1
00:00:00,000 --> 00:00:03,330


2
00:00:03,330 --> 00:00:04,830
Do you mind
introducing yourself?

3
00:00:04,830 --> 00:00:07,410
I'm Jon Doyle.

4
00:00:07,410 --> 00:00:08,910
What sort of introduction?

5
00:00:08,910 --> 00:00:12,930
So what are you doing right now?

6
00:00:12,930 --> 00:00:17,290
I'm pursuing both
how to represent

7
00:00:17,290 --> 00:00:20,440
decision [INAUDIBLE]
preference information,

8
00:00:20,440 --> 00:00:22,810
which economists
typically have not

9
00:00:22,810 --> 00:00:25,450
had good representations
for, but we'll

10
00:00:25,450 --> 00:00:27,520
need good representations
if you want

11
00:00:27,520 --> 00:00:31,370
to have proxies or intelligent
agents that do what you want.

12
00:00:31,370 --> 00:00:35,630
And I am also
working on mechanics,

13
00:00:35,630 --> 00:00:39,930
and I just published
a book that extends

14
00:00:39,930 --> 00:00:44,490
the modern axioms of mechanics
to cover minds as well

15
00:00:44,490 --> 00:00:45,290
as bodies.

16
00:00:45,290 --> 00:00:48,130


17
00:00:48,130 --> 00:00:50,850
If you want to think back
to the time of DesCartes,

18
00:00:50,850 --> 00:00:52,440
where there was both
the physical body

19
00:00:52,440 --> 00:00:54,120
and the mental substance.

20
00:00:54,120 --> 00:00:56,400
You now can have a theory
in which both parts of this

21
00:00:56,400 --> 00:00:59,380
are satisfying the same
set of mechanical laws.

22
00:00:59,380 --> 00:01:01,040
So that's the other part of my--

23
00:01:01,040 --> 00:01:03,010
Well, we'll get back to that.

24
00:01:03,010 --> 00:01:05,810
How did you first get into
artificial intelligence?

25
00:01:05,810 --> 00:01:09,580


26
00:01:09,580 --> 00:01:13,120
Well, the proximate reason was
I had a teacher at University

27
00:01:13,120 --> 00:01:19,920
of Houston who thought
that understanding the mind

28
00:01:19,920 --> 00:01:24,580
was going to be the next big
thing in the next 20 years.

29
00:01:24,580 --> 00:01:29,940
And that got me very
interested in it.

30
00:01:29,940 --> 00:01:31,500
I went into the field.

31
00:01:31,500 --> 00:01:35,310
Cool, so what other kind of
research was-- when was that?

32
00:01:35,310 --> 00:01:42,600
That was '72 or '73,
somewhere around that.

33
00:01:42,600 --> 00:01:45,580
So where did you go from there?

34
00:01:45,580 --> 00:01:49,040
Well, I went to
grad school at MIT,

35
00:01:49,040 --> 00:01:54,960
and conceivably,
I'm the first person

36
00:01:54,960 --> 00:01:58,260
to actually get a degree
with specification.

37
00:01:58,260 --> 00:02:00,120
My PhD is in artificial
intelligence.

38
00:02:00,120 --> 00:02:00,872
Wow.

39
00:02:00,872 --> 00:02:02,040
Congratulations.

40
00:02:02,040 --> 00:02:02,940
They offered it.

41
00:02:02,940 --> 00:02:06,630
I don't know-- maybe somebody
must have had a thesis in that

42
00:02:06,630 --> 00:02:12,960
before, but I picked it, and
I work with Jerry Sussman, who

43
00:02:12,960 --> 00:02:13,980
was my advisor.

44
00:02:13,980 --> 00:02:19,080
And after that, I worked at
Stanford University for a year

45
00:02:19,080 --> 00:02:23,880
with John McCarthy, and I was
at CMU for seven or so years.

46
00:02:23,880 --> 00:02:24,760
All the big places.

47
00:02:24,760 --> 00:02:28,050
And I'm back at MIT
for another 13 or so.

48
00:02:28,050 --> 00:02:30,060
And now I'm at North
Carolina State University.

49
00:02:30,060 --> 00:02:30,560
Great.

50
00:02:30,560 --> 00:02:31,000
Great.

