1
00:00:00,000 --> 00:00:04,940


2
00:00:04,940 --> 00:00:06,790
Professor at MIT.

3
00:00:06,790 --> 00:00:08,710
That's correct.

4
00:00:08,710 --> 00:00:13,360
So start with the
starting question.

5
00:00:13,360 --> 00:00:15,670
What got you into the field
of artificial intelligence?

6
00:00:15,670 --> 00:00:24,600
I've been interested in how I
think since I was a small kid.

7
00:00:24,600 --> 00:00:28,030
And for many years I was
searching for that field which

8
00:00:28,030 --> 00:00:31,390
would provide some answers to
the questions I was asking.

9
00:00:31,390 --> 00:00:34,030
So I tried neurobiology.

10
00:00:34,030 --> 00:00:35,800
I tried information theory.

11
00:00:35,800 --> 00:00:39,040
And then one day I attended
a lecture by Marvin Minsky,

12
00:00:39,040 --> 00:00:41,110
and that was the
end of the story.

13
00:00:41,110 --> 00:00:41,920
I knew where I was.

14
00:00:41,920 --> 00:00:44,660


15
00:00:44,660 --> 00:00:48,360
I guess, what was not there
in the other disciplines

16
00:00:48,360 --> 00:00:49,450
you investigated?

17
00:00:49,450 --> 00:00:53,410
A computational understanding
of the nature of intelligence.

18
00:00:53,410 --> 00:00:57,100
Not just a catalog of behavior,
but an explanation of it.

19
00:00:57,100 --> 00:00:59,470
Something that reaches
all the way down

20
00:00:59,470 --> 00:01:01,990
from the highest
levels of reflective

21
00:01:01,990 --> 00:01:06,640
thought to the kind of thinking
that's closely aligned with

22
00:01:06,640 --> 00:01:10,210
and touching of our perceptions.

23
00:01:10,210 --> 00:01:13,390
OK, that makes sense.

24
00:01:13,390 --> 00:01:15,478
So you took a course
with Marvin Minsky?

25
00:01:15,478 --> 00:01:16,770
Or you just attended a lecture?

26
00:01:16,770 --> 00:01:18,082
What course?

27
00:01:18,082 --> 00:01:19,540
Oh, I've forgotten
the name, but it

28
00:01:19,540 --> 00:01:21,910
was the graduate artificial
intelligence course,

29
00:01:21,910 --> 00:01:25,990
which I subsequently translated
into an undergraduate version

30
00:01:25,990 --> 00:01:27,700
and have been
teaching ever since.

31
00:01:27,700 --> 00:01:34,180
Cool, so I guess you ended
up working in the AI lab?

32
00:01:34,180 --> 00:01:40,342
Yes, on the basis of the
term paper I wrote for that--

33
00:01:40,342 --> 00:01:41,800
something that
Marvin was teaching,

34
00:01:41,800 --> 00:01:46,370
he invited me to
spend the summer

35
00:01:46,370 --> 00:01:47,990
in the artificial
Intelligence lab,

36
00:01:47,990 --> 00:01:50,660
and I was of course
thrilled beyond description

37
00:01:50,660 --> 00:01:51,750
and never left.

38
00:01:51,750 --> 00:01:53,750
What sort of things were
going on when you first

39
00:01:53,750 --> 00:01:55,520
arrived at the AI lab?

40
00:01:55,520 --> 00:02:01,260
Building robots,
making computers see,

41
00:02:01,260 --> 00:02:03,810
building descriptions
of things, trying

42
00:02:03,810 --> 00:02:06,090
to understand how
computers can learn

43
00:02:06,090 --> 00:02:08,490
in ways that were
different from the way rats

44
00:02:08,490 --> 00:02:11,008
learn from repeated trials.

