1
00:00:00,000 --> 00:00:04,310


2
00:00:04,310 --> 00:00:06,060
Thanks for visiting us today.

3
00:00:06,060 --> 00:00:07,910
[INAUDIBLE] introduce yourself?

4
00:00:07,910 --> 00:00:12,710
I'm Manuela Veloso a
computer science faculty

5
00:00:12,710 --> 00:00:14,566
of Carnegie Mellon University.

6
00:00:14,566 --> 00:00:15,440
Cool.

7
00:00:15,440 --> 00:00:19,490
So I guess I want to
start off with asking you

8
00:00:19,490 --> 00:00:22,610
when you got into the field of
AI, and how are got into it?

9
00:00:22,610 --> 00:00:26,660
OK, so I got into
the field of the AI

10
00:00:26,660 --> 00:00:28,700
when I became more
of a grad student.

11
00:00:28,700 --> 00:00:34,670
So my undergraduate degree
is in electrical engineer.

12
00:00:34,670 --> 00:00:39,290
And then I got interested on
the information processing.

13
00:00:39,290 --> 00:00:42,380
And I did a master's
thesis in databases.

14
00:00:42,380 --> 00:00:46,640
And then my PhD thesis
starting in 1986

15
00:00:46,640 --> 00:00:50,930
was on this problem of analogy,
so planning by analogy.

16
00:00:50,930 --> 00:00:52,850
And I got very
interested on the problem

17
00:00:52,850 --> 00:00:56,360
of actually problem solving
in general, and in particular

18
00:00:56,360 --> 00:00:57,120
through analogy.

19
00:00:57,120 --> 00:01:01,460
So I've been working in AI
since 1986, I would say.

20
00:01:01,460 --> 00:01:02,120
Good deal.

21
00:01:02,120 --> 00:01:05,420
And you just like the
problem solving aspect?

22
00:01:05,420 --> 00:01:09,830
So I was always fascinated by
the problem solving aspect.

23
00:01:09,830 --> 00:01:14,810
So even now as of
today, I am a person

24
00:01:14,810 --> 00:01:17,420
that likes to research in
terms of the integration

25
00:01:17,420 --> 00:01:20,270
of perception,
cognition, and action, so

26
00:01:20,270 --> 00:01:23,720
the actual understanding
of the world,

27
00:01:23,720 --> 00:01:26,390
but all the way to
really solving problems,

28
00:01:26,390 --> 00:01:27,870
to actually doing.

29
00:01:27,870 --> 00:01:32,050
So not just labeling where you
are, but actually take action.

30
00:01:32,050 --> 00:01:37,070
So I think that I still
have that in my background.

31
00:01:37,070 --> 00:01:38,390
It's very interesting.

32
00:01:38,390 --> 00:01:40,200
So when you--

33
00:01:40,200 --> 00:01:43,110
I guess you did your
thesis at Carnegie Mellon.

34
00:01:43,110 --> 00:01:43,610
Yes.

35
00:01:43,610 --> 00:01:46,338
And you still work there.

36
00:01:46,338 --> 00:01:46,880
That's right.

37
00:01:46,880 --> 00:01:50,900
I guess when you first entered
in the field as a PhD student,

38
00:01:50,900 --> 00:01:52,790
what else was going
on at Carnegie Mellon?

39
00:01:52,790 --> 00:01:57,410
Well, it was a very, very
interesting and rich period

40
00:01:57,410 --> 00:01:59,150
of the field of AI.

41
00:01:59,150 --> 00:02:02,360
So at Carnegie Mellon
in the late '80s,

42
00:02:02,360 --> 00:02:10,889
there were like these giants of
AI, Herb Simon, Allen Newell,

43
00:02:10,889 --> 00:02:18,560
[INAUDIBLE],, Tom Mitchell, Jeff
Hinton, Mark [INAUDIBLE],, Scott

44
00:02:18,560 --> 00:02:19,515
[INAUDIBLE].

45
00:02:19,515 --> 00:02:22,490


46
00:02:22,490 --> 00:02:29,030
And there were many people that
were all very enthusiastically

47
00:02:29,030 --> 00:02:31,430
pushing the field of AI forward.

48
00:02:31,430 --> 00:02:34,070
I find myself to
be a real product

49
00:02:34,070 --> 00:02:40,700
of those days in which problem
solving was really underlying

50
00:02:40,700 --> 00:02:43,580
a lot of the research there.

