1
00:00:00,000 --> 00:00:03,110


2
00:00:03,110 --> 00:00:05,050
Thanks for visiting today.

3
00:00:05,050 --> 00:00:05,550
OK.

4
00:00:05,550 --> 00:00:06,140
Great.

5
00:00:06,140 --> 00:00:07,850
So yes, good afternoon.

6
00:00:07,850 --> 00:00:10,490
So my name is Chuck Rich.

7
00:00:10,490 --> 00:00:12,335
I got interested
in AI when I was

8
00:00:12,335 --> 00:00:14,450
an undergraduate at the
University of Toronto

9
00:00:14,450 --> 00:00:15,900
around 19--

10
00:00:15,900 --> 00:00:17,960
early 1970s.

11
00:00:17,960 --> 00:00:21,710
I came to Tech Square
as an incoming graduate

12
00:00:21,710 --> 00:00:25,710
student in 1973 to the AI Lab.

13
00:00:25,710 --> 00:00:30,280
And I left in 1991.

14
00:00:30,280 --> 00:00:31,730
I only managed to kind of make--

15
00:00:31,730 --> 00:00:33,150
I didn't make escape velocity.

16
00:00:33,150 --> 00:00:35,670
I'm still sort of in
orbit around Tech Square.

17
00:00:35,670 --> 00:00:38,570
I'm currently working at
Mitsubishi Electric Research

18
00:00:38,570 --> 00:00:41,390
Labs at Cambridge and in
Kendall Square, about a block

19
00:00:41,390 --> 00:00:43,205
and a half from Tech Square.

20
00:00:43,205 --> 00:00:44,780
I've been there 15 years.

21
00:00:44,780 --> 00:00:49,320
I was one of the founding
members of that laboratory.

22
00:00:49,320 --> 00:00:49,820
Let me see.

23
00:00:49,820 --> 00:00:54,460
The kind of work I did in
the field was basically--

24
00:00:54,460 --> 00:01:02,720
so the first 7 and 3/4 years, I
was doing my master's and PhD.

25
00:01:02,720 --> 00:01:05,030
Both of which were concerned
with a project that

26
00:01:05,030 --> 00:01:09,020
was called the Programmer's
Apprentice project that

27
00:01:09,020 --> 00:01:11,810
was started by initially
myself and Howie Shrobe,

28
00:01:11,810 --> 00:01:13,820
and later joined by Dick waters.

29
00:01:13,820 --> 00:01:16,718
And that project was
really overlap between AI

30
00:01:16,718 --> 00:01:17,760
and software engineering.

31
00:01:17,760 --> 00:01:21,170
So it was using
software engineering

32
00:01:21,170 --> 00:01:23,450
as a domain for doing
basic research and AI,

33
00:01:23,450 --> 00:01:25,280
so [INAUDIBLE], reasoning.

34
00:01:25,280 --> 00:01:29,450
And also showing that AI could
be applied to making better

35
00:01:29,450 --> 00:01:31,190
software engineering tools.

36
00:01:31,190 --> 00:01:34,430
So it's an early project in the
sort of intersection between AI

37
00:01:34,430 --> 00:01:36,980
and software engineering.

38
00:01:36,980 --> 00:01:42,350
And so basically,
a couple of little

39
00:01:42,350 --> 00:01:45,020
anecdotes along the
way about the time

40
00:01:45,020 --> 00:01:49,820
and about Tech Square,
about my experience there.

41
00:01:49,820 --> 00:01:52,790
Because one of the first
interesting claims to fame

42
00:01:52,790 --> 00:01:56,660
that Howie Shrobe and I had is--

43
00:01:56,660 --> 00:01:59,510
so our supervisor
was Gerry Sussman.

44
00:01:59,510 --> 00:02:04,460
And I started talking to Gerry
for a master's thesis topic.

45
00:02:04,460 --> 00:02:09,020
And I was talking about AI
and software engineering.

46
00:02:09,020 --> 00:02:10,400
I was interested in both.

47
00:02:10,400 --> 00:02:13,340
And I was going to his office
as graduate students do

48
00:02:13,340 --> 00:02:17,000
from time to time and trying
to find an advisor and so on.

49
00:02:17,000 --> 00:02:19,520
And it turns out that Howie
Shrobe, who's also here--

50
00:02:19,520 --> 00:02:22,970
he's also a fellow--

51
00:02:22,970 --> 00:02:25,130
had independently
been talking to Gerry

52
00:02:25,130 --> 00:02:26,550
about fairly similar things.

53
00:02:26,550 --> 00:02:29,058
And so Jerry said
to both of us--

54
00:02:29,058 --> 00:02:30,350
he introduced us to each other.

55
00:02:30,350 --> 00:02:31,267
He said, listen, guys.

56
00:02:31,267 --> 00:02:33,500
You're both talking about
the same sort of thing.

57
00:02:33,500 --> 00:02:35,480
Why don't you do a
master's thesis together?

58
00:02:35,480 --> 00:02:39,620
Gerry's a very
collaborative person.

59
00:02:39,620 --> 00:02:40,703
And we said, fine.

60
00:02:40,703 --> 00:02:41,870
And he's a professor, right.

61
00:02:41,870 --> 00:02:45,743
Well it turns out it was
a great idea for Howie.

62
00:02:45,743 --> 00:02:47,660
We got along great, and
we've worked together,

63
00:02:47,660 --> 00:02:49,490
and have been friends
for a very long time.

