1
00:00:00,000 --> 00:00:04,508
Audrow Nash: Is there a sense
we've been missing in robotics?

2
00:00:04,508 --> 00:00:09,239
What about smell? I talked with
Kordel France, who's the founder

3
00:00:09,239 --> 00:00:13,378
of Scentience AI. Where they're
working to make smell as

4
00:00:13,378 --> 00:00:17,590
accessible to robots as vision
is today. But why smell? I

5
00:00:17,590 --> 00:00:21,803
hadn't thought of it much before
this interview. A lot of

6
00:00:21,803 --> 00:00:25,868
information is communicated
through smell. Is the fruit

7
00:00:25,868 --> 00:00:30,007
ripe? Does the bag have an
explosive in it? Does someone

8
00:00:30,007 --> 00:00:34,442
have lung cancer or other health
problems? It's amazing what

9
00:00:34,442 --> 00:00:38,876
kinds of things you can infer
about the world. If you have a

10
00:00:38,876 --> 00:00:43,533
good nose, and I think we'll see
a wave of startups in the next

11
00:00:43,533 --> 00:00:47,524
few years building robot
companies around the sense of

12
00:00:47,524 --> 00:00:51,441
smell from agriculture to
security to health care and

13
00:00:51,441 --> 00:00:55,580
more, and enabling that is
exactly what Kordel France is

14
00:00:55,580 --> 00:01:00,088
working on with Scentience AI.
You'll like this interview. If

15
00:01:00,088 --> 00:01:04,818
you are curious about a new type
of sensor and the big impact it

16
00:01:04,818 --> 00:01:08,957
could have on robotics, and
especially if you want to be

17
00:01:08,957 --> 00:01:13,244
part of the first wave of
robotics startups that comes out

18
00:01:13,244 --> 00:01:17,900
around applications leveraging a
good sense of smell. So grab a

19
00:01:17,900 --> 00:01:21,965
snack. Get comfortable. And
here's my conversation with

20
00:01:21,965 --> 00:01:23,000
Kordel France.

21
00:01:23,000 --> 00:01:24,000
Audrow Nash: Hi. Would you
introduce yourself?

22
00:01:24,000 --> 00:01:29,069
Kordel France: Hello. I'm Kordel
France. I am a roboticist, area

23
00:01:29,069 --> 00:01:32,000
engineer, and, founder of
Scentience.

24
00:01:32,000 --> 00:01:35,980
Audrow Nash: And Scentience is a
new thing for you. Will you tell

25
00:01:35,980 --> 00:01:39,837
me about where you were just a
little bit ago, and then how you

26
00:01:39,837 --> 00:01:41,000
came to Scentience?

27
00:01:41,000 --> 00:01:44,687
Kordel France: Sure. So I've
started, a few years ago, a

28
00:01:44,687 --> 00:01:48,703
machine learning company that
was focused on olfaction or the

29
00:01:48,703 --> 00:01:52,587
sense of smell and if you can
think, you can think of this,

30
00:01:52,587 --> 00:01:56,537
that what we are building is
like a camera is like a digital

31
00:01:56,537 --> 00:02:00,750
vision sensor, and microphone is
a digital audio sensor. We were

32
00:02:00,750 --> 00:02:04,964
trying to build a digital scent
or smell sensor and particularly

33
00:02:04,964 --> 00:02:08,453
engage towards medical
applications. And that company

34
00:02:08,453 --> 00:02:12,666
was called Data Diagnostics. And
they built we built a I was one

35
00:02:12,666 --> 00:02:16,419
of the founders, and we built a
product that could detect

36
00:02:16,419 --> 00:02:20,698
different chemicals on the human
body, so indicative of different

37
00:02:20,698 --> 00:02:24,056
medical conditions. So
coronavirus was one of them.

38
00:02:24,056 --> 00:02:27,940
Pneumonia was another one lung
cancer was another one. And,

39
00:02:27,940 --> 00:02:31,627
yeah, since then I've moved on
to kind of outside of the

40
00:02:31,627 --> 00:02:35,774
medical field into, the company
Scentience and working to build

41
00:02:35,774 --> 00:02:39,000
what, really build the sense of
smell for robots.

42
00:02:39,000 --> 00:02:43,975
Audrow Nash: I like the name, so
it's like “Sentience”, but it's

43
00:02:43,975 --> 00:02:48,795
“Scentience” with the “scent”
part. Oh, yeah. That's great. So

44
00:02:48,795 --> 00:02:52,993
tell me about smell more
generally. Why is this, like,

45
00:02:52,993 --> 00:02:57,890
how does it work? And then why
is this kind of an unappreciated

46
00:02:57,890 --> 00:03:01,000
sense, especially in robotics?
It seems.

47
00:03:01,000 --> 00:03:05,096
Kordel France: So the sense of
smell is really nothing, new to

48
00:03:05,096 --> 00:03:08,993
kind of the world as far as
sensors go. I mean, we all have

49
00:03:08,993 --> 00:03:13,089
carbon monoxide detectors in our
home. Smoke alarms. Right? So

50
00:03:13,089 --> 00:03:16,722
you can think of those as
electronic noses that look at

51
00:03:16,722 --> 00:03:20,356
one chemical in particular.
However, they're really not

52
00:03:20,356 --> 00:03:24,319
super sensitive. You really need
a lot of smoke in order for

53
00:03:24,319 --> 00:03:28,613
those to go off. And if you look
at a lot of the other influences

54
00:03:28,613 --> 00:03:32,643
around us in nature, like moths,
dogs, insects, even cellular

55
00:03:32,643 --> 00:03:36,408
organisms, they all have,
olfactory senses that help them

56
00:03:36,408 --> 00:03:39,910
navigate. And, you know,
especially dogs, they, their

57
00:03:39,910 --> 00:03:43,939
primal sense is or their primary
sense is olfaction. And so I

58
00:03:43,939 --> 00:03:47,506
don't we don't really have
mechanisms to replicate the

59
00:03:47,506 --> 00:03:51,734
accuracy and the capability that
dogs have to navigate. And, to,

60
00:03:51,734 --> 00:03:55,896
to interpret the world and the
air around us and olfaction is a

61
00:03:55,896 --> 00:03:59,463
really interesting sense
because, the chemicals around

62
00:03:59,463 --> 00:04:03,493
you in the air, you can observe
with three and three ways you

63
00:04:03,493 --> 00:04:07,655
can see them optically, you can
hear the acoustics, and you can

64
00:04:07,655 --> 00:04:11,024
react them with, chemical
reactions. Now, there are

65
00:04:11,024 --> 00:04:14,789
different theories for how
humans, smell, but, most argue

66
00:04:14,789 --> 00:04:18,885
that we react well, there's a
chemical reaction that occurs in

67
00:04:18,885 --> 00:04:22,783
the nose that, introduces a
signal to the brain, that helps

68
00:04:22,783 --> 00:04:25,954
you understand as you're
smelling baked bread or

69
00:04:25,954 --> 00:04:29,389
strawberries or something like
that. Now, we're with

70
00:04:29,389 --> 00:04:33,286
Scentience. We're trying to
build sensors that really allow

71
00:04:33,286 --> 00:04:37,448
you to observe, to quantify the
data around you in a very, very

72
00:04:37,448 --> 00:04:41,544
sensitive manner. So looking at
a parts per trillion, which or

73
00:04:41,544 --> 00:04:45,772
part per quadrillion resolution,
and to give you a scale, humans

74
00:04:45,772 --> 00:04:49,273
can detect it a part per
thousand or part per million

75
00:04:49,273 --> 00:04:53,435
depending on a compound. So very
we're going much, much deeper.

76
00:04:53,435 --> 00:04:57,266
And the, trying to make it so
that's very rapid. So we can

77
00:04:57,266 --> 00:05:01,428
instantly observe what the air
around us is telling us. There's

78
00:05:01,428 --> 00:05:05,656
so much data in air right in the
air around us. And so, a lot of

79
00:05:05,656 --> 00:05:09,752
the sensors right now are not
super fast. They take minutes or

80
00:05:09,752 --> 00:05:13,847
seconds or sometimes even hours
to really get a, response back

81
00:05:13,847 --> 00:05:17,877
from a stimulus. And so we're
trying to fix that problem with

82
00:05:17,877 --> 00:05:19,000
the, our sensors.

83
00:05:19,000 --> 00:05:22,871
Audrow Nash: That's awesome.
Okay. I feel like there's a lot

84
00:05:22,871 --> 00:05:26,806
to unpack there. But so why why
we interested in this. Why is

85
00:05:26,806 --> 00:05:30,355
scent like what kinds of
problems can you solve. If you

86
00:05:30,355 --> 00:05:34,355
had good sense in theory I would
think all the things dogs are

87
00:05:34,355 --> 00:05:38,290
trained for like bomb detection
or drugs or whatever it might

88
00:05:38,290 --> 00:05:42,097
be. But what what are some of
the problems that you imagine

89
00:05:42,097 --> 00:05:46,097
being able to go after? If you
have really good, I don't know,

90
00:05:46,097 --> 00:05:47,000
sense sensors?

91
00:05:47,000 --> 00:05:50,626
Kordel France: Sure. So you you,
made a perfect foundation for

92
00:05:50,626 --> 00:05:54,019
it. So a lot of the things that
dogs are used for, they're

93
00:05:54,019 --> 00:05:57,411
detecting explosives or drugs,
and for border security. So

94
00:05:57,411 --> 00:06:01,038
that's pretty low hanging fruit
for us. That's one thing we're

95
00:06:01,038 --> 00:06:04,781
going after. If we can enable a
drone or a robot to. Yeah, to go

96
00:06:04,781 --> 00:06:07,881
to basically if you can tell
them go find go find the

97
00:06:07,881 --> 00:06:11,215
explosive substance. They could
clear a soccer stadium or

98
00:06:11,215 --> 00:06:14,900
football stadium or a school or
whatever and not endanger a dog

99
00:06:14,900 --> 00:06:18,351
or human, potentially, and
especially if you can swarm them

100
00:06:18,351 --> 00:06:21,977
together. And then, some other
exciting applications are like,

101
00:06:21,977 --> 00:06:25,662
in aerospace. Right? So a lot of
the ways to determine signs of

102
00:06:25,662 --> 00:06:29,347
life, they send soil samples or
rovers or some, samplers to try

103
00:06:29,347 --> 00:06:32,798
to collect sand and bring it
back or soil and bring it back

104
00:06:32,798 --> 00:06:35,840
and observe the chemical
composition, or they try to

105
00:06:35,840 --> 00:06:39,232
analyze it, you know. Right on
the side of itself. And the

106
00:06:39,232 --> 00:06:42,800
current instruments are very
heavy. They take a lot of power.

107
00:06:42,800 --> 00:06:46,426
And, they're very, they're very
sensitive to kind of, space is

108
00:06:46,426 --> 00:06:49,760
not their the best environment
they've been in design for

109
00:06:49,760 --> 00:06:52,977
because they've been designed
for the lab. So, with our

110
00:06:52,977 --> 00:06:56,545
sensors, you can shrink things
down into a much smaller size,

111
00:06:56,545 --> 00:07:00,172
weight, and power package that
can allow you to get data back,

112
00:07:00,172 --> 00:07:03,447
very quickly. Every kilogram
that goes up on a rocket is

113
00:07:03,447 --> 00:07:07,074
literally millions of dollars.
And so, you know, if you say if

114
00:07:07,074 --> 00:07:10,583
you send something up, that's
the payload of a fridge and we

115
00:07:10,583 --> 00:07:13,800
can compress it down to
something that's the payload of

116
00:07:13,800 --> 00:07:17,485
a, you know, an energy drink can
or a pop can or something like

117
00:07:17,485 --> 00:07:20,000
that. And that's that's much
better, right?