51
00:02:43,580 --> 00:02:50,360
And I learned a lot from Herb
Simon and Alan on this idea

52
00:02:50,360 --> 00:02:55,700
that indeed systems could
learn from experience.

53
00:02:55,700 --> 00:02:58,070
But they could organize
their problem-solving

54
00:02:58,070 --> 00:03:01,790
through this concept of what's
the state of the world, what

55
00:03:01,790 --> 00:03:04,160
are all the possible
actions you can take?

56
00:03:04,160 --> 00:03:08,810
And you faced this
decision making

57
00:03:08,810 --> 00:03:11,190
challenge of making choices.

58
00:03:11,190 --> 00:03:14,060
So you are going always to
face this problem of, OK, I

59
00:03:14,060 --> 00:03:18,290
could do one out of n things,
and how do you choose?

60
00:03:18,290 --> 00:03:20,960
And whatever you
do choosing, you

61
00:03:20,960 --> 00:03:24,590
may find sooner or later in
your research, in your choosing,

62
00:03:24,590 --> 00:03:26,580
that it was not the right thing.

63
00:03:26,580 --> 00:03:28,360
And so the question
becomes in AI,

64
00:03:28,360 --> 00:03:31,940
if you are doing the
choice management

65
00:03:31,940 --> 00:03:35,870
and when you find it's
not the right thing, what

66
00:03:35,870 --> 00:03:38,260
do you do about that?

67
00:03:38,260 --> 00:03:40,338
And one of the things
that Allen Newell and Herb

68
00:03:40,338 --> 00:03:42,380
Simon were pushing in
those days was this concept

69
00:03:42,380 --> 00:03:46,100
of chunking, this concept
of, OK, get that impasse

70
00:03:46,100 --> 00:03:49,430
and save that, and we
learn that up there,

71
00:03:49,430 --> 00:03:50,970
when you had to do
that choice, that

72
00:03:50,970 --> 00:03:52,670
doesn't work in this situation.

73
00:03:52,670 --> 00:03:57,150
And I tried to build upon that
trying to use that by analogy.

74
00:03:57,150 --> 00:03:59,720
So you would save
some experience

75
00:03:59,720 --> 00:04:01,550
of the choice
you've made, and you

76
00:04:01,550 --> 00:04:04,760
tried to transfer
it to new problems

77
00:04:04,760 --> 00:04:07,070
and reproduce the same
successful choices.

78
00:04:07,070 --> 00:04:09,770
That's-- wow, OK.

79
00:04:09,770 --> 00:04:14,830
So I guess, this might
be too technical,

80
00:04:14,830 --> 00:04:18,829
but what were the most
challenging parts of not just

81
00:04:18,829 --> 00:04:22,640
your work, but the general work
in problem solving, that part

82
00:04:22,640 --> 00:04:23,630
of the field?

83
00:04:23,630 --> 00:04:28,820
So just to reiterate a
little bit, so in those days,

84
00:04:28,820 --> 00:04:35,960
there was this problem
solving fever going on at CMU.

85
00:04:35,960 --> 00:04:39,980
And there was a little
bit of the beginning

86
00:04:39,980 --> 00:04:44,660
of the neural net paradigms at
CMU, all the non-symbolic kind

87
00:04:44,660 --> 00:04:46,550
of reasoning, all
the statistical,

88
00:04:46,550 --> 00:04:52,020
the speech recognisers being
highly based on statistics.

89
00:04:52,020 --> 00:04:56,660
So there has been from
those days these, well

90
00:04:56,660 --> 00:05:02,010
how much do you put a symbolic
knowledge into a system,

91
00:05:02,010 --> 00:05:05,490
or how much you just look
at a large amount of data

92
00:05:05,490 --> 00:05:08,740
and gather statistical
information and just use that.

93
00:05:08,740 --> 00:05:11,580
So it's interesting
that I don't think

94
00:05:11,580 --> 00:05:21,970
we reached a solution to
these two ways of approaching.

95
00:05:21,970 --> 00:05:25,060
But those started in those
days, that discussion.

96
00:05:25,060 --> 00:05:29,560
And all of that is from
the mid-80s, late '80s.

97
00:05:29,560 --> 00:05:31,420
And we are still
trying to resolve

98
00:05:31,420 --> 00:05:33,010
how to put this all together.