64
00:02:49,490 --> 00:02:53,330
And it was very good from the
standpoint of the research.

65
00:02:53,330 --> 00:02:56,210
But it was a disaster from an
administrative point of view.

66
00:02:56,210 --> 00:02:59,330
It turns out that really
one of the big purposes

67
00:02:59,330 --> 00:03:00,540
of the master's program--

68
00:03:00,540 --> 00:03:01,790
at least the time I was there.

69
00:03:01,790 --> 00:03:02,960
And maybe it's still here.

70
00:03:02,960 --> 00:03:04,970
You had to do a master's
thesis before a PhD.

71
00:03:04,970 --> 00:03:08,360
It was to see if your master's
thesis was good enough

72
00:03:08,360 --> 00:03:09,770
to get into the PhD program.

73
00:03:09,770 --> 00:03:11,300
Because you really weren't
admitted into the PhD program

74
00:03:11,300 --> 00:03:13,520
until you finished
your master's thesis.

75
00:03:13,520 --> 00:03:15,650
And the department
was very concerned

76
00:03:15,650 --> 00:03:18,350
about if they had a
joint master's thesis,

77
00:03:18,350 --> 00:03:21,720
how are they going to know
there wasn't a weak sister.

78
00:03:21,720 --> 00:03:24,830
No feminist offense intended.

79
00:03:24,830 --> 00:03:25,610
That was a term.

80
00:03:25,610 --> 00:03:28,250
How to know who
did what and they

81
00:03:28,250 --> 00:03:31,560
wouldn't be able to decide who
should go to the PhD program.

82
00:03:31,560 --> 00:03:34,340
And so Gerry was
very, very good.

83
00:03:34,340 --> 00:03:36,475
He apparently stood on the desk.

84
00:03:36,475 --> 00:03:37,100
I wasn't there.

85
00:03:37,100 --> 00:03:38,960
But I'm told he actually stood
on the desk of the department

86
00:03:38,960 --> 00:03:39,585
chair's office.

87
00:03:39,585 --> 00:03:41,450
And he would resign
if they didn't let us

88
00:03:41,450 --> 00:03:42,815
both in the PhD program.

89
00:03:42,815 --> 00:03:45,050
That was an excellent
thesis and so on.

90
00:03:45,050 --> 00:03:46,640
And in fact, we
did some research.

91
00:03:46,640 --> 00:03:49,880
We were the only
joint master's thesis

92
00:03:49,880 --> 00:03:53,120
that had been done in several
decades in the EECS Department.

93
00:03:53,120 --> 00:03:56,840
And this was 1975, '76
that we actually had it in.

94
00:03:56,840 --> 00:03:59,390
And I bet you a nickel
that we're probably after--

95
00:03:59,390 --> 00:04:01,430
all the other students
saw what we went through.

96
00:04:01,430 --> 00:04:03,710
And they said, oh, god.

97
00:04:03,710 --> 00:04:06,620
You guys almost murdered
by this experience.

98
00:04:06,620 --> 00:04:10,370
I'm sure no one ever attempted
a joint master's thesis again.

99
00:04:10,370 --> 00:04:14,540
So an interesting recollection
there or claim to fame.

100
00:04:14,540 --> 00:04:16,040
I guess a couple
other things to say

101
00:04:16,040 --> 00:04:19,709
what people that inspired you.

102
00:04:19,709 --> 00:04:22,010
So Gerry Sussman, who
is my thesis advisor,

103
00:04:22,010 --> 00:04:25,050
both for a master's
and a PhD, certainly

104
00:04:25,050 --> 00:04:26,660
is a tremendous inspiration.

105
00:04:26,660 --> 00:04:28,910
And particularly the one
thing that I find just really

106
00:04:28,910 --> 00:04:34,890
admirable about Gerry, GJS,
and his research approach,

107
00:04:34,890 --> 00:04:38,090
which I confess I haven't
been able to duplicate,

108
00:04:38,090 --> 00:04:44,030
is he was, and probably
is, a master at

109
00:04:44,030 --> 00:04:48,210
embedding a really big idea
in a really small program.

110
00:04:48,210 --> 00:04:49,130
So when I was there--

111
00:04:49,130 --> 00:04:50,422
things like maintenance system.

112
00:04:50,422 --> 00:04:52,820
So you'd be there one
afternoon and we'd

113
00:04:52,820 --> 00:04:55,548
be talking about something,
he and John Doyle

114
00:04:55,548 --> 00:04:56,840
and some of the other students.

115
00:04:56,840 --> 00:04:59,660
And he stayed up all night
with one of the other students.

116
00:04:59,660 --> 00:05:02,180
The next morning he'd have
this program that did it.

117
00:05:02,180 --> 00:05:05,690
And it would be a
paper or a thesis.

118
00:05:05,690 --> 00:05:09,200
It was just really
a real knack for how

119
00:05:09,200 --> 00:05:11,690
to-- and you see
it in the scheme

120
00:05:11,690 --> 00:05:13,100
book and the whole approach.

121
00:05:13,100 --> 00:05:16,340
He has a way of
distilling ideas down

122
00:05:16,340 --> 00:05:18,785
into a really tight
computational framework.

123
00:05:18,785 --> 00:05:20,060
I, on the other hand--

124
00:05:20,060 --> 00:05:22,160
well I hope I have big ideas.

