might cost $2 or $3 per person.

261
00:13:47,500 --> 00:13:49,990
At the Lab for Computer Science,
a team of undergraduates

262
00:13:49,990 --> 00:13:53,170
is building a virtual
Compassion Corps

263
00:13:53,170 --> 00:13:55,198
that I discuss in What Will Be.

264
00:13:55,198 --> 00:13:56,740
This is nothing more
and nothing less

265
00:13:56,740 --> 00:14:01,180
than a gigantic want
ad website where

266
00:14:01,180 --> 00:14:04,480
people who wish to give
human help and people who

267
00:14:04,480 --> 00:14:07,630
wish to receive human help
find each other according

268
00:14:07,630 --> 00:14:09,220
to the categories of help.

269
00:14:09,220 --> 00:14:11,638
And they simply match
with each other and help.

270
00:14:11,638 --> 00:14:13,180
This is the first
time in our history

271
00:14:13,180 --> 00:14:15,130
we could do this worldwide.

272
00:14:15,130 --> 00:14:19,850


273
00:14:19,850 --> 00:14:27,040
So, to make it
short, the vision is

274
00:14:27,040 --> 00:14:30,070
to let people do
more by doing less,

275
00:14:30,070 --> 00:14:33,800
and by this, I
mean three things.

276
00:14:33,800 --> 00:14:37,120
First, bring the
technology into our lives.

277
00:14:37,120 --> 00:14:41,290
Second, improve human
productivity and ease of use.

278
00:14:41,290 --> 00:14:45,730
And third, include in this
vision all people of the world.

279
00:14:45,730 --> 00:14:49,990
Good enough for top-level
vision, but now,

280
00:14:49,990 --> 00:14:54,410
over what computing model
do we see this taking place?

281
00:14:54,410 --> 00:14:57,370
Well, for the last
20 years, you've

282
00:14:57,370 --> 00:15:00,950
heard me talking about the
information marketplace.

283
00:15:00,950 --> 00:15:04,660
And I'm not going to change
my tune because we haven't yet

284
00:15:04,660 --> 00:15:06,320
realized it.

285
00:15:06,320 --> 00:15:11,720
So the model I am proposing is
the same model, the information

286
00:15:11,720 --> 00:15:15,640
marketplace, which
I define as follows.

287
00:15:15,640 --> 00:15:19,480
People and their computers,
about half a billion

288
00:15:19,480 --> 00:15:24,370
to a billion in the next decade,
plus appliances, gadgets,

289
00:15:24,370 --> 00:15:27,850
embedded computers, and
the like that are unmanned,

290
00:15:27,850 --> 00:15:33,030
interconnected together
doing three things,

291
00:15:33,030 --> 00:15:38,520
buying, selling, and
freely exchanging

292
00:15:38,520 --> 00:15:45,080
two things, information
and information services.

293
00:15:45,080 --> 00:15:47,450
Now, with respect to
buying and selling,

294
00:15:47,450 --> 00:15:52,800
currently we are at $50 billion
per year on the internet.

295
00:15:52,800 --> 00:15:58,610
My estimate is $4 trillion
in the next 30 years or so,

296
00:15:58,610 --> 00:16:01,940
doing so by going
predominantly down.

297
00:16:01,940 --> 00:16:04,730
That half of the
industrial economy

298
00:16:04,730 --> 00:16:08,510
is called the office sector,
previously white collar work.

299
00:16:08,510 --> 00:16:12,890
Half of our economy
is the office sector.

300
00:16:12,890 --> 00:16:15,980
And that is where the
bulk of it will come.

301
00:16:15,980 --> 00:16:20,090
Now, a big part of that is going
to be a new kind of activity

302
00:16:20,090 --> 00:16:25,290
we haven't seen yet, which I
like to call information work.

303
00:16:25,290 --> 00:16:27,890
So we deal with
information as a noun.

304
00:16:27,890 --> 00:16:33,620
Information as a noun is
text, sounds, images, videos.

305
00:16:33,620 --> 00:16:37,340
Information as a verb
is what a doctor does.

306
00:16:37,340 --> 00:16:40,400
It's what a thousand accountants
do from Beijing, China

307
00:16:40,400 --> 00:16:43,250
when they sell their services
to General Motors in Detroit

308
00:16:43,250 --> 00:16:44,780
for $2 an hour.

309
00:16:44,780 --> 00:16:46,970
It is the processing, the
management, the handling

310
00:16:46,970 --> 00:16:48,515
of information, the service.

311
00:16:48,515 --> 00:16:51,370


312
00:16:51,370 --> 00:16:54,400
That is going to be, I
believe, a very large part

313
00:16:54,400 --> 00:16:56,270
of the information marketplace.

314
00:16:56,270 --> 00:16:59,140
You add to that the
non-dollar component,

315
00:16:59,140 --> 00:17:02,860
the family message links,
the political discussions,

316
00:17:02,860 --> 00:17:07,089
the publishing, the
administration of health freely

317
00:17:07,089 --> 00:17:10,660
where it happens, do-gooder
activities, chat groups,

318
00:17:10,660 --> 00:17:14,740
neighborhoods, all that huge
non-dollar component which

319
00:17:14,740 --> 00:17:16,220
cannot be measured.

320
00:17:16,220 --> 00:17:18,290
We cannot put a $4
trillion stamp on it,

321
00:17:18,290 --> 00:17:20,200
but it's just as important.

322
00:17:20,200 --> 00:17:25,270
And you have the picture of
this information marketplace.

323
00:17:25,270 --> 00:17:26,599
We're not yet there.

324
00:17:26,599 --> 00:17:29,410
There's no information work
on the current marketplace.

325
00:17:29,410 --> 00:17:30,910
There is no automation.

326
00:17:30,910 --> 00:17:32,740
There's no capability
to do collaboration.

327
00:17:32,740 --> 00:17:37,900
All we're doing is having
voyeurs and exhibitionists.

328
00:17:37,900 --> 00:17:39,280
The voyeurs are browsing.

329
00:17:39,280 --> 00:17:41,920
The exhibitionists are saying,
look at what I'm selling.

330
00:17:41,920 --> 00:17:44,150
And there are a few little
transactions between them.

331
00:17:44,150 --> 00:17:46,930
So, we have a little bit of
ways to go before we get there.

332
00:17:46,930 --> 00:17:49,450


333
00:17:49,450 --> 00:17:55,240
So, let this be the vision
of doing more by doing less.

334
00:17:55,240 --> 00:17:59,620
And let this be the
model over which

335
00:17:59,620 --> 00:18:03,980
we will do more by doing less
in the information marketplace.

336
00:18:03,980 --> 00:18:08,900
Now, that's still
pretty general.

337
00:18:08,900 --> 00:18:10,780
So how might we do that?

338
00:18:10,780 --> 00:18:14,750


339
00:18:14,750 --> 00:18:19,190
What I would like to do next
is to announce a major research

340
00:18:19,190 --> 00:18:23,600
project that we
are launching today

341
00:18:23,600 --> 00:18:26,480
at the MIT Laboratory
for Computer Science.

342
00:18:26,480 --> 00:18:30,620
We call it Oxygen. It's
a research project.

343
00:18:30,620 --> 00:18:36,980
It involves a radically
new system for tomorrow

344
00:18:36,980 --> 00:18:41,150
that contains a lot of this
stuff that I talked about.

345
00:18:41,150 --> 00:18:47,030
I'm very happy to say that
Oxygen went as a $38.5 million

346
00:18:47,030 --> 00:18:50,150
proposal to DARPA, and
I'm very pleased to say

347
00:18:50,150 --> 00:18:52,580
that DARPA awarded it to us.

348
00:18:52,580 --> 00:18:55,190
But they haven't
awarded the money yet,

349
00:18:55,190 --> 00:18:57,770
so we're negotiating that.

350
00:18:57,770 --> 00:18:59,570
While I'm at it, let
me say, we always

351
00:18:59,570 --> 00:19:01,170
say bad things about government.

352
00:19:01,170 --> 00:19:03,890
Let me say that in the last 25
years I've led this laboratory,

353
00:19:03,890 --> 00:19:06,530
and the 10 previous
years DARPA has

354
00:19:06,530 --> 00:19:10,010
been with us and with
other major universities,

355
00:19:10,010 --> 00:19:14,570
they've carved computer science
and technology on the map.

356
00:19:14,570 --> 00:19:18,110
And let me say that the current
efforts of the administration

357
00:19:18,110 --> 00:19:20,210
to increase the funding
in computer science

358
00:19:20,210 --> 00:19:22,700
and the various activities
that have happened there

359
00:19:22,700 --> 00:19:23,960
are also laudable.

360
00:19:23,960 --> 00:19:26,480
So we shouldn't always
complain about them.

361
00:19:26,480 --> 00:19:28,310
They've been with
us and this birthday

362
00:19:28,310 --> 00:19:31,940
is largely due to them.

363
00:19:31,940 --> 00:19:38,880
All right, Oxygen. Let me
make one more qualification

364
00:19:38,880 --> 00:19:42,320
before I describe it to you.

365
00:19:42,320 --> 00:19:46,580
We're not a corporation
with a smooth presentation

366
00:19:46,580 --> 00:19:48,840
about the new system.

367
00:19:48,840 --> 00:19:51,060
We have a framework.

368
00:19:51,060 --> 00:19:53,460
Now, I'm not going
to be surprised

369
00:19:53,460 --> 00:19:55,350
because I'm certain
that when we're done,

370
00:19:55,350 --> 00:19:57,225
the thing will be totally
different from what

371
00:19:57,225 --> 00:19:59,050
I described to you today.

372
00:19:59,050 --> 00:20:00,810
As you hear the talks
of my colleagues

373
00:20:00,810 --> 00:20:02,760
who will speak about
various pieces of this,

374
00:20:02,760 --> 00:20:04,540
keep that in mind.

375
00:20:04,540 --> 00:20:06,630
And as you hear the
talks of the others, who

376
00:20:06,630 --> 00:20:09,450
will talk about other
things while pretending

377
00:20:09,450 --> 00:20:12,540
to talk about the
pieces, keep in mind

378
00:20:12,540 --> 00:20:16,740
that this laboratory does a
lot of independent research.

379
00:20:16,740 --> 00:20:19,950
Keep in mind, too, that as
we do all the system's work,

380
00:20:19,950 --> 00:20:22,950
our eyeballs are
focused on trying

381
00:20:22,950 --> 00:20:25,890
to understand the
fundamentals and the theories

382
00:20:25,890 --> 00:20:29,550
behind this brave
new field of ours.

383
00:20:29,550 --> 00:20:32,350
With that as background,
what is Oxygen?

384
00:20:32,350 --> 00:20:38,790
Oxygen is eight technologies,
which I will now enumerate

385
00:20:38,790 --> 00:20:40,900
and describe very briefly.

386
00:20:40,900 --> 00:20:49,940
Number one, the Handy
21, for 21st century.

387
00:20:49,940 --> 00:21:00,260
You see it has a little display
right there, which you can see.

388
00:21:00,260 --> 00:21:03,350
If you're nearsighted, it helps.

389
00:21:03,350 --> 00:21:09,230
It's like a cellular phone,
except that it has in it

390
00:21:09,230 --> 00:21:11,360
a small screen, as you've seen.

391
00:21:11,360 --> 00:21:14,060
It has a very small analog part.

392
00:21:14,060 --> 00:21:15,710
The rest is all
digital, which means

393
00:21:15,710 --> 00:21:17,540
it can change like a chameleon.

394
00:21:17,540 --> 00:21:19,880
It's got an infrared detector.

395
00:21:19,880 --> 00:21:23,330
And it's got, of course, a
few other odds and ends which

396
00:21:23,330 --> 00:21:27,620
permit it, with
downloading, to alter

397
00:21:27,620 --> 00:21:30,950
its function from a
two-way radio talking

398
00:21:30,950 --> 00:21:34,520
to other such handhelds,
to a network node talking

399
00:21:34,520 --> 00:21:39,140
to the campus network, to a
cell phone which sniffs the air

400
00:21:39,140 --> 00:21:42,500
and decides, I'm in France so
I'm going to switch to GSM.

401
00:21:42,500 --> 00:21:46,840
Or it says, I'm in America using
CDMA, and on and on and on.

402
00:21:46,840 --> 00:21:48,890
We have done this early
prototype in connection

403
00:21:48,890 --> 00:21:52,940
with another DARPA contract that
Victor Zue and [? Raclowski ?]

404
00:21:52,940 --> 00:21:58,320
and Clark are involved in, and
maybe we'll hear more about it.

405
00:21:58,320 --> 00:22:02,300
But that is the first
technology of Oxygen. What

406
00:22:02,300 --> 00:22:05,450
we're trying to do
with that is make sure

407
00:22:05,450 --> 00:22:11,690
that you have a thing that is
person centric wherever you go.

408
00:22:11,690 --> 00:22:13,760
It's with you.

409
00:22:13,760 --> 00:22:16,670
In symmetric form, the
second technology of Oxygen

410
00:22:16,670 --> 00:22:22,490
is the Enviro 21, which
is environment centric.

411
00:22:22,490 --> 00:22:24,380
It sticks to the wall.

412
00:22:24,380 --> 00:22:27,590
Actually, it's big, so it may be
in the wall or under the desk.

413
00:22:27,590 --> 00:22:29,720
It's in the trunk of your car.

414
00:22:29,720 --> 00:22:31,720
It's in your home.

415
00:22:31,720 --> 00:22:36,970
It is, in every respect, like
the Handy 21 functionally,

416
00:22:36,970 --> 00:22:39,380
but it is much bigger.

417
00:22:39,380 --> 00:22:42,980
So it bears a relationship of
a power socket to a battery,

418
00:22:42,980 --> 00:22:45,230
if you wish.

419
00:22:45,230 --> 00:22:48,700
Also, the Enviro
21 has tentacles

420
00:22:48,700 --> 00:22:52,660
which are connected to
sensors and actuators

421
00:22:52,660 --> 00:22:55,580
and the famous appliances
we are talking about.

422
00:22:55,580 --> 00:22:57,160
And it can control them.

423
00:22:57,160 --> 00:23:01,180
And also, the Handy
has a little infrared

424
00:23:01,180 --> 00:23:03,500
in it in our current vision.

425
00:23:03,500 --> 00:23:08,500
And if you want to bring
this door into your world,

426
00:23:08,500 --> 00:23:10,480
you put an infrared
tag on the door.

427
00:23:10,480 --> 00:23:12,460
And when you point
the handy to it,

428
00:23:12,460 --> 00:23:15,160
the machine identity of the
door comes into your system.

429
00:23:15,160 --> 00:23:18,130
And then you could bring
up a model that says,

430
00:23:18,130 --> 00:23:19,910
these are the people
working behind it,

431
00:23:19,910 --> 00:23:21,190
and this is what they're
supposed to be doing.

432
00:23:21,190 --> 00:23:23,110
And you could go down
the corridor and check

433
00:23:23,110 --> 00:23:26,230
what's behind every
door, at least virtually.

434
00:23:26,230 --> 00:23:31,190
So, there are interesting
possibilities with Oxygen.

435
00:23:31,190 --> 00:23:36,350
The third technology
is N21, Network 21,

436
00:23:36,350 --> 00:23:39,100
which is what we have
to add to the internet

437
00:23:39,100 --> 00:23:43,870
to make possible the
incorporation of these highly

438
00:23:43,870 --> 00:23:48,190
mobile nodes called people,
because the internet goes

439
00:23:48,190 --> 00:23:49,660
to fixed addresses.

440
00:23:49,660 --> 00:23:54,300
And we want to go to resolve
according to intention.

441
00:23:54,300 --> 00:23:57,720
It also makes possible
networks that rise and collapse

442
00:23:57,720 --> 00:23:59,820
as people get together
for a meeting,

443
00:23:59,820 --> 00:24:02,280
makes possible networks
of unknown elements,

444
00:24:02,280 --> 00:24:04,500
fuzzy networks with
self-adaptation,

445
00:24:04,500 --> 00:24:08,100
and we'll hear more
about this later on.

446
00:24:08,100 --> 00:24:13,170
These are the three hardware
technologies of Oxygen,

447
00:24:13,170 --> 00:24:16,000
with a lot of software
in them, of course.

448
00:24:16,000 --> 00:24:19,530
The fourth technology
may surprise you

449
00:24:19,530 --> 00:24:23,920
because it is an inner
hardcore technology.

450
00:24:23,920 --> 00:24:28,510
It is built-in
speech understanding.

451
00:24:28,510 --> 00:24:32,720
We really want our
machines to be easy to use.

452
00:24:32,720 --> 00:24:34,730
We have to make them
interact with us naturally.

453
00:24:34,730 --> 00:24:39,510
None of us is born with keyboard
sockets or mouse sockets

454
00:24:39,510 --> 00:24:41,180
on our belly.

455
00:24:41,180 --> 00:24:44,860
We're born with mouths and ears,
and that's what we should use.

456
00:24:44,860 --> 00:24:48,940
Dr. Zeus technologies are
fantastic at the narrow domain

457
00:24:48,940 --> 00:24:51,370
level, as you heard
from Bill Gates.

458
00:24:51,370 --> 00:24:53,660
What we are doing now
is stitching together

459
00:24:53,660 --> 00:24:56,650
narrow domains to create
a broader listening

460
00:24:56,650 --> 00:25:00,130
domain with some human
help, so that you

461
00:25:00,130 --> 00:25:03,460
can help us navigate through
these various narrow places

462
00:25:03,460 --> 00:25:05,380
and still talk to it.

463
00:25:05,380 --> 00:25:09,190
The Handy 21 has no keyboard.

464
00:25:09,190 --> 00:25:12,020


465
00:25:12,020 --> 00:25:16,790
The fifth Oxygen technology
is individualized

466
00:25:16,790 --> 00:25:18,470
knowledge access.

467
00:25:18,470 --> 00:25:20,690
By this, I mean
that people should

468
00:25:20,690 --> 00:25:24,260
be able to access knowledge
in ways familiar to them.

469
00:25:24,260 --> 00:25:27,080
And we're talking about
low-hanging fruit here.

470
00:25:27,080 --> 00:25:29,215
We're not talking about
highfalutin concepts.

471
00:25:29,215 --> 00:25:30,590
You should be able
to say, get me

472
00:25:30,590 --> 00:25:33,110
that red book next to the blue
one, or next to the document

473
00:25:33,110 --> 00:25:35,360
that came from Paris yesterday.

474
00:25:35,360 --> 00:25:37,940
And then if you run
out from your own store

475
00:25:37,940 --> 00:25:42,290
of goods and knowledge, you go
to the store of your friends.

476
00:25:42,290 --> 00:25:45,880
And you say, assuming
they permit you,

477
00:25:45,880 --> 00:25:47,650
do you know the answer
to this question?

478
00:25:47,650 --> 00:25:49,670
Can I access this from you?

479
00:25:49,670 --> 00:25:52,990
And if that doesn't work,
you can go out to the web

480
00:25:52,990 --> 00:25:55,300
and get the 5,000 hits.

481
00:25:55,300 --> 00:25:59,050
But then in Oxygen, we
triangulate between those three

482
00:25:59,050 --> 00:26:02,050
to try to tailor big
hits with friends

483
00:26:02,050 --> 00:26:04,930
and individual knowledge,
as you will hear Professor

484
00:26:04,930 --> 00:26:07,750
Karger describe later on.

485
00:26:07,750 --> 00:26:12,620
The sixth Oxygen
technology is automation.

486
00:26:12,620 --> 00:26:14,900
Now, this is the
biggest workhorse

487
00:26:14,900 --> 00:26:16,970
of the Information Age.

488
00:26:16,970 --> 00:26:18,800
No revolution is
a true revolution

489
00:26:18,800 --> 00:26:22,220
unless the machines
can do our work.

490
00:26:22,220 --> 00:26:23,900
Take the shovels,
throw them away,

491
00:26:23,900 --> 00:26:25,550
bring in the
electronic bulldozers.

492
00:26:25,550 --> 00:26:29,350
That's what we want
with that technology.

493
00:26:29,350 --> 00:26:32,440
In Oxygen, we offer
scripts for you

494
00:26:32,440 --> 00:26:35,060
to automate what
you're going to do.

495
00:26:35,060 --> 00:26:38,960
And you can set up and
control procedures that will

496
00:26:38,960 --> 00:26:40,490
take place in a given time.