99
00:05:33,010 --> 00:05:35,740
And, for example,
I've been doing

100
00:05:35,740 --> 00:05:39,770
a lot of robot work, a
lot of robot soccer work.

101
00:05:39,770 --> 00:05:42,220
And from a technical
point of view, indeed.

102
00:05:42,220 --> 00:05:47,140
Because these robots need
to solve a concrete problem,

103
00:05:47,140 --> 00:05:51,100
like these scoring goals into
a goal, working as a teammate.

104
00:05:51,100 --> 00:05:54,970
There has always to be
some symbolic knowledge,

105
00:05:54,970 --> 00:05:56,740
but over the task.

106
00:05:56,740 --> 00:05:59,980
On the other hand, they do
have to intersect moving balls

107
00:05:59,980 --> 00:06:05,000
and to process pixels of images,
and to actuate their motors.

108
00:06:05,000 --> 00:06:07,900
And, therefore, they
have a lot, a lot,

109
00:06:07,900 --> 00:06:12,430
of low level processing
needed to be done.

110
00:06:12,430 --> 00:06:14,020
So from a technical
point of view,

111
00:06:14,020 --> 00:06:17,740
the challenges become how--
in my problem solving view--

112
00:06:17,740 --> 00:06:21,040
how do you combine all
that low level processing

113
00:06:21,040 --> 00:06:22,990
that needs to be done
at a perceptual level,

114
00:06:22,990 --> 00:06:26,270
at the atuator level with,
indeed, a lot of the task

115
00:06:26,270 --> 00:06:31,180
knowledge that is highly
symbolic or highly abstract,

116
00:06:31,180 --> 00:06:33,610
even to model what
the others are doing

117
00:06:33,610 --> 00:06:35,140
to come up with some teamwork.

118
00:06:35,140 --> 00:06:38,980
So though it's not maybe--

119
00:06:38,980 --> 00:06:41,230
that's the technical
challenges of problem

120
00:06:41,230 --> 00:06:43,540
solving from my point of view.

121
00:06:43,540 --> 00:06:46,210
This problem solving is not
done by a single creature,

122
00:06:46,210 --> 00:06:47,860
it's a teamwork aspect.

123
00:06:47,860 --> 00:06:50,590
Problem solving really
involves going all the way down

124
00:06:50,590 --> 00:06:54,460
to an execute, therefore the low
level challenges of processing

125
00:06:54,460 --> 00:06:58,000
and actuating very low level
information with actually

126
00:06:58,000 --> 00:07:02,650
solving the problem, which
involves the high level aspect.

127
00:07:02,650 --> 00:07:03,390
That makes sense.

128
00:07:03,390 --> 00:07:03,890
OK.

129
00:07:03,890 --> 00:07:04,810
Nice.

130
00:07:04,810 --> 00:07:06,190
Do you have any--

131
00:07:06,190 --> 00:07:09,010
I guess some people
have talked to us

132
00:07:09,010 --> 00:07:11,710
of a lot of favorite
stories or anecdotes

133
00:07:11,710 --> 00:07:13,540
that took place when they were--

134
00:07:13,540 --> 00:07:18,140
--I can tell you one that
was very striking to me.

135
00:07:18,140 --> 00:07:25,210
So in 1998 we were doing
robot soccer in Paris.

136
00:07:25,210 --> 00:07:28,720
And with the first
AIBO robots, which

137
00:07:28,720 --> 00:07:32,170
were these four legged
robots built by Sony.

138
00:07:32,170 --> 00:07:38,020
And the robots had all on board,
like a camera, a computer,

139
00:07:38,020 --> 00:07:41,030
and they actually
actuate their legs.

140
00:07:41,030 --> 00:07:46,550
And so they were playing in
a field that had landmarks so

141
00:07:46,550 --> 00:07:47,940
that they knew where they were.

142
00:07:47,940 --> 00:07:52,180
And so we solved the
localization problem

143
00:07:52,180 --> 00:07:58,390
by actually using a Bayesian
model of the map and then

144
00:07:58,390 --> 00:08:00,730
their sensors to predict
where they were on the field,

145
00:08:00,730 --> 00:08:02,020
it doesn't matter.