45
00:02:11,008 --> 00:02:13,050
All that sort of thing
was tremendously exciting,

46
00:02:13,050 --> 00:02:16,710
and especially so because
the kind of hardware

47
00:02:16,710 --> 00:02:19,560
that people like Tom Knight
and Richard Greenblatt

48
00:02:19,560 --> 00:02:22,720
were building and programming
was unique in the world,

49
00:02:22,720 --> 00:02:25,020
and we felt that nobody
else had anything

50
00:02:25,020 --> 00:02:29,850
like the kinds of tools we
had to work on the problems.

51
00:02:29,850 --> 00:02:32,150
List machines?

52
00:02:32,150 --> 00:02:34,250
Even before list machines, yeah.

53
00:02:34,250 --> 00:02:35,550
List machines came along.

54
00:02:35,550 --> 00:02:38,490
A chess machine was there.

55
00:02:38,490 --> 00:02:44,640
But even the beginning, those
guys knew how to put together.

56
00:02:44,640 --> 00:02:46,170
Didn't have displays
before anybody

57
00:02:46,170 --> 00:02:47,420
had heard of bit map displays.

58
00:02:47,420 --> 00:02:50,540


59
00:02:50,540 --> 00:02:55,490
Ethernet-like connectivity and
email systems, all of which

60
00:02:55,490 --> 00:02:58,592
were pioneered and
championed by the early days

61
00:02:58,592 --> 00:03:00,050
of the artificial
Intelligence lab.

62
00:03:00,050 --> 00:03:02,360
OK, sweet.

63
00:03:02,360 --> 00:03:04,370
I suppose I should add,
especially the time

64
00:03:04,370 --> 00:03:05,058
sharing system.

65
00:03:05,058 --> 00:03:06,600
It was a fabulous
time sharing system

66
00:03:06,600 --> 00:03:08,840
that made it possible
for a lot of people

67
00:03:08,840 --> 00:03:12,200
to do good work on a computer
that by today's standards,

68
00:03:12,200 --> 00:03:14,750
of course, would probably
be about the sort of thing

69
00:03:14,750 --> 00:03:18,030
you would think
to put in a watch.

70
00:03:18,030 --> 00:03:21,030
Fair enough.

71
00:03:21,030 --> 00:03:23,880
So I guess you--

72
00:03:23,880 --> 00:03:27,260
what was your first project
when you went to the AI lab?

73
00:03:27,260 --> 00:03:30,650
Oh, I did some things
on computer vision.

74
00:03:30,650 --> 00:03:34,130
I did an analysis of
what it would look like,

75
00:03:34,130 --> 00:03:36,440
how a computer could
recognize that the objects had

76
00:03:36,440 --> 00:03:38,010
a hole in it.

77
00:03:38,010 --> 00:03:41,390
But pretty quickly I
was on to my PhD thesis

78
00:03:41,390 --> 00:03:44,360
topic, which was on the subject
of how a computer can learn.

79
00:03:44,360 --> 00:03:46,220
Right.

80
00:03:46,220 --> 00:03:51,290
I suppose you probably
haven't seen our project page,

81
00:03:51,290 --> 00:03:54,020
but the first film in
the collection we have

82
00:03:54,020 --> 00:03:57,380
is Professor Minsky
talking about his thesis

83
00:03:57,380 --> 00:03:58,520
with the [INAUDIBLE].

84
00:03:58,520 --> 00:04:01,302


85
00:04:01,302 --> 00:04:02,760
Often is the case
that someone else

86
00:04:02,760 --> 00:04:04,190
can explain it better you can.

87
00:04:04,190 --> 00:04:06,890


88
00:04:06,890 --> 00:04:07,970
Yeah.

89
00:04:07,970 --> 00:04:09,508
Whenever we explain
anything, we're

90
00:04:09,508 --> 00:04:11,300
hallucinating that the
other person already

91
00:04:11,300 --> 00:04:12,800
knows what we're
about to tell them,

92
00:04:12,800 --> 00:04:15,020
so it's not necessarily
the best person

93
00:04:15,020 --> 00:04:16,541
to get the explanation from.