497
00:26:40,490 --> 00:26:42,260
Or when something
arrives you can

498
00:26:42,260 --> 00:26:45,230
encapsulate physical objects,
deal with them together

499
00:26:45,230 --> 00:26:48,410
with informational objects, and
create a world around you that

500
00:26:48,410 --> 00:26:53,870
caters to you by offloading
work from your head and eyes

501
00:26:53,870 --> 00:26:55,990
to the machines.

502
00:26:55,990 --> 00:27:00,640
The seventh Oxygen
technology is collaboration.

503
00:27:00,640 --> 00:27:02,530
And this you can
think of as a method

504
00:27:02,530 --> 00:27:06,310
of keeping track of meetings,
a method of keeping track

505
00:27:06,310 --> 00:27:11,170
of who said what, open
issues, and a way of zeroing

506
00:27:11,170 --> 00:27:15,510
in from human summaries, which
may be brief, to fragments

507
00:27:15,510 --> 00:27:18,480
of speech that took place
in the meeting, all the way

508
00:27:18,480 --> 00:27:21,990
to original work in the
meeting that will remind you

509
00:27:21,990 --> 00:27:23,940
if you come back
later of what you did.

510
00:27:23,940 --> 00:27:27,220


511
00:27:27,220 --> 00:27:31,740
The last, eighth Oxygen
technology is customization.

512
00:27:31,740 --> 00:27:34,020
One of the most valuable
resources in the world

513
00:27:34,020 --> 00:27:37,230
is human beings,
and Oxygen enables

514
00:27:37,230 --> 00:27:39,960
us to customize to
individual needs.

515
00:27:39,960 --> 00:27:43,170
Besides doing this in automation
and all the other technologies

516
00:27:43,170 --> 00:27:47,010
I mentioned earlier, it does
this with all the software.

517
00:27:47,010 --> 00:27:51,490
In Oxygen, there is no
shrinkwrap software.

518
00:27:51,490 --> 00:27:54,400
Everything is downloaded,
and everything

519
00:27:54,400 --> 00:27:57,820
is adjusted according to
whether a user asked for it,

520
00:27:57,820 --> 00:27:59,470
or there was an
error, or there was

521
00:27:59,470 --> 00:28:01,780
new software made available.

522
00:28:01,780 --> 00:28:03,350
This, then, ladies
and gentlemen,

523
00:28:03,350 --> 00:28:09,010
is Oxygen, Handy 21s,
Enrivo 21s in the wall,

524
00:28:09,010 --> 00:28:15,550
Network 21, built-in speech,
individualized knowledge

525
00:28:15,550 --> 00:28:19,810
access, automation,
collaboration, customization.

526
00:28:19,810 --> 00:28:22,600
Now, these eight are not
operating individually,

527
00:28:22,600 --> 00:28:24,500
they're operating together.

528
00:28:24,500 --> 00:28:26,890
And that's where
their power lies.

529
00:28:26,890 --> 00:28:30,530
And now, I'd to
make a bigger claim.

530
00:28:30,530 --> 00:28:34,220
When we envisioned that person
with the rings on his finger,

531
00:28:34,220 --> 00:28:36,440
trying to drive from
Boston to New York,

532
00:28:36,440 --> 00:28:38,350
we were wishing that
we could give him

533
00:28:38,350 --> 00:28:42,220
a gas pedal, a steering
wheel, and brakes

534
00:28:42,220 --> 00:28:44,260
to make it easier to drive.

535
00:28:44,260 --> 00:28:49,270
My claim is that the five
software technologies

536
00:28:49,270 --> 00:28:52,540
of speech, individualized
knowledge access, automation,

537
00:28:52,540 --> 00:28:56,470
collaboration, and
customization are essentially

538
00:28:56,470 --> 00:29:00,070
the new kids on the block
of the Information Age,

539
00:29:00,070 --> 00:29:01,930
that of the thousands
and millions of things

540
00:29:01,930 --> 00:29:06,220
we read that will be
possible, a great deal of them

541
00:29:06,220 --> 00:29:08,510
can be done with these five.

542
00:29:08,510 --> 00:29:12,550
I believe these are pretty close
to the gas pedal, the brakes,

543
00:29:12,550 --> 00:29:16,550
and the steering wheel, at least
for the next couple of decades,

544
00:29:16,550 --> 00:29:18,710
or maybe earlier.

545
00:29:18,710 --> 00:29:22,300
And if that is true, then
companies and individuals

546
00:29:22,300 --> 00:29:25,060
will do well to ask
in their companies

547
00:29:25,060 --> 00:29:27,250
and in their
activities, how can they

548
00:29:27,250 --> 00:29:32,470
maximize their own activities
under these five elements?

549
00:29:32,470 --> 00:29:35,290
Imagine a health
application under Oxygen.

550
00:29:35,290 --> 00:29:37,210
What's the individualized
knowledge access?

551
00:29:37,210 --> 00:29:39,760
Well, it's the doctor's
records, the patient records,

552
00:29:39,760 --> 00:29:40,630
it's Medline.

553
00:29:40,630 --> 00:29:44,570
All this has to be put
together, triangulated,

554
00:29:44,570 --> 00:29:45,998
and help the doctors.

555
00:29:45,998 --> 00:29:48,290
And then you can imagine the
collaboration when doctors

556
00:29:48,290 --> 00:29:50,330
get together, what
they will put on that,

557
00:29:50,330 --> 00:29:52,760
and the customization
for individual doctors

558
00:29:52,760 --> 00:29:56,450
and the automation of routine
tasks for x-rays, et cetera.

559
00:29:56,450 --> 00:29:59,930
So, applications will take
their color from Oxygen

560
00:29:59,930 --> 00:30:06,020
as applications have in the
past, if our dreams come true.

561
00:30:06,020 --> 00:30:10,530
Now, I would like to close,
ladies and gentlemen,

562
00:30:10,530 --> 00:30:15,510
by observing that both my
colleagues and I have been

563
00:30:15,510 --> 00:30:21,570
immensely grateful to work
on this, the information

564
00:30:21,570 --> 00:30:25,530
revolution, for the
last many, many years.

565
00:30:25,530 --> 00:30:27,360
And we're very happy
and excited that we're

566
00:30:27,360 --> 00:30:29,910
working on it right now,
at this very moment,

567
00:30:29,910 --> 00:30:33,330
at this very
junction and change.

568
00:30:33,330 --> 00:30:35,910
The vision of doing
more by doing less

569
00:30:35,910 --> 00:30:37,890
on tomorrow's
information marketplace,

570
00:30:37,890 --> 00:30:43,110
with Oxygen as our first
step, is where I see us today.

571
00:30:43,110 --> 00:30:45,060
And since we're talking
about revolutions,

572
00:30:45,060 --> 00:30:48,930
I'd like to close with an
observation and a wish.

573
00:30:48,930 --> 00:30:51,710


574
00:30:51,710 --> 00:30:56,490
All three revolutions
are based on things.

575
00:30:56,490 --> 00:30:59,580
The Agrarian Revolution
was based on the plow.

576
00:30:59,580 --> 00:31:02,580
The Information Revolution,
which is now happening,

577
00:31:02,580 --> 00:31:04,080
is based on the computer.

578
00:31:04,080 --> 00:31:06,900
And the one in the middle,
the Industrial Revolution,

579
00:31:06,900 --> 00:31:11,180
was based on the motor, the
chemicals, and so forth.

580
00:31:11,180 --> 00:31:14,330
Well, maybe the
time has come for us

581
00:31:14,330 --> 00:31:18,710
to start thinking of a fourth
revolution, one that is not

582
00:31:18,710 --> 00:31:22,100
based on things,
but one that tries

583
00:31:22,100 --> 00:31:27,540
to understand the most precious
resource on our planet,

584
00:31:27,540 --> 00:31:28,810
ourselves.

585
00:31:28,810 --> 00:31:29,310
Thank you.

586
00:31:29,310 --> 00:31:47,480


587
00:31:47,480 --> 00:31:50,435
And now that the chef has
presented us with the menu,

588
00:31:50,435 --> 00:31:54,560
we will now begin with a
series of short, nourishing

589
00:31:54,560 --> 00:31:58,010
presentations from the
researchers at the Laboratory

590
00:31:58,010 --> 00:31:59,750
for Computer Science.

591
00:31:59,750 --> 00:32:02,150
Now, this promises
to be a treat for us,

592
00:32:02,150 --> 00:32:04,130
a series of treats for
us, but I assure you

593
00:32:04,130 --> 00:32:07,490
it is agonizing for them.

594
00:32:07,490 --> 00:32:12,740
Imagine summarizing your
life's work in 13 minutes.

595
00:32:12,740 --> 00:32:16,820
And by the way, you
have agreed to summarize

596
00:32:16,820 --> 00:32:18,635
your life's work in 13 minutes.

597
00:32:18,635 --> 00:32:21,470


598
00:32:21,470 --> 00:32:24,380
Now, as we just heard,
at the heart of Oxygen

599
00:32:24,380 --> 00:32:29,480
are the hardware devices,
Handy 21 and Enviro 21.

600
00:32:29,480 --> 00:32:31,010
And we're about to
hear a novel way

601
00:32:31,010 --> 00:32:33,920
to get more out
of hardware, as we

602
00:32:33,920 --> 00:32:36,560
hear from Anant
Agarwal, who will talk

603
00:32:36,560 --> 00:32:39,290
to us about raw computation.

604
00:32:39,290 --> 00:32:48,690


605
00:32:48,690 --> 00:32:49,350
Thanks, Bob.

606
00:32:49,350 --> 00:32:51,880


607
00:32:51,880 --> 00:32:56,950
Oxygen has as its foundations
the premise that computation

608
00:32:56,950 --> 00:32:58,800
will become free.

609
00:32:58,800 --> 00:33:02,600
So in this talk, I will review
some technological trends

610
00:33:02,600 --> 00:33:09,610
that justify this belief and
discuss how we're rethinking

611
00:33:09,610 --> 00:33:13,720
computer architecture in
light of these technological

612
00:33:13,720 --> 00:33:15,710
advancements.

613
00:33:15,710 --> 00:33:20,680
So, as a quick review,
let's go back to 1987.

614
00:33:20,680 --> 00:33:26,710
And what you see here is roughly
1 square centimeter of silicon.

615
00:33:26,710 --> 00:33:29,980
And a little over
10 years ago, one

616
00:33:29,980 --> 00:33:36,370
could fit in approximately a
10 or 20 MIPS CPU, [? a la ?]

617
00:33:36,370 --> 00:33:39,990
the MIPS in that
small amount of area.

618
00:33:39,990 --> 00:33:47,330
And then you go forward 10
years and that same chip in 1997

619
00:33:47,330 --> 00:33:50,990
would have fit approximately
in that little corner

620
00:33:50,990 --> 00:33:55,470
of the same
1-square-centimeter sized chip.

621
00:33:55,470 --> 00:33:57,930
So let's continue to go forward.

622
00:33:57,930 --> 00:34:01,680
And 10 years from
now, we are going

623
00:34:01,680 --> 00:34:05,130
to have chips with billions
of transistors in them

624
00:34:05,130 --> 00:34:08,280
and the teeny
little MIPS chip is

625
00:34:08,280 --> 00:34:12,929
barely noticeable in the
bottom left-hand corner.

626
00:34:12,929 --> 00:34:15,710
So, we have an amazing
opportunity before us

627
00:34:15,710 --> 00:34:19,400
where we are going to be
getting an ungodly number

628
00:34:19,400 --> 00:34:24,090
of transistors in a square
centimeter of silicon.

629
00:34:24,090 --> 00:34:27,900
So we can fritter
away this opportunity.

630
00:34:27,900 --> 00:34:32,130
And one way to do that is
to continue living life

631
00:34:32,130 --> 00:34:37,590
as we have been in the past and
look to the big or small eyes

632
00:34:37,590 --> 00:34:40,590
of this world to continue
advancing our architecture

633
00:34:40,590 --> 00:34:44,610
and technologies, and
just do more of the same,

634
00:34:44,610 --> 00:34:48,300
build microprocessors that
are bigger of the same kind,

635
00:34:48,300 --> 00:34:51,570
and build chips microprocessors
with huge amounts of register

636
00:34:51,570 --> 00:34:54,760
files, huge amounts of
control and complexity.

637
00:34:54,760 --> 00:34:57,210
Well, the problem is that
this kind of architecture

638
00:34:57,210 --> 00:34:59,340
simply does not scale.

639
00:34:59,340 --> 00:35:04,040
And the reason is that
as technology advances,

640
00:35:04,040 --> 00:35:08,450
and certainly as we get
into the 2007s and on,

641
00:35:08,450 --> 00:35:10,760
we are going to find
that the transistors have

642
00:35:10,760 --> 00:35:12,350
become incredibly fast.

643
00:35:12,350 --> 00:35:14,400
They're much faster
than ever before.

644
00:35:14,400 --> 00:35:17,870
However, wires
simply do not scale.

645
00:35:17,870 --> 00:35:21,020
And this kind of architecture
requires long wires

646
00:35:21,020 --> 00:35:23,690
in the chip to manage the
fairly complicated hardware

647
00:35:23,690 --> 00:35:24,470
and control.

648
00:35:24,470 --> 00:35:28,820
And this class of architecture
simply doesn't scale.

649
00:35:28,820 --> 00:35:33,170
But in many senses,
we are already there.

650
00:35:33,170 --> 00:35:37,370
We've already hit a
complexity and a speed issue

651
00:35:37,370 --> 00:35:38,810
with the existing architectures.

652
00:35:38,810 --> 00:35:41,888
And if you look today at
where computing has gone,

653
00:35:41,888 --> 00:35:42,680
what do you notice?

654
00:35:42,680 --> 00:35:46,610
When you buy a PC, just
so you can call it a PC,

655
00:35:46,610 --> 00:35:49,320
you do have an X86 in there.

656
00:35:49,320 --> 00:35:51,890
That's what I guess
goes into a PC.

657
00:35:51,890 --> 00:35:55,090
But when you want to do
anything real, what do you do?

658
00:35:55,090 --> 00:35:57,550
You buy a graphics
card, a Photoshop card,

659
00:35:57,550 --> 00:36:01,060
you buy a math card, an
encryption card, a radio card.

660
00:36:01,060 --> 00:36:02,980
I just buy big cases
for my computer

661
00:36:02,980 --> 00:36:05,562
simply because I can
plug in a ton of cards.

662
00:36:05,562 --> 00:36:07,270
And you get a whole
bunch of these cards,

663
00:36:07,270 --> 00:36:10,570
and real performance
and streaming multimedia

664
00:36:10,570 --> 00:36:14,730
applications comes from these
specialized hardware units.

665
00:36:14,730 --> 00:36:17,370
So how do these give
you performance?

666
00:36:17,370 --> 00:36:21,270
These give you performance
because some poor souls sitting

667
00:36:21,270 --> 00:36:23,100
out somewhere have
very carefully

668
00:36:23,100 --> 00:36:26,920
handcrafted the wiring
through the application.

669
00:36:26,920 --> 00:36:30,220
They have customized the
wiring to applications

670
00:36:30,220 --> 00:36:32,800
such as encryption,
radio, and so on.

671
00:36:32,800 --> 00:36:36,190
They've hand-hacked the wires
on the board or on chips

672
00:36:36,190 --> 00:36:40,300
so that the application has been
very carefully choreographed

673
00:36:40,300 --> 00:36:43,210
on the VLSI chip so that
all the wires take data

674
00:36:43,210 --> 00:36:45,490
right from where it should
be to the right place,

675
00:36:45,490 --> 00:36:47,030
at exactly the right time.

676
00:36:47,030 --> 00:36:49,850
But a huge amount of effort
goes into that process.

677
00:36:49,850 --> 00:36:53,030
And that's how you
get performance.

678
00:36:53,030 --> 00:36:55,040
So, what do we do?

679
00:36:55,040 --> 00:36:57,010
This is a problem.

680
00:36:57,010 --> 00:36:58,690
You want to make
people's lives easy,

681
00:36:58,690 --> 00:37:04,870
but it certainly gives us pause
in terms of how we can actually

682
00:37:04,870 --> 00:37:05,860
achieve this.

683
00:37:05,860 --> 00:37:09,965
So the key insight here is
to observe the following.

684
00:37:09,965 --> 00:37:12,190
In about 10 years,
we're going to have

685
00:37:12,190 --> 00:37:16,545
so many gates on our chip,
billions of gates on our chip.

686
00:37:16,545 --> 00:37:17,920
We're going to
get to a situation

687
00:37:17,920 --> 00:37:21,340
where we're going to have
gates crawling on it for years,

688
00:37:21,340 --> 00:37:25,240
tons of gates on our chip
with ungodly amounts of logic.

689
00:37:25,240 --> 00:37:28,000
And what we're going
to do is we are

690
00:37:28,000 --> 00:37:32,140
going to solve our computing
problem by throwing gates

691
00:37:32,140 --> 00:37:33,910
at the problem.

692
00:37:33,910 --> 00:37:36,790
And what you do
here is, in the past

693
00:37:36,790 --> 00:37:40,270
while people hand-compiled these
applications by customizing

694
00:37:40,270 --> 00:37:42,640
the wiring of boards,
what we are going to do

695
00:37:42,640 --> 00:37:48,010
is take the application
and build a compiler,

696
00:37:48,010 --> 00:37:50,950
and build a compiler that
uses these excess gates

697
00:37:50,950 --> 00:37:52,540
to choreograph the wires.

698
00:37:52,540 --> 00:37:55,780
Again, the key is that
instead of having users spend

699
00:37:55,780 --> 00:37:58,295
huge amounts of time
orchestrating the wires

700
00:37:58,295 --> 00:38:00,670
and carefully laying out the
wires to get the performance

701
00:38:00,670 --> 00:38:02,650
from applications,
what we're going to do

702
00:38:02,650 --> 00:38:06,190
is build an architecture
and a compiler that

703
00:38:06,190 --> 00:38:09,310
uses these excess
amounts of gates

704
00:38:09,310 --> 00:38:11,290
to allow us to
configure the wires,

705
00:38:11,290 --> 00:38:15,220
thereby directly combining
applications into the logic.

706
00:38:15,220 --> 00:38:17,920
So what we get
here is a trashing

707
00:38:17,920 --> 00:38:20,230
of the traditional
instruction set.

708
00:38:20,230 --> 00:38:24,880
Raw is an architectural
approach that trashes existing

709
00:38:24,880 --> 00:38:26,230
instructions and architectures.

710
00:38:26,230 --> 00:38:30,010
And instead, what it does
is exposes the raw hardware

711
00:38:30,010 --> 00:38:32,010
to the compiler.

712
00:38:32,010 --> 00:38:34,530
So you get the raw hardware
on a chip completely exposed

713
00:38:34,530 --> 00:38:35,950
to the compiler.

714
00:38:35,950 --> 00:38:39,270
And raw hardware means
the wires and the compiler

715
00:38:39,270 --> 00:38:41,800
on the chip completely
exposed to the compiler.

716
00:38:41,800 --> 00:38:45,510
And what we're going to do
is use these excess gates.

717
00:38:45,510 --> 00:38:48,000
We're going to have a compiler
that causes those three

718
00:38:48,000 --> 00:38:51,240
gates to orchestrate
the wires on the chip

719
00:38:51,240 --> 00:38:54,000
so that we get the
effect of customized

720
00:38:54,000 --> 00:38:56,110
wiring to our applications.

721
00:38:56,110 --> 00:39:00,120
So what this does is lead to
a new model of computation.

722
00:39:00,120 --> 00:39:01,800
I love this new
model of computation

723
00:39:01,800 --> 00:39:05,260
because it is an
exercise in excess.

724
00:39:05,260 --> 00:39:06,870
You're going to
have lots of gates.

725
00:39:06,870 --> 00:39:08,703
So we're going to throw
gates at the problem

726
00:39:08,703 --> 00:39:11,140
and combine applications
directly into logic.

727
00:39:11,140 --> 00:39:12,990
And the key idea
here is that we are

728
00:39:12,990 --> 00:39:16,230
going to program the
wires, not a logic unit.

729
00:39:16,230 --> 00:39:19,770
This is a radical departure
from existing architectures.

730
00:39:19,770 --> 00:39:22,740
In existing architectures,
existing instruction sets,

731
00:39:22,740 --> 00:39:24,930
what is exposed to
the software system?

732
00:39:24,930 --> 00:39:27,900
What is exposed are logic
operations such as "and,"

733
00:39:27,900 --> 00:39:30,647
"subtract," "add," and so on.