146
00:08:02,020 --> 00:08:05,050
But I remember all the map
that had been done til then

147
00:08:05,050 --> 00:08:08,920
to the localization assuming
that the robot either moved--

148
00:08:08,920 --> 00:08:12,340
so only moved by itself
and sensed the wall.

149
00:08:12,340 --> 00:08:15,610
We were in Paris, and
in the middle of a game

150
00:08:15,610 --> 00:08:18,910
the robots started getting
all entangled on each other.

151
00:08:18,910 --> 00:08:21,820
And the referees
picked up a robot

152
00:08:21,820 --> 00:08:24,190
and put it somewhere else.

153
00:08:24,190 --> 00:08:31,810
And at that moment, the robot
was all lost because it was

154
00:08:31,810 --> 00:08:35,530
believing that it was somewhere
else, seeing something,

155
00:08:35,530 --> 00:08:41,409
its map would say, I cannot see
a yellow marker from down there

156
00:08:41,409 --> 00:08:42,090
in the field.

157
00:08:42,090 --> 00:08:44,410
So these yellow marker
needs to be noise.

158
00:08:44,410 --> 00:08:48,550
So his model was
completely destroyed

159
00:08:48,550 --> 00:08:50,530
because there was this pick up.

160
00:08:50,530 --> 00:08:53,020
And I remember
looking at the game

161
00:08:53,020 --> 00:08:56,530
with my student [INAUDIBLE],,
screaming and shouting,

162
00:08:56,530 --> 00:09:00,280
do not touch our robots,
because in some sense,

163
00:09:00,280 --> 00:09:04,180
we knew that that lift
up and put somewhere else

164
00:09:04,180 --> 00:09:07,330
was not modeled,
was absolutely not

165
00:09:07,330 --> 00:09:11,560
part of the assumptions
that were underlying

166
00:09:11,560 --> 00:09:14,410
the whole problem solving.

167
00:09:14,410 --> 00:09:19,210
So it was a panic,
because reality

168
00:09:19,210 --> 00:09:24,460
was so obviously defeating
the model that we were using.

169
00:09:24,460 --> 00:09:27,010
But that was-- I mean, that
was a big lesson for me.

170
00:09:27,010 --> 00:09:29,620
Because from then
on, I will always

171
00:09:29,620 --> 00:09:34,360
push this idea that no matter
what model you come up with,

172
00:09:34,360 --> 00:09:38,170
it might be defeated by reality.

173
00:09:38,170 --> 00:09:40,980
So an executor, a
problem solver has always

174
00:09:40,980 --> 00:09:46,540
to hypothesize not only that
information gathered is noisy

175
00:09:46,540 --> 00:09:48,040
and you might not
know how to solve,

176
00:09:48,040 --> 00:09:51,100
but that it's model
might be incorrect.

177
00:09:51,100 --> 00:09:53,620
And that led in several
pieces, and so forth.

178
00:09:53,620 --> 00:09:56,920
Because this is still
something I am very--

179
00:09:56,920 --> 00:10:01,090
I still relate
back to 1998 panic

180
00:10:01,090 --> 00:10:03,280
when I saw this robot
being picked up.

181
00:10:03,280 --> 00:10:07,390
So many things came out of
that, pick up the robot.

182
00:10:07,390 --> 00:10:09,940
But anyway.

183
00:10:09,940 --> 00:10:17,320
So switching gears a bit I guess
given the real world computers,

184
00:10:17,320 --> 00:10:20,950
they're related
by get noisy data,

185
00:10:20,950 --> 00:10:24,420
computer models are
almost always more simple

186
00:10:24,420 --> 00:10:27,420
than reality is,
how can computers

187
00:10:27,420 --> 00:10:28,940
help humans solve problems?

188
00:10:28,940 --> 00:10:30,030
That is a very good one.

189
00:10:30,030 --> 00:10:33,360
So I've been working
a lot on autonomous,

190
00:10:33,360 --> 00:10:35,640
like computers all by itself.

191
00:10:35,640 --> 00:10:37,680
But a lot of the
research now is, indeed,

192
00:10:37,680 --> 00:10:40,830
towards more of these
interaction, human robot

193
00:10:40,830 --> 00:10:43,620
interaction, human
computer interaction,

194
00:10:43,620 --> 00:10:50,250
and have the computer or the
robot be more of an assistant.

