1
00:00:00,000 --> 00:00:01,736


2
00:00:01,736 --> 00:00:06,050
I guess we'll start having
you introduce yourself.

3
00:00:06,050 --> 00:00:07,060
My name is Reid Simmons.

4
00:00:07,060 --> 00:00:10,750
I'm a research professor at
Carnegie Mellon University

5
00:00:10,750 --> 00:00:13,040
in robotics and
computer science.

6
00:00:13,040 --> 00:00:24,040
And I came to the AI Lab at MIT
in 1979 to do my graduate work.

7
00:00:24,040 --> 00:00:27,220
As an undergraduate, I
was very much interested

8
00:00:27,220 --> 00:00:30,107
in computer graphics
and that's basically

9
00:00:30,107 --> 00:00:31,690
where I thought I
was going to end up.

10
00:00:31,690 --> 00:00:36,310
And I took a course from
Stuart Shapiro, who is also

11
00:00:36,310 --> 00:00:41,740
a AAAI fellow now, and just kind
of was blown away by the fact

12
00:00:41,740 --> 00:00:45,970
that you could--

13
00:00:45,970 --> 00:00:47,410
there actually
was a field there.

14
00:00:47,410 --> 00:00:50,110
People weren't just
playing, making machines

15
00:00:50,110 --> 00:00:50,920
to be intelligent.

16
00:00:50,920 --> 00:00:54,880
But there are actually people
out there doing research,

17
00:00:54,880 --> 00:00:56,518
and you can actually
have a career

18
00:00:56,518 --> 00:00:57,560
doing that type of thing.

19
00:00:57,560 --> 00:00:59,770
So I say, wow, that's
a lot more interesting

20
00:00:59,770 --> 00:01:01,742
than doing computer graphics.

21
00:01:01,742 --> 00:01:03,700
Those were the days when
computer graphics were

22
00:01:03,700 --> 00:01:06,310
like Pong, not what it is now.

23
00:01:06,310 --> 00:01:12,970
But I think I picked the
field that didn't accelerate

24
00:01:12,970 --> 00:01:15,160
as much as the
other one, but it's

25
00:01:15,160 --> 00:01:16,660
been a lot more
interesting I think.

26
00:01:16,660 --> 00:01:22,420


27
00:01:22,420 --> 00:01:26,120
So I guess we ended
up at the MIT AI

28
00:01:26,120 --> 00:01:28,630
Lab, what kinds of
things were going

29
00:01:28,630 --> 00:01:29,970
on there that you remember?

30
00:01:29,970 --> 00:01:31,450
Any cool projects?

31
00:01:31,450 --> 00:01:34,110


32
00:01:34,110 --> 00:01:34,610
People?

33
00:01:34,610 --> 00:01:35,350
Yeah.

34
00:01:35,350 --> 00:01:36,070
There is a bit.

35
00:01:36,070 --> 00:01:39,010
I started out working with--

36
00:01:39,010 --> 00:01:42,100
for the first year
basically, I was

37
00:01:42,100 --> 00:01:44,200
working with Chuck
Rich and Dick Waters

38
00:01:44,200 --> 00:01:45,910
and the Programmer's Apprentice.

39
00:01:45,910 --> 00:01:50,740
And what they were trying to
do is get a software system

40
00:01:50,740 --> 00:01:55,480
to generate software
automatically.

41
00:01:55,480 --> 00:01:59,230
But after the first
year, I switched over

42
00:01:59,230 --> 00:02:01,990
to working with Randy Davis.

43
00:02:01,990 --> 00:02:04,660
He just started putting
together a project

44
00:02:04,660 --> 00:02:07,730
to do model-based reasoning.

45
00:02:07,730 --> 00:02:11,870
So the big rage
up until that time

46
00:02:11,870 --> 00:02:14,870
had really been expert
systems, role based systems.