734
00:39:30,647 --> 00:39:32,730
What this is going to do,
what Raw is going to do,

735
00:39:32,730 --> 00:39:35,800
is not expose the logic
or the logic operations

736
00:39:35,800 --> 00:39:38,550
to the compiler, but rather,
the most valuable resource,

737
00:39:38,550 --> 00:39:40,260
the most critical
resource, the wires.

738
00:39:40,260 --> 00:39:43,060


739
00:39:43,060 --> 00:39:45,190
So what does it mean
to program a wire?

740
00:39:45,190 --> 00:39:47,920
What does it mean to expose
the wires to the compiler

741
00:39:47,920 --> 00:39:49,850
and let the compiler
program them?

742
00:39:49,850 --> 00:39:52,150
So this is a
playground, if you will.

743
00:39:52,150 --> 00:39:56,560
And this is our most
critical resource, a wire.

744
00:39:56,560 --> 00:40:00,370
It is also the real problem in
terms of getting performance

745
00:40:00,370 --> 00:40:02,110
out of complicated chips.

746
00:40:02,110 --> 00:40:04,600
So what we're going to
do is build a compiler

747
00:40:04,600 --> 00:40:07,420
that is going to orchestrate
these wires in a way that

748
00:40:07,420 --> 00:40:10,420
customizes them to
the application.

749
00:40:10,420 --> 00:40:13,170
So the first step that
we're going to follow

750
00:40:13,170 --> 00:40:16,950
is that the compiler is going
to pipeline the wires so

751
00:40:16,950 --> 00:40:21,930
that long wires don't
incur long delays.

752
00:40:21,930 --> 00:40:24,600
Pipeline the wires,
use that excess logic

753
00:40:24,600 --> 00:40:27,330
to introduce registers to
pipeline the wires, so we

754
00:40:27,330 --> 00:40:31,290
can run at much higher clock
speeds than we could imagine,

755
00:40:31,290 --> 00:40:35,190
say around 10 to 15
gigahertz by the year 2010.

756
00:40:35,190 --> 00:40:38,760
The next step, why is it a
really valuable resource?

757
00:40:38,760 --> 00:40:41,610
So if you do not want to send
just one signal down a wire

758
00:40:41,610 --> 00:40:43,770
and bind it to that
signal for all time,

759
00:40:43,770 --> 00:40:48,220
rather multiplex the wires
among a large number of signals.

760
00:40:48,220 --> 00:40:50,400
And again, if you'd
like to use a compiler

761
00:40:50,400 --> 00:40:52,650
to introduce
multiplexers, again,

762
00:40:52,650 --> 00:40:55,050
use the excess
gates to orchestrate

763
00:40:55,050 --> 00:40:56,680
our most valuable resource.

764
00:40:56,680 --> 00:40:59,325
So use the gates to
build multiplexers

765
00:40:59,325 --> 00:41:02,550
to multiplex the wires among a
much larger number of signals

766
00:41:02,550 --> 00:41:06,480
and improve the
utilization of the wires.

767
00:41:06,480 --> 00:41:08,520
And finally, what
we would like to do

768
00:41:08,520 --> 00:41:12,810
is use a compiler to
customize and schedule

769
00:41:12,810 --> 00:41:15,660
and route the signals on
these wires at the right time

770
00:41:15,660 --> 00:41:18,450
so that the set of wires
takes on the identity

771
00:41:18,450 --> 00:41:20,110
of the application.

772
00:41:20,110 --> 00:41:22,410
So this is what it
means to program a wire.

773
00:41:22,410 --> 00:41:26,910
And we in fact like to call
these wires of the 21st century

774
00:41:26,910 --> 00:41:27,765
"soft wires."

775
00:41:27,765 --> 00:41:30,300


776
00:41:30,300 --> 00:41:33,120
So what we like to do with
this class of approach

777
00:41:33,120 --> 00:41:35,610
is enable the Handy 21.

778
00:41:35,610 --> 00:41:38,580
So we'd like to have a
single device into which we

779
00:41:38,580 --> 00:41:41,670
can directly map applications
using our compiler.

780
00:41:41,670 --> 00:41:47,580
So encryption, or speech
recognition of various forms,

781
00:41:47,580 --> 00:41:50,190
cell phones directly map
the appropriate software

782
00:41:50,190 --> 00:41:52,890
into Raw hardware, like games.

783
00:41:52,890 --> 00:41:55,740
And in collaboration with
the networking folks,

784
00:41:55,740 --> 00:41:56,910
we enable the following.

785
00:41:56,910 --> 00:42:00,060
We have a single device
that I can use and talk

786
00:42:00,060 --> 00:42:02,790
to and say, hey, turn
yourself into a cell phone.

787
00:42:02,790 --> 00:42:06,180
The device will go and locate
the appropriate configuration

788
00:42:06,180 --> 00:42:09,360
software, configuration
bits, download it,

789
00:42:09,360 --> 00:42:11,730
configure the wires
after our chip,

790
00:42:11,730 --> 00:42:14,610
and give it the
personality of a cell phone

791
00:42:14,610 --> 00:42:16,530
and then I talk to it.

792
00:42:16,530 --> 00:42:18,440
So what does that mean to me?

793
00:42:18,440 --> 00:42:20,090
That's pretty fantastic.

794
00:42:20,090 --> 00:42:26,030
Today, I carry a beeper,
I carry a cell phone,

795
00:42:26,030 --> 00:42:30,260
and I carry a Palm Pilot.

796
00:42:30,260 --> 00:42:32,860
So I'm going to take
all of these things

797
00:42:32,860 --> 00:42:34,730
and throw them away.

798
00:42:34,730 --> 00:42:38,810
I'm going to have a single
device, the Handy 21,

799
00:42:38,810 --> 00:42:41,030
and just carry
one little device.

800
00:42:41,030 --> 00:42:43,490
And just by talking
to it, I get it

801
00:42:43,490 --> 00:42:46,190
to download the appropriate
configuration bits

802
00:42:46,190 --> 00:42:49,130
so that it takes the personality
of pretty much any device

803
00:42:49,130 --> 00:42:51,660
that you want.

804
00:42:51,660 --> 00:42:55,590
So, the one other step,
and what I'd like to do.

805
00:42:55,590 --> 00:42:58,020
This is the little
device I'm talking about,

806
00:42:58,020 --> 00:43:00,930
and all it would
contain, in terms

807
00:43:00,930 --> 00:43:04,440
of digital logic, digital
hardware, a single Raw chip

808
00:43:04,440 --> 00:43:05,280
inside.

809
00:43:05,280 --> 00:43:07,170
And in addition to
that, it has a bunch

810
00:43:07,170 --> 00:43:08,620
of perceptual interfaces.

811
00:43:08,620 --> 00:43:13,600
It has cameras and perhaps
a little 3x3 DFT display.

812
00:43:13,600 --> 00:43:16,620
It has other perceptual
interfaces, like speech,

813
00:43:16,620 --> 00:43:18,180
both speakers and microphones.

814
00:43:18,180 --> 00:43:21,900
Of course it needs an antenna
and an A-to-D converter.

815
00:43:21,900 --> 00:43:25,530
And the A-to-D will probably
be integrated on the same chip

816
00:43:25,530 --> 00:43:29,460
itself, so that essentially
all the function that you

817
00:43:29,460 --> 00:43:32,130
buy custom hardware for
today will be accomplished

818
00:43:32,130 --> 00:43:34,350
within the Raw chip by
customizing applications

819
00:43:34,350 --> 00:43:38,280
directly into the
wires on the chip.

820
00:43:38,280 --> 00:43:40,560
By building a
device of this sort,

821
00:43:40,560 --> 00:43:43,750
we'd also like to paper
our walls with computation.

822
00:43:43,750 --> 00:43:46,320
We need to have computation
all over the map.

823
00:43:46,320 --> 00:43:47,767
You look and guess what?

824
00:43:47,767 --> 00:43:48,600
There's computation.

825
00:43:48,600 --> 00:43:50,250
It's going to be everywhere,
it's going to be free,

826
00:43:50,250 --> 00:43:52,980
and it's going to be readily
available like oxygen. You

827
00:43:52,980 --> 00:43:55,120
will be wondering how
you lived without it.

828
00:43:55,120 --> 00:43:58,560
So the key idea is that in
the walls of a new building,

829
00:43:58,560 --> 00:44:01,650
we have an array of Raw
like chips on the wall,

830
00:44:01,650 --> 00:44:04,980
so that these can give you a
massive amount of computation.

831
00:44:04,980 --> 00:44:08,130
As Mike pointed out, it's
the wall socket analogy

832
00:44:08,130 --> 00:44:11,520
or in contrast to
handheld little batteries.

833
00:44:11,520 --> 00:44:14,490
So I call these computational
fabrics the sheetrock

834
00:44:14,490 --> 00:44:16,710
of the 21st century.

835
00:44:16,710 --> 00:44:18,960
So the Raw architecture is
actually very, very simple.

836
00:44:18,960 --> 00:44:20,585
The challenge is in
the compiler system

837
00:44:20,585 --> 00:44:22,230
to orchestrate the wires.

838
00:44:22,230 --> 00:44:26,310
So the Raw chip itself
is an array of tiles,

839
00:44:26,310 --> 00:44:29,490
very simple, replicated,
simple to design and build.

840
00:44:29,490 --> 00:44:34,630
And the key in the tile
is the switching fabric.

841
00:44:34,630 --> 00:44:36,390
Each tile has a switch.

842
00:44:36,390 --> 00:44:39,760
What the switch does is
orchestrate the wires.

843
00:44:39,760 --> 00:44:41,760
The bunch of excess
gates that I talked about

844
00:44:41,760 --> 00:44:43,920
are devoted to the
switch and the compiler

845
00:44:43,920 --> 00:44:45,762
programs the switch.

846
00:44:45,762 --> 00:44:47,970
This is not a multiprocessor
in the traditional sense

847
00:44:47,970 --> 00:44:48,780
of the word.

848
00:44:48,780 --> 00:44:50,250
The bunch of tiles,
each of which

849
00:44:50,250 --> 00:44:53,940
has a memory which can be
programmed by the compiler,

850
00:44:53,940 --> 00:44:57,990
tells the switch exactly what
wire or what sort of wires

851
00:44:57,990 --> 00:45:01,860
to connect to what point on
every cycle of its operation.

852
00:45:01,860 --> 00:45:03,960
So the compiler, by
introducing a program

853
00:45:03,960 --> 00:45:06,300
into each of these
switching blocks,

854
00:45:06,300 --> 00:45:10,260
choreographs how data moves
around the entire chip.

855
00:45:10,260 --> 00:45:13,230
It's a wire-centric
model of computation.

856
00:45:13,230 --> 00:45:15,720
And, of course, you
need a bunch of logic.

857
00:45:15,720 --> 00:45:17,470
So there'll be a bunch
of logic out there.

858
00:45:17,470 --> 00:45:19,637
It really doesn't matter
exactly what it looks like.

859
00:45:19,637 --> 00:45:21,900
It's incredibly simple.

860
00:45:21,900 --> 00:45:24,690
And the other value of Raw
architecture of the sort

861
00:45:24,690 --> 00:45:27,870
is you can now bring
massive streams of data

862
00:45:27,870 --> 00:45:31,050
directly to where
computation happens upon it.

863
00:45:31,050 --> 00:45:33,450
And the chip that
you're building,

864
00:45:33,450 --> 00:45:36,420
by the end of next year,
in collaboration with IBM,

865
00:45:36,420 --> 00:45:40,380
is going to have on the
order of 1,500 [? I/Os. ?]

866
00:45:40,380 --> 00:45:43,230
Get massive amounts of data
right for the computation,

867
00:45:43,230 --> 00:45:45,480
get the wires orchestrated
to stream the data,

868
00:45:45,480 --> 00:45:50,950
and get much better performance
and power than we can today.

869
00:45:50,950 --> 00:45:55,800
One can view this, if you like,
in some respects, like an FPGA,

870
00:45:55,800 --> 00:45:59,250
but it's quite different
from FPGAs in that it's

871
00:45:59,250 --> 00:46:00,630
a much coarser grain.

872
00:46:00,630 --> 00:46:03,540
It has built-in support for
multiplexing all the wires.

873
00:46:03,540 --> 00:46:06,430
FPGA is [? bind ?]
hardware for one time.

874
00:46:06,430 --> 00:46:08,910
It takes a much longer time
to reprogram the wires.

875
00:46:08,910 --> 00:46:10,950
And I think the
most important part

876
00:46:10,950 --> 00:46:14,130
is that we are looking to
build an automatic compiler,

877
00:46:14,130 --> 00:46:16,572
a compiler that automatically
compiles applications

878
00:46:16,572 --> 00:46:18,030
into the chip, as
opposed to having

879
00:46:18,030 --> 00:46:20,590
users do hardware designs.

880
00:46:20,590 --> 00:46:22,110
So it's one example.

881
00:46:22,110 --> 00:46:25,350
This is not all hot air.

882
00:46:25,350 --> 00:46:28,230
My group, comprising Saman
Amarasinghe and myself

883
00:46:28,230 --> 00:46:31,360
and many of our students, have
combined many applications

884
00:46:31,360 --> 00:46:31,860
to this.

885
00:46:31,860 --> 00:46:36,780
We built a compiler that can
take sequential C applications

886
00:46:36,780 --> 00:46:41,250
and compile it directly into
a simulator of the Raw fabric.

887
00:46:41,250 --> 00:46:44,550
As one example, you
will hear a talk later

888
00:46:44,550 --> 00:46:52,500
on by John Guttag on a software
implementation of FM radio.

889
00:46:52,500 --> 00:46:55,140
So what we did is took
their sequential C code

890
00:46:55,140 --> 00:46:58,110
for the software area that
they've built and automatically

891
00:46:58,110 --> 00:47:00,455
compiled it onto a Raw machine.

892
00:47:00,455 --> 00:47:01,830
So I can kind of
give you a sense

893
00:47:01,830 --> 00:47:04,140
of what kinds of
performance and benefits

894
00:47:04,140 --> 00:47:06,970
are achievable with
something like that.

895
00:47:06,970 --> 00:47:10,090
So we've compiled that and it
has a channel selection module,

896
00:47:10,090 --> 00:47:14,010
the demodulation component, and
a low-pass filter component.

897
00:47:14,010 --> 00:47:19,110
And our projections are,
for the 128 tile Raw chip,

898
00:47:19,110 --> 00:47:22,350
that an automatically compiled
version of the sequential C

899
00:47:22,350 --> 00:47:25,590
application will run on the
order of 10 times faster

900
00:47:25,590 --> 00:47:29,560
than an X10 PC of that day.

901
00:47:29,560 --> 00:47:31,200
We also have some
wizard students

902
00:47:31,200 --> 00:47:35,940
and they hand-hacked
a version of the code

903
00:47:35,940 --> 00:47:38,558
and they got it for
the 10x improvement.

904
00:47:38,558 --> 00:47:40,350
So the real challenge
in their own research

905
00:47:40,350 --> 00:47:43,140
is to build a compiler that
can approach the 100x that you

906
00:47:43,140 --> 00:47:46,443
can get by hand customizing
the wiring of the application.

907
00:47:46,443 --> 00:47:48,360
We're a long way to go,
it's a real challenge.

908
00:47:48,360 --> 00:47:50,190
But the real key
is the compiler.

909
00:47:50,190 --> 00:47:52,980
And what have we
done here or what

910
00:47:52,980 --> 00:47:56,380
are we banking on that would
make our life fairly easy?

911
00:47:56,380 --> 00:47:59,950
The key is the observation
that 10 years from now,

912
00:47:59,950 --> 00:48:02,610
we are going to have an excess
amount of gates on a chip.

913
00:48:02,610 --> 00:48:03,810
The wire to the problem.

914
00:48:03,810 --> 00:48:05,107
Gates are free.

915
00:48:05,107 --> 00:48:07,440
And what we're going to do
is throw gates at the problem

916
00:48:07,440 --> 00:48:08,940
to solve a compilation problem.

917
00:48:08,940 --> 00:48:11,492
If you're not going to try
to be stingy about gates,

918
00:48:11,492 --> 00:48:13,200
we're going to throw
gates at the problem

919
00:48:13,200 --> 00:48:16,990
and use those to make our
compilation job a lot easier.

920
00:48:16,990 --> 00:48:21,990
So finally, to close, let
me talk about futures.

921
00:48:21,990 --> 00:48:25,470
If we can get our way and
build a single Raw chip that

922
00:48:25,470 --> 00:48:27,480
can take the place of
general-purpose computing

923
00:48:27,480 --> 00:48:31,800
chips, such as PC micros,
and at the same time

924
00:48:31,800 --> 00:48:34,650
eliminate specialized hardware
because the compiler combined

925
00:48:34,650 --> 00:48:36,630
these functions by
customizing wires,

926
00:48:36,630 --> 00:48:38,250
then we will have
a universal Raw

927
00:48:38,250 --> 00:48:41,640
chip that can replace both
general-purpose micros

928
00:48:41,640 --> 00:48:42,760
and special hardware.

929
00:48:42,760 --> 00:48:44,430
And so I make a
projection that it's

930
00:48:44,430 --> 00:48:46,675
quite likely that 10
or 15 years from now,

931
00:48:46,675 --> 00:48:48,300
there will be only
three kinds of chips

932
00:48:48,300 --> 00:48:52,170
in the world, DRAM chips,
Raw chips, and, of course,

933
00:48:52,170 --> 00:48:53,650
potato chips.

934
00:48:53,650 --> 00:48:54,150
Thank you.

935
00:48:54,150 --> 00:49:07,410


936
00:49:07,410 --> 00:49:09,840
But of course, we're
going to have to connect

937
00:49:09,840 --> 00:49:11,460
these computers together.

938
00:49:11,460 --> 00:49:13,500
We're going to need
networks, networks that

939
00:49:13,500 --> 00:49:14,790
go way beyond the internet.

940
00:49:14,790 --> 00:49:18,780
And to talk to us about
that is Hari Balakrishnan,

941
00:49:18,780 --> 00:49:23,490
who is the youngest member
of the faculty at LSC.

942
00:49:23,490 --> 00:49:24,970
Take it away.

943
00:49:24,970 --> 00:49:28,035
Thanks a lot, Bob, and good
morning, ladies and gentlemen.

944
00:49:28,035 --> 00:49:29,410
What I'm going to
describe to you

945
00:49:29,410 --> 00:49:33,400
is my vision for N21 Oxygen's
networking component, which

946
00:49:33,400 --> 00:49:35,230
I fondly hope will also
be a core component

947
00:49:35,230 --> 00:49:36,580
in the future internet.

948
00:49:36,580 --> 00:49:39,700
And this is a vision that has
two main themes and components

949
00:49:39,700 --> 00:49:42,220
in it that I'd like you
to take on from the stock.

950
00:49:42,220 --> 00:49:45,410
The first is the notion
of self-organization.

951
00:49:45,410 --> 00:49:48,220
And the second is the
notion of adaptation.

952
00:49:48,220 --> 00:49:50,050
By self-organization,
what I mean

953
00:49:50,050 --> 00:49:53,860
is the ability for our networks
and for our distributed network

954
00:49:53,860 --> 00:49:56,650
applications to organize
themselves and configure

955
00:49:56,650 --> 00:49:59,170
themselves and get
efficient work done

956
00:49:59,170 --> 00:50:01,480
without any manual
intervention whatsoever,

957
00:50:01,480 --> 00:50:03,790
and without any user frustration
in setting things up,

958
00:50:03,790 --> 00:50:05,810
which we're also familiar with.

959
00:50:05,810 --> 00:50:08,780
By adaptation, what I mean is
the ability for our protocols

960
00:50:08,780 --> 00:50:11,210
and applications to adapt to
all the changing conditions,

961
00:50:11,210 --> 00:50:13,580
from levels of congestion,
to errors on networks,

962
00:50:13,580 --> 00:50:17,240
to heterogeneity of
different types of links,

963
00:50:17,240 --> 00:50:20,820
and to transfer data as
efficiently as possible.

964
00:50:20,820 --> 00:50:23,230
And this is what I'm
going to talk about.

965
00:50:23,230 --> 00:50:26,218
So I'm going to start first
by asking you to imagine

966
00:50:26,218 --> 00:50:28,260
a very simple scenario,
not from science fiction,

967
00:50:28,260 --> 00:50:30,510
but this is something
we'd all like to do

968
00:50:30,510 --> 00:50:33,822
today, something which we
all really will find useful,

969
00:50:33,822 --> 00:50:36,030
but something that's really
hard to accomplish today.