94
00:04:16,541 --> 00:04:17,541
You're right about that.

95
00:04:17,541 --> 00:04:20,430


96
00:04:20,430 --> 00:04:26,010
So I guess, where did
your work go from there?

97
00:04:26,010 --> 00:04:29,810
Well, it's always
been a focus on--

98
00:04:29,810 --> 00:04:31,835
to some extent, accidentally,
I suppose-- always

99
00:04:31,835 --> 00:04:34,340
been focused on problems
having to do with learning.

100
00:04:34,340 --> 00:04:37,900


101
00:04:37,900 --> 00:04:40,750
And to back up a little bit and
add something to that earlier

102
00:04:40,750 --> 00:04:42,500
discussion-- you might
want to add it in--

103
00:04:42,500 --> 00:04:44,830
I think that the
main characteristic

104
00:04:44,830 --> 00:04:47,830
of those early days and in
the Artificial Intelligence

105
00:04:47,830 --> 00:04:51,160
Laboratory was a
tremendous excitement that

106
00:04:51,160 --> 00:04:56,380
was derived partly from
the monster brains who

107
00:04:56,380 --> 00:04:59,440
are working there, and partly
from the general mission.

108
00:04:59,440 --> 00:05:01,000
We really felt
that we were going

109
00:05:01,000 --> 00:05:03,790
to be able to develop
a computational theory

110
00:05:03,790 --> 00:05:06,610
of intelligence, and do some
very deep scientific work

111
00:05:06,610 --> 00:05:08,920
that no one would ever
be able to do again

112
00:05:08,920 --> 00:05:13,770
because once you discovered
it, you discovered it.

113
00:05:13,770 --> 00:05:14,610
Makes sense.

114
00:05:14,610 --> 00:05:19,560
I guess we've heard a lot about
the atmosphere of the AI lab

115
00:05:19,560 --> 00:05:23,070
in the early days
of its existence.

116
00:05:23,070 --> 00:05:24,990
Do you have any
particular anecdotes

117
00:05:24,990 --> 00:05:28,380
or cool stories you remember
from when you were there?

118
00:05:28,380 --> 00:05:28,880
No.

119
00:05:28,880 --> 00:05:30,040
No?

120
00:05:30,040 --> 00:05:31,460
Just general work.

121
00:05:31,460 --> 00:05:34,200
All of them are--

122
00:05:34,200 --> 00:05:36,004
all of them are good.

123
00:05:36,004 --> 00:05:38,896
Oh, OK.

124
00:05:38,896 --> 00:05:40,350
Fair enough.

125
00:05:40,350 --> 00:05:48,790
So after the arch
thesis, you still

126
00:05:48,790 --> 00:05:51,550
work on sort of
machine learning.

127
00:05:51,550 --> 00:05:54,070
Still do, yes.

128
00:05:54,070 --> 00:05:56,920
Of course, the field
has changed a great deal

129
00:05:56,920 --> 00:05:59,020
since those early days.

130
00:05:59,020 --> 00:06:02,680
Partly a natural
consequence of a field

131
00:06:02,680 --> 00:06:05,920
growing up and maturing,
people working in it,

132
00:06:05,920 --> 00:06:06,730
standards changing.

133
00:06:06,730 --> 00:06:09,550


134
00:06:09,550 --> 00:06:11,410
And to some extent, I
think that the field

135
00:06:11,410 --> 00:06:15,190
has suffered a bit from
the disappointments

136
00:06:15,190 --> 00:06:17,020
of the early days.

137
00:06:17,020 --> 00:06:21,250
The problems proved to be much
harder than anyone anticipated.

138
00:06:21,250 --> 00:06:24,040
Early progress was so rapid
that there was a general sense

139
00:06:24,040 --> 00:06:26,350
that we'd be able to
solve all the problems

140
00:06:26,350 --> 00:06:29,782
within a decade or two, but
we weren't able to do that.