47
00:02:14,870 --> 00:02:19,600
And there was the
common understanding

48
00:02:19,600 --> 00:02:22,060
that these systems
were very brittle,

49
00:02:22,060 --> 00:02:25,690
because they didn't really
capture didn't really

50
00:02:25,690 --> 00:02:26,530
capture knowledge.

51
00:02:26,530 --> 00:02:28,870
They just captured
rules of thumb.

52
00:02:28,870 --> 00:02:31,180
And so what we tried
to do is develop

53
00:02:31,180 --> 00:02:35,050
systems that had a deeper
understanding of whatever

54
00:02:35,050 --> 00:02:38,470
problem they were trying to
solve and more general ways

55
00:02:38,470 --> 00:02:39,820
of solving those problems.

56
00:02:39,820 --> 00:02:44,320
And most of his
group was working

57
00:02:44,320 --> 00:02:48,885
on analysis of circuits.

58
00:02:48,885 --> 00:02:51,930


59
00:02:51,930 --> 00:02:54,720
Because of my
background in graphics,

60
00:02:54,720 --> 00:02:57,580
I was interested more
in spatial reasoning.

61
00:02:57,580 --> 00:03:04,590
And he had some contacts
with Schlumberger,

62
00:03:04,590 --> 00:03:06,240
who was doing oil exploration.

63
00:03:06,240 --> 00:03:07,800
And they were very
much interested

64
00:03:07,800 --> 00:03:08,850
in spatial reasoning.

65
00:03:08,850 --> 00:03:10,470
They were trying to
understand what's

66
00:03:10,470 --> 00:03:14,250
happening underneath the ground
by looking at the data that

67
00:03:14,250 --> 00:03:17,110
comes back, and trying to figure
out what was happening there,

68
00:03:17,110 --> 00:03:21,030
and what processes must have
taken place millions of years

69
00:03:21,030 --> 00:03:21,870
ago.

70
00:03:21,870 --> 00:03:24,037
Because they were interested
in trying to figure out

71
00:03:24,037 --> 00:03:28,380
where oil was, so it
was a nice combination--

72
00:03:28,380 --> 00:03:32,370
it was a nice
combination, my interest

73
00:03:32,370 --> 00:03:37,210
in both model-based reasoning
and also spatial reasoning.

74
00:03:37,210 --> 00:03:41,682
So I started working with
them on that problem of doing

75
00:03:41,682 --> 00:03:42,765
geological interpretation.

76
00:03:42,765 --> 00:03:48,960
And trying to understand from a
picture of the geologic region

77
00:03:48,960 --> 00:03:50,640
what actually occurred.

78
00:03:50,640 --> 00:03:58,350
And this was related to
work in planning, basically

79
00:03:58,350 --> 00:03:59,580
interpretation.

80
00:03:59,580 --> 00:04:01,890
Trying to figure out what
sequence of events happened

81
00:04:01,890 --> 00:04:03,060
is the opposite.

82
00:04:03,060 --> 00:04:04,800
The inverse of the
problem of planning

83
00:04:04,800 --> 00:04:06,960
was just to generate
a sequence of events

84
00:04:06,960 --> 00:04:08,380
to make something happen.

85
00:04:08,380 --> 00:04:11,966
So I got involved in
the planning committee,

86
00:04:11,966 --> 00:04:16,170
at which this was a time
when you a lot of things

87
00:04:16,170 --> 00:04:17,070
were going on there.

88
00:04:17,070 --> 00:04:22,790


89
00:04:22,790 --> 00:04:25,340
Drew McDermott was
very much involved

90
00:04:25,340 --> 00:04:26,890
in doing work and planning.

91
00:04:26,890 --> 00:04:30,230
And he was at Yale at the time
but had done his work at MIT.

92
00:04:30,230 --> 00:04:35,340
And he was a big influence
on the work I've done.

93
00:04:35,340 --> 00:04:44,040
Gerald Sussman, his early
work on planning and debugging

94
00:04:44,040 --> 00:04:49,580
was influential when
I was working on it.