970
00:50:36,030 --> 00:50:37,860
And it starts first
with a group of people

971
00:50:37,860 --> 00:50:40,170
who gather together in a
meeting room to collaborate.

972
00:50:40,170 --> 00:50:42,553
This is just a simple meeting
we're all so used to doing.

973
00:50:42,553 --> 00:50:44,970
We have a variety of different
computing and communication

974
00:50:44,970 --> 00:50:48,480
devices, including, of course,
the H21, or the Handy 21

975
00:50:48,480 --> 00:50:50,802
that Michael and Anant
talked to you about.

976
00:50:50,802 --> 00:50:52,260
During the process
of this meeting,

977
00:50:52,260 --> 00:50:54,000
we would like to bring
these devices in and have

978
00:50:54,000 --> 00:50:55,810
them networked and
have an application

979
00:50:55,810 --> 00:50:57,510
to share information
with each other

980
00:50:57,510 --> 00:50:59,160
without any
intervention on our part

981
00:50:59,160 --> 00:51:01,710
and any frustration of our own.

982
00:51:01,710 --> 00:51:04,170
Of course, we also find that
in any environment that we

983
00:51:04,170 --> 00:51:06,305
move in and live in, there
are a number of services

984
00:51:06,305 --> 00:51:08,430
and a number of different
things in the environment

985
00:51:08,430 --> 00:51:11,490
that we have that we would
like to control and like

986
00:51:11,490 --> 00:51:14,182
to gain information access
from, ranging from coffee

987
00:51:14,182 --> 00:51:16,140
machines that would be
networked in the future,

988
00:51:16,140 --> 00:51:21,360
to projection equipment, to copy
machines to temperature sensors

989
00:51:21,360 --> 00:51:25,080
and other types of sensors and
actuators in our environment.

990
00:51:25,080 --> 00:51:27,240
Of course, we also
have the Enviro 21,

991
00:51:27,240 --> 00:51:29,610
which are embedded sensors
and actuators in the wall

992
00:51:29,610 --> 00:51:32,990
that we would like to
control from our network.

993
00:51:32,990 --> 00:51:34,610
In our meeting, we
also have a person

994
00:51:34,610 --> 00:51:36,290
who's been lucky enough
to be at home today,

995
00:51:36,290 --> 00:51:38,300
but he would like to
participate in the meeting.

996
00:51:38,300 --> 00:51:41,780
And he does so by using
his wide area wireless

997
00:51:41,780 --> 00:51:44,700
link from his home to
connect to our corporation.

998
00:51:44,700 --> 00:51:47,300
And we would like to do this
again without any setting up

999
00:51:47,300 --> 00:51:48,120
of our own.

1000
00:51:48,120 --> 00:51:49,670
And again, during the
process of our meeting,

1001
00:51:49,670 --> 00:51:52,045
we find that a colleague of
ours is vacationing somewhere

1002
00:51:52,045 --> 00:51:54,350
in the Mediterranean and we
would like our system to,

1003
00:51:54,350 --> 00:51:57,650
in a secure and private way,
track her and have her join in

1004
00:51:57,650 --> 00:52:00,560
and participate via
satellite connection.

1005
00:52:00,560 --> 00:52:03,763
We'd like all this to
happen automatically.

1006
00:52:03,763 --> 00:52:05,430
During the meeting,
we find that there's

1007
00:52:05,430 --> 00:52:07,458
lots of information,
lots of services

1008
00:52:07,458 --> 00:52:09,000
from the worldwide
web, that we would

1009
00:52:09,000 --> 00:52:11,160
like to garner in
and actually use

1010
00:52:11,160 --> 00:52:12,900
to make our meeting
more productive,

1011
00:52:12,900 --> 00:52:14,850
lots of different types of data.

1012
00:52:14,850 --> 00:52:17,010
What we would like to do
is to have all this happen

1013
00:52:17,010 --> 00:52:19,020
automatically and
efficiently to create

1014
00:52:19,020 --> 00:52:21,570
this notion of collaboration,
of collaborative regions that

1015
00:52:21,570 --> 00:52:25,200
aren't people that automatically
can be spontaneously created

1016
00:52:25,200 --> 00:52:30,163
and when the meeting is done,
be turned on automatically.

1017
00:52:30,163 --> 00:52:31,580
What are some
technical challenges

1018
00:52:31,580 --> 00:52:33,635
in enabling this vision?

1019
00:52:33,635 --> 00:52:34,760
The first is heterogeneity.

1020
00:52:34,760 --> 00:52:36,410
There's a wide
variety and diversity

1021
00:52:36,410 --> 00:52:38,150
of devices, of
types of networks,

1022
00:52:38,150 --> 00:52:39,950
of applications and
data and services,

1023
00:52:39,950 --> 00:52:44,740
that we somehow have to overcome
and have our system handle.

1024
00:52:44,740 --> 00:52:46,490
The second is the large
amount of dynamism

1025
00:52:46,490 --> 00:52:48,920
and spontaneous operation
and mobility that we need.

1026
00:52:48,920 --> 00:52:51,080
We need to have these
networks and applications

1027
00:52:51,080 --> 00:52:53,810
rapidly deployable
and organized without

1028
00:52:53,810 --> 00:52:55,580
any manual intervention.

1029
00:52:55,580 --> 00:52:58,880
And the third is a much
higher and much better

1030
00:52:58,880 --> 00:53:01,220
degree of fault tolerance
and decentralization

1031
00:53:01,220 --> 00:53:02,300
than what we have today.

1032
00:53:02,300 --> 00:53:04,217
We absolutely need our
networks to be far more

1033
00:53:04,217 --> 00:53:06,610
robust than they are today.

1034
00:53:06,610 --> 00:53:10,020
I'm going to describe to you how
the ideas of self-organization

1035
00:53:10,020 --> 00:53:14,340
and the ideas of adaptation
built into Oxygen's protocols

1036
00:53:14,340 --> 00:53:18,068
and middleware will
enable this to happen.

1037
00:53:18,068 --> 00:53:19,610
Before we get into
Oxygen, let's look

1038
00:53:19,610 --> 00:53:20,818
at how current networks work.

1039
00:53:20,818 --> 00:53:24,380
And this is a highly simplified,
high-level view of networks,

1040
00:53:24,380 --> 00:53:27,030
which are typically built
around clients and servers.

1041
00:53:27,030 --> 00:53:29,150
Think of them as web
clients and web servers.

1042
00:53:29,150 --> 00:53:32,570
And packets get routed from
clients to servers and vice

1043
00:53:32,570 --> 00:53:36,247
versa via routers using
IP, the internet protocol.

1044
00:53:36,247 --> 00:53:38,330
One of the most interesting
and important services

1045
00:53:38,330 --> 00:53:39,920
that exists in
the internet today

1046
00:53:39,920 --> 00:53:43,580
is the domain name system, which
is a fairly static mapping that

1047
00:53:43,580 --> 00:53:46,700
converts host names, such
as microsoft.com or cnn.com

1048
00:53:46,700 --> 00:53:49,350
or mit.edu to IP addresses.

1049
00:53:49,350 --> 00:53:50,900
And this is a fairly
static mapping

1050
00:53:50,900 --> 00:53:52,970
that's not dynamic,
doesn't get updated

1051
00:53:52,970 --> 00:53:57,480
or changed very spontaneously
and doesn't happen often.

1052
00:53:57,480 --> 00:54:00,800
So the problem is that
the internet today

1053
00:54:00,800 --> 00:54:03,680
is tuned to work with mostly
static topology and services.

1054
00:54:03,680 --> 00:54:06,230
It isn't tuned to handle
a high degree of dynamism

1055
00:54:06,230 --> 00:54:07,787
and spontaneity.

1056
00:54:07,787 --> 00:54:10,370
The second problem is that our
applications cannot learn about

1057
00:54:10,370 --> 00:54:10,870
the network.

1058
00:54:10,870 --> 00:54:13,520
We've been used to a highly
layered view of networking

1059
00:54:13,520 --> 00:54:16,040
architecture, and that
information doesn't easily

1060
00:54:16,040 --> 00:54:18,120
pass between different
layers of the protocol

1061
00:54:18,120 --> 00:54:19,333
stack up to the application.

1062
00:54:19,333 --> 00:54:21,000
So it's very hard
today for applications

1063
00:54:21,000 --> 00:54:23,450
to learn about the network, to
learn that bandwidth may have

1064
00:54:23,450 --> 00:54:24,908
changed, or to
learn that there are

1065
00:54:24,908 --> 00:54:26,480
wireless links along the path.

1066
00:54:26,480 --> 00:54:28,970
The third is the high cost
of operational management,

1067
00:54:28,970 --> 00:54:32,290
of setting these networks up,
of configuring networks to work.

1068
00:54:32,290 --> 00:54:34,790
And we would like to eliminate
that, or at least reduce that

1069
00:54:34,790 --> 00:54:38,420
to something insignificant.

1070
00:54:38,420 --> 00:54:40,140
How are we going
to achieve this?

1071
00:54:40,140 --> 00:54:41,450
Well, the first thing
to do is, rather than

1072
00:54:41,450 --> 00:54:44,030
think of the internet and future
networks as transmitting bits,

1073
00:54:44,030 --> 00:54:46,490
we should really think of them
as transmitting information.

1074
00:54:46,490 --> 00:54:48,200
We should view
ourselves as having

1075
00:54:48,200 --> 00:54:49,940
a large number of
devices and sensors

1076
00:54:49,940 --> 00:54:52,295
and actuators in the world,
a large number of providers,

1077
00:54:52,295 --> 00:54:55,400
a large number of
information providers.

1078
00:54:55,400 --> 00:54:57,380
And at any point in
time, let's imagine

1079
00:54:57,380 --> 00:55:00,890
we have an application
running on our Handy 21.

1080
00:55:00,890 --> 00:55:02,480
And we want to control
our environment

1081
00:55:02,480 --> 00:55:05,450
or retrieve some information
from the worldwide web

1082
00:55:05,450 --> 00:55:06,890
or from the internet.

1083
00:55:06,890 --> 00:55:09,320
Rather than think
of this as bits,

1084
00:55:09,320 --> 00:55:11,580
let's think of us as a
consumer of information.

1085
00:55:11,580 --> 00:55:14,360
Let's imagine that we want
some piece of information.

1086
00:55:14,360 --> 00:55:15,740
We have a distributed
application

1087
00:55:15,740 --> 00:55:17,157
that's running on
our machine that

1088
00:55:17,157 --> 00:55:19,580
requires some data
or some functionality

1089
00:55:19,580 --> 00:55:23,310
or some service that exists
somewhere in the network.

1090
00:55:23,310 --> 00:55:25,610
The point is that
often our application

1091
00:55:25,610 --> 00:55:28,100
or often we can
express what we want,

1092
00:55:28,100 --> 00:55:29,810
we can describe what
we want in words,

1093
00:55:29,810 --> 00:55:32,340
or our applications can
express what they want.

1094
00:55:32,340 --> 00:55:34,340
Unfortunately, they can't
often tell the network

1095
00:55:34,340 --> 00:55:36,302
where to find it.

1096
00:55:36,302 --> 00:55:37,760
With the way the
domain name system

1097
00:55:37,760 --> 00:55:39,800
works today, unless
we tell the network

1098
00:55:39,800 --> 00:55:45,470
where to find this information,
we don't get the information.

1099
00:55:45,470 --> 00:55:48,390
We can express this information
in interesting ways.

1100
00:55:48,390 --> 00:55:50,715
And that's how N21
is going to work.

1101
00:55:50,715 --> 00:55:52,340
So we're going to
think of this network

1102
00:55:52,340 --> 00:55:54,590
as having a set of consumers
and a set of providers

1103
00:55:54,590 --> 00:55:56,840
of information, of any piece
of information or service

1104
00:55:56,840 --> 00:55:58,280
or functionality that we want.

1105
00:55:58,280 --> 00:56:01,100
And the task of the network,
the first task of the network,

1106
00:56:01,100 --> 00:56:03,920
is to efficiently match up
these consumers and providers.

1107
00:56:03,920 --> 00:56:06,530
The consumer expresses the
information or the service

1108
00:56:06,530 --> 00:56:08,700
or functionality that it wants.

1109
00:56:08,700 --> 00:56:11,900
And the job of the network is
to figure out the best provider

1110
00:56:11,900 --> 00:56:14,750
to get this information from.

1111
00:56:14,750 --> 00:56:18,440
The second job of N21, of the
future network infrastructure,

1112
00:56:18,440 --> 00:56:21,013
is to enable applications,
to learn about conditions

1113
00:56:21,013 --> 00:56:23,180
in the network, to learn
about changes in bandwidth,

1114
00:56:23,180 --> 00:56:25,555
to learn about different types
of [? error conditions, ?]

1115
00:56:25,555 --> 00:56:28,130
to learn about failure modes,
and adapt well to what's

1116
00:56:28,130 --> 00:56:30,260
going on, adapt well
to heterogeneity,

1117
00:56:30,260 --> 00:56:33,490
and adapt well to everything
that changes in the network.

1118
00:56:33,490 --> 00:56:36,872
And our hope is that Oxygen's
N21 middleware protocols that

1119
00:56:36,872 --> 00:56:38,830
will achieve this will
also be a core component

1120
00:56:38,830 --> 00:56:42,207
of the future internet and
enable a much better way

1121
00:56:42,207 --> 00:56:44,290
in which applications and
distributed applications

1122
00:56:44,290 --> 00:56:46,300
can be set up to
organize themselves.

1123
00:56:46,300 --> 00:56:50,682
This is the notion
of self-organization.

1124
00:56:50,682 --> 00:56:52,640
What I'm going to describe
now is one component

1125
00:56:52,640 --> 00:56:54,470
of the self-organization puzzle.

1126
00:56:54,470 --> 00:56:56,450
And this is built
around the naming system

1127
00:56:56,450 --> 00:56:59,940
that we call an
intentional naming system.

1128
00:56:59,940 --> 00:57:02,760
It departs greatly from the
way current naming systems are

1129
00:57:02,760 --> 00:57:04,602
set up, which are
fairly static and which

1130
00:57:04,602 --> 00:57:06,060
usually involve
things like mapping

1131
00:57:06,060 --> 00:57:07,830
host names to addresses.

1132
00:57:07,830 --> 00:57:11,280
In N21's naming system, the INS,
or intentional naming system,

1133
00:57:11,280 --> 00:57:14,490
names do not express just
network locations or host

1134
00:57:14,490 --> 00:57:15,390
names.

1135
00:57:15,390 --> 00:57:16,960
Names express
desired properties.

1136
00:57:16,960 --> 00:57:18,960
They're intentional because
applications can now

1137
00:57:18,960 --> 00:57:21,000
express that intent
in what they want

1138
00:57:21,000 --> 00:57:23,697
by describing what they want.

1139
00:57:23,697 --> 00:57:25,280
So, for example,
names are descriptive

1140
00:57:25,280 --> 00:57:27,320
and information providers
could periodically

1141
00:57:27,320 --> 00:57:29,900
disseminate this descriptive
piece of information,

1142
00:57:29,900 --> 00:57:32,460
describing the service that
they provide into the network.

1143
00:57:32,460 --> 00:57:35,555
So for example, if I have a
camera in room 504 of the MIT

1144
00:57:35,555 --> 00:57:37,460
LCS building, I
could periodically

1145
00:57:37,460 --> 00:57:40,800
advertise this in this
fashion, for example,

1146
00:57:40,800 --> 00:57:44,690
a set of attributes and values
or a set of other properties

1147
00:57:44,690 --> 00:57:48,455
in some sort of
defined language.

1148
00:57:48,455 --> 00:57:49,830
When I want to
look up something,

1149
00:57:49,830 --> 00:57:51,830
and I want to retrieve a
piece of functionality,

1150
00:57:51,830 --> 00:57:53,930
names become query
expressions, again,

1151
00:57:53,930 --> 00:57:56,450
in a carefully defined language
with a well-defined set

1152
00:57:56,450 --> 00:58:00,050
of operators that we need to be
thinking about quite carefully.

1153
00:58:00,050 --> 00:58:01,723
For example, these
query expressions

1154
00:58:01,723 --> 00:58:03,890
could involve exact matches
of attributes to values,

1155
00:58:03,890 --> 00:58:06,800
they could involve qualities
of ranges and wildcard matches.

1156
00:58:06,800 --> 00:58:09,530
And for example, if I'm standing
in Tech Square in Cambridge

1157
00:58:09,530 --> 00:58:11,930
and I want to find the
closest cafe, what I could do

1158
00:58:11,930 --> 00:58:13,670
is to have my
application, that's

1159
00:58:13,670 --> 00:58:16,340
my cafe application or
my location application,

1160
00:58:16,340 --> 00:58:18,890
send a message out that wants
the cafe within a quarter mile

1161
00:58:18,890 --> 00:58:20,680
from here.

1162
00:58:20,680 --> 00:58:22,300
Presumably, the
cafes in the area

1163
00:58:22,300 --> 00:58:24,040
are proxies acting
on their behalf,

1164
00:58:24,040 --> 00:58:26,808
and will have periodically
broadcast or periodically

1165
00:58:26,808 --> 00:58:29,350
disseminated this information
describing themselves as a cafe

1166
00:58:29,350 --> 00:58:30,170
into the network.

1167
00:58:30,170 --> 00:58:32,830
And what we would like to do
is to have our system match

1168
00:58:32,830 --> 00:58:35,440
these queries with
the properties

1169
00:58:35,440 --> 00:58:37,690
that the providers
have disseminated

1170
00:58:37,690 --> 00:58:40,000
to the network, allowing,
of course, for fuzzy matches

1171
00:58:40,000 --> 00:58:43,030
because we may not always
expect an exact match to happen,

1172
00:58:43,030 --> 00:58:44,740
with different
providers of information

1173
00:58:44,740 --> 00:58:48,582
being different from the
consumers of information.

1174
00:58:48,582 --> 00:58:51,040
The way we're going to do this
is to first build it around,

1175
00:58:51,040 --> 00:58:53,110
of course, the system
architecture, because clearly,

1176
00:58:53,110 --> 00:58:55,402
just defining the syntax and
the semantics of the names

1177
00:58:55,402 --> 00:58:58,060
doesn't really achieve
the goal that we want.

1178
00:58:58,060 --> 00:59:00,673
And the first thing that we
need in any naming system

1179
00:59:00,673 --> 00:59:02,590
is, of course, a way of
resolving these names,

1180
00:59:02,590 --> 00:59:05,680
to where in the network
these services can be found.

1181
00:59:05,680 --> 00:59:10,173
And we do this by defining INRs,
or intentional name resolvers.

1182
00:59:10,173 --> 00:59:12,340
The way these intentional
names resolvers are set up

1183
00:59:12,340 --> 00:59:14,980
are that they're set up to
form a distributed overlay

1184
00:59:14,980 --> 00:59:17,740
network on top of the IP,
on top of the internet,

1185
00:59:17,740 --> 00:59:20,350
running over the
internet protocol.

1186
00:59:20,350 --> 00:59:23,240
And these resolvers are set
up so they can automatically

1187
00:59:23,240 --> 00:59:26,110
spawn up when
applications desire

1188
00:59:26,110 --> 00:59:29,710
the ability to resolve names
that they have to addresses.

1189
00:59:29,710 --> 00:59:32,410
They also form an overlay
topology of their own

1190
00:59:32,410 --> 00:59:34,330
using proper protocols
to discover neighbors

1191
00:59:34,330 --> 00:59:36,423
and also to perform
load balancing.

1192
00:59:36,423 --> 00:59:38,590
And one of the important
issues that always comes up

1193
00:59:38,590 --> 00:59:39,850
when one does
networking research

1194
00:59:39,850 --> 00:59:41,500
is how you're going
to deploy the scheme.

1195
00:59:41,500 --> 00:59:42,970
And we've carefully
thought about this,

1196
00:59:42,970 --> 00:59:44,020
and the way we're
going to deploy it

1197
00:59:44,020 --> 00:59:46,030
is by starting with
today's domain name system

1198
00:59:46,030 --> 00:59:48,010
and evolving that
in a direction that

1199
00:59:48,010 --> 00:59:49,510
allows us to make
it more expressive

1200
00:59:49,510 --> 00:59:51,980
and make it more intentional.