141
00:06:29,782 --> 00:06:31,240
And as a consequence,
I think there

142
00:06:31,240 --> 00:06:34,240
was a kind of general
and glacial drift

143
00:06:34,240 --> 00:06:37,030
away from addressing
scientific questions,

144
00:06:37,030 --> 00:06:42,220
and toward developing leading
edge applications in computer

145
00:06:42,220 --> 00:06:44,020
science.

146
00:06:44,020 --> 00:06:49,160
And that glacial drift over a
long period of time, of course,

147
00:06:49,160 --> 00:06:53,800
has a big impact on who goes
into the field and what kinds

148
00:06:53,800 --> 00:06:55,390
of things that they do.

149
00:06:55,390 --> 00:07:01,750
I'm pleased to say, though, that
I think that these days there

150
00:07:01,750 --> 00:07:05,710
is emerging a new
sense of enthusiasm,

151
00:07:05,710 --> 00:07:12,130
or a new sense of expectation,
a new belief that we can grapple

152
00:07:12,130 --> 00:07:14,830
with the scientific questions
and develop what we all

153
00:07:14,830 --> 00:07:16,480
originally set out
to develop, which

154
00:07:16,480 --> 00:07:19,865
is a computational theory
of human level intelligence

155
00:07:19,865 --> 00:07:20,365
and beyond.

156
00:07:20,365 --> 00:07:23,240


157
00:07:23,240 --> 00:07:28,300
So what do you think of--

158
00:07:28,300 --> 00:07:30,940
I mean, so at the
Triple AI conference

159
00:07:30,940 --> 00:07:34,900
there was sort of a fair
amount of complaining

160
00:07:34,900 --> 00:07:38,300
about everything's
statistical these days.

161
00:07:38,300 --> 00:07:41,860
There are all sorts of fads that
have appeared and disappeared

162
00:07:41,860 --> 00:07:44,140
in artificial intelligence
over the past 50 years,

163
00:07:44,140 --> 00:07:46,510
and that will continue
to happen, I think,

164
00:07:46,510 --> 00:07:49,660
until we have a much
better sense of what

165
00:07:49,660 --> 00:07:51,890
the answers really look like.

166
00:07:51,890 --> 00:07:57,010
Certainly those statistical
approaches have merit.

167
00:07:57,010 --> 00:07:59,380
They're especially
valuable in applications.

168
00:07:59,380 --> 00:08:01,720
They shed light on some
scientific questions,

169
00:08:01,720 --> 00:08:08,680
but like all the other
popular approaches,

170
00:08:08,680 --> 00:08:10,640
they're a tool, not the answer.

171
00:08:10,640 --> 00:08:13,120
So OK, makes sense.

172
00:08:13,120 --> 00:08:15,430
The real answer lies
in the observation

173
00:08:15,430 --> 00:08:19,325
that intelligence is
the product of thinking

174
00:08:19,325 --> 00:08:21,700
with our perceptions, and
thinking with our descriptions,

175
00:08:21,700 --> 00:08:25,000
and a blurry line that
separates the two.

176
00:08:25,000 --> 00:08:27,970
But thinking that is backed in
all cases up and down the line

177
00:08:27,970 --> 00:08:33,340
by tying a couple of loops,
by an incredible capability

178
00:08:33,340 --> 00:08:39,090
for accumulating, analyzing,
and making use of regularities

179
00:08:39,090 --> 00:08:42,970
in what we experience,
and not all of that

180
00:08:42,970 --> 00:08:46,270
is going to be
susceptible to attack

181
00:08:46,270 --> 00:08:49,490
by any one particular method.

182
00:08:49,490 --> 00:08:52,780
So professor McCarthy
said his opinion was

183
00:08:52,780 --> 00:08:55,030
that within the
hardware of the machines

184
00:08:55,030 --> 00:09:00,390
we had in the 70s, human level
intelligence is possible.