118
00:07:20,000 --> 00:07:21,000
Audrow Nash: Yeah.

119
00:07:21,000 --> 00:07:24,452
Kordel France: And then
additionally, another exciting

120
00:07:24,452 --> 00:07:28,224
domain is agriculture because,
crops, food, drinks, quality

121
00:07:28,224 --> 00:07:32,188
control and plants, etc. they
all emit chemicals that indicate

122
00:07:32,188 --> 00:07:35,960
the relative state, and their,
kind of their general health

123
00:07:35,960 --> 00:07:39,476
status. So like, for example,
flowers and we'll admit a

124
00:07:39,476 --> 00:07:42,865
certain chemical indicating
they're about to, they're

125
00:07:42,865 --> 00:07:46,253
approaching their end of life,
right? They're lacking

126
00:07:46,253 --> 00:07:50,217
chlorophyl or they're lacking
certain chemicals that that make

127
00:07:50,217 --> 00:07:53,925
them healthy. And so if we can
detect that, you can detect

128
00:07:53,925 --> 00:07:57,314
through food spoilage more
accurately, better quality

129
00:07:57,314 --> 00:08:01,214
control on, on crops in general,
and over groceries. And then

130
00:08:01,214 --> 00:08:04,602
similarly with the cosmetic
industry as well, quality

131
00:08:04,602 --> 00:08:08,246
control, different perfumes,
cosmetic products, etc.. So,

132
00:08:08,246 --> 00:08:12,210
yeah, there's a lot of different
applications there. But I can

133
00:08:12,210 --> 00:08:14,000
zoom in on any one of those.

134
00:08:14,000 --> 00:08:18,039
Audrow Nash: Cool. I really like
the idea of quality control,

135
00:08:18,039 --> 00:08:22,144
because I guess, like a lot of
quality control when I think of

136
00:08:22,144 --> 00:08:25,852
it as visual. And with this you
get, you get a very rich

137
00:08:25,852 --> 00:08:29,626
additional sensor that you can
use that, I mean, for like

138
00:08:29,626 --> 00:08:33,731
produce or something would be
unbelievable. And I imagine also

139
00:08:33,731 --> 00:08:37,638
for like, skin products and
stuff where they made correctly

140
00:08:37,638 --> 00:08:41,478
this kind of thing. That's so
cool. Okay. So yeah, I mean,

141
00:08:41,478 --> 00:08:45,517
those areas to zoom in on or
those areas are very cool, but I

142
00:08:45,517 --> 00:08:49,688
want to keep going deeper into
to understand the whole thing of

143
00:08:49,688 --> 00:08:53,794
sense or so when you keep saying
chemicals, when we're talking

144
00:08:53,794 --> 00:08:58,097
about scents, how does detection
work? Like what are what are you

145
00:08:58,097 --> 00:09:01,938
doing? And I understand, okay,
we have the carbon monoxide

146
00:09:01,938 --> 00:09:05,844
sensor in the house and it is
looking for one compound. But

147
00:09:05,844 --> 00:09:09,685
how do you make it general. So
that you can smell multiple

148
00:09:09,685 --> 00:09:13,723
different things. And then how
do you, how do these detection

149
00:09:13,723 --> 00:09:17,829
mechanisms work to identify like
one part and trillion, as you

150
00:09:17,829 --> 00:09:21,669
were saying, in this kind of
thing, like how does all that

151
00:09:21,669 --> 00:09:22,000
work?

152
00:09:22,000 --> 00:09:25,383
Kordel France: So certain
chemicals favored different

153
00:09:25,383 --> 00:09:29,213
methods of detection. And I
mentioned three of them. There's

154
00:09:29,213 --> 00:09:32,723
optical detection, acoustic
detection and then chemical

155
00:09:32,723 --> 00:09:36,681
reactive detection. So to talk
about optical for a moment, the

156
00:09:36,681 --> 00:09:40,319
certain chemicals in the air,
they will, for more or less

157
00:09:40,319 --> 00:09:44,021
terms refract with varying
wavelengths of light. And so if

158
00:09:44,021 --> 00:09:47,723
you can find that specific
wavelength and you get a signal

159
00:09:47,723 --> 00:09:51,872
back, from that refraction, then
it's an, it's indicative of what

160
00:09:51,872 --> 00:09:55,830
chemical that might be. Right.
So, alcohols might refract in a

161
00:09:55,830 --> 00:09:59,149
very certain way. Organic
compounds in general might

162
00:09:59,149 --> 00:10:03,043
refract this very specific way
or a very specific wavelength.

163
00:10:03,043 --> 00:10:06,936
And so this, spectroscopy or
these, these infrared techniques

164
00:10:06,936 --> 00:10:10,511
are one way to detect certain
chemicals. Now, some favor

165
00:10:10,511 --> 00:10:13,894
optical detection, others
favorite chemical, favorite

166
00:10:13,894 --> 00:10:17,915
chemical Co detection. So as an
example for chemical detection,

167
00:10:17,915 --> 00:10:22,000
if you think of a battery right,
there's an electrode or there's

168
00:10:22,000 --> 00:10:25,638
a anode and a cathode and
there's an electrolyte and the,

169
00:10:25,638 --> 00:10:28,894
electrolyte is charged. Now
it's, it's charged. The

170
00:10:28,894 --> 00:10:32,787
electrolyte is designed in a
certain way such that if it hits

171
00:10:32,787 --> 00:10:36,553
a certain charge, then it will
react with specific group of

172
00:10:36,553 --> 00:10:40,638
compounds, for example, alcohols
or, aldehydes or something like

173
00:10:40,638 --> 00:10:44,596
that. So it basically the as the
chemicals, interface with the

174
00:10:44,596 --> 00:10:48,234
electrolyte, they send a charge
back depending on how the

175
00:10:48,234 --> 00:10:51,936
electrolyte, the electrons
interface with the electrolyte.

176
00:10:51,936 --> 00:10:55,957
And so it gives an indication of
what chemicals are around you.

177
00:10:55,957 --> 00:10:59,532
And then similarly with,
acoustic sensors, the vibration

178
00:10:59,532 --> 00:11:03,298
of chemicals at a quantum level,
you can you can detect the

179
00:11:03,298 --> 00:11:07,319
frequencies of those vibrations
and those different frequencies

180
00:11:07,319 --> 00:11:10,000
determine what chemical you're
looking at.

181
00:11:10,000 --> 00:11:11,000
Audrow Nash: That's cool. Yeah.

182
00:11:11,000 --> 00:11:14,957
Kordel France: So it's a matter
of like the, you know, the the

183
00:11:14,957 --> 00:11:19,043
mediums are all there. The ways
and detection are all there. The

184
00:11:19,043 --> 00:11:22,745
problem is really how do we
compress these lab instruments

185
00:11:22,745 --> 00:11:26,255
that have been used to
illustrate these techniques down

186
00:11:26,255 --> 00:11:30,277
to such a small form factor that
they can be used on, you know,

187
00:11:30,277 --> 00:11:32,000
handheld devices or robots?

188
00:11:32,000 --> 00:11:36,735
Audrow Nash: Yeah. So how do you
go about that. Because that does

189
00:11:36,735 --> 00:11:41,325
sound like the big problem. So I
guess you're saying that these

190
00:11:41,325 --> 00:11:45,914
are, it's well-established in a
sense to use site acoustics and

191
00:11:45,914 --> 00:11:49,921
chemical reactions to find
things that are to basically

192
00:11:49,921 --> 00:11:54,146
infer about what is in the air
and from that you can infer

193
00:11:54,146 --> 00:11:58,517
something like scent and but how
and but so these are fridge

194
00:11:58,517 --> 00:12:02,742
sized things or large devices.
How are you going about and

195
00:12:02,742 --> 00:12:06,894
combining and minimizing these
like I imagine that that's

196
00:12:06,894 --> 00:12:11,338
probably the bulk or that that's
a huge challenge that you've

197
00:12:11,338 --> 00:12:15,417
been working on with the
previous company and no current

198
00:12:15,417 --> 00:12:16,000
company.

199
00:12:16,000 --> 00:12:19,700
Kordel France: Yeah, a lot of my
PhD research went into it as

200
00:12:19,700 --> 00:12:23,462
well. To try to understand, you
have to understand the current

201
00:12:23,462 --> 00:12:27,101
state of the art and how these
systems work. And, there have

202
00:12:27,101 --> 00:12:30,438
been a lot of great engineers
that have worked on these

203
00:12:30,438 --> 00:12:34,320
things, and so they've optimized
them really well. And so it's a

204
00:12:34,320 --> 00:12:38,203
matter of fact, it's a matter of
really can we optimize further?

205
00:12:38,203 --> 00:12:41,175
And if we can't, what
constraints can we relax in

206
00:12:41,175 --> 00:12:44,815
order to get a smaller size
weight and power factor. So like

207
00:12:44,815 --> 00:12:48,091
for example, there's an
instrument called the GCA CMS,

208
00:12:48,091 --> 00:12:51,306
which stands for Gas
chromatography Mass spectrometry

209
00:12:51,306 --> 00:12:55,189
machine. And it's about the size
of a fridge. And it takes hours

210
00:12:55,189 --> 00:12:58,889
to run an air sample. Now it's
very accurate and it gives you

211
00:12:58,889 --> 00:13:02,590
very detailed results. But if
you can imagine something like,

212
00:13:02,590 --> 00:13:05,926
a dog which has like a 100Hz
refresh capability and its

213
00:13:05,926 --> 00:13:09,445
olfactory senses, that's not
very competitive to something

214
00:13:09,445 --> 00:13:13,084
that takes hours to run. Right?
So maybe if we can relax the

215
00:13:13,084 --> 00:13:16,542
accuracy and the number of
compounds we're looking at, we

216
00:13:16,542 --> 00:13:20,000
can get a smaller, you know,
size weighting power factor.

217
00:13:20,000 --> 00:13:24,073
Audrow Nash: Okay. And so you
did that for those three domains

218
00:13:24,073 --> 00:13:27,949
of sensing. How do you, how do
you make it so that there is

219
00:13:27,949 --> 00:13:32,088
because I imagine and maybe it
depends on the use case that you

220
00:13:32,088 --> 00:13:36,226
want. Okay. I imagine it's quite
specific on the use case. Tell

221
00:13:36,226 --> 00:13:38,000
me about that a little bit.

222
00:13:38,000 --> 00:13:41,313
Kordel France: Yeah. So if you
want just like passive

223
00:13:41,313 --> 00:13:45,313
monitoring like for example with
like quality control of food or

224
00:13:45,313 --> 00:13:48,938
drink or produce or or for or
you don't need 100Hz refresh

225
00:13:48,938 --> 00:13:52,438
capability because you really
it's just it's the problem

226
00:13:52,438 --> 00:13:56,125
doesn't define that type of a
solution. So you can get away

227
00:13:56,125 --> 00:13:59,750
with more, a longer, refresh
rate, a longer detection time

228
00:13:59,750 --> 00:14:03,063
and have more accurate
responses. Now, something like

229
00:14:03,063 --> 00:14:07,125
olfactory navigation. Like what?
How a moth or how a dog actually

230
00:14:07,125 --> 00:14:10,875
navigates the world. That's
where a really fast refresh rate

231
00:14:10,875 --> 00:14:14,563
comes in handy, because where
you really need to be able to

232
00:14:14,563 --> 00:14:18,063
you need to be able to update
your world state much more

233
00:14:18,063 --> 00:14:21,375
quickly and respond to
actuators. Right. If I need to

234
00:14:21,375 --> 00:14:25,000
drive left because the scent
plume is moving left, I don't

235
00:14:25,000 --> 00:14:29,000
want to wait an hour to do that
because the scent plume changes.