51
00:02:31,000 --> 00:02:31,500
That's cool.

52
00:02:31,500 --> 00:02:34,740
So tell me about, since
our project is originally

53
00:02:34,740 --> 00:02:37,635
about some film reels
we found at Tech Square.

54
00:02:37,635 --> 00:02:39,510
Could you tell us a
little bit about what was

55
00:02:39,510 --> 00:02:41,210
going on when you got there?

56
00:02:41,210 --> 00:02:45,480
And maybe you remember
about Sussman?

57
00:02:45,480 --> 00:02:48,300
Well, I know various
stories he told me,

58
00:02:48,300 --> 00:02:52,160
but I saw the thing on the list.

59
00:02:52,160 --> 00:02:56,900
The two best stories I know,
that I can remember immediately

60
00:02:56,900 --> 00:03:01,180
from my grad school
days, both actually

61
00:03:01,180 --> 00:03:03,020
involved Guy Steele,
who was a student there

62
00:03:03,020 --> 00:03:04,390
at the same time as myself.

63
00:03:04,390 --> 00:03:10,883
And in one of them, near the--

64
00:03:10,883 --> 00:03:12,050
I forget whether it was on--

65
00:03:12,050 --> 00:03:14,175
I guess it was while he
was working on his master's

66
00:03:14,175 --> 00:03:15,740
thesis, maybe his PhD.

67
00:03:15,740 --> 00:03:18,980
But he had a bug which depended
upon the phase of the moon,

68
00:03:18,980 --> 00:03:23,490
literally, which was a
joke up until that point.

69
00:03:23,490 --> 00:03:26,750
But [INAUDIBLE]
at that point had

70
00:03:26,750 --> 00:03:29,870
a thing where you could print
out the time and the date.

71
00:03:29,870 --> 00:03:32,090
In fact, you could also
print the phase of the moon,

72
00:03:32,090 --> 00:03:35,330
and it turned out
that, in his program,

73
00:03:35,330 --> 00:03:37,400
it would compile
things into LISP.

74
00:03:37,400 --> 00:03:40,670
And it would have these
comments up at the top

75
00:03:40,670 --> 00:03:43,040
saying when it was
done, and then there

76
00:03:43,040 --> 00:03:45,590
was a sentence about when it
was done, and then at the end,

77
00:03:45,590 --> 00:03:47,715
there would be a period at
the end of the sentence.

78
00:03:47,715 --> 00:03:52,640
Well, it turns out there was
word wrapping in the output

79
00:03:52,640 --> 00:03:55,490
routines, and if
the face of the moon

80
00:03:55,490 --> 00:03:57,800
was just right, that
period would wind up

81
00:03:57,800 --> 00:04:02,010
isolated on a [INAUDIBLE] bug.

82
00:04:02,010 --> 00:04:04,340
When this happened to him,
and he realized what it was,

83
00:04:04,340 --> 00:04:06,020
he took a screenshot,
and for years, he

84
00:04:06,020 --> 00:04:07,920
had that screenshot
on his office door.

85
00:04:07,920 --> 00:04:10,580


86
00:04:10,580 --> 00:04:13,910
The other thing is that
I had taken, I'd say,

87
00:04:13,910 --> 00:04:19,100
quantum mechanics before I
came to MIT, and Guy, I think,

88
00:04:19,100 --> 00:04:20,810
was taking a quantum
mechanics class.

89
00:04:20,810 --> 00:04:25,120
And there was this
funny article--

90
00:04:25,120 --> 00:04:28,030
maybe not so funny article
in the paper, or a joke,

91
00:04:28,030 --> 00:04:33,220
at the time about an aircraft
windshield manufacturer who

92
00:04:33,220 --> 00:04:36,350
had been testing windshields.

93
00:04:36,350 --> 00:04:39,220
They would shoot chickens at
them at 800 miles an hour,

94
00:04:39,220 --> 00:04:41,020
and the joke was
that somebody finally

95
00:04:41,020 --> 00:04:44,870
realized they didn't need to
use live chickens to do it.