185
00:09:00,390 --> 00:09:03,930
He said that it's entirely
not hardware dependent.

186
00:09:03,930 --> 00:09:05,820
All the problems
remaining to be solved

187
00:09:05,820 --> 00:09:08,530
are software breakthroughs.

188
00:09:08,530 --> 00:09:09,930
What do you think?

189
00:09:09,930 --> 00:09:18,100
I get-- I think that we
will need some new hardware,

190
00:09:18,100 --> 00:09:20,680
but we don't know what
it will look like.

191
00:09:20,680 --> 00:09:23,410
But we do know that
whatever it will look like,

192
00:09:23,410 --> 00:09:24,430
Tom Knight can build it.

193
00:09:24,430 --> 00:09:27,230


194
00:09:27,230 --> 00:09:27,730
Fair enough.

195
00:09:27,730 --> 00:09:30,370
I think our thinking
about the problem

196
00:09:30,370 --> 00:09:33,160
has not been
severely handicapped

197
00:09:33,160 --> 00:09:37,390
by the hardware of the day,
and I pick my words carefully

198
00:09:37,390 --> 00:09:41,980
because in the early days, I
used to do a little computer

199
00:09:41,980 --> 00:09:42,800
vision research.

200
00:09:42,800 --> 00:09:45,490
And in those early days, it
was an expensive operation

201
00:09:45,490 --> 00:09:49,420
to drive a 3 by 3 filter
across a television image.

202
00:09:49,420 --> 00:09:51,970
And so you didn't think about
driving a 30 by 30 filter

203
00:09:51,970 --> 00:09:55,390
across a much larger range.

204
00:09:55,390 --> 00:09:58,540
That's easy with the kind
of hardware we have today,

205
00:09:58,540 --> 00:10:01,180
but we didn't think about it
then because it was so far

206
00:10:01,180 --> 00:10:05,110
beyond what we could
anticipate having

207
00:10:05,110 --> 00:10:08,200
within our graduate
students' lifetime.

208
00:10:08,200 --> 00:10:12,070
That is to say within a time a
student will spend [INAUDIBLE]..

209
00:10:12,070 --> 00:10:13,920
So it did influence
our thinking a bit,

210
00:10:13,920 --> 00:10:20,740
but aside from that
small footnote,

211
00:10:20,740 --> 00:10:24,250
I think that our integrated
artificial intelligence has

212
00:10:24,250 --> 00:10:28,540
never been severely handicapped
by computational limitations

213
00:10:28,540 --> 00:10:32,800
because we know that if
we need massive amounts

214
00:10:32,800 --> 00:10:36,100
of parallel look-up,
we can get it.

215
00:10:36,100 --> 00:10:38,350
After all, one of the
great reasons for optimism

216
00:10:38,350 --> 00:10:40,690
these days is that
computing has become free,

217
00:10:40,690 --> 00:10:43,240
as Tom Knight likes to say.

218
00:10:43,240 --> 00:10:47,110
In the computer I
have on my side,

219
00:10:47,110 --> 00:10:50,570
I have a gigabyte of memory.

220
00:10:50,570 --> 00:10:53,350
And when I came into the field
of artificial intelligence,

221
00:10:53,350 --> 00:10:54,580
memory was a dollar of bytes.

222
00:10:54,580 --> 00:10:57,700
So there's a billion dollars,
unadjusted for inflation,

223
00:10:57,700 --> 00:10:59,890
over there in my laptop.

224
00:10:59,890 --> 00:11:03,950
And years from now
when this videotape

225
00:11:03,950 --> 00:11:08,140
is seen, having a gigabyte
of memory only in a laptop

226
00:11:08,140 --> 00:11:10,870
will seem quaint, just
as thinking of memory

227
00:11:10,870 --> 00:11:14,260
as costing $1 a byte
seems quaint today.