236
00:14:29,000 --> 00:14:30,000
Audrow Nash: Every one.

237
00:14:29,000 --> 00:14:32,868
Kordel France: Second. So you've
got to be able to react in a

238
00:14:32,868 --> 00:14:36,863
very, very quick way. So, those
types of things that are taking

239
00:14:36,863 --> 00:14:40,541
into consideration, for the
products that we build. So, we

240
00:14:40,541 --> 00:14:43,966
have two, one that is two
products, one that is a fast

241
00:14:43,966 --> 00:14:47,644
responding, very low sized
weight and power, product that,

242
00:14:47,644 --> 00:14:51,259
really is tuned for olfactory
navigation and one that's a

243
00:14:51,259 --> 00:14:54,937
little larger, that, about the
size of a coffee maker that

244
00:14:54,937 --> 00:14:58,678
allows you to profile the air
around you, or an air sample,

245
00:14:58,678 --> 00:15:02,483
and yet, at the expense of a
longer detection time, get much

246
00:15:02,483 --> 00:15:06,034
more higher accuracy and a much
larger array of chemical

247
00:15:06,034 --> 00:15:08,000
compounds that can be analyzed.

248
00:15:08,000 --> 00:15:10,983
Audrow Nash: Okay. Very cool. I
would love to hear about the use

249
00:15:10,983 --> 00:15:13,966
cases you're targeting for both
of these. The small one and then

250
00:15:13,966 --> 00:15:16,809
the large one. Did you have the
small one nearby? You held up

251
00:15:16,809 --> 00:15:19,000
something earlier? I don't know
if that was it.

252
00:15:19,000 --> 00:15:22,614
Kordel France: I do, yes. So
we've got, one of them here.

253
00:15:22,614 --> 00:15:26,672
This is a smaller one. And, it's
really built to, it's, a little

254
00:15:26,672 --> 00:15:30,604
shrink in size as the company
grows, but, we've built it to be

255
00:15:30,604 --> 00:15:33,837
able to be used, by robots,
right, so that they can

256
00:15:33,837 --> 00:15:37,769
integrate this with a very low
sized weight power factor. Have

257
00:15:37,769 --> 00:15:41,700
an API that they can they can,
talk to, very easily and really

258
00:15:41,700 --> 00:15:45,568
navigate the world. I mean, if
you're if you're familiar with

259
00:15:45,568 --> 00:15:49,626
what's going on with, like Sega
or Tesla robots or App Tronic or

260
00:15:49,626 --> 00:15:53,620
Fourier or any of these really
big robotics companies that have

261
00:15:53,620 --> 00:15:57,425
gotten billions of dollars in
funding. They all have vision,

262
00:15:57,425 --> 00:16:01,419
they'll have audio, they'll have
speech capabilities. They have

263
00:16:01,419 --> 00:16:05,541
haptic capabilities and tactical
capabilities since they can walk

264
00:16:05,541 --> 00:16:09,345
around, but they're missing the
sense of smell. And when you

265
00:16:09,345 --> 00:16:13,340
mention that to people, they're
like, okay, well, why do I need

266
00:16:13,340 --> 00:16:17,081
that? Right. Well, if you think
about the applications, you

267
00:16:17,081 --> 00:16:21,076
really expand the capabilities
that that robot can do by giving

268
00:16:21,076 --> 00:16:24,817
it the sense of olfaction or
smell. So being able to detect

269
00:16:24,817 --> 00:16:28,558
when there are high traces of
carbon dioxide in the home or

270
00:16:28,558 --> 00:16:32,235
being able to quality control
food and manipulate, produce

271
00:16:32,235 --> 00:16:36,040
around to, you know, fit in your
grocery store, right. Like,

272
00:16:36,040 --> 00:16:39,781
those are things that a robot
currently won't be able to do

273
00:16:39,781 --> 00:16:42,000
unless it's given a sense of
smell.

274
00:16:42,000 --> 00:16:45,963
Audrow Nash: Yeah. It's so
funny. Imagine imagining a robot

275
00:16:45,963 --> 00:16:49,591
personal shopper picking out
fruit or something in the

276
00:16:49,591 --> 00:16:53,689
grocery store. With that one,
what do you imagine are some of

277
00:16:53,689 --> 00:16:57,853
the like the low hanging fruit
for applications? Because like,

278
00:16:57,853 --> 00:17:01,951
my impression is humanoids are
pretty far away, from being in

279
00:17:01,951 --> 00:17:06,183
homes and being, like, generally
like they might be very useful

280
00:17:06,183 --> 00:17:09,945
in manufacturing and very small,
or maybe not small, but

281
00:17:09,945 --> 00:17:13,774
particular niches, I imagine.
First, but there's a lot of

282
00:17:13,774 --> 00:17:17,939
robots that are not humanoids.
What would you imagine would be

283
00:17:17,939 --> 00:17:21,768
some of the like, startups?
We'll see. Using this kind of

284
00:17:21,768 --> 00:17:25,664
sensor in the next, like say
three years or something like

285
00:17:25,664 --> 00:17:26,000
this.

286
00:17:26,000 --> 00:17:29,897
Kordel France: So Aero Aerial
Robotics are a interesting one

287
00:17:29,897 --> 00:17:33,535
to start with. If we can imagine
a drone that has vision

288
00:17:33,535 --> 00:17:37,432
capabilities and navigates by
vision and GPS, if you go into

289
00:17:37,432 --> 00:17:41,264
like a tunnel or if you go
somewhere where you can't really

290
00:17:41,264 --> 00:17:45,356
get a good GPS signal and then
it's dark, you're you're options

291
00:17:45,356 --> 00:17:49,059
become limited on how you
navigate. Right. So and you can

292
00:17:49,059 --> 00:17:53,086
navigate by there are ultrasonic
sensors. There's radar. There

293
00:17:53,086 --> 00:17:56,463
are other ways to do it. But
those radars very power

294
00:17:56,463 --> 00:18:00,426
consuming and and if you could
put an olfactory sensor on the

295
00:18:00,426 --> 00:18:04,193
robot and you can tell it to
basically get as close to the

296
00:18:04,193 --> 00:18:08,025
ethanol as you can or get as
close to whatever compound you

297
00:18:08,025 --> 00:18:11,987
can, it literally acts like a
dog and tracking the scent and,

298
00:18:11,987 --> 00:18:16,209
you know, trying to find its way
through, so that's I mean, there

299
00:18:16,209 --> 00:18:19,652
are a lot of olfactory
navigation capabilities that I

300
00:18:19,652 --> 00:18:23,679
feel like, you know, robots can
do with between aero robotics,

301
00:18:23,679 --> 00:18:27,576
ground robotics, like, with
spot. Right. And then underwater

302
00:18:27,576 --> 00:18:30,824
as well, because, surface
dwelling, or ocean floor

303
00:18:30,824 --> 00:18:34,591
dwelling, creatures partially
navigate by scent. Right? So

304
00:18:34,591 --> 00:18:38,683
some crabs navigate by chemical
traces that they, they actually

305
00:18:38,683 --> 00:18:42,776
can observe in currents in the
water. And, GPS is not an option

306
00:18:42,776 --> 00:18:46,543
when you're in the middle when
you're like for submarines,

307
00:18:46,543 --> 00:18:50,505
right. They currently use sonar,
which is, is very useful and

308
00:18:50,505 --> 00:18:54,467
works, but it'd be helpful to
have, another, another modality

309
00:18:54,467 --> 00:18:58,429
for redundancy for navigation.
Right. If you can maybe, like,

310
00:18:58,429 --> 00:19:02,002
if you can imagine a scenario
where you have to track a

311
00:19:02,002 --> 00:19:05,834
current or you have to track, a
certain chemical, that, you

312
00:19:05,834 --> 00:19:09,796
know, is emitted from a, like,
emitted from a current, from a

313
00:19:09,796 --> 00:19:13,694
specific landmark. Then you can
just track that chemical the

314
00:19:13,694 --> 00:19:17,591
whole way there and then know
where you are. Right. So there

315
00:19:17,591 --> 00:19:20,904
are a lot of, interesting
applications in olfactory

316
00:19:20,904 --> 00:19:24,606
navigation, which is, to be
honest, the north star of the

317
00:19:24,606 --> 00:19:28,698
company. We want to make these
sensors as small and lightweight

318
00:19:28,698 --> 00:19:32,725
as possible. But, we in order to
do that, there's, there are a

319
00:19:32,725 --> 00:19:36,687
lot of interesting applications
that we can address along the

320
00:19:36,687 --> 00:19:40,000
way with, that aren't
necessarily navigation based.

321
00:19:40,000 --> 00:19:45,748
Audrow Nash: Yeah. For sure. And
just for completeness, with the,

322
00:19:45,748 --> 00:19:51,143
larger coffee maker sized one,
what are some applications you

323
00:19:51,143 --> 00:19:53,000
imagine for that one?

324
00:19:53,000 --> 00:19:57,247
Kordel France: Agriculture is a
really interesting, application

325
00:19:57,247 --> 00:20:01,292
there because there is there are
a lot of compounds that are

326
00:20:01,292 --> 00:20:05,404
emitted, during or whether it's,
you know, growing crops in a

327
00:20:05,404 --> 00:20:09,247
field or in, a flower shop or
something like that. And so

328
00:20:09,247 --> 00:20:13,292
being able to monitor chemicals
there is pretty interesting.

329
00:20:13,292 --> 00:20:17,135
Because they give a very good
state on the health of the,

330
00:20:17,135 --> 00:20:21,382
actual, produce being grown or
the or the crop being grown. And

331
00:20:21,382 --> 00:20:25,427
can give you an idea of what
diseases might be affecting it.

332
00:20:25,427 --> 00:20:28,933
What it's like, what its
longevity will be, what its

333
00:20:28,933 --> 00:20:33,045
yield will be. Wow. Which is
yeah, there's a there's a lot of

334
00:20:33,045 --> 00:20:36,955
interesting, insights that can
be gathered there. And then

335
00:20:36,955 --> 00:20:40,798
similarly, the, within the
cosmetic industry, they, and I

336
00:20:40,798 --> 00:20:45,045
I'm not kidding you. They have
trained human professionals that

337
00:20:45,045 --> 00:20:48,955
quality control, like, for
example, Chanel five or awkward

338
00:20:48,955 --> 00:20:52,798
these, you know, or whatever.
And they've been trained to

339
00:20:52,798 --> 00:20:56,978
observe the smell and say, yes,
this matches what the standard

340
00:20:56,978 --> 00:21:01,360
should be. Quality control test.
Now, I don't know about you, but

341
00:21:01,360 --> 00:21:05,067
I've lost my sense of smell,
during the last few years.

342
00:21:05,067 --> 00:21:09,315
Throughout the pandemic. And so,
if I was a quality control, if

343
00:21:09,315 --> 00:21:13,022
you have a cold, right. It
really hinders your sense of

344
00:21:13,022 --> 00:21:17,135
smell. And so humans can be a
hard standard to match against.