96
00:04:44,870 --> 00:04:48,550
But Guy decided to
exercise his skills,

97
00:04:48,550 --> 00:04:54,950
and he computed the wavelength
of an 800 mile an hour chicken

98
00:04:54,950 --> 00:04:59,110
and then put up, on
his wall, frequencies,

99
00:04:59,110 --> 00:05:01,510
various things about the
electromagnetic spectrum

100
00:05:01,510 --> 00:05:04,210
and then x-rays and
gamma rays and then 800--

101
00:05:04,210 --> 00:05:07,390
a big separation mark
and then the frequency

102
00:05:07,390 --> 00:05:09,676
of an 800 mile an hour chicken.

103
00:05:09,676 --> 00:05:10,900
That's fabulous.

104
00:05:10,900 --> 00:05:13,390
Was that representative of
the general culture at MIT--

105
00:05:13,390 --> 00:05:15,670
People thought about
all sorts of stuff

106
00:05:15,670 --> 00:05:17,660
and talked about
all sorts of stuff.

107
00:05:17,660 --> 00:05:20,230
There was a gang of us, of
varying composition, that

108
00:05:20,230 --> 00:05:25,150
would walk over to a
sub shop near the labs,

109
00:05:25,150 --> 00:05:29,440
and there would always
be some discussion.

110
00:05:29,440 --> 00:05:32,200
During my time, there were a
lot of discussions of philosophy

111
00:05:32,200 --> 00:05:34,510
and logic and all
sorts of things,

112
00:05:34,510 --> 00:05:37,390
and those were some of the
most interesting discussions

113
00:05:37,390 --> 00:05:38,210
of the day.

114
00:05:38,210 --> 00:05:38,710
That's cool.

115
00:05:38,710 --> 00:05:41,190
Were you part of the
Chinese Food Sect?

116
00:05:41,190 --> 00:05:43,840
I used to hear that was
a very popular option.

117
00:05:43,840 --> 00:05:49,000
Yeah, [INAUDIBLE] small eating
place was down the street,

118
00:05:49,000 --> 00:05:52,450
and they had very good food.

119
00:05:52,450 --> 00:05:56,560
They had a hot and sour wonton
soup which was extremely spicy,

120
00:05:56,560 --> 00:05:59,500
and if you had a
cold or something,

121
00:05:59,500 --> 00:06:01,540
it was always a good
idea to go there,

122
00:06:01,540 --> 00:06:04,060
and your entire
respiratory system

123
00:06:04,060 --> 00:06:06,670
would be cleaned
out in short order

124
00:06:06,670 --> 00:06:08,980
by eating some hot
and sour wonton soup.

125
00:06:08,980 --> 00:06:11,640
I'm wondering how
similar it was to now.

126
00:06:11,640 --> 00:06:15,790
What were people's
hours like at the lab?

127
00:06:15,790 --> 00:06:17,980
A lot of people worked late.

128
00:06:17,980 --> 00:06:19,850
I've always been
a morning person,

129
00:06:19,850 --> 00:06:25,010
so I would, a lot of the
time, get there 4:00 or 5:00

130
00:06:25,010 --> 00:06:27,860
in the morning, and I'd have--

131
00:06:27,860 --> 00:06:30,140
some people would just
be leaving at that point,

132
00:06:30,140 --> 00:06:31,880
and I would have a lot of time.

133
00:06:31,880 --> 00:06:33,435
And well, this was also--

134
00:06:33,435 --> 00:06:36,060
I mean, I probably wouldn't have
gotten there quite that early,

135
00:06:36,060 --> 00:06:39,080
except this was the time
of shared mainframes.

136
00:06:39,080 --> 00:06:39,770
Oh, absolutely.

137
00:06:39,770 --> 00:06:42,710
And also, not every--

138
00:06:42,710 --> 00:06:44,720
there was a terminal
in each office or so,

139
00:06:44,720 --> 00:06:46,380
but not everybody--

140
00:06:46,380 --> 00:06:47,750
So if you're much faster--

141
00:06:47,750 --> 00:06:50,090
So if you wanted time
to spend on a computer,

142
00:06:50,090 --> 00:06:52,687
either you had to stay late,
or later than everybody else,

143
00:06:52,687 --> 00:06:54,770
for have to get there
earlier than everybody else.

144
00:06:54,770 --> 00:06:55,580
Interesting.

145
00:06:55,580 --> 00:06:56,650
I got there earlier.