228
00:11:14,260 --> 00:11:17,590
So computing in
massive quantities

229
00:11:17,590 --> 00:11:21,640
is no longer a cost
obstacle, so that's

230
00:11:21,640 --> 00:11:23,200
one of the great
reasons why we can

231
00:11:23,200 --> 00:11:24,790
be optimistic about the future.

232
00:11:24,790 --> 00:11:29,830
It no longer requires a big
laboratory, a giant contract,

233
00:11:29,830 --> 00:11:37,270
and vast teams of people to
do good work in the field.

234
00:11:37,270 --> 00:11:43,630
So people tend to hate
questions that are specific,

235
00:11:43,630 --> 00:11:47,443
but I guess, where do
you think we're headed,

236
00:11:47,443 --> 00:11:49,110
and when do you think
we're going to be?

237
00:11:49,110 --> 00:11:51,360
I'm very optimistic.

238
00:11:51,360 --> 00:11:53,070
I think if you look
at where the field

239
00:11:53,070 --> 00:11:54,820
of artificial
intelligence is today,

240
00:11:54,820 --> 00:11:58,800
you can say it's at the point
where molecular biology was

241
00:11:58,800 --> 00:11:59,940
in about 1950.

242
00:11:59,940 --> 00:12:03,330
So we're about a half
a century behind.

243
00:12:03,330 --> 00:12:05,820
But what I mean by that is
that we've been in business now

244
00:12:05,820 --> 00:12:08,070
for 50 years.

245
00:12:08,070 --> 00:12:11,460
We've done some fabulous things
in the world of applications,

246
00:12:11,460 --> 00:12:15,570
just as molecular biology had,
or as biology had in 1950.

247
00:12:15,570 --> 00:12:17,220
They had antibiotics,
and we have

248
00:12:17,220 --> 00:12:19,230
various kinds of
applications that

249
00:12:19,230 --> 00:12:22,460
populate all bit systems today.

250
00:12:22,460 --> 00:12:23,940
But we're still
waiting for analog

251
00:12:23,940 --> 00:12:27,030
to the discovery
of structured DNA.

252
00:12:27,030 --> 00:12:32,310
And I think we have accumulated
enough knowledge in our field

253
00:12:32,310 --> 00:12:35,310
and in the [INAUDIBLE] fields
that that kind of breakthrough

254
00:12:35,310 --> 00:12:38,010
can now be expected.

255
00:12:38,010 --> 00:12:40,500
Whether it will come tomorrow
or a decade from now,

256
00:12:40,500 --> 00:12:43,620
I can't say for sure, but
I have every confidence

257
00:12:43,620 --> 00:12:51,020
that we are ready for that
kind of a great discovery,

258
00:12:51,020 --> 00:12:54,170
after which the world will
never be able to look back.

259
00:12:54,170 --> 00:12:56,510
What do you see a human
level intelligence

260
00:12:56,510 --> 00:13:01,050
doing for us as either
individuals or as a society?

261
00:13:01,050 --> 00:13:06,800
We don't want to get hysterical
about those possibilities.

262
00:13:06,800 --> 00:13:08,653
That kind of hysteria
comes and goes

263
00:13:08,653 --> 00:13:10,320
in the field of
artificial intelligence,

264
00:13:10,320 --> 00:13:18,720
and what we generally find
is that the commercial value

265
00:13:18,720 --> 00:13:23,590
of new technology, especially
artificial intelligence

266
00:13:23,590 --> 00:13:25,427
technology, is not
in displacing people,

267
00:13:25,427 --> 00:13:27,010
but making it possible
to do something

268
00:13:27,010 --> 00:13:30,160
you couldn't do before.

269
00:13:30,160 --> 00:13:33,340
Companies that go into business
to replace people using

270
00:13:33,340 --> 00:13:35,380
artificial
intelligence generally

271
00:13:35,380 --> 00:13:36,970
go broke within a
short period of time

272
00:13:36,970 --> 00:13:40,700
because people are
pretty hard to replace.