125
00:05:22,160 --> 00:05:25,790
But I'm lucky if I get them
into medium size programs.

126
00:05:25,790 --> 00:05:27,213
More often, I have big ideas.

127
00:05:27,213 --> 00:05:28,130
They're huge programs.

128
00:05:28,130 --> 00:05:30,713
And it's a lot more work and a
lot less successful to do that.

129
00:05:30,713 --> 00:05:35,000
So Gerry was just a terrific
example in that regard.

130
00:05:35,000 --> 00:05:37,460
I guess the other comment
that I was thinking

131
00:05:37,460 --> 00:05:38,840
about in this interview--

132
00:05:38,840 --> 00:05:42,800
I wanted to make about
those years, those 18 years

133
00:05:42,800 --> 00:05:46,886
at the eighth floor
in Tech Square.

134
00:05:46,886 --> 00:05:49,970
It was a different social
sense, particularly

135
00:05:49,970 --> 00:05:54,420
in the early time, late
'70s and early '80s.

136
00:05:54,420 --> 00:05:54,920
Yeah.

137
00:05:54,920 --> 00:05:57,140
At that point, I'm going
to sound old fogey,

138
00:05:57,140 --> 00:05:58,520
but there's something that
was really special then.

139
00:05:58,520 --> 00:06:00,690
And maybe it was just a
part of being a small lab.

140
00:06:00,690 --> 00:06:03,230
So first of all, at
that point, each year

141
00:06:03,230 --> 00:06:05,330
there were a handful of
new graduate students, two

142
00:06:05,330 --> 00:06:06,230
or three.

143
00:06:06,230 --> 00:06:07,107
It was a big deal.

144
00:06:07,107 --> 00:06:08,690
You never know who's
going to show up.

145
00:06:08,690 --> 00:06:13,220
I remember going into
Marvin's secretary's office

146
00:06:13,220 --> 00:06:15,680
and looking, peeking
at her desk to see

147
00:06:15,680 --> 00:06:18,380
the resumes of the incoming
graduate students this year,

148
00:06:18,380 --> 00:06:20,660
that kind of thing.

149
00:06:20,660 --> 00:06:23,540
Every person seemed
to really matter more.

150
00:06:23,540 --> 00:06:27,230
You also had the feeling
that each new PhD

151
00:06:27,230 --> 00:06:29,915
thesis was like a whole
new paradigm, which

152
00:06:29,915 --> 00:06:30,913
is good and bad.

153
00:06:30,913 --> 00:06:32,330
In many ways, in
sceince, you need

154
00:06:32,330 --> 00:06:33,710
to build on other people's work.

155
00:06:33,710 --> 00:06:36,110
But there was a certain
naivete at that point

156
00:06:36,110 --> 00:06:38,885
that the next big thing
was the next PhD thesis.

157
00:06:38,885 --> 00:06:40,760
In fact, one thing I
really remember strongly

158
00:06:40,760 --> 00:06:43,060
is in the playroom,
which was the open area

159
00:06:43,060 --> 00:06:45,440
on the eighth floor,
which was kind

160
00:06:45,440 --> 00:06:48,080
of the heart of the AI Lab.

161
00:06:48,080 --> 00:06:49,943
There were beanbag
chairs, and director

162
00:06:49,943 --> 00:06:51,110
chairs, and stuff like that.

163
00:06:51,110 --> 00:06:52,970
There was a table,
a big plant table.

164
00:06:52,970 --> 00:06:56,930
And one thing that I have
a really strong memory of,

165
00:06:56,930 --> 00:06:59,180
particularly when I was a
first year graduate student,

166
00:06:59,180 --> 00:07:04,980
was that on the table-- and I
would go in there at lunchtime.

167
00:07:04,980 --> 00:07:07,290
We would each lunch in there,
sandwiches or whatever.

168
00:07:07,290 --> 00:07:10,860
And on the table would be a
copy of Scott Fahlman's master's

169
00:07:10,860 --> 00:07:12,630
thesis that he just
finished working on.

170
00:07:12,630 --> 00:07:14,070
He hadn't quite
handed it in yet,

171
00:07:14,070 --> 00:07:16,140
like the second, or
third, or last draft.

172
00:07:16,140 --> 00:07:18,480
And he put it there,
and we would read it.

173
00:07:18,480 --> 00:07:20,210
It was like an important thing.

174
00:07:20,210 --> 00:07:21,960
And you'd pick up
Scott's thesis and you'd

175
00:07:21,960 --> 00:07:23,800
read it, and talk about
it, and talk to him.

176
00:07:23,800 --> 00:07:29,250
And it was a real sense of
important things happening.

177
00:07:29,250 --> 00:07:32,222
I remember reading Howard
Goldstein's PhD thesis

178
00:07:32,222 --> 00:07:33,180
when it first came out.

179
00:07:33,180 --> 00:07:34,763
And there it was
sitting on the table.

180
00:07:34,763 --> 00:07:36,000
And we had the advanced copy.

181
00:07:36,000 --> 00:07:37,770
And it was something
we looked forward to.

182
00:07:37,770 --> 00:07:41,760
We really felt part of a
process and excitement.

183
00:07:41,760 --> 00:07:45,750
And this was important
to understand that--

184
00:07:45,750 --> 00:07:48,480
later on, about the
time I left in '91,

185
00:07:48,480 --> 00:07:52,890
it was much more
business as usual.