146
00:06:56,650 --> 00:06:59,023


147
00:06:59,023 --> 00:07:01,440
So what do you think some of
the most challenging problems

148
00:07:01,440 --> 00:07:06,510
in artificial intelligence were
and remain today in the field?

149
00:07:06,510 --> 00:07:08,790
Or what intrigues you?

150
00:07:08,790 --> 00:07:10,410
Maybe that's too
general a question.

151
00:07:10,410 --> 00:07:14,500
Well, there's still--
we still don't

152
00:07:14,500 --> 00:07:19,960
know how to represent
lots of information,

153
00:07:19,960 --> 00:07:25,310
and part of the reason
is that everything

154
00:07:25,310 --> 00:07:26,900
that we know how
to do, we basically

155
00:07:26,900 --> 00:07:29,360
have to filter through
individual human minds.

156
00:07:29,360 --> 00:07:31,820
Somebody has to
understand how to do that.

157
00:07:31,820 --> 00:07:34,850
And intelligence
is so complicated,

158
00:07:34,850 --> 00:07:37,800
and the mind is so
complicated, that it's--

159
00:07:37,800 --> 00:07:40,130
we can only understand
little bits at a time,

160
00:07:40,130 --> 00:07:44,780
and there's a finite number
of people in the field.

161
00:07:44,780 --> 00:07:52,320
And it's not clear that you can
get something of the complexity

162
00:07:52,320 --> 00:07:55,050
and size that would look
like a regular person

163
00:07:55,050 --> 00:07:57,560
without spending a
very long time on it.

164
00:07:57,560 --> 00:08:00,960
If I look at mathematics
or some other fields

165
00:08:00,960 --> 00:08:04,140
and see how long
it took to develop

166
00:08:04,140 --> 00:08:07,800
what are thought of as
basic everyday tools today,

167
00:08:07,800 --> 00:08:11,560
it was hundreds of years.

168
00:08:11,560 --> 00:08:12,960
If you look at physics--

169
00:08:12,960 --> 00:08:17,730
I mean, Newton's famous
for these three postulates.

170
00:08:17,730 --> 00:08:20,610
F equals MA was
not one of those.

171
00:08:20,610 --> 00:08:23,700
F equals MA is an equation
published by Euler

172
00:08:23,700 --> 00:08:27,070
50 years later.

173
00:08:27,070 --> 00:08:28,470
There was a long period.

174
00:08:28,470 --> 00:08:29,940
Now, there weren't
very many people

175
00:08:29,940 --> 00:08:31,357
working on this
compared to today,

176
00:08:31,357 --> 00:08:36,570
but there was a long period to
get from initial ideas of how

177
00:08:36,570 --> 00:08:40,110
something would work
to the sort of things

178
00:08:40,110 --> 00:08:42,900
you could use every day
to calculate more things

179
00:08:42,900 --> 00:08:43,659
or to do stuff.

180
00:08:43,659 --> 00:08:45,960
And I think we're in that
same sort of position in AI,

181
00:08:45,960 --> 00:08:48,720
that we have all
sorts of good ideas,

182
00:08:48,720 --> 00:08:52,710
and it takes a long
time to set the table,

183
00:08:52,710 --> 00:08:56,100
if you want to think of it that
way, for all the tools you need

184
00:08:56,100 --> 00:09:00,280
and collecting all the
information and trying it out.

185
00:09:00,280 --> 00:09:05,225
And since it's a long time,
invariably, somewhere, you're

186
00:09:05,225 --> 00:09:07,350
part of the way through,
and you have a better idea

187
00:09:07,350 --> 00:09:09,073
about some part, and
so you don't quite

188
00:09:09,073 --> 00:09:11,490
finish what you were doing,
and you try to keep adjusting.

189
00:09:11,490 --> 00:09:13,815
And the time keeps
moving off a little bit.

190
00:09:13,815 --> 00:09:15,690
I mean, there's a lot
of progress being made,

191
00:09:15,690 --> 00:09:20,260
but it's a very
difficult plotline.