195
00:10:50,250 --> 00:10:55,320
I think it's a much better
way of looking at, eventually,

196
00:10:55,320 --> 00:10:58,740
the future, when we have
robots moving around.

197
00:10:58,740 --> 00:11:03,810
Because there is
a kind of a hope

198
00:11:03,810 --> 00:11:07,390
that capabilities
will be complementary.

199
00:11:07,390 --> 00:11:12,660
So that like in humans, I mean,
I might play squash better

200
00:11:12,660 --> 00:11:13,410
than someone else.

201
00:11:13,410 --> 00:11:16,770
But that person maybe plays
basketball much better.

202
00:11:16,770 --> 00:11:22,260
Or I might be a very
good a cook and you might

203
00:11:22,260 --> 00:11:24,630
be a very good poem writer.

204
00:11:24,630 --> 00:11:28,710
So there is not a
real human that's

205
00:11:28,710 --> 00:11:30,940
omnipotent, in some sense.

206
00:11:30,940 --> 00:11:34,140
So when we create the
robot, we create a robot

207
00:11:34,140 --> 00:11:37,800
or an AI creature that
has its limitations

208
00:11:37,800 --> 00:11:40,740
and can be complemented either
by humans or by other AI

209
00:11:40,740 --> 00:11:42,150
creatures.

210
00:11:42,150 --> 00:11:45,110
I really think that that's
a very nice way to go.

211
00:11:45,110 --> 00:11:47,130
So indeed computers
and robots can

212
00:11:47,130 --> 00:11:50,790
assist people, both on
their physical limitations

213
00:11:50,790 --> 00:11:53,160
but also on their
known limitations

214
00:11:53,160 --> 00:11:56,350
or their execution limitations.

215
00:11:56,350 --> 00:12:00,890
So it's a very, very
healthy way of looking at AI

216
00:12:00,890 --> 00:12:03,930
as more creatures in the world
that will have limitations

217
00:12:03,930 --> 00:12:05,020
like humans do.

218
00:12:05,020 --> 00:12:07,230
So they're not-- they
don't necessarily

219
00:12:07,230 --> 00:12:09,180
need to be the ones
that beat everyone

220
00:12:09,180 --> 00:12:12,240
and be the best chess
player, the best robot soccer

221
00:12:12,240 --> 00:12:14,160
player, or this or that.

222
00:12:14,160 --> 00:12:17,250
But they can be more of
like a co-- they just

223
00:12:17,250 --> 00:12:19,200
coexist with humans
in the same way

224
00:12:19,200 --> 00:12:21,480
that we all coexist
with each other.

225
00:12:21,480 --> 00:12:22,300
Look at my accent.

226
00:12:22,300 --> 00:12:24,592
I mean, it's a limitation
that other people don't have.

227
00:12:24,592 --> 00:12:27,330
But I speak many languages, too,
which other people might not.

228
00:12:27,330 --> 00:12:30,550
So it's like we all
complement each other.

229
00:12:30,550 --> 00:12:31,710
And that's very nice.

230
00:12:31,710 --> 00:12:34,920
So sort of give a
robot or computer

231
00:12:34,920 --> 00:12:37,360
a specific context
within which it

232
00:12:37,360 --> 00:12:38,830
can assist in problem solving.

233
00:12:38,830 --> 00:12:39,390
Exactly.

234
00:12:39,390 --> 00:12:40,348
Yeah, that makes sense.

235
00:12:40,348 --> 00:12:43,710
And it will-- again,
the system will grow.

236
00:12:43,710 --> 00:12:45,450
I mean, we have this
feeling that we all

237
00:12:45,450 --> 00:12:46,800
learn from each other.

238
00:12:46,800 --> 00:12:48,720
And these robots
or these computers

239
00:12:48,720 --> 00:12:50,940
need also to have
that ability to not

240
00:12:50,940 --> 00:12:53,640
be static in terms
of what they've done,

241
00:12:53,640 --> 00:12:57,090
but evolve with their
interaction with either humans

242
00:12:57,090 --> 00:12:59,160
and with the environment.

243
00:12:59,160 --> 00:13:02,580
So they should be able to
change their parameters,

244
00:13:02,580 --> 00:13:04,560
their weights, their
knowledge, whatever

245
00:13:04,560 --> 00:13:07,320
they have inside, their
databases, their access

246
00:13:07,320 --> 00:13:08,440
functions.