95
00:04:49,580 --> 00:04:52,790
And then some of
the fellow students,

96
00:04:52,790 --> 00:04:54,980
Phil Adrian and Dave
Chapman, came along

97
00:04:54,980 --> 00:04:58,640
and turned the whole
thing on its ear

98
00:04:58,640 --> 00:05:02,200
by saying, well, intelligence
wasn't really planning

99
00:05:02,200 --> 00:05:02,700
any more.

100
00:05:02,700 --> 00:05:08,900
It was just more
reacting to the moment.

101
00:05:08,900 --> 00:05:11,870
So it was a time when there
was a lot of big arguments

102
00:05:11,870 --> 00:05:18,050
in the AI Lab about
model-based reasoning

103
00:05:18,050 --> 00:05:24,350
processes versus more reactive
embedded type situations.

104
00:05:24,350 --> 00:05:27,570


105
00:05:27,570 --> 00:05:30,950
For a long time, the two
camps were pretty separate.

106
00:05:30,950 --> 00:05:32,870
You're either in one
camp or the other.

107
00:05:32,870 --> 00:05:35,900
And now after 20
years have passed,

108
00:05:35,900 --> 00:05:37,880
there's been a
lot more blending.

109
00:05:37,880 --> 00:05:41,840
And most people, I think,
will agree that neither camp

110
00:05:41,840 --> 00:05:43,920
is right or wrong.

111
00:05:43,920 --> 00:05:45,690
There's aspects of both of them.

112
00:05:45,690 --> 00:05:49,160
And the trick has been to see
how we can all put it together.

113
00:05:49,160 --> 00:05:54,320
But in those days, people had
very strong opinions of me.

114
00:05:54,320 --> 00:06:00,200
And they really tried to
stake out very rigid positions

115
00:06:00,200 --> 00:06:02,584
and in order to make points.

116
00:06:02,584 --> 00:06:03,440
OK.

117
00:06:03,440 --> 00:06:04,430
That make sense.

118
00:06:04,430 --> 00:06:07,570
So I guess where did
you head after that,

119
00:06:07,570 --> 00:06:08,980
the next stage in your career?

120
00:06:08,980 --> 00:06:14,930


121
00:06:14,930 --> 00:06:19,130
I had every expectation of
continuing on in the area

122
00:06:19,130 --> 00:06:20,570
that I've done my thesis work.

123
00:06:20,570 --> 00:06:27,810
And I was looking for
position, and, by chance,

124
00:06:27,810 --> 00:06:30,020
I got a call from
Tom Mitchell, who was

125
00:06:30,020 --> 00:06:32,060
at Carnegie Mellon at the time.

126
00:06:32,060 --> 00:06:35,960
And he said that he had just--

127
00:06:35,960 --> 00:06:39,410
he was involved in a robotics
project in which there

128
00:06:39,410 --> 00:06:43,610
was Red Whitaker, who
was in charge of building

129
00:06:43,610 --> 00:06:47,660
the robots, Takeo Kanade, who
was in charge of doing machine

130
00:06:47,660 --> 00:06:52,370
perception, the vision,
and Tom was basically

131
00:06:52,370 --> 00:06:54,530
in charge of making
the robot smart.

132
00:06:54,530 --> 00:06:59,570
And he needed someone to
come and basically work

133
00:06:59,570 --> 00:07:03,650
as postdoc to do that.

134
00:07:03,650 --> 00:07:07,310
I'd never done any
work in robotics.

135
00:07:07,310 --> 00:07:13,100
There was a lot of work in
the AI Lab being done there.

136
00:07:13,100 --> 00:07:18,404
Rod Brooks was there,
Tomas Lozano-Perez,

137
00:07:18,404 --> 00:07:23,210
and a lot of Brooks's students.

138
00:07:23,210 --> 00:07:30,290
Ian Horswell and John Cannell
were doing interesting work

139
00:07:30,290 --> 00:07:32,875
there.