192
00:09:20,260 --> 00:09:24,270
So if they were a
subarea or paradigm

193
00:09:24,270 --> 00:09:26,910
of artificial intelligence
in the future that

194
00:09:26,910 --> 00:09:29,880
was developed and had
a law named after you,

195
00:09:29,880 --> 00:09:31,830
what area would you
like that to be?

196
00:09:31,830 --> 00:09:32,410
Well, aw, no.

197
00:09:32,410 --> 00:09:34,730
I don't know
anything about that.

198
00:09:34,730 --> 00:09:37,380
What area do you
find most fascinating

199
00:09:37,380 --> 00:09:41,050
within artificial intelligence?

200
00:09:41,050 --> 00:09:44,170
I don't know how-- well, I--

201
00:09:44,170 --> 00:09:46,360
What do you think the
biggest potentials are?

202
00:09:46,360 --> 00:09:49,840
Well, one of my
favorite activities

203
00:09:49,840 --> 00:09:56,770
is trying to formulate the ideas
that we don't understand well

204
00:09:56,770 --> 00:10:01,150
yet in mathematical terms,
not because the formalization

205
00:10:01,150 --> 00:10:02,710
always gives you the--

206
00:10:02,710 --> 00:10:04,600
that you can always
succeed, but just

207
00:10:04,600 --> 00:10:08,050
making that attempt
raises a lot of problems

208
00:10:08,050 --> 00:10:10,960
that you hadn't quite put
your finger on before.

209
00:10:10,960 --> 00:10:14,920
And I think one of the
exciting things about the field

210
00:10:14,920 --> 00:10:17,020
is that we have figured
out how to formalize

211
00:10:17,020 --> 00:10:19,060
various things that
nobody has figured out

212
00:10:19,060 --> 00:10:22,870
how to say precisely
before, and I

213
00:10:22,870 --> 00:10:26,785
think that's an
exciting thing for me.

214
00:10:26,785 --> 00:10:29,160
It's not to everybody's taste,
but that's exciting to me,

215
00:10:29,160 --> 00:10:30,780
and that's going to
continue for a long while.

216
00:10:30,780 --> 00:10:33,120
I think if we're going to
make really big progress,

217
00:10:33,120 --> 00:10:35,810
there has to be a lot of
additional work on formalizing

218
00:10:35,810 --> 00:10:36,310
that.

219
00:10:36,310 --> 00:10:38,240
Can you give us some
examples of things that

220
00:10:38,240 --> 00:10:39,900
have been formalized already?

221
00:10:39,900 --> 00:10:45,060
Well, the one that
I've been involved with

222
00:10:45,060 --> 00:10:49,320
is non-monotonic
logics, which is

223
00:10:49,320 --> 00:10:52,350
a real contribution to logic,
as far as I'm concerned.

224
00:10:52,350 --> 00:10:55,050
And it's the first
theory anybody

225
00:10:55,050 --> 00:10:59,730
had about how to talk about
assumptions that you might want

226
00:10:59,730 --> 00:11:02,160
to make and rules for
making assumptions

227
00:11:02,160 --> 00:11:04,950
when you want to
have them fit well

228
00:11:04,950 --> 00:11:08,430
or cohere with other
information that you have.

229
00:11:08,430 --> 00:11:10,770
It's not probabilistic.

230
00:11:10,770 --> 00:11:12,660
There are probabilistic
theories of it,

231
00:11:12,660 --> 00:11:14,190
but in fact, I
think those are not

232
00:11:14,190 --> 00:11:16,560
the right way of
thinking about it.

233
00:11:16,560 --> 00:11:21,270
It's a theory that's combining
both preferences about what you

234
00:11:21,270 --> 00:11:24,660
think you'd rather conclude
in a state of ignorance

235
00:11:24,660 --> 00:11:28,027
with some sort of inference
about, well, what do I know

236
00:11:28,027 --> 00:11:29,610
and how does that
constrain what I can

237
00:11:29,610 --> 00:11:32,470
do to fulfill my preferences?

238
00:11:32,470 --> 00:11:34,950
So it's a quite
interesting logic.

239
00:11:34,950 --> 00:11:37,920
It's the basis for a
huge literature right now

240
00:11:37,920 --> 00:11:42,090
on logic programming.