186
00:07:52,890 --> 00:07:56,100
It was lab, and
there were offices,

187
00:07:56,100 --> 00:07:59,310
and people working with
their advisors and so on.

188
00:07:59,310 --> 00:08:03,000
But there wasn't that sense
of focus and this feeling

189
00:08:03,000 --> 00:08:04,590
that this was the next--

190
00:08:04,590 --> 00:08:06,310
that every thesis was
the next big thing.

191
00:08:06,310 --> 00:08:08,700
And I sort of miss that.

192
00:08:08,700 --> 00:08:10,320
And maybe it's just
the way the field

193
00:08:10,320 --> 00:08:13,540
itself has changed, for
better and for worse.

194
00:08:13,540 --> 00:08:16,770
Why do you think that is
that the shift took place?

195
00:08:16,770 --> 00:08:19,440
I guess, I think
there are two things.

196
00:08:19,440 --> 00:08:22,140
One is I think that the
beginnings of things

197
00:08:22,140 --> 00:08:22,790
are special.

198
00:08:22,790 --> 00:08:24,040
You see that in a lot of ways.

199
00:08:24,040 --> 00:08:26,100
And that was closer
to the beginning,

200
00:08:26,100 --> 00:08:28,530
and things are smaller,
and that's just true.

201
00:08:28,530 --> 00:08:31,260
In the lab I'm in now, I was
one of the founders of the lab.

202
00:08:31,260 --> 00:08:33,960
And the first three,
or four, or five years,

203
00:08:33,960 --> 00:08:36,270
there were five people, 10
people, and then 15 people.

204
00:08:36,270 --> 00:08:38,159
It was really special.

205
00:08:38,159 --> 00:08:40,020
And now we're like 75 people.

206
00:08:40,020 --> 00:08:42,037
And it's not a bad
place or anything.

207
00:08:42,037 --> 00:08:42,870
It's just different.

208
00:08:42,870 --> 00:08:44,070
Anybody who studies
organizations

209
00:08:44,070 --> 00:08:45,000
will tell you that.

210
00:08:45,000 --> 00:08:48,510
Also, there's this
whole issue of trying

211
00:08:48,510 --> 00:08:49,650
to be more scientific.

212
00:08:49,650 --> 00:08:52,240
And that is good.

213
00:08:52,240 --> 00:08:57,150
The idea that each PhD should
be a whole new paradigm is fun.

214
00:08:57,150 --> 00:08:58,517
But it's not exactly--

215
00:08:58,517 --> 00:08:59,100
it's not good.

216
00:08:59,100 --> 00:09:01,830
It's got its bad side to it
too, because you could never--

217
00:09:01,830 --> 00:09:03,455
we aren't complaining--
you could never

218
00:09:03,455 --> 00:09:05,610
build anybody else's code
or anybody else's work.

219
00:09:05,610 --> 00:09:07,480
And that was a problem.

220
00:09:07,480 --> 00:09:09,750
Also another big effect,
just at that time,

221
00:09:09,750 --> 00:09:13,830
was around 1980 David Maher
arrived at the AI Lab.

222
00:09:13,830 --> 00:09:16,530
I was there.

223
00:09:16,530 --> 00:09:18,360
If you're a researcher,
his name looms

224
00:09:18,360 --> 00:09:20,830
large in the history
of the field.

225
00:09:20,830 --> 00:09:25,980
He really had a big
effect on the lab,

226
00:09:25,980 --> 00:09:28,010
I think positive and negative.

227
00:09:28,010 --> 00:09:31,680


228
00:09:31,680 --> 00:09:34,057
Up to that point,
vision research--

229
00:09:34,057 --> 00:09:36,640
a little bit except for the work
that Berthold Horn was doing,

230
00:09:36,640 --> 00:09:38,848
which was always very
mathematical and very rigorous.

231
00:09:38,848 --> 00:09:42,270
But a lot of the vision research
was really quick and dirty

232
00:09:42,270 --> 00:09:44,460
and fairly ad hoc,
like all AI research

233
00:09:44,460 --> 00:09:49,680
was really, good old
fashioned AI style.

234
00:09:49,680 --> 00:09:53,790
And David showed how you
could do it differently, how

235
00:09:53,790 --> 00:09:56,190
you could really be scientific.

236
00:09:56,190 --> 00:10:00,533
And there's a lot of talk that
science was like a big hammer

237
00:10:00,533 --> 00:10:02,700
that you were going to be
banging people in the head

238
00:10:02,700 --> 00:10:03,200
with.

239
00:10:03,200 --> 00:10:04,800
And there's a lot,
I think, of envy

240
00:10:04,800 --> 00:10:08,550
of how much success and
progress, clear progress,

241
00:10:08,550 --> 00:10:10,980
David's students were
making, because he was really

242
00:10:10,980 --> 00:10:15,000
making them be very rigorous,
and very mathematical

243
00:10:15,000 --> 00:10:15,557
and son on.

244
00:10:15,557 --> 00:10:16,890
They pushed the vision problems.

245
00:10:16,890 --> 00:10:19,050
They were making progress
by leaps and bounds.

246
00:10:19,050 --> 00:10:21,240
And there's a certain
sense of superiority.

247
00:10:21,240 --> 00:10:24,420
They were more scientific than
the rest of us scruffy people.