140
00:07:32,875 --> 00:07:34,250
But I had never
done any of that.

141
00:07:34,250 --> 00:07:37,940
And in fact, the opposite,
I was in the camp

142
00:07:37,940 --> 00:07:40,400
where you do a lot of--
it's a lot of reasoning,

143
00:07:40,400 --> 00:07:43,193
and you think about
things, and that's it.

144
00:07:43,193 --> 00:07:45,110
So now it's like you're
reasoning and thinking

145
00:07:45,110 --> 00:07:47,330
about things, and you have
to actually connect it

146
00:07:47,330 --> 00:07:48,860
to the real world.

147
00:07:48,860 --> 00:07:50,820
And that was a real eye opener.

148
00:07:50,820 --> 00:07:53,390
So that basically
completely changed

149
00:07:53,390 --> 00:07:55,550
the nature of my
research by actually

150
00:07:55,550 --> 00:07:58,310
having to make robots work.

151
00:07:58,310 --> 00:08:02,720
Rather than just sticking with
a very high level reasoning

152
00:08:02,720 --> 00:08:06,230
aspects we had to look at how to
actually deal with uncertainty,

153
00:08:06,230 --> 00:08:08,870
how to deal with the fact that
the world is really messy.

154
00:08:08,870 --> 00:08:14,880
And as I said, it's helped--

155
00:08:14,880 --> 00:08:18,960
it brought those two
competing factions

156
00:08:18,960 --> 00:08:20,310
closer and closer together.

157
00:08:20,310 --> 00:08:22,620
Because it's clear
that to actually do

158
00:08:22,620 --> 00:08:26,130
things in the world, you
have to react to the moment

159
00:08:26,130 --> 00:08:28,260
and what's there.

160
00:08:28,260 --> 00:08:30,150
But on the other hand,
to be intelligent,

161
00:08:30,150 --> 00:08:32,490
you have to take a
step back and look back

162
00:08:32,490 --> 00:08:35,039
at the higher level reasoning.

163
00:08:35,039 --> 00:08:35,850
OK.

164
00:08:35,850 --> 00:08:38,220
That makes a lot of sense.

165
00:08:38,220 --> 00:08:41,450
So I guess I'll ask what some
of the most challenging things

166
00:08:41,450 --> 00:08:42,782
you're facing right now are.

167
00:08:42,782 --> 00:08:46,730


168
00:08:46,730 --> 00:08:50,960
So one of the things
that I've been interested

169
00:08:50,960 --> 00:08:53,600
in for the last couple of years,
and it's kind of consuming

170
00:08:53,600 --> 00:08:57,680
is human robot
social interaction.

171
00:08:57,680 --> 00:09:11,810
And the interesting thing is
that as I start getting into,

172
00:09:11,810 --> 00:09:16,220
there's more and more
evidence that intelligence

173
00:09:16,220 --> 00:09:20,900
and being intelligent and being
social are inextricably linked.

174
00:09:20,900 --> 00:09:25,250
So a lot of what we have what
we consider intelligence,

175
00:09:25,250 --> 00:09:27,800
like language and
communication, are

176
00:09:27,800 --> 00:09:30,950
there more for social
reasons than anything else.

177
00:09:30,950 --> 00:09:33,770
And we're intrinsically
social creatures.

178
00:09:33,770 --> 00:09:38,960
And you can't
separate-- you really

179
00:09:38,960 --> 00:09:41,780
can't separate the social
aspects from the intelligence

180
00:09:41,780 --> 00:09:44,970
aspects in many ways.

181
00:09:44,970 --> 00:09:50,780
And it's fascinating because
people are very, very good

182
00:09:50,780 --> 00:09:54,170
at both generating and
reading social cues,

183
00:09:54,170 --> 00:09:59,840
from gaze, from gesture, from
tone of voice, even posture.

184
00:09:59,840 --> 00:10:02,450


185
00:10:02,450 --> 00:10:08,060
If I want to become assertive,
I can lean in like this.