241
00:11:42,090 --> 00:11:48,030
It still has not worked its way
through most AI representations

242
00:11:48,030 --> 00:11:51,300
or mechanisms partly because
of some complexity issues

243
00:11:51,300 --> 00:11:55,830
that we're still working
out, but it's a new idea,

244
00:11:55,830 --> 00:11:59,960
as far as new formalized idea,
and it didn't exist before.

245
00:11:59,960 --> 00:12:03,220


246
00:12:03,220 --> 00:12:05,800
I think we're going
to have more of those,

247
00:12:05,800 --> 00:12:07,505
but we don't really have them.

248
00:12:07,505 --> 00:12:08,880
The work I'm doing
on preferences

249
00:12:08,880 --> 00:12:11,050
right now is just
trying to say, well,

250
00:12:11,050 --> 00:12:14,015
if you talk about generic
preference information, what

251
00:12:14,015 --> 00:12:14,640
does that mean?

252
00:12:14,640 --> 00:12:21,003
If I say I like red cars better
than blue cars, to economists,

253
00:12:21,003 --> 00:12:23,170
that doesn't necessarily
make any sense because they

254
00:12:23,170 --> 00:12:26,500
want to talk about
decision preferences

255
00:12:26,500 --> 00:12:29,980
that order the specific
alternatives you had,

256
00:12:29,980 --> 00:12:32,050
not properties of those.

257
00:12:32,050 --> 00:12:35,020
And there are a lot of possible
interpretations for what

258
00:12:35,020 --> 00:12:38,050
it could mean to say, I like
red cars better than blue cars,

259
00:12:38,050 --> 00:12:41,530
and we still are just
exploring the first couple

260
00:12:41,530 --> 00:12:42,410
simplest of those.

261
00:12:42,410 --> 00:12:44,800
And we have no
idea whether, when

262
00:12:44,800 --> 00:12:47,740
you or I talk about
what we like or dislike,

263
00:12:47,740 --> 00:12:50,200
what's involved in
the meaning of this.

264
00:12:50,200 --> 00:12:52,150
So have you been--

265
00:12:52,150 --> 00:12:54,740
how do you feel about the
development of the field?

266
00:12:54,740 --> 00:12:57,500
What have been
some of the biggest

267
00:12:57,500 --> 00:13:00,570
disappointments, surprises,
what you're particularly pleased

268
00:13:00,570 --> 00:13:01,070
about?

269
00:13:01,070 --> 00:13:03,620
The most surprising
thing, I guess, to me,

270
00:13:03,620 --> 00:13:07,940
is that there's been an
enormous amount of progress

271
00:13:07,940 --> 00:13:11,670
both in the computers that
underlie what we can do

272
00:13:11,670 --> 00:13:16,010
and in what people have
actually done with it.

273
00:13:16,010 --> 00:13:17,660
What's really
surprising about that

274
00:13:17,660 --> 00:13:20,210
isn't that all this
progress has happened?

275
00:13:20,210 --> 00:13:23,870
It's that almost all of it's
invisible to most of the world,

276
00:13:23,870 --> 00:13:27,440
because right now, huge
sections of society

277
00:13:27,440 --> 00:13:31,040
are dependent on various
sorts of expert systems

278
00:13:31,040 --> 00:13:34,010
or knowledge based systems or
systems that were synthesized

279
00:13:34,010 --> 00:13:36,980
by knowledge based
systems that control

280
00:13:36,980 --> 00:13:43,415
all sorts of interesting things,
and these are right out of AI.

281
00:13:43,415 --> 00:13:44,900
They were so used to the--

282
00:13:44,900 --> 00:13:47,180
Well, most of them are hidden.

283
00:13:47,180 --> 00:13:50,450
So in fact, if you look at
AI in the popular world,

284
00:13:50,450 --> 00:13:52,360
you see novels and
whatnot and that--

285
00:13:52,360 --> 00:13:53,930
I should mention
that some of those

286
00:13:53,930 --> 00:13:55,925
were things that got
me interested in,

287
00:13:55,925 --> 00:13:58,220
or at least set
the stage for me.

288
00:13:58,220 --> 00:14:02,210
But you see commercials
where people--

289
00:14:02,210 --> 00:14:04,550
or ads where people
are claiming to have

290
00:14:04,550 --> 00:14:07,230
AI in their toaster or whatever,
or in their rice cooker.