273
00:13:40,700 --> 00:13:43,820
On the other hand, if you can
combine people and computers,

274
00:13:43,820 --> 00:13:46,970
and put them together
in combinations

275
00:13:46,970 --> 00:13:49,220
that can do something that
couldn't be done before,

276
00:13:49,220 --> 00:13:51,530
then you've got a
serious business.

277
00:13:51,530 --> 00:13:54,930
And that's the way I think
artificial intelligence will

278
00:13:54,930 --> 00:13:59,760
be making itself felt
in the years to come--

279
00:13:59,760 --> 00:14:00,720
many years to come.

280
00:14:00,720 --> 00:14:05,850
And by the time computers are
really as smart as people,

281
00:14:05,850 --> 00:14:09,240
we will have had many
generations to adjust.

282
00:14:09,240 --> 00:14:11,270
Fair enough.

283
00:14:11,270 --> 00:14:14,270
Do you read science fiction?

284
00:14:14,270 --> 00:14:17,322
No, I make it.

285
00:14:17,322 --> 00:14:18,030
Very good answer.

286
00:14:18,030 --> 00:14:19,220
What do you read?

287
00:14:19,220 --> 00:14:19,970
What do I do?

288
00:14:19,970 --> 00:14:21,480
Do you read?

289
00:14:21,480 --> 00:14:22,560
I like to read history.

290
00:14:22,560 --> 00:14:23,160
History?

291
00:14:23,160 --> 00:14:23,780
Yeah.

292
00:14:23,780 --> 00:14:24,655
What kind of history?

293
00:14:24,655 --> 00:14:27,860
American Civil War history
in particular, but just

294
00:14:27,860 --> 00:14:33,230
about any history from--

295
00:14:33,230 --> 00:14:41,660
histories focused
on Roman emperors

296
00:14:41,660 --> 00:14:49,720
to more modern histories
of Peter the Great.

297
00:14:49,720 --> 00:14:52,680
Anything that is
similar to that.

298
00:14:52,680 --> 00:14:55,058
Fair enough.

299
00:14:55,058 --> 00:14:55,600
Just curious.

300
00:14:55,600 --> 00:15:00,972
Sometimes people have
very crazy tastes.

301
00:15:00,972 --> 00:15:02,340
Oh, artificial intelligences.

302
00:15:02,340 --> 00:15:02,850
That's all they read about.

303
00:15:02,850 --> 00:15:04,800
A lot of people in
artificial intelligence

304
00:15:04,800 --> 00:15:06,510
are addicted to science fiction.

305
00:15:06,510 --> 00:15:16,653


306
00:15:16,653 --> 00:15:17,810
Fair enough, OK.

307
00:15:17,810 --> 00:15:20,550
Well, I think that
should probably do it.

308
00:15:20,550 --> 00:15:24,900
OK, let me think of anything
else I want to touch on.

309
00:15:24,900 --> 00:15:26,880
Can I just say a few
things you might want

310
00:15:26,880 --> 00:15:30,090
to edit in if you feel like it?

311
00:15:30,090 --> 00:15:34,035
Everyone is always curious about
how long it was going to take,

312
00:15:34,035 --> 00:15:38,490
and JCR Licklider
always said that we

313
00:15:38,490 --> 00:15:41,820
tend to overestimate
what we can do in a year,

314
00:15:41,820 --> 00:15:44,970
and underestimate what
we can do in a decade.

315
00:15:44,970 --> 00:15:47,380
And I suppose if we
take this problem

316
00:15:47,380 --> 00:15:51,570
of artificial intelligence,
we can usefully

317
00:15:51,570 --> 00:15:54,360
paraphrase what
Licklider said, and argue

318
00:15:54,360 --> 00:15:58,200
that we tend to overestimate
what we can do in a decade

319
00:15:58,200 --> 00:16:01,800
and underestimate what
we can do in a century.