248
00:10:24,420 --> 00:10:28,710
And there's an envy and desire
to imitate, that kind of thing.

249
00:10:28,710 --> 00:10:31,170
Marvin complained bitterly.

250
00:10:31,170 --> 00:10:32,155
I think he regretted.

251
00:10:32,155 --> 00:10:33,780
He said to me, well
at least in secret,

252
00:10:33,780 --> 00:10:36,580
that he brought David Maher in,
because he thought very highly

253
00:10:36,580 --> 00:10:37,080
of him.

254
00:10:37,080 --> 00:10:39,330
I don't know what the
context was exactly.

255
00:10:39,330 --> 00:10:42,150
And I think he regretted it.

256
00:10:42,150 --> 00:10:45,420
Because basically, everyone
tried to become a David Maher.

257
00:10:45,420 --> 00:10:49,390
And that wasn't maybe the
best thing for the field.

258
00:10:49,390 --> 00:10:51,022
You need a little
bit of diversity.

259
00:10:51,022 --> 00:10:52,980
And we shouldn't all try
to do things that way.

260
00:10:52,980 --> 00:10:54,813
That might have worked
at that point in time

261
00:10:54,813 --> 00:10:58,830
for that kind of a
problem in that area.

262
00:10:58,830 --> 00:11:02,740
But it was--

263
00:11:02,740 --> 00:11:04,450
I think that was the
period when there

264
00:11:04,450 --> 00:11:07,510
began to be a shift
towards trying to make

265
00:11:07,510 --> 00:11:10,480
AI more of a normal science.

266
00:11:10,480 --> 00:11:14,980
And I think David's presence
and David's example,

267
00:11:14,980 --> 00:11:18,560
positive example in most
ways, in many ways, I think,

268
00:11:18,560 --> 00:11:21,460
had a lot to do with
that shift, as well

269
00:11:21,460 --> 00:11:23,625
as just the
inevitable developing

270
00:11:23,625 --> 00:11:26,470
and the maturing of the field.

271
00:11:26,470 --> 00:11:29,200
So shifting gears
a little, what's

272
00:11:29,200 --> 00:11:32,020
the most challenging thing
you've worked on personally

273
00:11:32,020 --> 00:11:33,640
since you've been here?

274
00:11:33,640 --> 00:11:35,640
I guess the most challenging
area I've worked on

275
00:11:35,640 --> 00:11:37,150
is theorem proving.

276
00:11:37,150 --> 00:11:40,060
I guess as part of the
Programmer's Apprentice

277
00:11:40,060 --> 00:11:43,120
project, we were doing how
to represent knowledge,

278
00:11:43,120 --> 00:11:45,850
basically what programmers
know about programs, the kind

279
00:11:45,850 --> 00:11:47,270
of architectures and so on.

280
00:11:47,270 --> 00:11:49,360
And then we wanted to build
a system, a tool, that

281
00:11:49,360 --> 00:11:51,730
would help programmers
analyze their programs

282
00:11:51,730 --> 00:11:53,200
and synthesize them and so on.

283
00:11:53,200 --> 00:11:55,910
And that involved
automated deduction.

284
00:11:55,910 --> 00:12:00,130
And I ended up writing a kind
of an automated deduction

285
00:12:00,130 --> 00:12:01,240
theory proving system.

286
00:12:01,240 --> 00:12:05,800
And I must say it
basically scared me off

287
00:12:05,800 --> 00:12:07,570
of that for the
rest of my career.

288
00:12:07,570 --> 00:12:10,540
Because what I
discovered was that--

289
00:12:10,540 --> 00:12:13,160
and people just probably to
do this better than I do.

290
00:12:13,160 --> 00:12:17,590
But my attempts to do it was--

291
00:12:17,590 --> 00:12:21,790
you write a bunch of rules,
or axioms, or procedures

292
00:12:21,790 --> 00:12:22,400
interacted.

293
00:12:22,400 --> 00:12:25,505
And you'd be focused
on the path that you

294
00:12:25,505 --> 00:12:27,880
were hoping the system would
take on some examples you're

295
00:12:27,880 --> 00:12:28,960
working on.

296
00:12:28,960 --> 00:12:32,620
And you know the system,
the computer, whatever

297
00:12:32,620 --> 00:12:34,390
you want to call
it, was just so good

298
00:12:34,390 --> 00:12:37,480
at finding another path
which you could just count.

299
00:12:37,480 --> 00:12:38,620
It would just count.

300
00:12:38,620 --> 00:12:39,395
Oh, I got one.

301
00:12:39,395 --> 00:12:41,770
Successor, how about this
successor to the successor one?

302
00:12:41,770 --> 00:12:44,395
How about this successor to the
successor to the successor one?

303
00:12:44,395 --> 00:12:46,300
And somewhere in
your framework, there

304
00:12:46,300 --> 00:12:49,870
were some axioms that
could count essentially.

305
00:12:49,870 --> 00:12:52,005
Then you system would spend
all this time counting

306
00:12:52,005 --> 00:12:53,800
and not do any
useful work at all.

307
00:12:53,800 --> 00:12:57,520
So there's this feeling that
these systems, theorem proving

308
00:12:57,520 --> 00:13:00,890
systems, could just run
away from you really easily.