345
00:21:17,135 --> 00:21:20,978
And there's not really a great
machine out there that can

346
00:21:20,978 --> 00:21:24,685
observe a wide profile of
chemical compounds, in a very

347
00:21:24,685 --> 00:21:28,730
fast manner. So that coffee
maker size machine could be used

348
00:21:28,730 --> 00:21:32,506
to quality control over
cosmetics, perfumes and the like

349
00:21:32,506 --> 00:21:35,000
paints, like in automotive
compounds.

350
00:21:35,000 --> 00:21:36,000
Audrow Nash: Really?

351
00:21:35,000 --> 00:21:38,831
Kordel France: Yeah. Yeah. And
automotive or or aerospace when

352
00:21:38,831 --> 00:21:42,787
you, when you paint or when you
anodized specific materials the,

353
00:21:42,787 --> 00:21:46,247
the paint needs to be in the
primer. It needs to be at a

354
00:21:46,247 --> 00:21:49,955
certain chemical setting and
there's a, there's a shelf life

355
00:21:49,955 --> 00:21:53,725
for it. If the humidity is too
high or the temperature is too

356
00:21:53,725 --> 00:21:56,691
high or there's a slight
imbalance of benzene or

357
00:21:56,691 --> 00:22:00,461
something within the paint of
the bonding will be off, and it

358
00:22:00,461 --> 00:22:04,416
won't be as, as as efficient as
it was really designed. So being

359
00:22:04,416 --> 00:22:08,000
able to quality control that is
extremely helpful as well.

360
00:22:08,000 --> 00:22:11,828
Audrow Nash: Yeah it's really
cool. So what I suspect is that

361
00:22:11,828 --> 00:22:15,593
for a lot of the things you're
talking about, there are very

362
00:22:15,593 --> 00:22:19,421
specialized, probably large and
expensive devices that can do

363
00:22:19,421 --> 00:22:22,935
this kind of thing or like
people that are trained and I

364
00:22:22,935 --> 00:22:26,889
bet you there's labor shortages
there for that kind of thing. I

365
00:22:26,889 --> 00:22:30,842
bet you also like, I mean, I've
gone and tried to smell candles

366
00:22:30,842 --> 00:22:34,670
or perfumes or whatever it is,
and after a few they all smell

367
00:22:34,670 --> 00:22:38,435
the same. And so I, I imagine
there's like heavy fatigue for

368
00:22:38,435 --> 00:22:42,200
these where they have to, like,
do something to, clear their

369
00:22:42,200 --> 00:22:45,966
nose or something so they could
smell again. But yeah. So it

370
00:22:45,966 --> 00:22:49,856
strikes me that there's a lot of
maybe specialized things that

371
00:22:49,856 --> 00:22:53,872
exist that serve these use cases
that you're describing. And I'm

372
00:22:53,872 --> 00:22:57,700
sure that this is not for all of
the things that you could do

373
00:22:57,700 --> 00:23:01,779
with what you're describing. But
it sounds like what you're going

374
00:23:01,779 --> 00:23:05,545
after is a much more general
approach, that can be leveraged

375
00:23:05,545 --> 00:23:09,059
to do a lot of these tasks. Is
that fair, or how are you

376
00:23:09,059 --> 00:23:10,000
thinking of it?

377
00:23:10,000 --> 00:23:13,435
Kordel France: Yeah. I mean, I
think olfactory sensing in

378
00:23:13,435 --> 00:23:17,172
general is largely an untapped,
as largely an untapped market.

379
00:23:17,172 --> 00:23:21,089
Right. I don't think we've we've
nearly hit the potential that we

380
00:23:21,089 --> 00:23:24,825
that we can. So, like I said,
the North Star, the, the general

381
00:23:24,825 --> 00:23:28,502
application we're building for
would be navigation, olfactory

382
00:23:28,502 --> 00:23:32,419
based navigation. But if we open
up people's eyes to observe that

383
00:23:32,419 --> 00:23:36,336
there's a whole other domain of,
I, you know, faction that should

384
00:23:36,336 --> 00:23:39,771
be introduced and observed,
then, that's that's that'd be

385
00:23:39,771 --> 00:23:43,146
fantastic. Because if you look
at like AI right now, the

386
00:23:43,146 --> 00:23:46,702
different modalities that are
receiving a lot of investment

387
00:23:46,702 --> 00:23:50,498
are like vision and language and
speech. So ChatGPT is language

388
00:23:50,498 --> 00:23:54,355
and vision and, and now speech,
and they have all these chemical

389
00:23:54,355 --> 00:23:57,550
capabilities or they have all of
these, these sensing

390
00:23:57,550 --> 00:24:00,985
capabilities, but they're
missing the sense of olfaction.

391
00:24:00,985 --> 00:24:04,661
Now, now, one might argue, was
like, why would I want ChatGPT

392
00:24:04,661 --> 00:24:08,277
to smell? Right? It's like,
well, maybe we have an investor,

393
00:24:08,277 --> 00:24:09,000
but that was

394
00:24:09,000 --> 00:24:11,561
Audrow Nash: Yeah, it could be
like, is this fruit bad? Is this

395
00:24:11,561 --> 00:24:14,000
food bad? Whatever. That's. And
a lot of other things. Yeah.

396
00:24:14,000 --> 00:24:17,159
Kordel France: Yeah. That's
that's an excellent example.

397
00:24:17,159 --> 00:24:20,656
Right. If you send it a picture
right now, maybe it can, maybe

398
00:24:20,656 --> 00:24:24,210
it can describe. Well the apple
doesn't look super great. Maybe

399
00:24:24,210 --> 00:24:27,651
it's okay. Maybe it's not a eat,
but maybe it looks okay, but

400
00:24:27,651 --> 00:24:30,867
it's off gassing certain
alcohols, right. Which come from

401
00:24:30,867 --> 00:24:34,138
fruit. And if the olfactory
sensors can pick that up, that

402
00:24:34,138 --> 00:24:37,579
can be like, well, it's actually
below the healthy level too,

403
00:24:37,579 --> 00:24:41,133
right? You should. Yeah. Yeah,
it's too ripe. You shouldn't eat

404
00:24:41,133 --> 00:24:44,518
them. Which is which is really
interesting. Right. So if, we

405
00:24:44,518 --> 00:24:47,959
have like, language models, we
have a vision language models,

406
00:24:47,959 --> 00:24:51,456
and which combine both of those
senses. So I'm waiting for, an

407
00:24:51,456 --> 00:24:55,010
incentive to building these, I'd
like to see more investment in

408
00:24:55,010 --> 00:24:58,677
olfaction vision language models
where you combine all the senses

409
00:24:58,677 --> 00:25:02,062
together and you're just fuzing
multiple modalities, vision,

410
00:25:02,062 --> 00:25:05,277
language, olfaction into a
single, model that can be that

411
00:25:05,277 --> 00:25:08,605
can be really observe used to
observe the world in a better

412
00:25:08,605 --> 00:25:09,000
manner.

413
00:25:09,000 --> 00:25:12,733
Audrow Nash: Yeah. I think
that's so cool. So one thing

414
00:25:12,733 --> 00:25:16,737
that strikes me thinking of like
thinking as an application

415
00:25:16,737 --> 00:25:20,741
developer for this kind of
thing, I don't know how to think

416
00:25:20,741 --> 00:25:24,949
and smell. Like one thing that
has been interesting is like in

417
00:25:24,949 --> 00:25:28,817
the past, I've used a thermal
camera. So we put this on a

418
00:25:28,817 --> 00:25:32,957
drone and I was using this for
detection of different things.

419
00:25:32,957 --> 00:25:37,029
And I thought this was really
cool because there's all sorts

420
00:25:37,029 --> 00:25:40,965
of new ways you can perceive
with it. So, for example, you

421
00:25:40,965 --> 00:25:45,037
can look on the ground where
someone was standing and you'll

422
00:25:45,037 --> 00:25:49,042
see little warm footprints. You
can tell if a mug is hot or

423
00:25:49,042 --> 00:25:53,114
where it was like, you can see
all sorts of cool things. You

424
00:25:53,114 --> 00:25:56,914
can tell if appliances are
running from a distance, like

425
00:25:56,914 --> 00:26:00,783
all sorts of things that I
wouldn't think of until I look

426
00:26:00,783 --> 00:26:04,787
through one. For an extended
period of time and kind of see

427
00:26:04,787 --> 00:26:09,130
how it inter or how it perceives
the world. I imagine that smell

428
00:26:09,130 --> 00:26:13,270
is very similar in that it's
hard to understand without going

429
00:26:13,270 --> 00:26:17,546
and trying it a lot. And I know
we have our senses of smell, so

430
00:26:17,546 --> 00:26:21,414
maybe that's a good analog, but
if you're going one for a

431
00:26:21,414 --> 00:26:25,758
trillion particle, one particle
in a trillion and picking it up,

432
00:26:25,758 --> 00:26:29,830
it might, we might our sense of
smell. No, may not be a good

433
00:26:29,830 --> 00:26:33,970
analog to it. Tell me how to
think and smell. The weird thing

434
00:26:33,970 --> 00:26:37,702
to say, but, like, how would you
go about developing an

435
00:26:37,702 --> 00:26:41,367
application or trying to
understand what this would be

436
00:26:41,367 --> 00:26:45,507
good for? How would you be an
application developer trying to

437
00:26:45,507 --> 00:26:47,000
learn how smell works?

438
00:26:47,000 --> 00:26:50,689
Kordel France: It's a really
it's a terrific question. And

439
00:26:50,689 --> 00:26:54,696
it's really the, in my opinion,
probably the one of the biggest

440
00:26:54,696 --> 00:26:58,575
hindrances from olfaction moving
at the pace, the rest of the

441
00:26:58,575 --> 00:27:02,455
modalities. Right. The vision in
language in audio. So if you

442
00:27:02,455 --> 00:27:05,762
imagine an image it has
basically every pixel, every

443
00:27:05,762 --> 00:27:09,515
color can be quantified with
red, green and blue. There are

444
00:27:09,515 --> 00:27:13,458
three colors that represent all
of the colors in an image. And

445
00:27:13,458 --> 00:27:17,020
the standard for images are
PNGs, JPEGs, etc. those if I

446
00:27:17,020 --> 00:27:20,899
want to. As an AI or machine
learning engineers, I want to go

447
00:27:20,899 --> 00:27:24,524
build a machine learning model.
I can go online and I can

448
00:27:24,524 --> 00:27:28,468
download literally terabytes of
open source data sets of these

449
00:27:28,468 --> 00:27:32,347
standardized vision formats and
PNG, Jpeg, etc. that all root

450
00:27:32,347 --> 00:27:35,718
down to RGB pixels. And a
similar method applies with

451
00:27:35,718 --> 00:27:39,661
different frequencies in audio.
So there are standards for all

452
00:27:39,661 --> 00:27:43,541
these different data modalities
that we train AI on. But if I

453
00:27:43,541 --> 00:27:47,294
want to build an olfaction
model, there are no open source.

454
00:27:47,294 --> 00:27:50,728
There's not there aren't
terabytes of data sets online

455
00:27:50,728 --> 00:27:52,000
that I can probably.

456
00:27:52,000 --> 00:27:53,000
Audrow Nash: None or almost
none.