1201
00:59:51,980 --> 00:59:54,190
Let's say I'm on
my Handy 21 and I'd

1202
00:59:54,190 --> 00:59:59,020
like to find out from the
security camera located

1203
00:59:59,020 --> 01:00:01,198
in my office what
it's currently seeing.

1204
01:00:01,198 --> 01:00:02,740
In today's world,
what I'd have to do

1205
01:00:02,740 --> 01:00:04,930
is originally to
have set up my camera

1206
01:00:04,930 --> 01:00:06,910
to be something like
camera 510, which

1207
01:00:06,910 --> 01:00:09,880
happens to be my room
number in lcs.mit.edu.

1208
01:00:09,880 --> 01:00:12,520
And then go in and mess
with the domain name system

1209
01:00:12,520 --> 01:00:14,740
architecture, have a system
administrator do that,

1210
01:00:14,740 --> 01:00:17,873
and then plug in the name
to IP address mapping.

1211
01:00:17,873 --> 01:00:19,790
Of course, let's imagine
that this is actually

1212
01:00:19,790 --> 01:00:21,740
a mobile camera
network that are maybe

1213
01:00:21,740 --> 01:00:25,010
robotic cameras or other
mobile elements that are

1214
01:00:25,010 --> 01:00:27,140
moving these cameras around.

1215
01:00:27,140 --> 01:00:29,372
Then this approach
doesn't work very well,

1216
01:00:29,372 --> 01:00:30,830
because the moment
something moves,

1217
01:00:30,830 --> 01:00:32,240
its network location changes.

1218
01:00:32,240 --> 01:00:33,823
And the name [INAUDIBLE]
we're mapping

1219
01:00:33,823 --> 01:00:36,800
may need to be
reinvoked and remapped.

1220
01:00:36,800 --> 01:00:40,280
So rather than go with a
static mapping of host names,

1221
01:00:40,280 --> 01:00:44,978
let's express what we want
using an intentional name.

1222
01:00:44,978 --> 01:00:46,520
And the intentional
name in this case

1223
01:00:46,520 --> 01:00:49,100
would say that I want
to go to room 510

1224
01:00:49,100 --> 01:00:51,320
in NE43, which is
the MIT LSC building,

1225
01:00:51,320 --> 01:00:52,975
to an entity that's a camera.

1226
01:00:52,975 --> 01:00:54,350
And I would send
this packet out.

1227
01:00:54,350 --> 01:00:56,210
My application would
send this packet out.

1228
01:00:56,210 --> 01:00:59,030
And this name would get looked
up in the intentional name

1229
01:00:59,030 --> 01:01:00,692
resolvers that
look up these names

1230
01:01:00,692 --> 01:01:02,150
and then decide
what to do with it.

1231
01:01:02,150 --> 01:01:03,817
They might decide
that they know exactly

1232
01:01:03,817 --> 01:01:06,680
where to find it, in which
case they tell me where it is.

1233
01:01:06,680 --> 01:01:09,260
And they send the packet to the
service that I'm looking for.

1234
01:01:09,260 --> 01:01:11,052
Or they decide they
don't know where it is,

1235
01:01:11,052 --> 01:01:12,470
but they think
they know who else

1236
01:01:12,470 --> 01:01:15,050
might know where it is and
have these packets routed

1237
01:01:15,050 --> 01:01:17,785
via the name resolver network
to the eventual location

1238
01:01:17,785 --> 01:01:18,410
of the service.

1239
01:01:18,410 --> 01:01:21,290


1240
01:01:21,290 --> 01:01:24,335
When the camera gets my request,
it sends an image out and back

1241
01:01:24,335 --> 01:01:26,210
through the intentional
name resolvers to me.

1242
01:01:26,210 --> 01:01:28,252
And the important thing
here is that these images

1243
01:01:28,252 --> 01:01:30,950
get routed back, not with an IP
address that's pre-configured,

1244
01:01:30,950 --> 01:01:33,980
but with a name that expresses
who the consumer was,

1245
01:01:33,980 --> 01:01:36,680
which in this case,
was me on my Handy 21.

1246
01:01:36,680 --> 01:01:39,440
What this buys us actually
is automatic mobility,

1247
01:01:39,440 --> 01:01:41,660
because if I were to
move from this location

1248
01:01:41,660 --> 01:01:46,090
to some other location, because
the response isn't coming back

1249
01:01:46,090 --> 01:01:47,950
with the hot
pre-configured IP address

1250
01:01:47,950 --> 01:01:50,048
but coming back with a
name, if I move around

1251
01:01:50,048 --> 01:01:52,090
and I periodically send
this piece of information

1252
01:01:52,090 --> 01:01:54,040
that I've moved around,
the intentional name

1253
01:01:54,040 --> 01:01:55,720
resolvers and elsewhere
in the network

1254
01:01:55,720 --> 01:01:57,400
will cache this
information and know

1255
01:01:57,400 --> 01:02:00,710
how to route the packets
to me at my new location.

1256
01:02:00,710 --> 01:02:02,920
The other piece
that I get for free

1257
01:02:02,920 --> 01:02:04,930
is group communication,
because a name can now

1258
01:02:04,930 --> 01:02:08,282
map on to multiple services
or multiple people.

1259
01:02:08,282 --> 01:02:10,240
So for example, if there's
a network of cameras

1260
01:02:10,240 --> 01:02:13,810
that I want to get to, I can
use the name and use something

1261
01:02:13,810 --> 01:02:15,340
like a wild-card
entry to match it

1262
01:02:15,340 --> 01:02:18,250
to all of the cameras that exist
to have the packet returned

1263
01:02:18,250 --> 01:02:18,970
to them directly.

1264
01:02:18,970 --> 01:02:21,850


1265
01:02:21,850 --> 01:02:24,340
The key idea that enables
this is the integration

1266
01:02:24,340 --> 01:02:26,797
of name resolution
and message routing.

1267
01:02:26,797 --> 01:02:29,380
We've departed from the current
view of name resolution, which

1268
01:02:29,380 --> 01:02:31,263
is something that
the DNS does, which

1269
01:02:31,263 --> 01:02:32,680
is that you try
to get a hostname,

1270
01:02:32,680 --> 01:02:35,170
you try to resolve a hostname,
and it returns an IP address

1271
01:02:35,170 --> 01:02:35,470
to you.

1272
01:02:35,470 --> 01:02:37,000
And then you go
ahead and separately

1273
01:02:37,000 --> 01:02:38,990
establish a connection
to that IP address.

1274
01:02:38,990 --> 01:02:41,410
So when things move or the
world changes around you,

1275
01:02:41,410 --> 01:02:44,050
end up with stale,
outdated information.

1276
01:02:44,050 --> 01:02:46,120
In our architecture, we
integrate the process

1277
01:02:46,120 --> 01:02:47,650
of name resolution,
the looking up

1278
01:02:47,650 --> 01:02:49,602
and resolution of
names to routing.

1279
01:02:49,602 --> 01:02:51,310
So the name resolver
is actually actively

1280
01:02:51,310 --> 01:02:53,320
participate in
routing, and this way,

1281
01:02:53,320 --> 01:02:55,360
enables the applications
to actually control

1282
01:02:55,360 --> 01:02:56,710
what they get from the network.

1283
01:02:56,710 --> 01:02:59,938


1284
01:02:59,938 --> 01:03:00,730
This is the vision.

1285
01:03:00,730 --> 01:03:02,650
And I've described one
piece of the puzzle

1286
01:03:02,650 --> 01:03:04,167
that we're actively working on.

1287
01:03:04,167 --> 01:03:06,250
But there are some significant
research challenges

1288
01:03:06,250 --> 01:03:08,590
that we need to overcome
before this vision becomes

1289
01:03:08,590 --> 01:03:12,070
a complete reality and we have
truly self-organizing network

1290
01:03:12,070 --> 01:03:13,080
systems.

1291
01:03:13,080 --> 01:03:14,830
The first, of course,
is to make the scale

1292
01:03:14,830 --> 01:03:17,500
to large networks and large
numbers of applications.

1293
01:03:17,500 --> 01:03:20,110
We're working on a scalable
wide area naming architecture

1294
01:03:20,110 --> 01:03:22,240
based on intentional names
and actively thinking

1295
01:03:22,240 --> 01:03:24,010
of how to deploy
this on the internet,

1296
01:03:24,010 --> 01:03:26,440
and how to deploy this first
within the context of N21,

1297
01:03:26,440 --> 01:03:29,173
and then transition
it to the internet.

1298
01:03:29,173 --> 01:03:30,840
The second thing that
we need to look at

1299
01:03:30,840 --> 01:03:33,150
is an end-to-end
adaptation framework,

1300
01:03:33,150 --> 01:03:36,510
one that exposes network
information from underlying

1301
01:03:36,510 --> 01:03:38,520
layers of the protocol
stack to applications,

1302
01:03:38,520 --> 01:03:41,020
so that they can learn about
what's going on in the network.

1303
01:03:41,020 --> 01:03:43,530
And the second is coming
up with new algorithms

1304
01:03:43,530 --> 01:03:45,870
and new protocols to react
to this changing knowledge,

1305
01:03:45,870 --> 01:03:49,357
to do clever things
that use bandwidth well.

1306
01:03:49,357 --> 01:03:51,690
You're also going to hear
from Professor [INAUDIBLE] who

1307
01:03:51,690 --> 01:03:53,232
is going to talk
about an idea called

1308
01:03:53,232 --> 01:03:56,320
radioactive networks that allow
applications to have networks

1309
01:03:56,320 --> 01:03:57,430
adapt to what they desire.

1310
01:03:57,430 --> 01:03:59,030
So you can use
this, for example,

1311
01:03:59,030 --> 01:04:01,030
to change the available
bandwidth on the network

1312
01:04:01,030 --> 01:04:03,922
by changing the underlying
wireless modulation.

1313
01:04:03,922 --> 01:04:05,380
In this context,
we're also working

1314
01:04:05,380 --> 01:04:07,960
on a new congestion
management architecture

1315
01:04:07,960 --> 01:04:10,135
for internet applications.

1316
01:04:10,135 --> 01:04:12,010
And the third piece that
we need to deal with

1317
01:04:12,010 --> 01:04:14,723
is some very hard security
and privacy issues.

1318
01:04:14,723 --> 01:04:16,390
We've enabled a lot
of new functionality

1319
01:04:16,390 --> 01:04:19,030
and a lot of new,
cool things, where

1320
01:04:19,030 --> 01:04:21,520
we can have services
express what they want.

1321
01:04:21,520 --> 01:04:24,420
We have consumers express
what they want and get to it.

1322
01:04:24,420 --> 01:04:26,170
But this leads to some
significant privacy

1323
01:04:26,170 --> 01:04:28,795
and security issues, and you'll
hear from Professor Rivest talk

1324
01:04:28,795 --> 01:04:31,300
about some of this.

1325
01:04:31,300 --> 01:04:34,750
So to conclude, my vision of
the future is very simple.

1326
01:04:34,750 --> 01:04:36,820
It's networks that learn.

1327
01:04:36,820 --> 01:04:38,980
What we would like, and
what I think we will see,

1328
01:04:38,980 --> 01:04:40,960
is better application
and user control

1329
01:04:40,960 --> 01:04:42,220
over what networks provide.

1330
01:04:42,220 --> 01:04:44,032
And things like our
intention naming system

1331
01:04:44,032 --> 01:04:45,240
are a step in that direction.

1332
01:04:45,240 --> 01:04:46,810
And a large number
of other pieces

1333
01:04:46,810 --> 01:04:49,300
of this puzzle that will
be fit and will make this

1334
01:04:49,300 --> 01:04:51,020
happen in the future.

1335
01:04:51,020 --> 01:04:53,110
The second is a
significantly higher degree

1336
01:04:53,110 --> 01:04:56,290
of dynamism, mobility, and
spontaneity, as well as

1337
01:04:56,290 --> 01:05:00,420
decentralization and robustness
in our network infrastructure.

1338
01:05:00,420 --> 01:05:02,880
No longer will it be the case
that a distributed system is

1339
01:05:02,880 --> 01:05:05,807
defined as, "if a machine that
you've never heard of fails,

1340
01:05:05,807 --> 01:05:07,140
you'll never get any work done."

1341
01:05:07,140 --> 01:05:09,557
And this is something Leslie
Lamport said a long time ago.

1342
01:05:09,557 --> 01:05:12,060
What we would like is to
eliminate that altogether

1343
01:05:12,060 --> 01:05:15,300
and build really robust
distributed network systems.

1344
01:05:15,300 --> 01:05:17,040
And the third point
is that I think

1345
01:05:17,040 --> 01:05:19,508
we're going to see a significant
degree of integration

1346
01:05:19,508 --> 01:05:22,050
with the real world, with the
real, physical world of devices

1347
01:05:22,050 --> 01:05:24,930
and sensors and actuators
that are already

1348
01:05:24,930 --> 01:05:26,520
embedded in our real world.

1349
01:05:26,520 --> 01:05:28,440
And we'll see them
more tightly integrated

1350
01:05:28,440 --> 01:05:31,080
with our networked world
so that we can actually

1351
01:05:31,080 --> 01:05:32,820
get useful work done.

1352
01:05:32,820 --> 01:05:35,370
And my belief is that the
self-organization middleware

1353
01:05:35,370 --> 01:05:38,760
and adaptation protocols, such
as the one I just described,

1354
01:05:38,760 --> 01:05:40,870
are key to making
this a reality.

1355
01:05:40,870 --> 01:05:41,370
Thank you.

1356
01:05:41,370 --> 01:05:52,890


1357
01:05:52,890 --> 01:05:55,740
Last week, the New York
Times called Dave Clark

1358
01:05:55,740 --> 01:05:59,880
"the father of the internet."

1359
01:05:59,880 --> 01:06:02,710
It's not exactly true.

1360
01:06:02,710 --> 01:06:07,800
I mean there was Licklider and
Taylor and Roberts and Kahn

1361
01:06:07,800 --> 01:06:10,110
and Serf and, of
course, Al Gore.

1362
01:06:10,110 --> 01:06:13,960
Now, don't ruin my
jokes like that.

1363
01:06:13,960 --> 01:06:17,770
Dave himself denies paternity.

1364
01:06:17,770 --> 01:06:22,690
He admits to being the first
cousin of the internet.

1365
01:06:22,690 --> 01:06:25,480
During the early
'80s, Dave had a lot

1366
01:06:25,480 --> 01:06:30,100
to do with putting personal
computers onto the internet.

1367
01:06:30,100 --> 01:06:31,780
And so I'm anxious
to hear what he's

1368
01:06:31,780 --> 01:06:35,330
going to say about
the post-PC internet.

1369
01:06:35,330 --> 01:06:35,830
Dave.

1370
01:06:35,830 --> 01:06:42,540


1371
01:06:42,540 --> 01:06:44,520
You're stealing my jokes, too.

1372
01:06:44,520 --> 01:06:47,160


1373
01:06:47,160 --> 01:06:51,038
So let me explain
this "post-PC" phrase.

1374
01:06:51,038 --> 01:06:53,580
I think several of us started
using it for different reasons.

1375
01:06:53,580 --> 01:06:55,495
I picked it up because
I wanted to tweak

1376
01:06:55,495 --> 01:06:57,870
my friends at Intel, but of
course now they're saying it,

1377
01:06:57,870 --> 01:06:58,370
too.

1378
01:06:58,370 --> 01:07:01,380
So the fun is going out of it.

1379
01:07:01,380 --> 01:07:02,970
There's a spectrum we're on.

1380
01:07:02,970 --> 01:07:05,400
I'm old enough to
remember timesharing.

1381
01:07:05,400 --> 01:07:07,650
And so I want to go back
to the good old days

1382
01:07:07,650 --> 01:07:09,990
when the CPU was this
monstrously expensive thing

1383
01:07:09,990 --> 01:07:12,120
and we all gathered in
a circle and bowed down.

1384
01:07:12,120 --> 01:07:16,200
So you have all these displays
clustered around the computer.

1385
01:07:16,200 --> 01:07:19,162
It filled the entire ninth
floor of the building.

1386
01:07:19,162 --> 01:07:19,870
It was wonderful.

1387
01:07:19,870 --> 01:07:20,787
It was called Multics.

1388
01:07:20,787 --> 01:07:23,250


1389
01:07:23,250 --> 01:07:27,390
I'm old enough to remember CTSS.

1390
01:07:27,390 --> 01:07:30,460
So what happened is the CPU got
cheaper as we kept doing this.

1391
01:07:30,460 --> 01:07:38,100
And as the CPU got cheaper,
we reached this magic point

1392
01:07:38,100 --> 01:07:42,047
where there was a 1 to 1 ratio
between the CPU and the PC.

1393
01:07:42,047 --> 01:07:43,380
1 to 1 is a very powerful ratio.

1394
01:07:43,380 --> 01:07:45,870
It lets you forget about
all sorts of things, like,

1395
01:07:45,870 --> 01:07:47,820
we forgot about
the Windows system.

1396
01:07:47,820 --> 01:07:50,190
We forgot how to split an
application into two halves.

1397
01:07:50,190 --> 01:07:52,080
We're going to have
to relearn all this.

1398
01:07:52,080 --> 01:07:54,300
And because that's a
very powerful ratio,

1399
01:07:54,300 --> 01:07:56,740
we've sat at that
point for a long time.

1400
01:07:56,740 --> 01:07:58,720
But that hasn't
stopped this trend,

1401
01:07:58,720 --> 01:08:00,622
the one that Anant
was referring to,

1402
01:08:00,622 --> 01:08:01,830
other people have alluded to.

1403
01:08:01,830 --> 01:08:04,500
The CPUs are still
getting cheaper, faster

1404
01:08:04,500 --> 01:08:06,180
than the displays
are getting cheaper.

1405
01:08:06,180 --> 01:08:08,770
So that ratio is eventually
going to unstick.

1406
01:08:08,770 --> 01:08:10,770
And when it does, it's
going to do it violently.

1407
01:08:10,770 --> 01:08:12,805
We can see this happening now.

1408
01:08:12,805 --> 01:08:15,180
When it does, of course, what
you're going to end up with

1409
01:08:15,180 --> 01:08:17,430
is this alternative
vision here, in which

1410
01:08:17,430 --> 01:08:21,990
there's one display
surrounded by a bunch of CPUs.

1411
01:08:21,990 --> 01:08:25,260
You don't want to think
of that as PC1, PC2, PC3.

1412
01:08:25,260 --> 01:08:26,520
That's stupid.

1413
01:08:26,520 --> 01:08:28,109
Those are all of
the computers that

1414
01:08:28,109 --> 01:08:29,640
are embedded in
all of the objects

1415
01:08:29,640 --> 01:08:32,920
that we encounter as we
walk through the world.

1416
01:08:32,920 --> 01:08:35,640
So what are the essential
characteristics here?

1417
01:08:35,640 --> 01:08:37,020
Processors are getting cheaper.

1418
01:08:37,020 --> 01:08:38,312
That's sort of straightforward.

1419
01:08:38,312 --> 01:08:40,500
I hope Anant talked
to you into that.

1420
01:08:40,500 --> 01:08:43,290
The second is user
interfaces go with people,

1421
01:08:43,290 --> 01:08:45,210
they don't go with processors.

1422
01:08:45,210 --> 01:08:47,340
I have a display on my desk.

1423
01:08:47,340 --> 01:08:50,609
I don't want to have one
display for computer.

1424
01:08:50,609 --> 01:08:52,620
I want to have one display.

1425
01:08:52,620 --> 01:08:55,202
I'd like the small one
with me when I walk around.

1426
01:08:55,202 --> 01:08:56,910
Do I have to throw
this across the floor?

1427
01:08:56,910 --> 01:08:59,130
I'm not sure.

1428
01:08:59,130 --> 01:09:01,229
I'd like a small one with
me when I walk around.

1429
01:09:01,229 --> 01:09:03,779
I'd like a big one in my office,
maybe a huge one someplace

1430
01:09:03,779 --> 01:09:04,279
else.

1431
01:09:04,279 --> 01:09:06,990
I'd like one in my car
that whispers in my ear.

1432
01:09:06,990 --> 01:09:08,430
So that's the second point.

1433
01:09:08,430 --> 01:09:10,350
And the third point.

1434
01:09:10,350 --> 01:09:13,000
Notice my picture had
this downward trend.

1435
01:09:13,000 --> 01:09:15,870
Let's see, there's one button
that they said don't use.

1436
01:09:15,870 --> 01:09:17,160
I got away with it.