186
00:10:08,060 --> 00:10:13,610
And that tells you a lot
about what I'm thinking about

187
00:10:13,610 --> 00:10:16,730
and how you should
react to that.

188
00:10:16,730 --> 00:10:19,422
And when I back off, It's
a lot less threatening

189
00:10:19,422 --> 00:10:20,255
because I'm relaxed.

190
00:10:20,255 --> 00:10:23,360


191
00:10:23,360 --> 00:10:26,390
And our robot systems can't
do any of that, in general.

192
00:10:26,390 --> 00:10:30,110
There's some work
that's going on in that,

193
00:10:30,110 --> 00:10:33,060
but it's still fairly
rudimentary I think.

194
00:10:33,060 --> 00:10:35,740


195
00:10:35,740 --> 00:10:39,290
So the interesting
thing is if we

196
00:10:39,290 --> 00:10:44,247
make these systems, these robots
in particular, more social,

197
00:10:44,247 --> 00:10:45,830
are they going to
be more intelligent,

198
00:10:45,830 --> 00:10:48,260
just by virtue of the fact
that they're more social?

199
00:10:48,260 --> 00:10:49,760
That's a really
challenging problem.

200
00:10:49,760 --> 00:10:52,520


201
00:10:52,520 --> 00:10:55,340
Something that we're
hoping is actually

202
00:10:55,340 --> 00:10:57,680
going to bear a lot of fruit.

203
00:10:57,680 --> 00:10:59,030
That makes sense.

204
00:10:59,030 --> 00:11:01,310
OK.

205
00:11:01,310 --> 00:11:02,463
Let's see.

206
00:11:02,463 --> 00:11:03,630
Do you read science fiction?

207
00:11:03,630 --> 00:11:06,860


208
00:11:06,860 --> 00:11:07,370
Yeah.

209
00:11:07,370 --> 00:11:09,080
I do read science fiction.

210
00:11:09,080 --> 00:11:12,360
What are some of your favorite
things you read recently?

211
00:11:12,360 --> 00:11:14,030
Oh.

212
00:11:14,030 --> 00:11:15,320
Let's see.

213
00:11:15,320 --> 00:11:19,948
One of my favorite authors
is Orson Scott Card,

214
00:11:19,948 --> 00:11:20,740
the Ender's series.

215
00:11:20,740 --> 00:11:24,320


216
00:11:24,320 --> 00:11:30,340
And more recently, I've become
enamored with Philip Dick.

217
00:11:30,340 --> 00:11:35,947
And I've never really had much
of an interest in reading,

218
00:11:35,947 --> 00:11:37,405
in general, reading
science fiction

219
00:11:37,405 --> 00:11:39,200
that had to do with robots.

220
00:11:39,200 --> 00:11:43,550
But I started reading
some of his stuff,

221
00:11:43,550 --> 00:11:45,740
and it's just very engrossing.

222
00:11:45,740 --> 00:11:48,590


223
00:11:48,590 --> 00:11:54,200
Yeah I wouldn't-- a lot of
people I talked to say that

224
00:11:54,200 --> 00:11:57,680
either reading science
fiction about robots or AI,

225
00:11:57,680 --> 00:12:02,570
or seeing movies like 2001 where
the things that kind of drove

226
00:12:02,570 --> 00:12:04,010
them into the field.

227
00:12:04,010 --> 00:12:07,130
And I can honestly
say that I don't

228
00:12:07,130 --> 00:12:09,080
think that had anything
to do with the fact

229
00:12:09,080 --> 00:12:11,940
that I got involved in this.

230
00:12:11,940 --> 00:12:19,310
It really was just this notion
back in the mid '70s that you

231
00:12:19,310 --> 00:12:23,000
could have-- and these
days, the computers--

232
00:12:23,000 --> 00:12:24,740
we had a computer
at our university,

233
00:12:24,740 --> 00:12:26,035
which I remember them.