457
00:27:53,000 --> 00:27:56,609
Kordel France: Right. And while
there are different data sets,

458
00:27:56,609 --> 00:28:00,217
they're not standardized. Some
people think, yeah, some people

459
00:28:00,217 --> 00:28:03,884
think that, olfaction should be
viewed as a multi-spectral time

460
00:28:03,884 --> 00:28:07,202
series. Others think it should
be viewed as a graph. This

461
00:28:07,202 --> 00:28:10,927
because it views a spectrogram.
So there's not even an agreement

462
00:28:10,927 --> 00:28:14,419
on how we should represent the
sense of smell. So I think it

463
00:28:14,419 --> 00:28:18,028
really part of what we're trying
to do is Scentience is really

464
00:28:18,028 --> 00:28:21,462
develop a common consensus
between industry leaders to say,

465
00:28:21,462 --> 00:28:25,070
what should that be? What's the
state of the art? What is, you

466
00:28:25,070 --> 00:28:28,271
know, everyone for people that
have been working in the

467
00:28:28,271 --> 00:28:31,938
industry for decades, why do you
think it should be represented

468
00:28:31,938 --> 00:28:35,431
by a graph? Why do you think is
represented by multispectral

469
00:28:35,431 --> 00:28:39,097
time series? As a researcher, I
can argue for any one of those,

470
00:28:39,097 --> 00:28:42,648
but it's it's it's we need to
get common consensus before, we

471
00:28:42,648 --> 00:28:45,849
can move forward to actually
develop something that, is

472
00:28:45,849 --> 00:28:49,167
standardized, like vision,
language, etc.. And so that's,

473
00:28:49,167 --> 00:28:52,950
that's part of what we're trying
to do with our data is we've had

474
00:28:52,950 --> 00:28:56,151
to try to standardize, in
multiple formats. But I think

475
00:28:56,151 --> 00:28:59,818
that as we start to progress and
move forward, we'll be able to

476
00:28:59,818 --> 00:29:03,368
standardize how people should
think about the sense of smell.

477
00:29:03,368 --> 00:29:06,802
Because we aren't going to be
able the quality control, the

478
00:29:06,802 --> 00:29:10,527
sense of smell as we get down to
like a part per trillion level,

479
00:29:10,527 --> 00:29:13,845
because humans can't really
detect to that level, similar

480
00:29:13,845 --> 00:29:17,395
with like, images or sounds that
have frequencies beyond with

481
00:29:17,395 --> 00:29:20,655
which we can hear. Right. If
you'd so like dogs can hear

482
00:29:20,655 --> 00:29:24,263
about above a certain frequency
that then humans can and, it's

483
00:29:24,263 --> 00:29:27,872
the same thing with olfaction.
So we have to have sensors that

484
00:29:27,872 --> 00:29:31,481
can validate whether or not a
chemical compound is present. So

485
00:29:31,481 --> 00:29:34,915
it kind of there are these
different foundational, I guess,

486
00:29:34,915 --> 00:29:38,116
pillars that need to be
developed before developers can

487
00:29:38,116 --> 00:29:41,841
really springboard and make, all
of these olfactory applications

488
00:29:41,841 --> 00:29:45,159
in the world. And I hope that
we're moving in a direction

489
00:29:45,159 --> 00:29:48,534
where we can define that for the
state of olfaction moving

490
00:29:48,534 --> 00:29:49,000
forward.

491
00:29:49,000 --> 00:29:53,228
Audrow Nash: Oh, yeah. Yeah, it
seems like so for the robot

492
00:29:53,228 --> 00:29:57,598
operating system. ROS we at the
end I wasn't involved, but at

493
00:29:57,598 --> 00:30:01,754
this time, but a lot of the
benefit that I think the robot

494
00:30:01,754 --> 00:30:05,552
operating system has done for
the robot, the robotics

495
00:30:05,552 --> 00:30:09,421
community, is that we have
standardized things in some

496
00:30:09,421 --> 00:30:13,076
respects. So we've made it so
that there is a robot

497
00:30:13,076 --> 00:30:17,375
description language, and we
have a bunch of common messages

498
00:30:17,375 --> 00:30:21,961
that, are used for all sorts of
different things. And that to me

499
00:30:21,961 --> 00:30:26,546
has been really valuable because
just seeing from when I started

500
00:30:26,546 --> 00:30:30,846
in robotics to now, it's like
it's much easier to share data

501
00:30:30,846 --> 00:30:34,643
and, share software and all
sorts of things. And so I

502
00:30:34,643 --> 00:30:39,014
imagine it will be a somewhat
similar journey with scent. But

503
00:30:39,014 --> 00:30:43,027
I wonder also if because of
robotics and where it is and

504
00:30:43,027 --> 00:30:47,398
where all the interest in AI and
stuff, there might be a much

505
00:30:47,398 --> 00:30:51,912
faster convergence in scent. And
also given that it's like from

506
00:30:51,912 --> 00:30:56,068
my naive perspective, it is a
smaller dimension of data to

507
00:30:56,068 --> 00:31:00,439
represent, but maybe that's not
true at all. But you have the

508
00:31:00,439 --> 00:31:04,809
different classes to represent,
different smells. And I guess

509
00:31:04,809 --> 00:31:09,109
there's a lot of discussion on
how to do that because it's a

510
00:31:09,109 --> 00:31:13,694
complex scent. But, I wonder if
it will go faster, I guess is my

511
00:31:13,694 --> 00:31:17,707
point. And that would be
exciting if you guys were to be

512
00:31:17,707 --> 00:31:20,000
probably the big pusher in this.

513
00:31:20,000 --> 00:31:23,880
Kordel France: Yeah, I agree, I
think it will go faster because

514
00:31:23,880 --> 00:31:27,883
we've we've standardized so many
different things within the last

515
00:31:27,883 --> 00:31:31,332
few years that I guess we have
that fluency now as, as a

516
00:31:31,332 --> 00:31:35,335
community, as an engineering and
research community. I think that

517
00:31:35,335 --> 00:31:38,846
one thing I'd like to see, and
it seems like we're moving

518
00:31:38,846 --> 00:31:42,233
forward in that direction, is
we're looking to build an

519
00:31:42,233 --> 00:31:46,236
analogy to the RGB image format.
So all images can be represented

520
00:31:46,236 --> 00:31:49,685
by three different colors.
Perhaps I'll olfactory senses

521
00:31:49,685 --> 00:31:53,134
can be, can be detected or
interpreted by a, an array of

522
00:31:53,134 --> 00:31:56,952
specific values. Right. So maybe
alcohols can be quantified as

523
00:31:56,952 --> 00:32:00,401
one aldehydes, aromatic
compounds. You know, if you have

524
00:32:00,401 --> 00:32:03,973
a level of these maybe that
indicates the level of benzene

525
00:32:03,973 --> 00:32:07,791
that's present, right. Or the
level of ethanol that's present.

526
00:32:07,791 --> 00:32:11,302
And so instead of having to
build out these large exotic,

527
00:32:11,302 --> 00:32:14,935
you know, sensing applications,
we have an abbreviation for

528
00:32:14,935 --> 00:32:18,569
what, every chemical around us
looks like in a much simpler

529
00:32:18,569 --> 00:32:19,000
format.

530
00:32:19,000 --> 00:32:22,000
Audrow Nash: That would be so
cool. Okay.

531
00:32:22,000 --> 00:32:25,320
Kordel France: And then if you
move forward with that, right,

532
00:32:25,320 --> 00:32:28,694
then you can publish, you know,
these, these values, right. So

533
00:32:28,694 --> 00:32:31,959
you can publish these arrays
into a data set that others can

534
00:32:31,959 --> 00:32:35,170
use to train and others can use
to download, and build, you

535
00:32:35,170 --> 00:32:38,218
know, large scale machine
learning models similar to how

536
00:32:38,218 --> 00:32:41,429
humans don't, like we have
vision data sets, right? We have

537
00:32:41,429 --> 00:32:44,857
snapshots of vision, which are
images. We can have snapshots of

538
00:32:44,857 --> 00:32:48,014
chemicals which might be, you
know, graphs or like I said,

539
00:32:48,014 --> 00:32:50,952
multispectral time series
snippets that can be used to

540
00:32:50,952 --> 00:32:54,218
represent a compound and its
relation image, etc.. So that's

541
00:32:54,218 --> 00:32:57,537
I think you're right. I think I
think we'll, we'll see a much

542
00:32:57,537 --> 00:33:00,476
faster acceleration with
olfaction because we have the

543
00:33:00,476 --> 00:33:02,000
fluency with other synonyms.

544
00:33:02,000 --> 00:33:06,147
Audrow Nash: Oh yeah. Do you
think, if you break it down into

545
00:33:06,147 --> 00:33:10,022
something that looks kind of
like RGB, where you have the

546
00:33:10,022 --> 00:33:14,101
different compounds, could you
imagine seeing something like

547
00:33:14,101 --> 00:33:18,384
and you mentioned image, image
recognition and image data sets.

548
00:33:18,384 --> 00:33:22,123
Would you imagine seeing
something like image net which

549
00:33:22,123 --> 00:33:26,474
is a big data set. And like what
do they do? They do comparisons

550
00:33:26,474 --> 00:33:30,621
for accuracy, or something like
this. Like, could you imagine

551
00:33:30,621 --> 00:33:33,000
having something similar for
scent?

552
00:33:33,000 --> 00:33:36,469
Kordel France: Absolutely. So
one thing we are working to

553
00:33:36,469 --> 00:33:40,242
develop, because we have so many
different sensing instruments

554
00:33:40,242 --> 00:33:44,198
and different chemical compounds
for which we're using to develop

555
00:33:44,198 --> 00:33:47,971
our sensors, is it's not called
sensor net, but we'll say it's

556
00:33:47,971 --> 00:33:51,622
called sensor net. It is an
analogy to ImageNet, right? It's

557
00:33:51,622 --> 00:33:55,152
a vast array of olfaction data
that can be used to train a

558
00:33:55,152 --> 00:33:58,438
large model, like like a
foundation model, in parallel

559
00:33:58,438 --> 00:34:01,846
with like ChatGPT size level
application. So hundreds of

560
00:34:01,846 --> 00:34:05,254
billions, trillion parameter
models, they need extremely

561
00:34:05,254 --> 00:34:08,967
large data sets. And if we can
build that foundation data set

562
00:34:08,967 --> 00:34:12,679
that sense. Net, if you will,
then that gives researchers the

563
00:34:12,679 --> 00:34:16,209
ability to go infuse that
modality into these really large

564
00:34:16,209 --> 00:34:17,000
scale models.

565
00:34:17,000 --> 00:34:21,460
Audrow Nash: For all of this, it
sounds like there's more of an

566
00:34:21,460 --> 00:34:25,779
industry focus than academic
focus in this or how how are you

567
00:34:25,779 --> 00:34:29,319
balancing the different
communities that you might

568
00:34:29,319 --> 00:34:33,000
interface with and provide value
to for all of this?