309
00:13:00,890 --> 00:13:03,220
And so since then
I sort of ended up

310
00:13:03,220 --> 00:13:10,780
adopting various techniques
for incomplete but efficient

311
00:13:10,780 --> 00:13:12,167
bounded inference techniques.

312
00:13:12,167 --> 00:13:13,750
And I put one more
plug in for someone

313
00:13:13,750 --> 00:13:17,500
who influenced me on that
one actually, which is--

314
00:13:17,500 --> 00:13:18,230
since he's here.

315
00:13:18,230 --> 00:13:20,740
It's David McAllester, who
had the office next to me

316
00:13:20,740 --> 00:13:21,910
as a graduate student.

317
00:13:21,910 --> 00:13:34,060
And I am still using code that
David wrote in 1979, '78, '79,

318
00:13:34,060 --> 00:13:35,680
originally in Lisp.

319
00:13:35,680 --> 00:13:38,920
And I translated it to
another version of Lisp.

320
00:13:38,920 --> 00:13:40,840
And I've translated
it into Java.

321
00:13:40,840 --> 00:13:44,680
And I'm still using his
code for the equality

322
00:13:44,680 --> 00:13:46,510
system for Boolean TMS.

323
00:13:46,510 --> 00:13:49,660
Actually, it's a combination
of a Boolean TMS,

324
00:13:49,660 --> 00:13:51,670
truth maintenance
system and equality

325
00:13:51,670 --> 00:13:54,760
system, which still
is actually down

326
00:13:54,760 --> 00:13:57,110
deep in the guts of
the work I'm doing now,

327
00:13:57,110 --> 00:14:03,400
which is more on interfaces
and task-based, collaborative

328
00:14:03,400 --> 00:14:04,670
interfaces.

329
00:14:04,670 --> 00:14:10,060
But down deep in the system is
that same algorithm and almost

330
00:14:10,060 --> 00:14:12,790
the same code translated
line by line--

331
00:14:12,790 --> 00:14:15,730
I've made a few improvements--
that David McAllester

332
00:14:15,730 --> 00:14:19,030
wrote in the late '70s.

333
00:14:19,030 --> 00:14:20,650
So I'm a big fan of David's.

334
00:14:20,650 --> 00:14:22,480
What does it do?

335
00:14:22,480 --> 00:14:28,970
It's a Boolean truth
mainteance system of equality.

336
00:14:28,970 --> 00:14:32,770
So what it does is it does unit
propositional resolution, which

337
00:14:32,770 --> 00:14:35,140
is a sound but
incomplete version

338
00:14:35,140 --> 00:14:38,860
of propositional
resolution combined

339
00:14:38,860 --> 00:14:43,330
with a complete decision
procedure for equality

340
00:14:43,330 --> 00:14:44,210
reasoning.

341
00:14:44,210 --> 00:14:48,190
So it's basically a
little inference engine

342
00:14:48,190 --> 00:14:52,210
that is a trade off between
efficiency and completeness.

343
00:14:52,210 --> 00:14:54,580
It's incomplete, but
it's very efficient.

344
00:14:54,580 --> 00:14:56,590
And it turns out for
a lot of the things

345
00:14:56,590 --> 00:14:59,890
that I use it for,
subsequently, it seemed

346
00:14:59,890 --> 00:15:01,740
to be just the right thing.

347
00:15:01,740 --> 00:15:05,467
And the code has just
been following me

348
00:15:05,467 --> 00:15:07,300
through revisions, and
through translations,

349
00:15:07,300 --> 00:15:08,320
and through translation.

350
00:15:08,320 --> 00:15:09,700
I tell David that.

351
00:15:09,700 --> 00:15:11,095
He cringes.

352
00:15:11,095 --> 00:15:12,235
I say, David, it's great.

353
00:15:12,235 --> 00:15:13,440
The code still works.

354
00:15:13,440 --> 00:15:14,990
It was good code
from the beginning.

355
00:15:14,990 --> 00:15:15,948
So that's kind of cool.

356
00:15:15,948 --> 00:15:17,870
So I don't know if
you work with it much,

357
00:15:17,870 --> 00:15:19,940
but do you have any strong
feelings about Prolog?

358
00:15:19,940 --> 00:15:20,440
Yeah.

359
00:15:20,440 --> 00:15:21,580
I do actually.

360
00:15:21,580 --> 00:15:24,970
I think that the
right approach is

361
00:15:24,970 --> 00:15:28,330
to have a prolog interpretor
written in some other language.

362
00:15:28,330 --> 00:15:29,950
So that for those
things for which

363
00:15:29,950 --> 00:15:35,410
you want to use that
backtracking logical algorithm,

364
00:15:35,410 --> 00:15:36,620
you could do that.

365
00:15:36,620 --> 00:15:39,443
But you could also do
conventional programming

366
00:15:39,443 --> 00:15:40,360
in the same framework.

367
00:15:40,360 --> 00:15:42,777
Because you don't end up doing
this ridiculous thing where

368
00:15:42,777 --> 00:15:45,380
you're using little tags to
make things sequentialize.

369
00:15:45,380 --> 00:15:48,025
It's always odd, the
things you're forced into.

370
00:15:48,025 --> 00:15:51,280


371
00:15:51,280 --> 00:15:55,630
So I think it's not a good
general purpose programming

372
00:15:55,630 --> 00:15:56,240
language.