247
00:13:08,440 --> 00:13:10,680
So that the whole thing
has to be flexible.

248
00:13:10,680 --> 00:13:16,050
Not as we would have thought
that it was going to converge,

249
00:13:16,050 --> 00:13:18,685
but as a function of the
interaction with other people,

250
00:13:18,685 --> 00:13:19,560
with the environment.

251
00:13:19,560 --> 00:13:22,590
And they just have to be
built with the ability

252
00:13:22,590 --> 00:13:24,450
to have these
things being changed

253
00:13:24,450 --> 00:13:27,370
as a function of
the interaction.

254
00:13:27,370 --> 00:13:29,340
Very interesting.

255
00:13:29,340 --> 00:13:33,487
So I guess people
usually don't like it

256
00:13:33,487 --> 00:13:35,070
when you ask these
kinds of questions,

257
00:13:35,070 --> 00:13:39,180
but where do you see
us going, concretely?

258
00:13:39,180 --> 00:13:41,860
Well, I am not very good
at handwaving, that's true.

259
00:13:41,860 --> 00:13:44,940
But on the other hand,
I do believe that we are

260
00:13:44,940 --> 00:13:47,815
going in the right direction.

261
00:13:47,815 --> 00:13:49,440
And I tell you there
are two directions

262
00:13:49,440 --> 00:13:50,460
in which we are going.

263
00:13:50,460 --> 00:13:53,580
One, which is like addressing
more challenging problems.

264
00:13:53,580 --> 00:13:55,950
Look at the DARPA challenge
in the desert, the Urban

265
00:13:55,950 --> 00:14:00,490
Challenge, the RoboCub
challenge, a lot of challenges,

266
00:14:00,490 --> 00:14:00,990
problems.

267
00:14:00,990 --> 00:14:05,020
I mean, the understanding
language technology,

268
00:14:05,020 --> 00:14:05,830
it's a challenge.

269
00:14:05,830 --> 00:14:09,640
And then the other direction
is like the developing

270
00:14:09,640 --> 00:14:13,290
more and more sophisticated
tools to address this problem,

271
00:14:13,290 --> 00:14:16,650
and understanding how the
math of some control theory,

272
00:14:16,650 --> 00:14:18,960
or how the statistics
of some approach,

273
00:14:18,960 --> 00:14:21,750
or how the psychology
models of something.

274
00:14:21,750 --> 00:14:26,220
So how do these other
sciences contribute

275
00:14:26,220 --> 00:14:32,520
for these very high level goal
of creating these human level

276
00:14:32,520 --> 00:14:33,150
intelligence?

277
00:14:33,150 --> 00:14:38,285
So that we kind of
understand that we

278
00:14:38,285 --> 00:14:40,410
have to learn the lessons
of physics from the math,

279
00:14:40,410 --> 00:14:43,050
from here, from there, from
there, from the psychology,

280
00:14:43,050 --> 00:14:45,550
from who knows, from
all the medicine.

281
00:14:45,550 --> 00:14:48,480
So this is all
these people that--

282
00:14:48,480 --> 00:14:50,620
all these research on.

283
00:14:50,620 --> 00:14:52,290
So we are going to
pursue this research

284
00:14:52,290 --> 00:14:55,530
on building the tools,
the really core tools.

285
00:14:55,530 --> 00:14:58,800
But then there's also the
problem to challenges.

286
00:14:58,800 --> 00:15:02,560
And, yes, so that's the
directions I think will happen.

287
00:15:02,560 --> 00:15:08,370
And you are always going
to towards problems.

288
00:15:08,370 --> 00:15:09,480
Tools to build tools.

289
00:15:09,480 --> 00:15:11,680
To build to the-- yeah, exactly.

290
00:15:11,680 --> 00:15:12,850
Interesting.

291
00:15:12,850 --> 00:15:14,410
OK.

292
00:15:14,410 --> 00:15:17,110
I guess that'd
probably do it for now.

293
00:15:17,110 --> 00:15:18,148
OK, thanks.

294
00:15:18,148 --> 00:15:19,690
Thanks very much
for talking with us.

295
00:15:19,690 --> 00:15:20,350
You are welcome.

296
00:15:20,350 --> 00:15:22,100
OK, thanks for putting
this together very.

297
00:15:22,100 --> 00:15:24,130
Nice good.