569
00:34:33,000 --> 00:34:36,496
Kordel France: That's where I
really appreciate the skills I

570
00:34:36,496 --> 00:34:39,992
was taught, in my PhD, skills.
I've taught my PhD because it

571
00:34:39,992 --> 00:34:43,487
there the there's a very niche
community for olfaction. They

572
00:34:43,487 --> 00:34:46,867
exist, but there are a few
researchers that have literally

573
00:34:46,867 --> 00:34:50,421
devoted their life to advancing
the state of amazing. And the

574
00:34:50,421 --> 00:34:54,208
amount of knowledge they know is
so, so under bonkers. In fact, a

575
00:34:54,208 --> 00:34:57,820
lot of the research papers that
I use to cite my own work were

576
00:34:57,820 --> 00:35:01,257
very difficult to find and they
had very low citations, but

577
00:35:01,257 --> 00:35:04,870
there literally foundational for
the field of olfaction. And I

578
00:35:04,870 --> 00:35:08,307
just feel like they are low
cited because they, there's not

579
00:35:08,307 --> 00:35:11,453
a huge appreciation for the
sense of smell yet because

580
00:35:11,453 --> 00:35:15,066
people don't quite understand
the applications for what can be

581
00:35:15,066 --> 00:35:18,503
done by observing the air around
us. But the work is, is is

582
00:35:18,503 --> 00:35:21,882
extremely foundational. And I
mean literally, I think that

583
00:35:21,882 --> 00:35:25,495
some of these papers will look
back in a few years and they'll

584
00:35:25,495 --> 00:35:28,000
be viewed as the attention is
all you need.

585
00:35:28,000 --> 00:35:29,000
Audrow Nash: Seminal.

586
00:35:29,000 --> 00:35:32,463
Kordel France: For, you know,
Transformers or the lottery

587
00:35:32,463 --> 00:35:36,047
ticket hypothesis for printing
neural networks or something

588
00:35:36,047 --> 00:35:39,267
like that. So yeah, the
community is very niche. It's

589
00:35:39,267 --> 00:35:42,852
growing. But I'd like to see a
larger research effort and a

590
00:35:42,852 --> 00:35:46,679
larger investment, opportunity
with or effort within olfaction,

591
00:35:46,679 --> 00:35:50,203
because there are a lot of
startups and language models, a

592
00:35:50,203 --> 00:35:53,423
lot of startups in vision,
Transformers and all these

593
00:35:53,423 --> 00:35:57,251
things. But there's I mean, name
I can name two startups at top

594
00:35:57,251 --> 00:36:00,957
of my head that I would say are
really advancing state of the

595
00:36:00,957 --> 00:36:04,541
art in action. And they're
yours. Yeah. One and another one

596
00:36:04,541 --> 00:36:07,943
that's called Osmo. The
interesting thing about Osmo, so

597
00:36:07,943 --> 00:36:11,163
they have a heavy research
backing as well. But, Jeff

598
00:36:11,163 --> 00:36:15,112
Hinton just trying to join their
board. Nobel Prize winner in AI.

599
00:36:15,112 --> 00:36:18,515
Right. So that's pretty
impressive. But we have an equal

600
00:36:18,515 --> 00:36:21,917
passion. I feel like in the
mission that we're trying to

601
00:36:21,917 --> 00:36:25,380
pursue, but they're looking at
scent generation and we're

602
00:36:25,380 --> 00:36:28,904
looking at scent detection. So
they have a really, really,

603
00:36:28,904 --> 00:36:32,063
really compelling AI product
where they can generate

604
00:36:32,063 --> 00:36:35,890
different scents based on what
you describe it. So visually and

605
00:36:35,890 --> 00:36:39,232
verbally and verbally and
they're working with a lot of

606
00:36:39,232 --> 00:36:42,938
perfume companies to develop,
you know, the next Chanel five,

607
00:36:42,938 --> 00:36:46,583
the next Aqua Digital, the next
really compelling Cologne or

608
00:36:46,583 --> 00:36:50,289
perfume that people are going to
love and they have, a really

609
00:36:50,289 --> 00:36:53,813
good team that's as receiving a
lot of investment. And I'm

610
00:36:53,813 --> 00:36:56,000
really excited to see where they
go.

611
00:36:56,000 --> 00:36:59,800
Audrow Nash: Super cool. Yeah.
It's funny that it's such a

612
00:36:59,800 --> 00:37:03,796
small thing like just from my
perspective looking at scent as

613
00:37:03,796 --> 00:37:07,465
a new sensing modality for
robots. It seems like there's

614
00:37:07,465 --> 00:37:11,461
probably I don't know when it
will be yet. I actually love to

615
00:37:11,461 --> 00:37:15,523
get your opinion on this, but
there will be a wave of startups

616
00:37:15,523 --> 00:37:19,519
that all because like mobility
in robotics, indoor outdoor is

617
00:37:19,519 --> 00:37:23,450
becoming if it's not already
solved for a lot of things. And

618
00:37:23,450 --> 00:37:27,512
so it's like, okay, well, what
can we strap on to these robots

619
00:37:27,512 --> 00:37:31,246
to make them useful? And sense
is something that like the

620
00:37:31,246 --> 00:37:35,242
ability to smell. Would they had
there's so many applications

621
00:37:35,242 --> 00:37:39,108
that I can imagine there. When
do you think there will be a

622
00:37:39,108 --> 00:37:43,169
wave of startups around this? I
know you probably hope as soon

623
00:37:43,169 --> 00:37:47,035
as possible, but what would you
think? Like two years? Five

624
00:37:47,035 --> 00:37:49,000
years? What are we looking at?

625
00:37:49,000 --> 00:37:52,667
Kordel France: I think it's on
the order of five years. And I

626
00:37:52,667 --> 00:37:56,273
think in ten years the wave,
will start to not collapse, but

627
00:37:56,273 --> 00:38:00,000
relax. But I think years five
through ten, we'll see. Is that.

628
00:38:00,000 --> 00:38:02,884
Audrow Nash: S-curve? Yeah. You
get to the top of the s at the

629
00:38:02,884 --> 00:38:04,000
ten years kind of thing.

630
00:38:04,000 --> 00:38:07,100
Kordel France: Exactly. Yeah.
That's that's the way the

631
00:38:07,100 --> 00:38:10,595
current landscape looks like.
And I mean, I really think in my

632
00:38:10,595 --> 00:38:14,090
ten years we'll look back and be
like, how did we not? We have

633
00:38:14,090 --> 00:38:17,415
cameras everywhere. How do we
not observe the air around us

634
00:38:17,415 --> 00:38:20,685
more? There's so much I mean,
even from a medical or human

635
00:38:20,685 --> 00:38:24,123
perspective, there's so much
data in breath. And now it's not

636
00:38:24,123 --> 00:38:27,336
even being explored to its full
potential, let alone like

637
00:38:27,336 --> 00:38:30,267
environmental air capabilities
or factory navigation

638
00:38:30,267 --> 00:38:31,000
capabilities.

639
00:38:31,000 --> 00:38:33,000
Audrow Nash: And.

640
00:38:33,000 --> 00:38:35,122
Kordel France: I think there
will be a lot of health care and

641
00:38:35,122 --> 00:38:37,000
medical startups that, look at
olfactory capabilities.

642
00:38:37,000 --> 00:38:38,000
Audrow Nash: Really.

643
00:38:38,000 --> 00:38:43,333
Kordel France: In breath. Yeah.
And then in robotic applications

644
00:38:43,333 --> 00:38:44,000
as well.

645
00:38:44,000 --> 00:38:47,375
Audrow Nash: Yeah. So
interesting. I think that it

646
00:38:47,375 --> 00:38:51,222
will be a big wave. So now
you're saying you said earlier

647
00:38:51,222 --> 00:38:55,339
that you want an API for smell.
And so I'm imagining that, if

648
00:38:55,339 --> 00:38:59,524
you can break things down into
like RGB for compounds that you

649
00:38:59,524 --> 00:39:03,506
identify, that would be very
helpful. But I'm wondering how

650
00:39:03,506 --> 00:39:07,353
it would work for like, I don't
know, I'm trying to smell

651
00:39:07,353 --> 00:39:11,403
something and it gives me some
compound values, then I don't

652
00:39:11,403 --> 00:39:15,250
really know how to map to
anything in the real world. And

653
00:39:15,250 --> 00:39:19,030
I can go collect a bunch of
data, but it's it's, I guess

654
00:39:19,030 --> 00:39:23,215
labeling that data becomes a bit
of a challenge. And I imagine

655
00:39:23,215 --> 00:39:27,265
that for many things like, say,
fruit detection or something

656
00:39:27,265 --> 00:39:31,382
like you could, you could have a
public library that would do

657
00:39:31,382 --> 00:39:35,567
this kind of thing. Well, and
tell you, is that apple too ripe

658
00:39:35,567 --> 00:39:39,684
or not? How would you imagine
having that or I guess, tell me

659
00:39:39,684 --> 00:39:43,936
about your API ideas and tell me
about how high up in the logic

660
00:39:43,936 --> 00:39:47,716
do you think you're going to be
going. And maybe it's an

661
00:39:47,716 --> 00:39:51,968
incremental keep improving, keep
getting higher. Kind of like a

662
00:39:51,968 --> 00:39:56,018
luxon s but for smell. But tell
me about that. Tell me about

663
00:39:56,018 --> 00:40:00,000
your API. And tell me about how
you expose things to users.

664
00:40:00,000 --> 00:40:03,281
Kordel France: So one thing
we're trying to do to make it

665
00:40:03,281 --> 00:40:06,446
really is, is easy and
compelling to use as possible is

666
00:40:06,446 --> 00:40:09,899
focus on fusion, fusion with
other modalities. So at to your

667
00:40:09,899 --> 00:40:12,892
point, olfactory senses by
itself maybe aren't super

668
00:40:12,892 --> 00:40:16,058
valuable, right. If, if,
someone's blind and they can't

669
00:40:16,058 --> 00:40:19,511
hear, or like, a animal's only
navigating by scent, it might

670
00:40:19,511 --> 00:40:22,849
have a, harder tip or harder
time of trying to observe the

671
00:40:22,849 --> 00:40:26,475
world around it. But if you add
in vision at an audio, you have

672
00:40:26,475 --> 00:40:30,043
the ability to touch and move
around the world. You get a much

673
00:40:30,043 --> 00:40:33,324
different, you get a much
different perspective. So we're

674
00:40:33,324 --> 00:40:36,604
trying to, with our olfaction
vision language models that

675
00:40:36,604 --> 00:40:40,288
we're building, we're trying to
make it so that with the API you

676
00:40:40,288 --> 00:40:43,626
get the, the olfaction data
back. So I know what the smell

677
00:40:43,626 --> 00:40:47,309
is. But now if I want to send a
request and interact with like a

678
00:40:47,309 --> 00:40:50,647
another camera or audio or
whatever, then I'm able to fuze

679
00:40:50,647 --> 00:40:54,043
that data together and say,
okay, I see that there's carbon

680
00:40:54,043 --> 00:40:57,612
monoxide in the scene. There's
nothing that I see in the scene

681
00:40:57,612 --> 00:41:01,237
right now that, gives off carbon
dioxide. But I do see a garage

682
00:41:01,237 --> 00:41:04,288
that I'm looking at lets
instructor move the robot to

683
00:41:04,288 --> 00:41:07,971
move left or right. So if I move
left, I start to see a vehicle.