291
00:14:07,230 --> 00:14:10,140


292
00:14:10,140 --> 00:14:12,590
Why are you looking at me?

293
00:14:12,590 --> 00:14:16,430
It's not untrue, but it's not--

294
00:14:16,430 --> 00:14:19,490
it makes people think that,
well, that's what AI is.

295
00:14:19,490 --> 00:14:22,055


296
00:14:22,055 --> 00:14:25,550
I mean, from that point of
view, my rice cooker knowing

297
00:14:25,550 --> 00:14:27,800
to turn itself off
is intelligence,

298
00:14:27,800 --> 00:14:30,620
and that's not really
a very good notion.

299
00:14:30,620 --> 00:14:32,690
What we're really
interested in is

300
00:14:32,690 --> 00:14:34,520
people who know a
lot about something,

301
00:14:34,520 --> 00:14:37,430
or people who know
something about everything.

302
00:14:37,430 --> 00:14:39,980
We're in the process of
formalizing the entire world's

303
00:14:39,980 --> 00:14:41,720
knowledge, and this
has not been done,

304
00:14:41,720 --> 00:14:44,840
not even the expert knowledge
that you find in any university

305
00:14:44,840 --> 00:14:46,100
curriculum.

306
00:14:46,100 --> 00:14:49,580
It's going to ripple
through all the disciplines,

307
00:14:49,580 --> 00:14:51,750
and eventually, we'll be
able to make use of that.

308
00:14:51,750 --> 00:14:54,800
But this is a long process
because trying to figure out

309
00:14:54,800 --> 00:14:57,440
how to say things formally
in a reasonable way

310
00:14:57,440 --> 00:14:58,430
is something that--

311
00:14:58,430 --> 00:15:01,710


312
00:15:01,710 --> 00:15:03,310
we know how to
get started on it.

313
00:15:03,310 --> 00:15:07,000
We have no idea whether the
tools we have right now are

314
00:15:07,000 --> 00:15:08,800
capable of completing the job.

315
00:15:08,800 --> 00:15:11,170
Building on that, what
kind of collaboration

316
00:15:11,170 --> 00:15:13,510
do you see with other
fields in the near future?

317
00:15:13,510 --> 00:15:15,670
Which fields are they?

318
00:15:15,670 --> 00:15:18,310
Oh, well, you can pick
any field and just

319
00:15:18,310 --> 00:15:23,890
try to say, how do we express
the knowledge in that field

320
00:15:23,890 --> 00:15:29,320
in formal terms that can be
used by a machine to reason

321
00:15:29,320 --> 00:15:32,470
or set up problems in that
field, or to do analysis

322
00:15:32,470 --> 00:15:33,250
or so forth.

323
00:15:33,250 --> 00:15:35,950
Right now, this is
all manual activity.

324
00:15:35,950 --> 00:15:40,210
It's some person sitting
down and worrying about it.

325
00:15:40,210 --> 00:15:42,790
In some cases, you can
have some data sets

326
00:15:42,790 --> 00:15:45,230
from which you can extract
some bits of the information,

327
00:15:45,230 --> 00:15:46,180
but generally not.

328
00:15:46,180 --> 00:15:48,130
I mean, you can't
extract from data

329
00:15:48,130 --> 00:15:51,370
sets the knowledge involved,
and if I'm a chemical engineer,

330
00:15:51,370 --> 00:15:52,630
how do I form--

331
00:15:52,630 --> 00:15:54,400
look at this power
plant that somebody

332
00:15:54,400 --> 00:15:57,610
wants me to come consult
on, and/or chemical plant,

333
00:15:57,610 --> 00:16:01,150
and figure out which parts
are relevant to the task

334
00:16:01,150 --> 00:16:03,640
that they're setting me,
keeping it from blowing up

335
00:16:03,640 --> 00:16:08,050
or getting the efficiency
higher on something?

336
00:16:08,050 --> 00:16:10,390
I have to know what
data to look at,

337
00:16:10,390 --> 00:16:12,670
how to think about the
problem, how to decompose it,

338
00:16:12,670 --> 00:16:17,520
and this is not something that
you can get out of a data set.