1437
01:09:17,160 --> 01:09:19,319
It crashes the computer
or something like that.

1438
01:09:19,319 --> 01:09:22,080
There's a trend that is going
back toward the top as well,

1439
01:09:22,080 --> 01:09:23,247
although they've renamed it.

1440
01:09:23,247 --> 01:09:26,010
They don't call it timesharing,
they call services in the net.

1441
01:09:26,010 --> 01:09:27,660
The idea that if you put
something in the net,

1442
01:09:27,660 --> 01:09:29,130
it actually may be
easier to manage.

1443
01:09:29,130 --> 01:09:31,047
You pay somebody else
and he takes care of it.

1444
01:09:31,047 --> 01:09:33,270
You don't have to install
it yourself for nine hours

1445
01:09:33,270 --> 01:09:35,920
so that you have an
excuse to fly to Asia.

1446
01:09:35,920 --> 01:09:40,470
And so there's a force in this
picture that is pushing us back

1447
01:09:40,470 --> 01:09:41,220
toward the top.

1448
01:09:41,220 --> 01:09:42,720
There's a force in
this picture that

1449
01:09:42,720 --> 01:09:44,130
is pushing us to the bottom.

1450
01:09:44,130 --> 01:09:47,670
And the PC is simply a way
station along that path.

1451
01:09:47,670 --> 01:09:51,779
Now, there are students I have
met-- it's really depressing--

1452
01:09:51,779 --> 01:09:54,360
who are so young
that they've never

1453
01:09:54,360 --> 01:09:56,550
seen anything except a PC.

1454
01:09:56,550 --> 01:09:59,100
And they think that's
the way it's always been.

1455
01:09:59,100 --> 01:10:02,220
So of course, that's the
way it will always be.

1456
01:10:02,220 --> 01:10:05,250
So there is a role
for old fogies.

1457
01:10:05,250 --> 01:10:07,320
If you've outlived the
present into the past,

1458
01:10:07,320 --> 01:10:09,750
maybe you can remind them that--

1459
01:10:09,750 --> 01:10:12,360
But we did TCP for
the PC in 1982.

1460
01:10:12,360 --> 01:10:14,010
That's a long time ago.

1461
01:10:14,010 --> 01:10:16,950
And this gives you a
false sense of stability.

1462
01:10:16,950 --> 01:10:20,220
And this traps you when you
think about what applications

1463
01:10:20,220 --> 01:10:21,900
are going to look like.

1464
01:10:21,900 --> 01:10:23,370
I'm sort of an
internet guy, so I

1465
01:10:23,370 --> 01:10:25,022
like to think of
applications that

1466
01:10:25,022 --> 01:10:26,730
have the word internet
in their name, so,

1467
01:10:26,730 --> 01:10:30,270
internet telephony, internet
telephone, internet radio,

1468
01:10:30,270 --> 01:10:33,060
internet music,
internet television.

1469
01:10:33,060 --> 01:10:35,340
And I had somebody
say to me, I don't

1470
01:10:35,340 --> 01:10:38,700
want to go to my PC
to make a phone call.

1471
01:10:38,700 --> 01:10:41,640
Wrong thought.

1472
01:10:41,640 --> 01:10:43,890
I have a telephone in almost
all rooms in my house.

1473
01:10:43,890 --> 01:10:46,740
I don't know, we
are sort of weird.

1474
01:10:46,740 --> 01:10:48,600
10 years from now,
what will you have

1475
01:10:48,600 --> 01:10:51,000
in every room in your house?

1476
01:10:51,000 --> 01:10:52,170
It won't be a telephone.

1477
01:10:52,170 --> 01:10:53,580
It will be a something.

1478
01:10:53,580 --> 01:10:56,040
It'll be on the
internet, we know that.

1479
01:10:56,040 --> 01:10:58,440
It'll probably have speech
input, speech output.

1480
01:10:58,440 --> 01:11:01,440
It'll have a display.

1481
01:11:01,440 --> 01:11:02,970
It probably won't
have a keyboard,

1482
01:11:02,970 --> 01:11:06,260
if you believe Victor Zue.

1483
01:11:06,260 --> 01:11:07,790
They'll be ubiquitous.

1484
01:11:07,790 --> 01:11:12,560
And that'll be where you do your
telephony, you listen to music.

1485
01:11:12,560 --> 01:11:15,050
Now, you might disagree
as to what that object is,

1486
01:11:15,050 --> 01:11:16,550
but envision it.

1487
01:11:16,550 --> 01:11:19,340
And say, that's the context
in which new applications are

1488
01:11:19,340 --> 01:11:21,560
going to emerge, because
that's the user interface

1489
01:11:21,560 --> 01:11:24,800
space in which they're
going to happen.

1490
01:11:24,800 --> 01:11:26,840
Now, there are a couple
of obvious things

1491
01:11:26,840 --> 01:11:29,810
that follow from
what I just said.

1492
01:11:29,810 --> 01:11:32,132
I don't want to dwell on this.

1493
01:11:32,132 --> 01:11:34,340
The consumer today thinks
they can go to a drugstore,

1494
01:11:34,340 --> 01:11:36,770
buy an RG11 and some wire,
screw it all together,

1495
01:11:36,770 --> 01:11:38,590
and actually put a new
phone extension in.

1496
01:11:38,590 --> 01:11:40,340
They don't have to run
HP OpenView for you

1497
01:11:40,340 --> 01:11:42,180
in order to do that.

1498
01:11:42,180 --> 01:11:44,600
So whatever tomorrow's
data networks are,

1499
01:11:44,600 --> 01:11:46,910
they have to be consumer grade.

1500
01:11:46,910 --> 01:11:50,150
You can't break it, and if
you do, it fixes itself.

1501
01:11:50,150 --> 01:11:53,060
You can buy it at the drugstore.

1502
01:11:53,060 --> 01:11:55,287
It has to be cheap.

1503
01:11:55,287 --> 01:11:57,120
There's an old saying
that somebody made up,

1504
01:11:57,120 --> 01:11:58,980
which is that the cost
of networking something

1505
01:11:58,980 --> 01:12:01,240
could be no more than 10% of
the cost of the something.

1506
01:12:01,240 --> 01:12:03,073
So if you have a dollar
object in your hand,

1507
01:12:03,073 --> 01:12:07,070
putting it on the net
shouldn't cost more than $0.10.

1508
01:12:07,070 --> 01:12:10,290
And that implies that networks
have to be heterogeneous.

1509
01:12:10,290 --> 01:12:12,920
Now, those of us who've lived
in internet for a long time,

1510
01:12:12,920 --> 01:12:14,930
have remembered that
we didn't always

1511
01:12:14,930 --> 01:12:16,550
have one networking answer.

1512
01:12:16,550 --> 01:12:18,950
As Bob will remind you,
if I were to drop my hat,

1513
01:12:18,950 --> 01:12:22,880
I'd try to invent something
called Token Ring.

1514
01:12:22,880 --> 01:12:23,760
Oh well.

1515
01:12:23,760 --> 01:12:27,560
But during the time that
we were debating this,

1516
01:12:27,560 --> 01:12:29,570
we could still talk
to each other because

1517
01:12:29,570 --> 01:12:31,400
of a set of protocols
that ran on top

1518
01:12:31,400 --> 01:12:34,100
of that mess called "internet."

1519
01:12:34,100 --> 01:12:36,230
Why is the future
also heterogeneous?

1520
01:12:36,230 --> 01:12:39,350
Well, you're going to have
somewhat expensive, like $1,

1521
01:12:39,350 --> 01:12:42,330
high-speed networks that are
good for entertainment, video,

1522
01:12:42,330 --> 01:12:44,210
high-speed audio, and
all that kind of stuff.

1523
01:12:44,210 --> 01:12:46,370
And you're going to have
really cheap nets, like 1/10

1524
01:12:46,370 --> 01:12:48,740
of a cent, which are going
to hook together little light

1525
01:12:48,740 --> 01:12:50,217
switches and things like this.

1526
01:12:50,217 --> 01:12:52,550
And the answer is that all
of this has to work together.

1527
01:12:52,550 --> 01:12:55,430
Somebody said to me,
consumer appliances

1528
01:12:55,430 --> 01:12:56,930
aren't taking off
because we haven't

1529
01:12:56,930 --> 01:12:59,360
figured out what the standard
is for networks in the home.

1530
01:12:59,360 --> 01:12:59,900
Wrong.

1531
01:12:59,900 --> 01:13:00,680
Think.

1532
01:13:00,680 --> 01:13:02,850
Multiple kinds,
glue them together.

1533
01:13:02,850 --> 01:13:04,340
They have to be glued together.

1534
01:13:04,340 --> 01:13:07,640
Great new market for Cisco.

1535
01:13:07,640 --> 01:13:10,700
But except heterogeneity,
I want to look

1536
01:13:10,700 --> 01:13:14,150
at some slightly more, I
think, interesting observations

1537
01:13:14,150 --> 01:13:15,200
about system structure.

1538
01:13:15,200 --> 01:13:18,140
And this relates a little
bit to what Anant was saying,

1539
01:13:18,140 --> 01:13:20,400
and Bill Gates a
little while ago.

1540
01:13:20,400 --> 01:13:22,070
If you think about
some of the devices--

1541
01:13:22,070 --> 01:13:23,810
I was going to
wave my cell phone,

1542
01:13:23,810 --> 01:13:25,820
but I can't because I
inadvertently destroyed it

1543
01:13:25,820 --> 01:13:26,510
yesterday.

1544
01:13:26,510 --> 01:13:27,878
I did throw it away.

1545
01:13:27,878 --> 01:13:28,670
It was an accident.

1546
01:13:28,670 --> 01:13:30,650
I have to buy another one.

1547
01:13:30,650 --> 01:13:33,290
A lot of the
devices we use today

1548
01:13:33,290 --> 01:13:36,230
are in the word processor stage
that Gates was illustrating,

1549
01:13:36,230 --> 01:13:38,720
in which you have a device
with a custom application built

1550
01:13:38,720 --> 01:13:39,220
into it.

1551
01:13:39,220 --> 01:13:41,300
My cell phone is really
only one application.

1552
01:13:41,300 --> 01:13:44,510
It does cell phone for a living.

1553
01:13:44,510 --> 01:13:46,160
Then we've got
this device that's

1554
01:13:46,160 --> 01:13:50,990
sort of the PC, the
amazing shrinking PC.

1555
01:13:50,990 --> 01:13:54,445
It turned from a desktop
to a laptop to a Palmtop.

1556
01:13:54,445 --> 01:13:55,820
I was reading a
magazine that was

1557
01:13:55,820 --> 01:13:58,830
talking about the micro,
ultra lightweight.

1558
01:13:58,830 --> 01:14:00,330
They're getting
thinner and thinner.

1559
01:14:00,330 --> 01:14:01,580
But they're still PCs.

1560
01:14:01,580 --> 01:14:04,160
They have complete
programmability in the way

1561
01:14:04,160 --> 01:14:08,180
that the shrinkwrap
presumption of today assumes.

1562
01:14:08,180 --> 01:14:10,190
So the first stage
in the process

1563
01:14:10,190 --> 01:14:12,920
of getting these
devices into the future

1564
01:14:12,920 --> 01:14:14,582
is to make them smaller.

1565
01:14:14,582 --> 01:14:16,040
But the second
stage, which I think

1566
01:14:16,040 --> 01:14:18,080
is actually more
interesting, is when

1567
01:14:18,080 --> 01:14:21,060
they start to be dependent
on services in the network

1568
01:14:21,060 --> 01:14:24,800
so that they're not
quite so self-sufficient.

1569
01:14:24,800 --> 01:14:29,060
The exciting applications
that run on your palm object

1570
01:14:29,060 --> 01:14:33,110
are the ones that have matching
services back in the net.

1571
01:14:33,110 --> 01:14:35,480
My assistant can make an
appointment in my calendar,

1572
01:14:35,480 --> 01:14:38,720
and if this thing were
only on the net, sigh,

1573
01:14:38,720 --> 01:14:41,390
it would be able to
tell me about it.

1574
01:14:41,390 --> 01:14:44,715
It's particularly interesting
when, as I did yesterday,

1575
01:14:44,715 --> 01:14:46,340
I destroy the contents
of this and have

1576
01:14:46,340 --> 01:14:48,260
to reload it from somewhere.

1577
01:14:48,260 --> 01:14:50,068
So what does this
say about software?

1578
01:14:50,068 --> 01:14:51,860
Well, it says we have
to do something we've

1579
01:14:51,860 --> 01:14:53,390
talked about for a long time.

1580
01:14:53,390 --> 01:14:56,210
We have to write
client server software.

1581
01:14:56,210 --> 01:14:57,890
Now, today the
server is on the PC,

1582
01:14:57,890 --> 01:15:01,400
but it's just going to diffuse
into the net as a set of things

1583
01:15:01,400 --> 01:15:03,530
that you can either
capitalize and purchase

1584
01:15:03,530 --> 01:15:05,780
or you can, in fact, rent.

1585
01:15:05,780 --> 01:15:08,998
There will be companies selling
you services on the net.

1586
01:15:08,998 --> 01:15:11,540
But then the next thing that's
going to happen to this device

1587
01:15:11,540 --> 01:15:16,440
that you carry with you is it's
going to blow up into parts.

1588
01:15:16,440 --> 01:15:18,330
Here's a cool device.

1589
01:15:18,330 --> 01:15:22,050
Can you see it in
the back of the room?

1590
01:15:22,050 --> 01:15:25,765
This is a little earphone
with a little microphone.

1591
01:15:25,765 --> 01:15:27,890
Other people are pulling
out their little earphones

1592
01:15:27,890 --> 01:15:28,515
and showing me.

1593
01:15:28,515 --> 01:15:30,890
You put it in your ear
like this, and you talk.

1594
01:15:30,890 --> 01:15:32,600
Now, what era does this live in?

1595
01:15:32,600 --> 01:15:36,897
Well, actually this lives in
the one-device, one-application

1596
01:15:36,897 --> 01:15:38,480
era, because the
plug on the other end

1597
01:15:38,480 --> 01:15:41,147
goes into my cell phone, and the
only thing you can use this for

1598
01:15:41,147 --> 01:15:43,155
is to make a cell phone call.

1599
01:15:43,155 --> 01:15:45,530
But as Victor is about to
explain to you as soon as I get

1600
01:15:45,530 --> 01:15:48,860
off the stage, we want
to talk to every one

1601
01:15:48,860 --> 01:15:50,820
of our applications.

1602
01:15:50,820 --> 01:15:53,480
So we're going to move from
a model of one device, one

1603
01:15:53,480 --> 01:15:57,030
application, to one
device, one function.

1604
01:15:57,030 --> 01:15:58,710
This is the listening function.

1605
01:15:58,710 --> 01:16:00,630
This is the talking function.

1606
01:16:00,630 --> 01:16:03,900
This is the two-line display
with four buttons function.

1607
01:16:03,900 --> 01:16:07,860
This is the slightly larger
display with a pen function.

1608
01:16:07,860 --> 01:16:11,180
And now when you want a new
application, for example,

1609
01:16:11,180 --> 01:16:13,875
you'd like to be able to do the
cell phone thing, what you're

1610
01:16:13,875 --> 01:16:15,500
going to do is you're
going to download

1611
01:16:15,500 --> 01:16:17,917
a piece of software that looks
around your world and says,

1612
01:16:17,917 --> 01:16:20,720
oh, he's asked me to
build him a cell phone.

1613
01:16:20,720 --> 01:16:21,930
Can I do that?

1614
01:16:21,930 --> 01:16:23,270
Well, I found an earpiece.

1615
01:16:23,270 --> 01:16:24,740
I found a microphone.

1616
01:16:24,740 --> 01:16:27,380
I found someplace that the
caller ID could show up.

1617
01:16:27,380 --> 01:16:29,330
I found his little black
book out on the net.

1618
01:16:29,330 --> 01:16:31,070
I found this service over here.

1619
01:16:31,070 --> 01:16:35,870
If I stir and mix,
that'll be a cell phone.

1620
01:16:35,870 --> 01:16:38,443
Where did my slide go?

1621
01:16:38,443 --> 01:16:40,610
Oh, I see, they're trying
to show you my microphone.

1622
01:16:40,610 --> 01:16:43,220
Cool, great.

1623
01:16:43,220 --> 01:16:46,250
So applications, as somebody
previously just said,

1624
01:16:46,250 --> 01:16:47,875
are not shrinkwrap.

1625
01:16:47,875 --> 01:16:49,250
What they have to
do is they have

1626
01:16:49,250 --> 01:16:51,260
to come down into
your world and they

1627
01:16:51,260 --> 01:16:52,790
have to organize your world.

1628
01:16:52,790 --> 01:16:56,210
And notice it gets organized
differently for each person,

1629
01:16:56,210 --> 01:16:59,510
because I might
like the caller ID

1630
01:16:59,510 --> 01:17:01,340
information on my wristwatch.

1631
01:17:01,340 --> 01:17:04,940
And you might like the caller
ID information gently whispered

1632
01:17:04,940 --> 01:17:05,840
in your ear.

1633
01:17:05,840 --> 01:17:07,970
And somebody else might
say, that's really stupid.

1634
01:17:07,970 --> 01:17:09,887
I just want this thing
in my pocket to vibrate

1635
01:17:09,887 --> 01:17:11,900
and I'll answer it anyway.

1636
01:17:11,900 --> 01:17:14,240
So there isn't a uniformity
of user interface.

1637
01:17:14,240 --> 01:17:17,300
There isn't a uniformity
of application function.

1638
01:17:17,300 --> 01:17:19,640
It's the opposite of shrinkwrap.

1639
01:17:19,640 --> 01:17:21,968
It's going to be
self-organizing.

1640
01:17:21,968 --> 01:17:23,510
And it's going to
be self-organizing,

1641
01:17:23,510 --> 01:17:26,992
both among the devices you
carry and the services that

1642
01:17:26,992 --> 01:17:27,950
are out on the network.

1643
01:17:27,950 --> 01:17:32,437


1644
01:17:32,437 --> 01:17:34,270
Let me make one more
point about this world.

1645
01:17:34,270 --> 01:17:39,170


1646
01:17:39,170 --> 01:17:40,793
Well, here's a
little, cute device.

1647
01:17:40,793 --> 01:17:42,460
I have no idea what
it is, but I bet you

1648
01:17:42,460 --> 01:17:43,750
it has a computer in it.

1649
01:17:43,750 --> 01:17:45,090
It says RadioShack.

1650
01:17:45,090 --> 01:17:48,140


1651
01:17:48,140 --> 01:17:51,340
Everything's going to
have a computer in it.

1652
01:17:51,340 --> 01:17:53,990
Everything's going
to be networked.

1653
01:17:53,990 --> 01:17:55,000
Why?

1654
01:17:55,000 --> 01:17:56,860
Do I have some grand vision?

1655
01:17:56,860 --> 01:17:58,570
No, I don't have a grand vision.

1656
01:17:58,570 --> 01:18:01,000
It makes the cost lower.

1657
01:18:01,000 --> 01:18:04,405
Lowering the cost
equals inevitable.

1658
01:18:04,405 --> 01:18:06,280
Somebody was explaining
to me today that when

1659
01:18:06,280 --> 01:18:08,283
you manufacture an appliance--

1660
01:18:08,283 --> 01:18:09,700
I don't mean a
computer appliance,

1661
01:18:09,700 --> 01:18:11,620
I mean something
like a microwave--

1662
01:18:11,620 --> 01:18:16,300
it's cheaper to put
a network interface

1663
01:18:16,300 --> 01:18:17,770
on the back of
the box than it is

1664
01:18:17,770 --> 01:18:19,360
to put a user interface
on the front of the box,

1665
01:18:19,360 --> 01:18:20,830
because the user interface
in the front of the box

1666
01:18:20,830 --> 01:18:22,480
has buttons and
displays and things.

1667
01:18:22,480 --> 01:18:26,100
And those cost money
and they break.

1668
01:18:26,100 --> 01:18:27,530
So all of these
computers that are

1669
01:18:27,530 --> 01:18:29,072
embedded in all the
objects around us

1670
01:18:29,072 --> 01:18:31,520
are going to get networked.

1671
01:18:31,520 --> 01:18:33,140
And here's the point.

1672
01:18:33,140 --> 01:18:36,980
Each one of those devices
represents a little piece

1673
01:18:36,980 --> 01:18:40,970
of cartilage that's linking the
real world to the cyber world.