684
00:41:07,971 --> 00:41:11,482
Okay, well, yeah, that makes
sense. There's a carbon monoxide

685
00:41:11,482 --> 00:41:14,878
emitting exhaust pipe from the
vehicle that's running right

686
00:41:14,878 --> 00:41:18,101
now. So, you kind of start to
interact with the mobility

687
00:41:18,101 --> 00:41:21,439
capabilities of the robot and
the vision capabilities. So,

688
00:41:21,439 --> 00:41:25,007
right now some vision language
models have the ability to say,

689
00:41:25,007 --> 00:41:28,748
like, if you give them if you if
you were to prompt them and say,

690
00:41:28,748 --> 00:41:32,144
I see carbon monoxide in this
image, where is it at some in

691
00:41:32,144 --> 00:41:35,655
our experience, they will force
fit something in the image to

692
00:41:35,655 --> 00:41:38,763
say like carbon monoxide is
here, whether or not it is

693
00:41:38,763 --> 00:41:42,273
right, but it can't actually
detect it. It's just saying, oh,

694
00:41:42,273 --> 00:41:45,669
I'm looking at a garage, you
know, maybe the floor are made

695
00:41:45,669 --> 00:41:49,122
in carbon dioxide, which is
nonsensical. But if you can take

696
00:41:49,122 --> 00:41:52,691
if you can incorporate spatial
and olfactory intelligence into

697
00:41:52,691 --> 00:41:55,971
it, say, look, I definitely see
carbon monoxide here, but

698
00:41:55,971 --> 00:41:59,367
there's nothing that's it looks
like it's emitting a strong

699
00:41:59,367 --> 00:42:02,532
chemical, chemical scent of
carbon monoxide. Let's move

700
00:42:02,532 --> 00:42:05,755
around and see if there's
something, you know, close by.

701
00:42:05,755 --> 00:42:09,381
So that logic of being able to
say, I see a garage, there might

702
00:42:09,381 --> 00:42:12,892
be a vehicle nearby, because I
know those emit car that those

703
00:42:12,892 --> 00:42:16,230
that emits carbon monoxide.
Let's look around and see if I

704
00:42:16,230 --> 00:42:19,799
can find a vehicle like it's a
very, very rudimentary example.

705
00:42:19,799 --> 00:42:23,022
But you can see how olfaction by
itself isn't that super

706
00:42:23,022 --> 00:42:26,475
interesting. But when you fuze
it with other modalities, you

707
00:42:26,475 --> 00:42:29,698
get a lot higher sense of
intelligence. And so if we can

708
00:42:29,698 --> 00:42:33,209
make our API able to fuze with
other modalities and, interact

709
00:42:33,209 --> 00:42:36,662
with other, other APIs on the
market that we get, we can get

710
00:42:36,662 --> 00:42:40,000
that level of intelligence that
we really want to achieve.

711
00:42:40,000 --> 00:42:44,106
Audrow Nash: Yeah. I feel like
saying is not interesting is

712
00:42:44,106 --> 00:42:48,281
kind of the wrong thing. It's
not you don't have context and

713
00:42:48,281 --> 00:42:52,526
because you don't have context
it's not as meaningful. That's

714
00:42:52,526 --> 00:42:56,562
first kind of thing. But I
wonder like to me what it seems

715
00:42:56,562 --> 00:43:00,250
like is you guys are doing
something like the way I'm

716
00:43:00,250 --> 00:43:04,425
thinking of, Scentience in my
head is similar to Luxonis. Do

717
00:43:04,425 --> 00:43:07,000
you know Luxonis, the camera
company?

718
00:43:07,000 --> 00:43:09,000
Kordel France: I do not.

719
00:43:09,000 --> 00:43:12,989
Audrow Nash: So it strikes me
that. So they do a lot of vision

720
00:43:12,989 --> 00:43:17,043
related work and a lot of depth
camera. I've talked to Bradley,

721
00:43:17,043 --> 00:43:20,968
their CEO, a number of times.
And he. Well, it seems like you

722
00:43:20,968 --> 00:43:24,765
guys are very well aligned, just
in different modalities. I

723
00:43:24,765 --> 00:43:28,432
wonder, but so it sounds like
your challenge becomes more

724
00:43:28,432 --> 00:43:32,422
multimodal. And I wonder because
that to me sounds like a much

725
00:43:32,422 --> 00:43:36,411
harder problem. Now, to make it
multimodal, as opposed to just

726
00:43:36,411 --> 00:43:40,272
exposing sent information to
this kind of thing, I wonder if

727
00:43:40,272 --> 00:43:44,197
a partnership with a company
like Luxonis, or something could

728
00:43:44,197 --> 00:43:48,251
be very valuable to help with,
heavy lift of, like, integrating

729
00:43:48,251 --> 00:43:51,919
with cameras or something like
this, or or maybe you just

730
00:43:51,919 --> 00:43:55,715
abstract it and say we can send
an image to anything like a

731
00:43:55,715 --> 00:43:59,447
ChatGPT and give our thing which
with context, how are you

732
00:43:59,447 --> 00:44:03,372
thinking about it? Because now
that you're saying multimodal,

733
00:44:03,372 --> 00:44:06,718
it sounds much harder to me,
which that might be the

734
00:44:06,718 --> 00:44:10,000
undertaking in a sense. But how
do you think of it?

735
00:44:10,000 --> 00:44:13,425
Kordel France: Yeah, I fully
agree. The what we're trying to

736
00:44:13,425 --> 00:44:16,963
do with the API is make it so
that we can we adopt interfacing

737
00:44:16,963 --> 00:44:20,103
standards, conventional
software, interfacing standards

738
00:44:20,103 --> 00:44:23,185
like, RESTful requests or
WebSockets or something like

739
00:44:23,185 --> 00:44:26,210
that that allows you or yeah,
other other interfacing

740
00:44:26,210 --> 00:44:29,521
standards that allow us to be
easily adopted by other, for

741
00:44:29,521 --> 00:44:32,831
example, camera companies,
because we don't want to become

742
00:44:32,831 --> 00:44:35,000
another camera company, other
company.

743
00:44:35,000 --> 00:44:36,000
Audrow Nash: Competitive.

744
00:44:36,000 --> 00:44:39,329
Kordel France: That. Right. And
we'll, we'll, we'll all starve

745
00:44:39,329 --> 00:44:42,497
by trying to build the next best
camera with no competitive

746
00:44:42,497 --> 00:44:44,000
advantage. So if we were to.

747
00:44:44,000 --> 00:44:46,000
Audrow Nash: Two hard things at
once.

748
00:44:46,000 --> 00:44:49,505
Kordel France: Right, right. So
if we can partner with, you

749
00:44:49,505 --> 00:44:52,891
know, the actual state of the
art in vision, in audio and

750
00:44:52,891 --> 00:44:56,634
speech that we're able to, and
they can interface with our API,

751
00:44:56,634 --> 00:45:00,139
we become just, you know, all
better together in the vector

752
00:45:00,139 --> 00:45:03,584
sum of kind of all of our
efforts, which is, which is much

753
00:45:03,584 --> 00:45:04,000
better.

754
00:45:04,000 --> 00:45:08,050
Audrow Nash: Hell yeah. Yeah,
that sounds really cool. And you

755
00:45:08,050 --> 00:45:11,839
think that the, like, the
transformer action models, like,

756
00:45:11,839 --> 00:45:15,759
you think that's the way to go
with all of this? Like that's

757
00:45:15,759 --> 00:45:19,286
where you guys are betting.
Like, to me, it seems like

758
00:45:19,286 --> 00:45:23,141
there's a lot of value before
that type of integration, but

759
00:45:23,141 --> 00:45:27,191
that does seem very exciting in
the long term. But how are how

760
00:45:27,191 --> 00:45:30,000
are you thinking about it? And
there are a.

761
00:45:30,000 --> 00:45:33,711
Kordel France: Lot of things,
I'm very passionate about this

762
00:45:33,711 --> 00:45:37,546
question. Because, you can get a
long ways of math. So my, my,

763
00:45:37,546 --> 00:45:41,258
my, my background is all in
math. Or at the beginning of it,

764
00:45:41,258 --> 00:45:44,660
it was all in math in my
undergrad and I fundamental is

765
00:45:44,660 --> 00:45:48,371
all math. But before you need to
even look at a transformer,

766
00:45:48,371 --> 00:45:52,206
there are a lot that you can do
before, like just conventional

767
00:45:52,206 --> 00:45:54,000
signal processing techniques.

768
00:45:54,000 --> 00:45:57,759
Audrow Nash: Oh, totally. And so
awesome machine learning on your

769
00:45:57,759 --> 00:46:01,228
labeling. Like you label your
data and then you're just it's

770
00:46:01,228 --> 00:46:04,235
like supervised machine
learning, where you can just

771
00:46:04,235 --> 00:46:07,935
tell if that fruit is rotten in
a very narrow case, this kind of

772
00:46:07,935 --> 00:46:11,000
thing where the product is bad,
whatever it might be.

773
00:46:11,000 --> 00:46:14,566
Kordel France: Yes, I totally
agree. And so, a lot of the

774
00:46:14,566 --> 00:46:18,257
efforts originally started in
that format. Now transformers

775
00:46:18,257 --> 00:46:22,198
become extremely compelling when
you have a lot of data and you

776
00:46:22,198 --> 00:46:26,140
have a lot of compute. I mean,
all of the large language models

777
00:46:26,140 --> 00:46:30,018
are stacked transformers and,
but they're trained on the whole

778
00:46:30,018 --> 00:46:33,522
internet. Companies have
hundreds of millions of dollars

779
00:46:33,522 --> 00:46:37,526
to train these models and a lot
of GPUs. So we we don't have the

780
00:46:37,526 --> 00:46:41,279
budget to do that. And we don't
we don't have the data to do

781
00:46:41,279 --> 00:46:45,033
that because we don't even have
a data standard. So to solve

782
00:46:45,033 --> 00:46:48,912
that problem, we are trying to
construct the data according to

783
00:46:48,912 --> 00:46:52,728
the most commonly accepted data
standards and then, build out

784
00:46:52,728 --> 00:46:56,419
the most intelligent model that
we can. Now the interesting

785
00:46:56,419 --> 00:47:00,423
thing with olfaction is we have
to use a lot of active learning.

786
00:47:00,423 --> 00:47:03,926
So we can't just build a model
and freeze it. It just it

787
00:47:03,926 --> 00:47:07,805
doesn't it doesn't work super
well. And here's why. Because if

788
00:47:07,805 --> 00:47:11,559
you have, you if you have a
chemical in one location and you

789
00:47:11,559 --> 00:47:15,312
change. So let's say we have a
Chanel five in, in Texas, and

790
00:47:15,312 --> 00:47:19,379
then we take that to Berlin, and
then we take that also to Dubai.

791
00:47:19,379 --> 00:47:23,070
It's going to look slightly
different at the most, granular

792
00:47:23,070 --> 00:47:27,074
level because there's different
air quality standards. There's a

793
00:47:27,074 --> 00:47:30,890
mixture of maybe a higher level
of carbon monoxide in Berlin.

794
00:47:30,890 --> 00:47:34,769
And there is Texas. There's, a
different level of humidity and

795
00:47:34,769 --> 00:47:38,647
temperature. So all these things
really evolve the fundamental

796
00:47:38,647 --> 00:47:41,963
way that the chemical is
observed. Chanel five hasn't

797
00:47:41,963 --> 00:47:45,967
changed, but the chemical scent
has changed, in the way it's the

798
00:47:45,967 --> 00:47:49,971
way it's interpreted. So we have
to use a lot of active learning

799
00:47:49,971 --> 00:47:53,161
to say, okay, well, I was
trained to look at carbon

800
00:47:53,161 --> 00:47:56,790
monoxide at this humidity level,
this temperature and this

801
00:47:56,790 --> 00:48:00,669
humidity or this pressure. But
how does it change according to

802
00:48:00,669 --> 00:48:04,297
variations in environmental
factors and it's different per

803
00:48:04,297 --> 00:48:08,364
compound. So, if you can imagine
the equation that you can map to

804
00:48:08,364 --> 00:48:12,117
each compound, each compound
will have a different equation.