339
00:16:17,520 --> 00:16:19,270
And the only way they
can get it right now

340
00:16:19,270 --> 00:16:21,340
is by reading textbooks
and talking to experts

341
00:16:21,340 --> 00:16:23,440
and doing a lot of manual
work, and it'd be nice

342
00:16:23,440 --> 00:16:25,060
if we could do that quicker.

343
00:16:25,060 --> 00:16:27,700
But that's something
which has not

344
00:16:27,700 --> 00:16:30,940
been done for virtually
any area for most

345
00:16:30,940 --> 00:16:32,740
areas of human activity,
and it's not just

346
00:16:32,740 --> 00:16:33,850
sciences and engineering.

347
00:16:33,850 --> 00:16:39,930
It's how to write well, how
to speak well, how to learn--

348
00:16:39,930 --> 00:16:42,880
do self-help, how to
manage your own life.

349
00:16:42,880 --> 00:16:45,010
So what do you think the
most important skills

350
00:16:45,010 --> 00:16:47,410
a new researcher in
the field should have?

351
00:16:47,410 --> 00:16:50,200


352
00:16:50,200 --> 00:16:54,572
One of them is to always keep
trying to understand something.

353
00:16:54,572 --> 00:16:56,280
Don't ever think that
just because you've

354
00:16:56,280 --> 00:16:59,050
read a paper or two that
you can then sit back,

355
00:16:59,050 --> 00:17:03,510
and you know what
you need to know.

356
00:17:03,510 --> 00:17:06,480
Always be working on a problem,
because if you have a problem,

357
00:17:06,480 --> 00:17:08,170
that gives you something
to think about.

358
00:17:08,170 --> 00:17:09,990
And if you are just
thinking of this

359
00:17:09,990 --> 00:17:13,020
as a task of understanding
and literature,

360
00:17:13,020 --> 00:17:16,490
you're not going
to be as motivated.

361
00:17:16,490 --> 00:17:21,023
I think the more mathematics
that you know, the better.

362
00:17:21,023 --> 00:17:23,190
And by that, I don't mean
just calculus or something

363
00:17:23,190 --> 00:17:25,357
like that, but there's a
lot of advanced mathematics

364
00:17:25,357 --> 00:17:28,290
that is about the structural
relations of things.

365
00:17:28,290 --> 00:17:30,960
And I think all that's
helpful in different ways

366
00:17:30,960 --> 00:17:36,070
to different people,
but I'm also trained--

367
00:17:36,070 --> 00:17:38,540
I have medical training so
that may be just a prejudice,

368
00:17:38,540 --> 00:17:40,970
but I think it helps a lot
of people in a lot of ways.

369
00:17:40,970 --> 00:17:44,570
And so you mentioned before that
there were some science fiction

370
00:17:44,570 --> 00:17:48,300
authors that had inspired
you at the beginning to get

371
00:17:48,300 --> 00:17:49,650
into artificial intelligence?

372
00:17:49,650 --> 00:17:52,000
Or think about
artificial intelligence?

373
00:17:52,000 --> 00:17:53,700
Who were they?

374
00:17:53,700 --> 00:17:59,010
The main ones were the
Asimov books on robots.

375
00:17:59,010 --> 00:18:03,450
Maybe even more so were
Heinlein's The Moon

376
00:18:03,450 --> 00:18:06,233
is a Harsh Mistress was
one of my favorite--

377
00:18:06,233 --> 00:18:07,650
still is one of
my favorite books.

378
00:18:07,650 --> 00:18:13,730


379
00:18:13,730 --> 00:18:15,900
Those are the main ones I
can think of, but I mean,

380
00:18:15,900 --> 00:18:18,920
I've always been
interested in thinking,

381
00:18:18,920 --> 00:18:26,270
and I was lucky that,
when I was in college,

382
00:18:26,270 --> 00:18:28,940
thinking was a topic that
looked like you could actually

383
00:18:28,940 --> 00:18:29,570
go and study.

384
00:18:29,570 --> 00:18:33,970


385
00:18:33,970 --> 00:18:34,470
Cool.

386
00:18:34,470 --> 00:18:35,540
Well, thank you so much.

387
00:18:35,540 --> 00:18:37,340
Oh, you're welcome.

388
00:18:37,340 --> 00:18:41,000