234
00:12:26,035 --> 00:12:30,740
They were going to cost $5
million and had 100K of memory.

235
00:12:30,740 --> 00:12:35,020
It was probably less powerful
than the computer in you

236
00:12:35,020 --> 00:12:37,940
cell phone.

237
00:12:37,940 --> 00:12:42,380
And this was what the
whole university was using.

238
00:12:42,380 --> 00:12:45,710
And it was incredible
to think at that time

239
00:12:45,710 --> 00:12:51,770
that something so relatively
slow, compared to today,

240
00:12:51,770 --> 00:12:56,420
was going to be able to emulate
the types of capabilities

241
00:12:56,420 --> 00:12:59,505
that units of humans exhibit.

242
00:12:59,505 --> 00:13:01,130
And you really had
to take it on faith.

243
00:13:01,130 --> 00:13:02,922
Because a lot of the
algorithms that people

244
00:13:02,922 --> 00:13:08,900
were developing would run in
minutes rather than in seconds

245
00:13:08,900 --> 00:13:10,100
to make certain decisions.

246
00:13:10,100 --> 00:13:13,610
And you say, this
can't be right.

247
00:13:13,610 --> 00:13:16,340
How could this be the way that--

248
00:13:16,340 --> 00:13:18,470
how could this be a
good model for how

249
00:13:18,470 --> 00:13:20,990
humans work if it
takes minutes to do

250
00:13:20,990 --> 00:13:22,610
what humans can do in seconds?

251
00:13:22,610 --> 00:13:26,300
And now of course, with
computers 1,000 times faster

252
00:13:26,300 --> 00:13:32,180
than 25 years, 30 years
ago when I started,

253
00:13:32,180 --> 00:13:35,540
the things that used to take
minutes now take milliseconds.

254
00:13:35,540 --> 00:13:38,030
And it's all feasible now.

255
00:13:38,030 --> 00:13:43,610
So it's interesting
how the technology

256
00:13:43,610 --> 00:13:48,020
has evolved to the point where
a lot of the ideas that people

257
00:13:48,020 --> 00:13:51,620
had years and years ago that
other people thought were just

258
00:13:51,620 --> 00:13:57,530
completely infeasible
have become

259
00:13:57,530 --> 00:14:00,030
feasible because of technology.

260
00:14:00,030 --> 00:14:02,780
So where do you
think we're headed?

261
00:14:02,780 --> 00:14:04,770
What's AI going to do?

262
00:14:04,770 --> 00:14:05,270
Or robotics?

263
00:14:05,270 --> 00:14:09,020


264
00:14:09,020 --> 00:14:13,130
So they've been in academic
institutions for a while now.

265
00:14:13,130 --> 00:14:16,190
Is Homer Simpson ever going
run into a situation where

266
00:14:16,190 --> 00:14:18,080
he says, I'm talking
to a computer

267
00:14:18,080 --> 00:14:22,280
or computers are directly
affecting my life as opposed

268
00:14:22,280 --> 00:14:23,450
to--

269
00:14:23,450 --> 00:14:26,440
for example, expert systems have
been in industry and planning

270
00:14:26,440 --> 00:14:27,840
and stuff for a long time.

271
00:14:27,840 --> 00:14:32,120
But they're not really
directly interacting--

272
00:14:32,120 --> 00:14:32,900
With people.

273
00:14:32,900 --> 00:14:33,860
Yeah.

274
00:14:33,860 --> 00:14:34,610
I think so.

275
00:14:34,610 --> 00:14:39,500
This is one of the reasons that
drove me into social robots.

276
00:14:39,500 --> 00:14:42,640
I'll give an anecdote from
my time at Carnegie Mellon.

277
00:14:42,640 --> 00:14:44,990
We had a robot named Xavier.

278
00:14:44,990 --> 00:14:48,230


279
00:14:48,230 --> 00:14:52,350
We were interested
in robust navigation.