1674
01:18:40,970 --> 01:18:43,470
When you have an object that's
got a computer inside it,

1675
01:18:43,470 --> 01:18:45,090
you can deal with
it in two ways.

1676
01:18:45,090 --> 01:18:47,630
You can walk up to it in
physical space like this,

1677
01:18:47,630 --> 01:18:50,330
and you can pat on the keyboard
and see if something happens,

1678
01:18:50,330 --> 01:18:53,090
or you can talk to
it in cyberspace.

1679
01:18:53,090 --> 01:18:55,205
So what's the Coke
machine of the future?

1680
01:18:55,205 --> 01:18:57,020
The Coke machine of
the future doesn't

1681
01:18:57,020 --> 01:18:59,730
have to have a place
where you put money in.

1682
01:18:59,730 --> 01:19:01,730
Boy, does that add cost,
because not only do you

1683
01:19:01,730 --> 01:19:03,230
have to have that
machinery that can tell

1684
01:19:03,230 --> 01:19:04,610
real quarters from
fake ones, but then you

1685
01:19:04,610 --> 01:19:05,900
have to send a human
out to enter it

1686
01:19:05,900 --> 01:19:07,400
and they have to
send a second human

1687
01:19:07,400 --> 01:19:08,920
to make sure it's not cheating.

1688
01:19:08,920 --> 01:19:09,710
No, no, no.

1689
01:19:09,710 --> 01:19:12,320
A Coke machine in the
future is a machine

1690
01:19:12,320 --> 01:19:17,960
for dispensing a Coke that just
has a little placard on front.

1691
01:19:17,960 --> 01:19:20,420
And the placard has some
sort of machine-readable name

1692
01:19:20,420 --> 01:19:21,630
across the bottom.

1693
01:19:21,630 --> 01:19:25,800
And then above it, it says,
scan and click one Coke.

1694
01:19:25,800 --> 01:19:26,550
So what do you do?

1695
01:19:26,550 --> 01:19:28,670
You take out of your
pocket some little object

1696
01:19:28,670 --> 01:19:31,070
that can scan that name.

1697
01:19:31,070 --> 01:19:32,810
It's on the little placard.

1698
01:19:32,810 --> 01:19:34,610
This is like a
mouse in real space.

1699
01:19:34,610 --> 01:19:35,660
You can point and click.

1700
01:19:35,660 --> 01:19:37,298
You can drag and drop.

1701
01:19:37,298 --> 01:19:38,840
And then you push
the "do it" button.

1702
01:19:38,840 --> 01:19:41,310
I don't know what the
Microsoft people call it.

1703
01:19:41,310 --> 01:19:44,090
But Windows have two
buttons, "do it" and "about."

1704
01:19:44,090 --> 01:19:46,820
And so you scan the name,
you push the "do it"

1705
01:19:46,820 --> 01:19:50,330
button, and the message goes
out around through the internet

1706
01:19:50,330 --> 01:19:53,330
and it goes to the place where
I pay the Coca-Cola franchisee

1707
01:19:53,330 --> 01:19:54,050
$1.

1708
01:19:54,050 --> 01:19:56,990
And then a Coca-Cola franchisee
sends a message around the net

1709
01:19:56,990 --> 01:20:00,050
to the Coke machine,
and it dumps a Coke out.

1710
01:20:00,050 --> 01:20:01,520
Why is this inevitable?

1711
01:20:01,520 --> 01:20:05,420
Because it's a much
cheaper Coke machine.

1712
01:20:05,420 --> 01:20:08,892
So what namespace links
real and cyberspace?

1713
01:20:08,892 --> 01:20:10,100
It's got to be very flexible.

1714
01:20:10,100 --> 01:20:11,660
It's got to be
machine scannable.

1715
01:20:11,660 --> 01:20:13,423
You don't want to
sit and type in a URL

1716
01:20:13,423 --> 01:20:15,590
in order to get a Coke out
of a Coke machine, right?

1717
01:20:15,590 --> 01:20:17,690
So this has got to be automatic.

1718
01:20:17,690 --> 01:20:21,830
All of these names
have to be scannable.

1719
01:20:21,830 --> 01:20:23,190
So what's the future look like?

1720
01:20:23,190 --> 01:20:31,330
Well, it's a network
full of services,

1721
01:20:31,330 --> 01:20:35,530
all kinds of services, calendar,
maintenance services, email

1722
01:20:35,530 --> 01:20:38,710
services, voicemail services.

1723
01:20:38,710 --> 01:20:41,560
It's a world in which
there are devices which

1724
01:20:41,560 --> 01:20:43,250
you use as user interfaces.

1725
01:20:43,250 --> 01:20:45,580
There are devices
you carry with you.

1726
01:20:45,580 --> 01:20:49,120
The Handy 21 is an example
of a device you carry.

1727
01:20:49,120 --> 01:20:51,820
By the way, this
is not a Handy 21.

1728
01:20:51,820 --> 01:20:53,350
This is not even
the eventual device

1729
01:20:53,350 --> 01:20:54,760
that my group is building.

1730
01:20:54,760 --> 01:20:57,370
This is a much bigger mockup
that you can use to play with.

1731
01:20:57,370 --> 01:20:58,960
Don't be put off by
the size of this.

1732
01:20:58,960 --> 01:21:03,500


1733
01:21:03,500 --> 01:21:05,810
There are devices
you carry with you.

1734
01:21:05,810 --> 01:21:08,520
There are devices you
encounter, like Enviro 21s.

1735
01:21:08,520 --> 01:21:11,150


1736
01:21:11,150 --> 01:21:13,040
The applications that
you choose to run

1737
01:21:13,040 --> 01:21:17,240
are constantly adapting
to these new devices.

1738
01:21:17,240 --> 01:21:19,190
As you walk into a
room, the application

1739
01:21:19,190 --> 01:21:21,800
has to reconfigure itself to
use the different devices you've

1740
01:21:21,800 --> 01:21:23,037
encountered.

1741
01:21:23,037 --> 01:21:25,370
And then there are other
devices that you don't actually

1742
01:21:25,370 --> 01:21:27,320
wish to make a part of
your application world.

1743
01:21:27,320 --> 01:21:28,910
You really want to
deal with them somehow.

1744
01:21:28,910 --> 01:21:31,040
You want to make the Coke
machine give you a Coke.

1745
01:21:31,040 --> 01:21:33,690
You don't want it to answer your
email because you're near it.

1746
01:21:33,690 --> 01:21:35,398
But you'd like to talk
to it in some way.

1747
01:21:35,398 --> 01:21:37,730
So you are constantly
walking through the world,

1748
01:21:37,730 --> 01:21:39,650
encountering these
devices, finding

1749
01:21:39,650 --> 01:21:42,590
the linkage between their
manifestation in real space

1750
01:21:42,590 --> 01:21:44,510
and the manifestation
in cyberspace,

1751
01:21:44,510 --> 01:21:47,600
and dealing with them using
the things you carry with you,

1752
01:21:47,600 --> 01:21:49,580
your interface devices.

1753
01:21:49,580 --> 01:21:51,920
And that is the
post-PC world that I

1754
01:21:51,920 --> 01:21:53,270
think we're about to move into.

1755
01:21:53,270 --> 01:21:56,810


1756
01:21:56,810 --> 01:21:57,995
14, no, I overshot.

1757
01:21:57,995 --> 01:22:09,660


1758
01:22:09,660 --> 01:22:16,970
David Clark, father
of the Token Ring,

1759
01:22:16,970 --> 01:22:20,060
said that future devices
won't have keyboards

1760
01:22:20,060 --> 01:22:23,360
if you believe Victor Zue.

1761
01:22:23,360 --> 01:22:26,510
Well, I do believe Victor Zue.

1762
01:22:26,510 --> 01:22:29,750
As Michael explained, the
natural ease of use of Oxygen

1763
01:22:29,750 --> 01:22:32,570
will be based on
built-in speech.

1764
01:22:32,570 --> 01:22:35,450
And we have a well-known
pioneer in interactive dialogue

1765
01:22:35,450 --> 01:22:38,750
systems, Associate
Director of LCS,

1766
01:22:38,750 --> 01:22:41,810
head of the laboratory Spoken
Language Systems Group,

1767
01:22:41,810 --> 01:22:45,410
and player of a mean
guitar, welcome Victor Zue.

1768
01:22:45,410 --> 01:22:52,910


1769
01:22:52,910 --> 01:22:55,520
Thank you, Bob.

1770
01:22:55,520 --> 01:22:58,400
Everybody wants to talk
to their computers.

1771
01:22:58,400 --> 01:23:01,610
For decades, science
fiction writers,

1772
01:23:01,610 --> 01:23:05,930
Hollywood has been
telling us, speech is it.

1773
01:23:05,930 --> 01:23:08,550
Is the ultimate interface.

1774
01:23:08,550 --> 01:23:10,510
Why is it?

1775
01:23:10,510 --> 01:23:12,940
I think part of this
is our strong desire

1776
01:23:12,940 --> 01:23:15,460
to make computers
anthropomorphic.

1777
01:23:15,460 --> 01:23:18,650
We want to make computers
behave like human beings.

1778
01:23:18,650 --> 01:23:21,490
But I think it's more than that.

1779
01:23:21,490 --> 01:23:22,960
Speech is natural.

1780
01:23:22,960 --> 01:23:26,610
We know how to speak before
we know how to read and write.

1781
01:23:26,610 --> 01:23:29,820
Speech is efficient.

1782
01:23:29,820 --> 01:23:32,070
Everybody can speak
about five times faster

1783
01:23:32,070 --> 01:23:34,770
than they can type, and probably
10 times faster than they

1784
01:23:34,770 --> 01:23:37,370
can write.

1785
01:23:37,370 --> 01:23:39,060
And speech is flexible.

1786
01:23:39,060 --> 01:23:40,440
We don't have to touch anything.

1787
01:23:40,440 --> 01:23:44,250
We don't have to see anything.

1788
01:23:44,250 --> 01:23:47,040
Now, over the last
two or three decades,

1789
01:23:47,040 --> 01:23:52,610
we're seeing the technology
for speech maturing such

1790
01:23:52,610 --> 01:23:55,940
that the first generation
of speech-based interfaces

1791
01:23:55,940 --> 01:23:58,010
are beginning to emerge.

1792
01:23:58,010 --> 01:24:00,830
And most of these interfaces
are based on the technology

1793
01:24:00,830 --> 01:24:03,210
of speech recognition.

1794
01:24:03,210 --> 01:24:05,880
So where are we now?

1795
01:24:05,880 --> 01:24:09,180
We're seeing high-performance,
speaker-independent speech

1796
01:24:09,180 --> 01:24:11,640
technology becoming possible.

1797
01:24:11,640 --> 01:24:15,600
For very large vocabularies
of tens of thousands of words,

1798
01:24:15,600 --> 01:24:17,820
you can use it if you
are a cooperative speaker

1799
01:24:17,820 --> 01:24:20,440
in a benign environment.

1800
01:24:20,440 --> 01:24:22,990
For moderate vocabulary
tasks, the system

1801
01:24:22,990 --> 01:24:25,960
can accept extemporaneously
generated speech

1802
01:24:25,960 --> 01:24:27,520
over the telephone.

1803
01:24:27,520 --> 01:24:31,570
In fact, you can go to
various computer stores

1804
01:24:31,570 --> 01:24:35,980
and buy speech-recognition
software for dictation.

1805
01:24:35,980 --> 01:24:40,180
Products offered by IBM,
Dragon Systems, [INAUDIBLE],,

1806
01:24:40,180 --> 01:24:41,500
and Philips.

1807
01:24:41,500 --> 01:24:43,660
You're also seeing
or experiencing

1808
01:24:43,660 --> 01:24:47,860
telephone-based transactions
pioneered by AT&T Bell Labs,

1809
01:24:47,860 --> 01:24:50,390
and now pursued by a
number of other places,

1810
01:24:50,390 --> 01:24:54,170
including Nuance,
Philips, and SpeechWorks.

1811
01:24:54,170 --> 01:24:57,340
So the bottom line is the
current-day day technology,

1812
01:24:57,340 --> 01:24:59,740
when well matched
to applications,

1813
01:24:59,740 --> 01:25:02,500
you can actually
put to good use.

1814
01:25:02,500 --> 01:25:04,600
But the point I would
like to make today

1815
01:25:04,600 --> 01:25:08,240
is that Oxygen needs far more.

1816
01:25:08,240 --> 01:25:09,710
So what do we need?

1817
01:25:09,710 --> 01:25:12,710
I think in the next generation,
our computer interfaces

1818
01:25:12,710 --> 01:25:15,170
are going to be the
kind of interfaces

1819
01:25:15,170 --> 01:25:19,670
that allow you to communicate
with computers very much

1820
01:25:19,670 --> 01:25:22,190
the same way you would
communicate with another human

1821
01:25:22,190 --> 01:25:23,300
being.

1822
01:25:23,300 --> 01:25:28,130
And therefore, the notion of
conversation is very important.

1823
01:25:28,130 --> 01:25:30,470
On the input side,
the machine must

1824
01:25:30,470 --> 01:25:33,950
be able to understand
verbal input.

1825
01:25:33,950 --> 01:25:37,280
That means the technology of
speech recognition converting

1826
01:25:37,280 --> 01:25:41,510
signal to symbols must be
augmented with language

1827
01:25:41,510 --> 01:25:44,620
understanding technology.

1828
01:25:44,620 --> 01:25:47,620
And you must be able to
understand within context.

1829
01:25:47,620 --> 01:25:50,140
So when someone
says, how far is it?

1830
01:25:50,140 --> 01:25:53,020
How much does it cost?

1831
01:25:53,020 --> 01:25:54,480
What about Tokyo, Japan?

1832
01:25:54,480 --> 01:25:56,850
Things like that you will
be able to understand

1833
01:25:56,850 --> 01:25:59,510
as part of a conversation.

1834
01:25:59,510 --> 01:26:01,430
On the output side,
the machine must

1835
01:26:01,430 --> 01:26:04,100
be able to verbalize
the information.

1836
01:26:04,100 --> 01:26:07,040
You have to take
documents from the web,

1837
01:26:07,040 --> 01:26:09,290
tabular form the
information, turn it

1838
01:26:09,290 --> 01:26:12,110
into well-formed sentences,
and eventually, utter

1839
01:26:12,110 --> 01:26:16,235
these sentences, perhaps
over the telephone.

1840
01:26:16,235 --> 01:26:17,610
Throughout this
process, you must

1841
01:26:17,610 --> 01:26:20,370
be able to engage in
dialogue so the machine

1842
01:26:20,370 --> 01:26:23,640
will be able to clarify
mistakes you might have made.

1843
01:26:23,640 --> 01:26:27,390
Did you say Boston,
Massachusetts or Austin, Texas?

1844
01:26:27,390 --> 01:26:29,850
And it'll try to help
you negotiate such

1845
01:26:29,850 --> 01:26:31,950
that you can conclude
a transaction

1846
01:26:31,950 --> 01:26:34,950
in a meaningful and
efficient manner.

1847
01:26:34,950 --> 01:26:37,147
Now, for the last decade,
we at the Laboratory

1848
01:26:37,147 --> 01:26:38,730
for Computer Science
have been working

1849
01:26:38,730 --> 01:26:42,030
on this kind of an interface,
conversational interface.

1850
01:26:42,030 --> 01:26:44,610
And this point, we're
seeing the emergence

1851
01:26:44,610 --> 01:26:47,490
of these kind of
technologies and systems

1852
01:26:47,490 --> 01:26:50,220
that can recognize
what a person said,

1853
01:26:50,220 --> 01:26:54,450
understand the meaning of
the utterances in context,

1854
01:26:54,450 --> 01:26:58,170
take appropriate
actions, and respond.

1855
01:26:58,170 --> 01:27:01,920
But as many of us have said this
morning, these kind of systems

1856
01:27:01,920 --> 01:27:04,620
are unfortunately kind of dumb.

1857
01:27:04,620 --> 01:27:08,550
They can only deal with limited
domains, such as weather,

1858
01:27:08,550 --> 01:27:12,810
flights, traffic, navigation,
banking, electronic commerce,

1859
01:27:12,810 --> 01:27:14,430
individual domains.

1860
01:27:14,430 --> 01:27:17,880
It is not omnipotent.

1861
01:27:17,880 --> 01:27:20,820
Nonetheless, they can
deal with real domains.

1862
01:27:20,820 --> 01:27:22,330
The applications are real.

1863
01:27:22,330 --> 01:27:24,510
The information is up-to-date.

1864
01:27:24,510 --> 01:27:28,640
And you can access this
information over the telephone.

1865
01:27:28,640 --> 01:27:30,230
It deals with
multiple languages.

1866
01:27:30,230 --> 01:27:32,360
The three that we paid
the most attention to

1867
01:27:32,360 --> 01:27:37,160
has been American English,
Spanish, and Mandarin Chinese,

1868
01:27:37,160 --> 01:27:39,170
among others.

1869
01:27:39,170 --> 01:27:42,080
And these systems can
perform information access

1870
01:27:42,080 --> 01:27:46,430
in almost real time, the
wait for the worldwide web

1871
01:27:46,430 --> 01:27:48,230
notwithstanding.

1872
01:27:48,230 --> 01:27:51,080
And using standard
PCs, machines come,

1873
01:27:51,080 --> 01:27:54,380
you pour software
into it, and it works.

1874
01:27:54,380 --> 01:27:56,180
The applications I'm
going to show you,

1875
01:27:56,180 --> 01:27:58,490
and the applications we
have been developing,

1876
01:27:58,490 --> 01:28:00,560
are all using an
architecture called

1877
01:28:00,560 --> 01:28:04,682
Galaxy, which was introduced
five years ago in our group.

1878
01:28:04,682 --> 01:28:06,140
And I just want to
quickly tell you

1879
01:28:06,140 --> 01:28:09,210
about some of the features
of this architecture.

1880
01:28:09,210 --> 01:28:12,540
First of all, it makes use of
a distributor architecture.

1881
01:28:12,540 --> 01:28:15,260
So you can have a very
lightweight client.

1882
01:28:15,260 --> 01:28:17,070
And the client can
be so lightweight,

1883
01:28:17,070 --> 01:28:18,800
it is only a telephone.

1884
01:28:18,800 --> 01:28:21,110
And the compute and
knowledge-intensive actions

1885
01:28:21,110 --> 01:28:24,440
are all taken on remote servers.

1886
01:28:24,440 --> 01:28:28,280
Galaxy can deal with
multiple domains of knowledge

1887
01:28:28,280 --> 01:28:30,260
and can access
them transparently.

1888
01:28:30,260 --> 01:28:33,660
And you can add
more, new domains.

1889
01:28:33,660 --> 01:28:35,700
It can deal with
multiple modalities.

1890
01:28:35,700 --> 01:28:38,700
You can point, click, type.

1891
01:28:38,700 --> 01:28:42,380
I can talk to my machines
using a telephone.

1892
01:28:42,380 --> 01:28:44,080
So what I'd like to
do is to give you

1893
01:28:44,080 --> 01:28:46,720
a very brief demonstration
in terms of where we are

1894
01:28:46,720 --> 01:28:48,610
and what we can do.

1895
01:28:48,610 --> 01:28:52,420
The first demonstration is
going to be without display.

1896
01:28:52,420 --> 01:28:56,230
In its simplest form, what the
Galaxy system can deal with

1897
01:28:56,230 --> 01:29:00,140
is a client, which is nothing
more than a telephone.

1898
01:29:00,140 --> 01:29:05,620
So telephone-based interaction
without any display

1899
01:29:05,620 --> 01:29:09,010
poses new challenges,
because if you only

1900
01:29:09,010 --> 01:29:10,810
have the auditory
channel, then you

1901
01:29:10,810 --> 01:29:13,570
can only absorb a small
amount of information

1902
01:29:13,570 --> 01:29:17,680
at a time, which means
that a dialogue management

1903
01:29:17,680 --> 01:29:21,910
component must work concurrently
with, cooperatively with,

1904
01:29:21,910 --> 01:29:24,790
language generation, so that
you can get information down

1905
01:29:24,790 --> 01:29:27,610
to a digestible
size for the user.

1906
01:29:27,610 --> 01:29:29,080
And the second
thing is, of course,

1907
01:29:29,080 --> 01:29:33,650
better-quality computer voice
response will be necessary.

1908
01:29:33,650 --> 01:29:36,160
We have been working in
the synthesis area also.