373
00:15:56,240 --> 00:15:58,540
And you know you
need to use it when--

374
00:15:58,540 --> 00:16:03,710
for example, there's a various
a Lisp Prolog hybrid system.

375
00:16:03,710 --> 00:16:05,860
Is it LispLog or ProLisp?

376
00:16:05,860 --> 00:16:06,950
I forget what it's called.

377
00:16:06,950 --> 00:16:07,970
ProLisp sounds familiar.

378
00:16:07,970 --> 00:16:08,470
Yeah.

379
00:16:08,470 --> 00:16:08,990
I'm not sure.

380
00:16:08,990 --> 00:16:10,782
I think it's actually
the other way around.

381
00:16:10,782 --> 00:16:13,120
But to me, having--

382
00:16:13,120 --> 00:16:16,570
just like you'd want to have
a standard limitation of astar

383
00:16:16,570 --> 00:16:19,120
search available
in a library, you'd

384
00:16:19,120 --> 00:16:21,520
want to have a Prolog engine
there in your library,

385
00:16:21,520 --> 00:16:22,910
whether it's in Java or in Lisp.

386
00:16:22,910 --> 00:16:24,710
And you want to use it
when it's appropriate.

387
00:16:24,710 --> 00:16:26,835
But I don't think you want
to just have it be like,

388
00:16:26,835 --> 00:16:28,540
my whole system's
written in the Prolog.

389
00:16:28,540 --> 00:16:30,340
Often, you end up doing things
that are much easier done

390
00:16:30,340 --> 00:16:31,170
in some other way.

391
00:16:31,170 --> 00:16:32,920
But because you want
to keep it in Prolog,

392
00:16:32,920 --> 00:16:34,490
you end up making
the code obscure.

393
00:16:34,490 --> 00:16:36,940
So I'm not a big
fan in that regard.

394
00:16:36,940 --> 00:16:38,500
OK.

395
00:16:38,500 --> 00:16:43,380
Skipping back a little bit to
the idea of the Tech Square

396
00:16:43,380 --> 00:16:46,270
and AI becoming more scientific.

397
00:16:46,270 --> 00:16:50,500
Besides the approach to problems
and the mathematical rigor,

398
00:16:50,500 --> 00:16:55,840
I guess, can you think of any
maybe specific, maybe not,

399
00:16:55,840 --> 00:17:01,900
overlaps we can incorporate
from other fields?

400
00:17:01,900 --> 00:17:05,319
What would we do with
that overlap exactly?

401
00:17:05,319 --> 00:17:08,319
Let's see how best
to formulate this.

402
00:17:08,319 --> 00:17:14,980
Ways we can apply concepts from
a different scientific field.

403
00:17:14,980 --> 00:17:17,349
That happening all
the time, a lot

404
00:17:17,349 --> 00:17:23,170
of stuff from utility
theory, from economics.

405
00:17:23,170 --> 00:17:26,140
A lot of people are
doing that kind of work.

406
00:17:26,140 --> 00:17:32,602
It's not been my own
particular forte or initiative,

407
00:17:32,602 --> 00:17:33,310
not particularly.

408
00:17:33,310 --> 00:17:35,370
So I don't have any great
insights on that one.

409
00:17:35,370 --> 00:17:36,670
That's OK.

410
00:17:36,670 --> 00:17:38,417
Let's see.

411
00:17:38,417 --> 00:17:40,000
Besides your own
work of course, which

412
00:17:40,000 --> 00:17:41,450
is the coolest thing out there.

413
00:17:41,450 --> 00:17:43,158
Because otherwise,
why would you be here?

414
00:17:43,158 --> 00:17:44,230
No, no, no, no.

415
00:17:44,230 --> 00:17:47,650
What do you think is the coolest
thing going on right now?

416
00:17:47,650 --> 00:17:49,815
If I were--

417
00:17:49,815 --> 00:17:51,190
Bill said, if I
were young again.

418
00:17:51,190 --> 00:17:53,357
I'm not as old as Neal, so
I don't have to say that.

419
00:17:53,357 --> 00:17:56,590
But if I was starting fresh
out a brand new fresh area,

420
00:17:56,590 --> 00:17:59,260
I guess I'm most
interested, and would

421
00:17:59,260 --> 00:18:06,460
be most interested in, the
intersection of information

422
00:18:06,460 --> 00:18:11,540
technology, primarily AI,
nanotechnology, and biology.

423
00:18:11,540 --> 00:18:14,440
It seems to me that if you
look at where those are coming

424
00:18:14,440 --> 00:18:17,080
together, I think that's
where, in some sense,

425
00:18:17,080 --> 00:18:19,520
science fiction is
really going to happen.

426
00:18:19,520 --> 00:18:24,220
So I just think that that's
a very cool place to be.

427
00:18:24,220 --> 00:18:26,880
It would require a lot tooling
up on my part to get there.

428
00:18:26,880 --> 00:18:28,720
And I don't know if I
have the flexibility

429
00:18:28,720 --> 00:18:30,637
or have the opportunity
interest, career wise,

430
00:18:30,637 --> 00:18:33,860
if someone want to give me year
to support me, get up to speed.

431
00:18:33,860 --> 00:18:36,580
But if I was
starting out fresh, I

432
00:18:36,580 --> 00:18:39,440
would jump into the middle
of that intersection.

433
00:18:39,440 --> 00:18:40,737
Can you elaborate?