280
00:14:52,350 --> 00:14:54,680
So in order to demonstrate
robust navigation,

281
00:14:54,680 --> 00:14:57,350
you wanted to navigate
for many hours

282
00:14:57,350 --> 00:14:58,950
over long periods of time.

283
00:14:58,950 --> 00:15:01,850
And so what we did was
rather than just have

284
00:15:01,850 --> 00:15:05,420
it wander the halls,
we had it to be

285
00:15:05,420 --> 00:15:07,520
able to go from room to room.

286
00:15:07,520 --> 00:15:10,375
You could, over the internet,
tell-- this was the early '90s.

287
00:15:10,375 --> 00:15:11,750
Over the internet,
you could tell

288
00:15:11,750 --> 00:15:14,340
it to go to different places.

289
00:15:14,340 --> 00:15:17,270
And so the robot would
basically wander the hall

290
00:15:17,270 --> 00:15:21,050
for four or five hours a day.

291
00:15:21,050 --> 00:15:25,520
And we kept this going for
quite a number of months.

292
00:15:25,520 --> 00:15:28,760
And it validated the
algorithms that we worked on

293
00:15:28,760 --> 00:15:33,140
but what was also interesting
was the social aspects.

294
00:15:33,140 --> 00:15:36,390
You could see how people
reacted to the robots.

295
00:15:36,390 --> 00:15:41,420
And these were people who were
in a technology University.

296
00:15:41,420 --> 00:15:43,640
They were in a place
that did robotics,

297
00:15:43,640 --> 00:15:45,740
but, by and large, they'd
never seen a robot.

298
00:15:45,740 --> 00:15:48,230
And they didn't know how
to react to the robot.

299
00:15:48,230 --> 00:15:50,810
And you could see
that there were

300
00:15:50,810 --> 00:15:52,790
kind of three types of people.

301
00:15:52,790 --> 00:15:54,700
There were the wall huggers.

302
00:15:54,700 --> 00:15:56,450
The robot would come
down, and they'd just

303
00:15:56,450 --> 00:15:58,450
stand against the wall
like this and wait for it

304
00:15:58,450 --> 00:16:00,710
to pass and let it go through.

305
00:16:00,710 --> 00:16:03,980
And those are the most
prevalent actually.

306
00:16:03,980 --> 00:16:06,530
And then there were people
who were kind of nonchalant,

307
00:16:06,530 --> 00:16:08,780
who would walk by it normally.

308
00:16:08,780 --> 00:16:11,610
And that was prob--
that was a small number.

309
00:16:11,610 --> 00:16:16,202
And then actually, the smallest
number were the antagonist.

310
00:16:16,202 --> 00:16:17,660
They would jump in
front of a robot

311
00:16:17,660 --> 00:16:20,660
and harass it and
say, if you think

312
00:16:20,660 --> 00:16:22,850
you deserve to be in
this corridor with me,

313
00:16:22,850 --> 00:16:24,920
you know you need
to get out of my way

314
00:16:24,920 --> 00:16:27,560
or be able to deal with the
fact that I'm in your face.

315
00:16:27,560 --> 00:16:29,780
And those are graduate students.

316
00:16:29,780 --> 00:16:33,590
But what was interesting is
that even with the people

317
00:16:33,590 --> 00:16:38,330
who are nonchalant about
it, there was always--

318
00:16:38,330 --> 00:16:40,327
you had the situation
where you're

319
00:16:40,327 --> 00:16:42,410
trying to go through a
doorway and someone else is

320
00:16:42,410 --> 00:16:45,050
coming through the doorway
and does the little dance.

321
00:16:45,050 --> 00:16:47,990
And so you go back and forth
a few times before you finally

322
00:16:47,990 --> 00:16:49,130
figure out how to pass.

323
00:16:49,130 --> 00:16:50,740
Well this happened all the time.

324
00:16:50,740 --> 00:16:54,260
And it was because the robot
didn't know the social rule.