320
00:16:01,800 --> 00:16:06,030
We ought to have a
great deal of optimism

321
00:16:06,030 --> 00:16:10,200
now because of all the
knowledge has been accumulated.

322
00:16:10,200 --> 00:16:11,700
We're a little
frustrated as a field

323
00:16:11,700 --> 00:16:17,465
because we haven't
solved a big problem yet.

324
00:16:17,465 --> 00:16:19,840
On the other hand, I think we
have a much clearer picture

325
00:16:19,840 --> 00:16:21,140
of the way forward.

326
00:16:21,140 --> 00:16:25,030
I think many of us
believe that to understand

327
00:16:25,030 --> 00:16:28,510
the nature of human
intelligence is

328
00:16:28,510 --> 00:16:31,450
going to require a
great deal of capability

329
00:16:31,450 --> 00:16:35,590
for understanding
intelligence at every level,

330
00:16:35,590 --> 00:16:37,570
from perceptions,
to descriptions,

331
00:16:37,570 --> 00:16:43,510
from the interactions of those,
from how we hallucinate images

332
00:16:43,510 --> 00:16:45,520
to help us answer questions.

333
00:16:45,520 --> 00:16:48,070
How, if I were to say
to you, John loves Mary.

334
00:16:48,070 --> 00:16:50,980
Did John kiss Mary?

335
00:16:50,980 --> 00:16:53,800
If I were to say to
you, John kissed Mary.

336
00:16:53,800 --> 00:16:54,850
Did John touch Mary?

337
00:16:54,850 --> 00:16:57,400
You would, I think,
imagine the scene

338
00:16:57,400 --> 00:16:59,950
and read the answer off of it.

339
00:16:59,950 --> 00:17:03,460
So we're beginning to understand
now that intelligence is not

340
00:17:03,460 --> 00:17:05,329
the product of a single
faculty, but it's

341
00:17:05,329 --> 00:17:10,790
the product of many
faculties in our brains

342
00:17:10,790 --> 00:17:11,750
all working together.

343
00:17:11,750 --> 00:17:12,800
We think with our eyes.

344
00:17:12,800 --> 00:17:14,250
We think with our hands.

345
00:17:14,250 --> 00:17:16,470
We think with language.

346
00:17:16,470 --> 00:17:23,520
We think with all
parts of our brains

347
00:17:23,520 --> 00:17:27,000
working together, and working in
the light of all the experience

348
00:17:27,000 --> 00:17:28,740
that they accumulate.

349
00:17:28,740 --> 00:17:30,330
So when we get the
answers, I think

350
00:17:30,330 --> 00:17:34,500
what we'll find is that
there's a fabulous engineering

351
00:17:34,500 --> 00:17:39,360
in the brain, and that
that engineering has a--

352
00:17:39,360 --> 00:17:43,572
rests on a foundation of
fabulous scientific principle.

353
00:17:43,572 --> 00:17:45,030
When we understand
the whole story,

354
00:17:45,030 --> 00:17:47,130
we'll see very
sophisticated mechanisms

355
00:17:47,130 --> 00:17:49,290
that are involved in there.

356
00:17:49,290 --> 00:17:51,310
And how long will it take?

357
00:17:51,310 --> 00:17:52,830
I have a great
deal of confidence

358
00:17:52,830 --> 00:17:56,580
now because I think that we
have such a much clearer idea

359
00:17:56,580 --> 00:18:00,180
of what needs to be
done than we used to,

360
00:18:00,180 --> 00:18:09,320
that we can expect in my
lifetime to substantially solve

361
00:18:09,320 --> 00:18:12,130
the key problems.

362
00:18:12,130 --> 00:18:12,870
Excellent.

363
00:18:12,870 --> 00:18:13,370
OK.

364
00:18:13,370 --> 00:18:15,520
Well, that's
encouraging to hear.