434
00:18:40,737 --> 00:18:42,070
Or can you explain specifically?

435
00:18:42,070 --> 00:18:42,610
No.

436
00:18:42,610 --> 00:18:44,030
It's not specific.

437
00:18:44,030 --> 00:18:50,890
It's more a general that there's
real things have to change.

438
00:18:50,890 --> 00:18:54,970
And the potential, the good
matter, is always change.

439
00:18:54,970 --> 00:18:57,260
Fundamentally, how
humans live in the world.

440
00:18:57,260 --> 00:18:59,090
I really do.

441
00:18:59,090 --> 00:19:06,310
So looking ahead, how do
you think it will affect us?

442
00:19:06,310 --> 00:19:08,170
I hate asking
people because they

443
00:19:08,170 --> 00:19:11,160
don't like to say
something concrete, but--

444
00:19:11,160 --> 00:19:13,320
You mean about AI in
general, or about--

445
00:19:13,320 --> 00:19:19,290
Yeah, or specific--
perhaps [INAUDIBLE]..

446
00:19:19,290 --> 00:19:20,510
Yeah-- no, I mean.

447
00:19:20,510 --> 00:19:23,135


448
00:19:23,135 --> 00:19:24,510
You know, what
gets me out of bed

449
00:19:24,510 --> 00:19:26,520
in the morning is
really I really

450
00:19:26,520 --> 00:19:29,710
want to see if an artificial
intelligence before I

451
00:19:29,710 --> 00:19:31,540
leave the planet.

452
00:19:31,540 --> 00:19:35,010
And I think I will.

453
00:19:35,010 --> 00:19:38,160
Assuming I live healthy
and so on, maybe another 50

454
00:19:38,160 --> 00:19:41,400
years or something,
I absolutely believe

455
00:19:41,400 --> 00:19:45,875
that I will see
a time when we'll

456
00:19:45,875 --> 00:19:48,000
be arguing about whether
artificial intelligence is

457
00:19:48,000 --> 00:19:48,500
around.

458
00:19:48,500 --> 00:19:51,390


459
00:19:51,390 --> 00:19:55,860
So artificial intelligence in
the form of say expert systems

460
00:19:55,860 --> 00:19:59,280
is incorporated into a lot--

461
00:19:59,280 --> 00:19:59,820
Oh sure.

462
00:19:59,820 --> 00:20:02,220
You can go to the innovative
applications of the AI

463
00:20:02,220 --> 00:20:02,823
conference.

464
00:20:02,823 --> 00:20:03,990
There's lots of great stuff.

465
00:20:03,990 --> 00:20:05,282
But I'm not talking about that.

466
00:20:05,282 --> 00:20:07,308
I'm talking about an artificial.

467
00:20:07,308 --> 00:20:08,850
That way I talk to
my students, and I

468
00:20:08,850 --> 00:20:11,400
had students that had recently
been in industry mostly, is--

469
00:20:11,400 --> 00:20:14,220


470
00:20:14,220 --> 00:20:16,650
basically a program,
or an entity,

471
00:20:16,650 --> 00:20:19,710
a physically embodied
robot, or a body agent

472
00:20:19,710 --> 00:20:23,730
on screen or something for
which there's something there.

473
00:20:23,730 --> 00:20:25,920
Sorry I was I should
have made it clear.

474
00:20:25,920 --> 00:20:28,530
I was switching gears.

475
00:20:28,530 --> 00:20:31,950
I know that these sort of
systems have been implemented.

476
00:20:31,950 --> 00:20:33,630
And they're behind
the scenes in a lot

477
00:20:33,630 --> 00:20:34,713
of the things you've seen.

478
00:20:34,713 --> 00:20:36,870


479
00:20:36,870 --> 00:20:40,050
Are people, do you
think, going to end up

480
00:20:40,050 --> 00:20:42,970
interacting with something that
specifically identify as AI,

481
00:20:42,970 --> 00:20:44,920
like this computer I'm talking.

482
00:20:44,920 --> 00:20:45,840
Yeah.

483
00:20:45,840 --> 00:20:48,080
I think so.

484
00:20:48,080 --> 00:20:51,240
Just look at avatar
and so on, the face.

485
00:20:51,240 --> 00:20:52,110
Yeah, absolutely.

486
00:20:52,110 --> 00:20:53,250
People want to.

487
00:20:53,250 --> 00:20:55,680
It's absolutely clear.

488
00:20:55,680 --> 00:20:57,990
Someone stood up in the
earlier session this morning

489
00:20:57,990 --> 00:21:00,540
and talked about your
books are getting

490
00:21:00,540 --> 00:21:04,770
some huge amount of money
in euros for building

491
00:21:04,770 --> 00:21:06,960
companions for older people.

492
00:21:06,960 --> 00:21:09,480
I think it's absolutely
going to happen,

493
00:21:09,480 --> 00:21:10,920
and sooner than we think.

494
00:21:10,920 --> 00:21:14,320


495
00:21:14,320 --> 00:21:17,260
I should get back to a session.

496
00:21:17,260 --> 00:21:18,610
Thanks for talking to us today.

497
00:21:18,610 --> 00:21:18,880
Yeah.

498
00:21:18,880 --> 00:21:19,380
Thanks.

499
00:21:19,380 --> 00:21:22,516
And I'll talk to you later.

500
00:21:22,516 --> 00:21:23,016