325
00:16:54,260 --> 00:16:56,640
The social rule is you're
coming down the corridor,

326
00:16:56,640 --> 00:16:57,620
you go to the right.

327
00:16:57,620 --> 00:16:59,078
The other person
goes to the right.

328
00:16:59,078 --> 00:16:59,930
You pass.

329
00:16:59,930 --> 00:17:05,300
And the robot-- because it
didn't have any inkling of that

330
00:17:05,300 --> 00:17:08,119
social rule and it was just
trying to find the best free

331
00:17:08,119 --> 00:17:09,919
path to go through--

332
00:17:09,919 --> 00:17:11,900
caused a lot of consternation.

333
00:17:11,900 --> 00:17:13,235
It was fine for the robot.

334
00:17:13,235 --> 00:17:14,170
The robot didn't care.

335
00:17:14,170 --> 00:17:15,170
It didn't have feelings.

336
00:17:15,170 --> 00:17:19,200
It didn't have to get any
place any particular time.

337
00:17:19,200 --> 00:17:22,319
But the people were always--
and so a lot of times the way

338
00:17:22,319 --> 00:17:25,472
they would react to it was,
rather than trying to out think

339
00:17:25,472 --> 00:17:26,930
the robot, what it
was going to do,

340
00:17:26,930 --> 00:17:30,565
they would just stand to this
side, let the robot pass,

341
00:17:30,565 --> 00:17:31,190
and then go on.

342
00:17:31,190 --> 00:17:32,607
And that's how
they dealt with it.

343
00:17:32,607 --> 00:17:36,260
And it was interesting
because they were basically

344
00:17:36,260 --> 00:17:40,760
adapting their behavior to
the robot, which is completely

345
00:17:40,760 --> 00:17:41,480
the wrong thing.

346
00:17:41,480 --> 00:17:43,940
If we want to put
robots in society,

347
00:17:43,940 --> 00:17:48,020
then the robots need to adapt
to humans rather than having

348
00:17:48,020 --> 00:17:49,640
humans adapt to robots.

349
00:17:49,640 --> 00:17:55,850
And I think until we get to that
point, where the robots can,

350
00:17:55,850 --> 00:17:59,840
both from a conversational
point of view,

351
00:17:59,840 --> 00:18:04,070
use cues like gaze and
gesture like we do,

352
00:18:04,070 --> 00:18:05,750
and also from a
spatial point of view,

353
00:18:05,750 --> 00:18:09,170
keep personal space,
knowing social rules,

354
00:18:09,170 --> 00:18:12,290
knowing things like, well,
if you want the robot to ride

355
00:18:12,290 --> 00:18:15,590
an elevator, it needs to
wait until people get off,

356
00:18:15,590 --> 00:18:18,750
just like a normal person would.

357
00:18:18,750 --> 00:18:23,990
These days robots
that ride elevators

358
00:18:23,990 --> 00:18:25,520
have no social graces.

359
00:18:25,520 --> 00:18:29,300
Basically, the elevator door
opens and they go right on.

360
00:18:29,300 --> 00:18:32,700
And it causes
consternation in people.

361
00:18:32,700 --> 00:18:37,190
So I think until we can solve
that problem of having robots

362
00:18:37,190 --> 00:18:39,890
behave more socially,
then there's really

363
00:18:39,890 --> 00:18:43,890
no hope for Homer Simpson to
be interacting with the robot

364
00:18:43,890 --> 00:18:45,950
and actually liking it.

365
00:18:45,950 --> 00:18:46,580
Sounds good.

366
00:18:46,580 --> 00:18:48,612
Unfortunately, we're
going to have to wrap up.

367
00:18:48,612 --> 00:18:49,570
But thanks for talking.

368
00:18:49,570 --> 00:18:49,750
OK.

369
00:18:49,750 --> 00:18:50,150
Great.

370
00:18:50,150 --> 00:18:50,650
Sure.

371
00:18:50,650 --> 00:18:52,270
My pleasure.

372
00:18:52,270 --> 00:18:55,000