1909
01:29:36,160 --> 01:29:38,390
But for the time being,
the demonstration

1910
01:29:38,390 --> 01:29:40,000
I'm going to show
you is actually

1911
01:29:40,000 --> 01:29:42,970
using a commercially
off-the-shelf synthesizer

1912
01:29:42,970 --> 01:29:46,810
from Compaq, formerly
Digital Equipment.

1913
01:29:46,810 --> 01:29:49,090
We have developed
several applications,

1914
01:29:49,090 --> 01:29:51,100
about half a dozen or so.

1915
01:29:51,100 --> 01:29:53,020
And there is a
system called Jupiter

1916
01:29:53,020 --> 01:29:56,260
that can deal with worldwide
weather information for 500

1917
01:29:56,260 --> 01:29:57,850
cities worldwide.

1918
01:29:57,850 --> 01:30:00,430
And Pegasus is a
system that can deal

1919
01:30:00,430 --> 01:30:03,730
with 4,000 flights flying in
the United States airspace

1920
01:30:03,730 --> 01:30:04,870
every day.

1921
01:30:04,870 --> 01:30:07,540
And we update the information
every two or three minutes.

1922
01:30:07,540 --> 01:30:10,040
And we don't lie to you.

1923
01:30:10,040 --> 01:30:13,190
Voyager can deal with
traffic and navigation.

1924
01:30:13,190 --> 01:30:15,770
And again, the information
about greater Boston traffic

1925
01:30:15,770 --> 01:30:19,210
comes in every
two minutes or so.

1926
01:30:19,210 --> 01:30:22,060
And one thing you'll
notice later on

1927
01:30:22,060 --> 01:30:24,790
is that in order for you to move
from one domain to the other,

1928
01:30:24,790 --> 01:30:26,650
you must explicitly switch.

1929
01:30:26,650 --> 01:30:28,300
I want to talk to Jupiter now.

1930
01:30:28,300 --> 01:30:31,060
I want to talk to
Voyager now, et cetera.

1931
01:30:31,060 --> 01:30:32,760
Well, because of
time constraints,

1932
01:30:32,760 --> 01:30:36,340
what I'm going to do is just
show you one demonstration

1933
01:30:36,340 --> 01:30:37,870
using Jupiter system.

1934
01:30:37,870 --> 01:30:39,370
Let's see whether
the machine is up.

1935
01:30:39,370 --> 01:30:40,703
Let me apologize for two things.

1936
01:30:40,703 --> 01:30:42,537
We don't have that many
phone lines and this

1937
01:30:42,537 --> 01:30:43,510
is a popular system.

1938
01:30:43,510 --> 01:30:44,920
Oftentimes people are calling.

1939
01:30:44,920 --> 01:30:47,530
If I get a busy
signal, I apologize.

1940
01:30:47,530 --> 01:30:51,160
And the second thing is in
the auditorium with feedback,

1941
01:30:51,160 --> 01:30:53,830
sometimes the system
may not work so well.

1942
01:30:53,830 --> 01:30:54,330
Let me try.

1943
01:30:54,330 --> 01:31:02,760


1944
01:31:02,760 --> 01:31:06,540
Hi, welcome to Jupiter, the
MIT Lab for Computer Science

1945
01:31:06,540 --> 01:31:08,220
weather information system.

1946
01:31:08,220 --> 01:31:11,070
This call is being recorded
for system development.

1947
01:31:11,070 --> 01:31:13,810
You may hang up or ask
for help at any time.

1948
01:31:13,810 --> 01:31:15,960
How can I help you?

1949
01:31:15,960 --> 01:31:17,960
What is the forecast
for Boston today?

1950
01:31:17,960 --> 01:31:27,520


1951
01:31:27,520 --> 01:31:30,790
In Boston this afternoon,
partly sunny and windy.

1952
01:31:30,790 --> 01:31:32,200
High around 50.

1953
01:31:32,200 --> 01:31:33,850
Tonight, clear.

1954
01:31:33,850 --> 01:31:35,530
Lows around 35.

1955
01:31:35,530 --> 01:31:38,680
What other information
can I give you?

1956
01:31:38,680 --> 01:31:39,690
Will it rain tomorrow?

1957
01:31:39,690 --> 01:31:45,010


1958
01:31:45,010 --> 01:31:47,890
To my knowledge, the forecast
calls for no rain in Boston

1959
01:31:47,890 --> 01:31:48,640
tomorrow.

1960
01:31:48,640 --> 01:31:51,340
Is there anything else?

1961
01:31:51,340 --> 01:31:52,120
How about Seattle?

1962
01:31:52,120 --> 01:31:54,960


1963
01:31:54,960 --> 01:31:57,480
The answer should be yes.

1964
01:31:57,480 --> 01:32:00,330
To my knowledge, the forecast
calls for no rain in Seattle

1965
01:32:00,330 --> 01:32:01,050
tomorrow.

1966
01:32:01,050 --> 01:32:04,230
Is there something else
you'd like to know?

1967
01:32:04,230 --> 01:32:06,450
Is it sunny anywhere
in the Caribbean?

1968
01:32:06,450 --> 01:32:10,950


1969
01:32:10,950 --> 01:32:15,000
Sorry, I'm not
sure what you said.

1970
01:32:15,000 --> 01:32:17,150
Is it sunny anywhere
in the Caribbean?

1971
01:32:17,150 --> 01:32:20,910


1972
01:32:20,910 --> 01:32:22,480
I am sorry.

1973
01:32:22,480 --> 01:32:25,000
I may not know the city
you're asking about.

1974
01:32:25,000 --> 01:32:26,740
Try asking for the
state or country.

1975
01:32:26,740 --> 01:32:31,030
For example, what cities do you
know about in Massachusetts?

1976
01:32:31,030 --> 01:32:33,150
What is the temperature
in Montego Bay?

1977
01:32:33,150 --> 01:32:38,540


1978
01:32:38,540 --> 01:32:42,110
In Montego Bay today,
high 86 and low 73.

1979
01:32:42,110 --> 01:32:44,340
Is this something else?

1980
01:32:44,340 --> 01:32:44,840
Goodbye.

1981
01:32:44,840 --> 01:32:54,300


1982
01:32:54,300 --> 01:32:57,750
This is a system that's
up almost all the time.

1983
01:32:57,750 --> 01:32:59,250
We will give you a
toll-free number.

1984
01:32:59,250 --> 01:33:02,460
You can call and donate
your voice to science,

1985
01:33:02,460 --> 01:33:05,330
because we record your voice.

1986
01:33:05,330 --> 01:33:07,580
Anyway, this is an example
of the kind of system

1987
01:33:07,580 --> 01:33:10,040
that's going to come
out pretty soon.

1988
01:33:10,040 --> 01:33:12,890
And I want to tell you
that I've been working

1989
01:33:12,890 --> 01:33:15,320
all my life in speech,
and it's logical

1990
01:33:15,320 --> 01:33:18,350
that I would think speech is
the solution to all problems.

1991
01:33:18,350 --> 01:33:21,960
If you have a hammer,
everything looks like nails.

1992
01:33:21,960 --> 01:33:24,390
But even I have to admit,
sometimes a picture

1993
01:33:24,390 --> 01:33:26,260
is kind of nice.

1994
01:33:26,260 --> 01:33:28,830
It can convey information
not available otherwise.

1995
01:33:28,830 --> 01:33:31,170
And sometimes typing,
pointing, and clicking,

1996
01:33:31,170 --> 01:33:35,320
all those kind of things,
can augment a speech input.

1997
01:33:35,320 --> 01:33:41,640
So, in fact, our system can
deal with a system with display.

1998
01:33:41,640 --> 01:33:44,430
And I'm going to show you a
brief video clip which will

1999
01:33:44,430 --> 01:33:48,050
show you how this thing works.

2000
01:33:48,050 --> 01:33:48,800
Let's take a look.

2001
01:33:48,800 --> 01:33:55,110


2002
01:33:55,110 --> 01:33:57,707
What happened?

2003
01:33:57,707 --> 01:33:59,290
Maybe I won't show
you the video clip.

2004
01:33:59,290 --> 01:34:03,670


2005
01:34:03,670 --> 01:34:04,210
Let's see.

2006
01:34:04,210 --> 01:34:08,110


2007
01:34:08,110 --> 01:34:09,850
It's not my fault.

2008
01:34:09,850 --> 01:34:13,907
Anyway, the good
news is that, you

2009
01:34:13,907 --> 01:34:16,420
can please check it out this
afternoon at the reception.

2010
01:34:16,420 --> 01:34:20,980


2011
01:34:20,980 --> 01:34:22,630
This will give
you an opportunity

2012
01:34:22,630 --> 01:34:25,310
to see the system
work in real life.

2013
01:34:25,310 --> 01:34:26,665
So let's try to do that.

2014
01:34:26,665 --> 01:34:28,540
Otherwise, I'm going to
get the hook from Bob

2015
01:34:28,540 --> 01:34:31,960
if I were to figure out what's
wrong with this display.

2016
01:34:31,960 --> 01:34:35,140
I want to spend the remainder
of my time trying to tie speech

2017
01:34:35,140 --> 01:34:39,270
to Oxygen. Now, as
Michael has said,

2018
01:34:39,270 --> 01:34:42,000
Oxygen has set forth
a radically new way

2019
01:34:42,000 --> 01:34:44,100
of human-machine interaction.

2020
01:34:44,100 --> 01:34:48,890
And it's going to be natural
using speech, no keyboard.

2021
01:34:48,890 --> 01:34:50,240
It's going to be mobile.

2022
01:34:50,240 --> 01:34:52,910
It's going to connect to
the rest of the world using

2023
01:34:52,910 --> 01:34:54,230
wireless networks.

2024
01:34:54,230 --> 01:34:57,620
And it's going to have
lightweight, hand-held devices

2025
01:34:57,620 --> 01:35:00,740
that will enable you to
communicate with machines

2026
01:35:00,740 --> 01:35:02,640
and with the environment.

2027
01:35:02,640 --> 01:35:05,450
I just want to mention that
the first-generation Handy

2028
01:35:05,450 --> 01:35:10,400
21 is under active development
on a separate DARPA project.

2029
01:35:10,400 --> 01:35:15,120
And the idea is that you are
going to have handheld devices.

2030
01:35:15,120 --> 01:35:17,570
And this is probably
the final version

2031
01:35:17,570 --> 01:35:23,280
of the design with a little bit
of navigation with the thumb.

2032
01:35:23,280 --> 01:35:24,830
And then you have camera.

2033
01:35:24,830 --> 01:35:29,060
And also you are going to have
speech interface and display.

2034
01:35:29,060 --> 01:35:31,040
It's going to go
through wireless network

2035
01:35:31,040 --> 01:35:35,505
and have the computing
done on remote servers.

2036
01:35:35,505 --> 01:35:36,880
Now, so far, I
have been spending

2037
01:35:36,880 --> 01:35:40,480
a lot of time talking about
speech as an interface,

2038
01:35:40,480 --> 01:35:44,450
but I really want to caution
you that spoken language,

2039
01:35:44,450 --> 01:35:48,280
the power of spoken language,
goes far beyond interfaces.

2040
01:35:48,280 --> 01:35:51,250
Increasingly, we're
seeing information

2041
01:35:51,250 --> 01:35:55,180
on the web in the form of audio,
radio broadcasts, for example,

2042
01:35:55,180 --> 01:35:59,660
and video, movies, and
video broadcasts, and also

2043
01:35:59,660 --> 01:36:01,970
a summary of meeting
notes, for example.

2044
01:36:01,970 --> 01:36:06,670
So in fact, we must
be able to index

2045
01:36:06,670 --> 01:36:10,570
the content in the form
of spoken language.

2046
01:36:10,570 --> 01:36:13,920
And what's even more important
is the fact that language

2047
01:36:13,920 --> 01:36:17,280
is what enables us to
communicate in such a subtle

2048
01:36:17,280 --> 01:36:19,670
and also efficient manner.

2049
01:36:19,670 --> 01:36:22,040
If I were to talk
to my computer--

2050
01:36:22,040 --> 01:36:27,280
let me know when the
Microsoft stock is above $160

2051
01:36:27,280 --> 01:36:31,240
or, call me half an hour before
the flight is going to arrive--

2052
01:36:31,240 --> 01:36:35,050
things like that, you can use
very simple human language

2053
01:36:35,050 --> 01:36:38,470
to convey the kind of commands
and have the computer really

2054
01:36:38,470 --> 01:36:42,070
serving as an agent to
accomplish tasks for you.

2055
01:36:42,070 --> 01:36:43,390
So look at it this way.

2056
01:36:43,390 --> 01:36:46,690
I think it's important to
realize that speech will

2057
01:36:46,690 --> 01:36:51,190
be a crucial part of the
infrastructure for Oxygen.

2058
01:36:51,190 --> 01:36:54,640
There are a lot of research
problems we need to deal with.

2059
01:36:54,640 --> 01:36:58,930
Now, we can deal with a
few domains, simple domains

2060
01:36:58,930 --> 01:36:59,950
and a few.

2061
01:36:59,950 --> 01:37:02,920
We want to be able to
deal with many domains

2062
01:37:02,920 --> 01:37:05,080
and complex domains.

2063
01:37:05,080 --> 01:37:10,630
You want to be able to
navigate transparently.

2064
01:37:10,630 --> 01:37:13,240
At the moment, you have to
say, I'm going from one domain

2065
01:37:13,240 --> 01:37:14,110
to the other.

2066
01:37:14,110 --> 01:37:17,110
And I'm going to have to
specifically say that.

2067
01:37:17,110 --> 01:37:18,730
When you have
multiple domains, this

2068
01:37:18,730 --> 01:37:21,200
will simply not be possible.

2069
01:37:21,200 --> 01:37:22,848
So what we want to
do, as Michael says,

2070
01:37:22,848 --> 01:37:24,640
is to take [? narrow ?]
domains, treat them

2071
01:37:24,640 --> 01:37:29,760
like little pieces of cloth,
and try to stitch it together

2072
01:37:29,760 --> 01:37:34,070
into a quilt. Allow people
to navigate transparently.

2073
01:37:34,070 --> 01:37:36,590
We can deal with a few
languages of the world,

2074
01:37:36,590 --> 01:37:40,400
and we must be able to deal
with many other languages.

2075
01:37:40,400 --> 01:37:42,200
At the moment,
pointing, clicking,

2076
01:37:42,200 --> 01:37:46,070
and those kind of things
are relatively primitive,

2077
01:37:46,070 --> 01:37:47,030
the coupling.

2078
01:37:47,030 --> 01:37:49,640
We want them to be
fully integrated.

2079
01:37:49,640 --> 01:37:53,610
And finally, if you really
want to use language,

2080
01:37:53,610 --> 01:37:58,170
then the machine must be able
to do more than just what

2081
01:37:58,170 --> 01:38:00,810
I say, but understand
the intention

2082
01:38:00,810 --> 01:38:04,170
and be able to do what I mean.

2083
01:38:04,170 --> 01:38:06,840
So these are some of the
important research issues

2084
01:38:06,840 --> 01:38:09,160
that we must address.

2085
01:38:09,160 --> 01:38:11,470
And I just want
to close by saying

2086
01:38:11,470 --> 01:38:15,550
that speech-based interfaces are
an integral part of Oxygen. It

2087
01:38:15,550 --> 01:38:18,590
is almost inevitable
for three reasons.

2088
01:38:18,590 --> 01:38:21,190
First of all, we
need to be mobile

2089
01:38:21,190 --> 01:38:23,540
and yet, we need
to be connected.

2090
01:38:23,540 --> 01:38:25,990
You can't carry a
keyboard around.

2091
01:38:25,990 --> 01:38:28,510
Machines are getting
smaller and smaller.

2092
01:38:28,510 --> 01:38:30,910
The only way you can type
to write in the future

2093
01:38:30,910 --> 01:38:33,130
will be using toothpicks.

2094
01:38:33,130 --> 01:38:38,080
Last, but not least,
we love to talk.

2095
01:38:38,080 --> 01:38:41,200
Tomorrow's interfaces,
however, must

2096
01:38:41,200 --> 01:38:45,040
be able to understand
and respond in context.

2097
01:38:45,040 --> 01:38:49,600
They must be able to
incorporate multiple modalities,

2098
01:38:49,600 --> 01:38:52,760
be increasingly untethered.

2099
01:38:52,760 --> 01:38:57,120
And speech is more
than an interface.

2100
01:38:57,120 --> 01:39:00,780
It is part of the
Oxygen infrastructure.

2101
01:39:00,780 --> 01:39:04,170
And there are a lot of problems
that need to be solved.

2102
01:39:04,170 --> 01:39:06,840
As my friend Fred
Jelinek once said,

2103
01:39:06,840 --> 01:39:10,370
it's going to keep my
family fed for many years.

2104
01:39:10,370 --> 01:39:10,870
Thank you.

2105
01:39:10,870 --> 01:39:28,020


2106
01:39:28,020 --> 01:39:28,520
Michael.

2107
01:39:28,520 --> 01:39:31,820


2108
01:39:31,820 --> 01:39:35,120
As we reach the midpoint
between the proximity talks,

2109
01:39:35,120 --> 01:39:37,370
I wanted to remind
you and tell you

2110
01:39:37,370 --> 01:39:39,140
something I didn't
tell you earlier,

2111
01:39:39,140 --> 01:39:44,180
namely, that this project,
for the first time

2112
01:39:44,180 --> 01:39:47,780
since timesharing, Multics,
and CTSS that you'll remember,

2113
01:39:47,780 --> 01:39:51,630
has coalesced some 30 members
of our faculty together.

2114
01:39:51,630 --> 01:39:53,690
So it's very, very widespread.

2115
01:39:53,690 --> 01:39:57,200
And we're collaborating with
our sister lab, the Artificial

2116
01:39:57,200 --> 01:40:00,000
Intelligence Laboratory,
who is extending

2117
01:40:00,000 --> 01:40:03,570
our speech into
vision, smart rooms,

2118
01:40:03,570 --> 01:40:05,820
and a lot of other things
we didn't talk this morning.

2119
01:40:05,820 --> 01:40:07,862
But we've written this
up, you can read about it,

2120
01:40:07,862 --> 01:40:09,920
and I just wanted to
remind you of that it's

2121
01:40:09,920 --> 01:40:11,240
a very, very large project.

2122
01:40:11,240 --> 01:40:14,330
Thank you, Bob.

2123
01:40:14,330 --> 01:40:18,080
I can only wonder what Oxygen
will think the next time

2124
01:40:18,080 --> 01:40:20,745
someone says that they have
excess gates coming out

2125
01:40:20,745 --> 01:40:21,245
their ears.

2126
01:40:21,245 --> 01:40:25,010


2127
01:40:25,010 --> 01:40:26,430
We're about to go to lunch.

2128
01:40:26,430 --> 01:40:28,430
We're going to have lunch
in the Johnson Center.

2129
01:40:28,430 --> 01:40:29,847
The Johnson Center,
by the way, is

2130
01:40:29,847 --> 01:40:32,390
where we're going to continue
this meeting tomorrow morning

2131
01:40:32,390 --> 01:40:35,360
at 8:30 on Vassar Street.

2132
01:40:35,360 --> 01:40:37,790
Tomorrow morning at
8:30 on Vassar Street

2133
01:40:37,790 --> 01:40:39,515
at Johnson Center.

2134
01:40:39,515 --> 01:40:41,390
And it's going to be
complicated to get there

2135
01:40:41,390 --> 01:40:45,170
because of the security
arrangements necessary for me

2136
01:40:45,170 --> 01:40:46,400
and the premier of China.

2137
01:40:46,400 --> 01:40:49,640


2138
01:40:49,640 --> 01:40:53,120
At lunch today, there are
110 tables, each of which

2139
01:40:53,120 --> 01:40:54,015
has a host.

2140
01:40:54,015 --> 01:40:55,640
And I believe you
have in your packages

2141
01:40:55,640 --> 01:40:59,660
this map of the tables.

2142
01:40:59,660 --> 01:41:01,790
And you should choose
someone that you'd

2143
01:41:01,790 --> 01:41:04,640
like to have lunch with, and
try to find room at their table.

2144
01:41:04,640 --> 01:41:08,450
For example, at my
table is number 29.

2145
01:41:08,450 --> 01:41:11,240
So please come over to the
Johnson Center for lunch.

2146
01:41:11,240 --> 01:41:15,115
Let's go promptly, and I'll
see you back here at 2:00 PM.

2147
01:41:15,115 --> 01:41:41,000