805
00:48:12,117 --> 00:48:15,934
Effectively carbon dioxide will
very differently than acetone

806
00:48:15,934 --> 00:48:19,437
than benzene when introduced
with humidity, temperature,

807
00:48:19,437 --> 00:48:20,000
pressure.

808
00:48:20,000 --> 00:48:23,933
Audrow Nash: Super super complex
mappings basically between

809
00:48:23,933 --> 00:48:28,067
compounds and what you sense.
Yeah, it's kind of thing because

810
00:48:28,067 --> 00:48:32,000
environmental factors and all
sorts of complex things okay.

811
00:48:32,000 --> 00:48:34,846
Kordel France: So we have to we
have to build this active

812
00:48:34,846 --> 00:48:37,491
learning capability that
basically allows the machine

813
00:48:37,491 --> 00:48:40,487
learning model to say, okay, I
know I have a I have a set of

814
00:48:40,487 --> 00:48:43,183
priors, I have a set of
knowledge, but I can update my

815
00:48:43,183 --> 00:48:46,427
knowledge as I interact with the
world. Now, you can imagine that

816
00:48:46,427 --> 00:48:49,323
that gets pretty dicey, because
you don't want the machine

817
00:48:49,323 --> 00:48:52,069
learning model to become a
runaway train. So how do you

818
00:48:52,069 --> 00:48:55,014
quality control it? Well, there
are a lot of mechanisms out

819
00:48:55,014 --> 00:48:58,159
there which we can get into if
you like, but, we you really you

820
00:48:58,159 --> 00:49:00,955
want to actively learn while
making sure that the models

821
00:49:00,955 --> 00:49:04,000
learning the right things and
not learning the wrong signals.

822
00:49:04,000 --> 00:49:05,000
Audrow Nash: Not overfitting.

823
00:49:05,000 --> 00:49:08,514
Kordel France: Exactly. Yeah.
Not overfitting. And, that can

824
00:49:08,514 --> 00:49:12,086
become very difficult because
especially if you think of like

825
00:49:12,086 --> 00:49:15,541
a privacy perspective, people
don't want models that are on

826
00:49:15,541 --> 00:49:19,290
that are always learning. And if
you're always sending data to a

827
00:49:19,290 --> 00:49:22,686
cloud to interact with, or to
update a model that can make

828
00:49:22,686 --> 00:49:25,966
people feel uncomfortable,
right? So it's a matter we're

829
00:49:25,966 --> 00:49:29,187
trying to harmonize the problem
of building a balanced,

830
00:49:29,187 --> 00:49:32,525
well-rounded data set that
encompasses chemical compounds

831
00:49:32,525 --> 00:49:35,863
and all different, some and all
different scenarios. Data

832
00:49:35,863 --> 00:49:39,084
privacy. How do we ensure that
we're, we're, protecting

833
00:49:39,084 --> 00:49:42,539
people's privacy with data and
not constantly sending, more

834
00:49:42,539 --> 00:49:46,229
model updates to, you know, edge
devices, etc.. And then how do

835
00:49:46,229 --> 00:49:49,508
we build this to scale to a
much, much larger, you know,

836
00:49:49,508 --> 00:49:53,256
standard that to the point where
we don't need to actively learn

837
00:49:53,256 --> 00:49:56,829
anymore because we have such a
good standard like ImageNet or

838
00:49:56,829 --> 00:49:58,000
something like that.

839
00:49:58,000 --> 00:50:01,875
Audrow Nash: Yeah. How do you so
do you have any sort of like

840
00:50:01,875 --> 00:50:05,878
weather sensor on your devices.
You do or. Because I'm thinking

841
00:50:05,878 --> 00:50:09,944
that would be really cool, have
temperature, have humidity and I

842
00:50:09,944 --> 00:50:14,009
don't know you have to like you
have to. So that is highly so it

843
00:50:14,009 --> 00:50:17,567
goes in with your vector of
data. And then you use it to

844
00:50:17,567 --> 00:50:21,379
interpret whatever it is, and
make more accurate assessments

845
00:50:21,379 --> 00:50:25,000
because it's so dependent on
those environmental factors.

846
00:50:25,000 --> 00:50:28,287
Kordel France: Yes. And in fact,
most chemical sensors I'd argue,

847
00:50:28,287 --> 00:50:30,966
are worthless unless you
condition on temperature and

848
00:50:30,966 --> 00:50:34,000
humidity pressure you can get
away with. Oh, sometimes, but.

849
00:50:34,000 --> 00:50:37,676
Audrow Nash: It sure varies so
much too. But maybe like, I

850
00:50:37,676 --> 00:50:41,415
mean, I'm just thinking, one
company I talked to was trying

851
00:50:41,415 --> 00:50:45,345
to use a barometer, which does
pressure to identify what floor

852
00:50:45,345 --> 00:50:49,211
they were on in a multi floor
building, and it didn't work at

853
00:50:49,211 --> 00:50:52,887
all, which was crazy. Because of
all that air conditioning

854
00:50:52,887 --> 00:50:56,183
systems and things in the
buildings that it just was

855
00:50:56,183 --> 00:51:00,049
meaningless data, which was so
interesting. So that was a big

856
00:51:00,049 --> 00:51:01,000
surprise to me.

857
00:51:01,000 --> 00:51:03,964
Kordel France: What's
interesting, that's a that's a

858
00:51:03,964 --> 00:51:07,498
really good point you bring up,
because one thing I definitely

859
00:51:07,498 --> 00:51:11,089
underestimated as we approached
these problems was how to model

860
00:51:11,089 --> 00:51:14,395
the air around you and how to
model plumes. Right? So from

861
00:51:14,395 --> 00:51:17,587
like, you know, factory
navigation perspective, you have

862
00:51:17,587 --> 00:51:20,778
to be able to model the plume
and understand how the air

863
00:51:20,778 --> 00:51:24,312
around you is changing. So, like
computational fluid dynamics,

864
00:51:24,312 --> 00:51:27,960
which are used to model like how
airplanes behave, that actually

865
00:51:27,960 --> 00:51:31,608
comes into play hugely with what
we're doing, Because we have to

866
00:51:31,608 --> 00:51:35,256
understand how the air around us
is moving and responding to its

867
00:51:35,256 --> 00:51:38,847
environment. So, like, if we're
outside, there's and we need to

868
00:51:38,847 --> 00:51:42,210
be able to know what the wind
speed is, the wind direction.

869
00:51:42,210 --> 00:51:45,687
And can we detect that, in real
time? And how is it changing?

870
00:51:45,687 --> 00:51:48,993
It's not as simple as just
pulling from like, weather APIs

871
00:51:48,993 --> 00:51:52,470
because from a local. Yeah. From
a local setting, you need to

872
00:51:52,470 --> 00:51:56,118
know exactly what's going on and
how it's changing. And, from if

873
00:51:56,118 --> 00:51:59,709
you're looking at like an indoor
perspective, the problem with,

874
00:51:59,709 --> 00:52:03,071
with olfactory navigation is
that you don't know if there's

875
00:52:03,071 --> 00:52:06,434
an Hvac system. And is that
causing the draft or is an open

876
00:52:06,434 --> 00:52:09,797
door. So that's where vision
comes in handy. It's like, how

877
00:52:09,797 --> 00:52:13,388
do I contextualize what's around
me to understand how the plume

878
00:52:13,388 --> 00:52:16,295
might behaving, to then
understand how the chemical

879
00:52:16,295 --> 00:52:19,658
signatures are changing? And,
that's a very hard problem to

880
00:52:19,658 --> 00:52:20,000
solve.

881
00:52:20,000 --> 00:52:24,063
Audrow Nash: Yeah. It seems like
a direct modeling approach would

882
00:52:24,063 --> 00:52:28,125
be just intractable. Completely,
because you need perfect sensing

883
00:52:28,125 --> 00:52:32,000
of your environment. And then
even if you did have that, like,

884
00:52:32,000 --> 00:52:35,813
say, you have a perfect model of
the building and all the air

885
00:52:35,813 --> 00:52:39,688
flows and everything else you
would like the fluid dynamics of

886
00:52:39,688 --> 00:52:43,313
the whole air, would probably be
very challenging and time

887
00:52:43,313 --> 00:52:46,000
consuming to model, in any real
time sense.

888
00:52:46,000 --> 00:52:49,000
Kordel France: Yes, it's yeah,
it's extremely challenging.

889
00:52:49,000 --> 00:52:51,000
Audrow Nash: And.

890
00:52:51,000 --> 00:52:54,484
Kordel France: Yeah, there
there's a bunch of great work

891
00:52:54,484 --> 00:52:58,217
out there. You know, honestly,
there's a lot of really great

892
00:52:58,217 --> 00:53:02,074
work that happened in like the
60s and 70s from like very old,

893
00:53:02,074 --> 00:53:06,055
DARPA research papers and AFR or
Air Force Research Lab research

894
00:53:06,055 --> 00:53:09,601
papers on plume modeling and,
how to model turbulence and

895
00:53:09,601 --> 00:53:13,023
everything that have been really
influential and, yeah,

896
00:53:13,023 --> 00:53:16,880
ultimately, I think it's going
to be all neural network based,

897
00:53:16,880 --> 00:53:18,000
kind of inspiring.

898
00:53:18,000 --> 00:53:19,000
Audrow Nash: Learned.

899
00:53:18,000 --> 00:53:21,368
Kordel France: Yeah, learned.
But, there's been a lot of

900
00:53:21,368 --> 00:53:24,977
really great work. That's been
published decades ago, that's

901
00:53:24,977 --> 00:53:26,000
been influential.

902
00:53:26,000 --> 00:53:30,123
Audrow Nash: Yeah. I imagine
with that kind of thing, you

903
00:53:30,123 --> 00:53:34,535
almost need an embodiment in a
sense, like you need something

904
00:53:34,535 --> 00:53:38,875
in the physical world that's
going to move around and learn.

905
00:53:38,875 --> 00:53:43,287
Or, I mean, maybe you could even
gather some information from

906
00:53:43,287 --> 00:53:47,265
simulation if you were
simulating these plumes. But, it

907
00:53:47,265 --> 00:53:51,532
feels hard from the perspective
of a company that makes the

908
00:53:51,532 --> 00:53:56,234
sensor, in a sense, like like an
alterna model or maybe something

909
00:53:56,234 --> 00:54:00,429
that you're pursuing or some
other company that could be a

910
00:54:00,429 --> 00:54:04,262
partner or something. I'm
thinking of like, the first

911
00:54:04,262 --> 00:54:08,819
interview for this new podcast
was with a company I did, called

912
00:54:08,819 --> 00:54:13,304
Electric Sheep. And what are you
doing? Yeah. They're awesome.

913
00:54:13,304 --> 00:54:18,005
But so their whole thing is that
it's embodied and it's providing

914
00:54:18,005 --> 00:54:21,983
some value, but it's clever
because over time, they are

915
00:54:21,983 --> 00:54:26,540
getting lots of data and they're
using that on a few platforms.

916
00:54:26,540 --> 00:54:31,169
And it gets to get more and more
sophisticated because they have

917
00:54:31,169 --> 00:54:35,509
a justification for having more
robots out there and getting

918
00:54:35,509 --> 00:54:39,993
more data, which is that it's
already providing value and what

919
00:54:39,993 --> 00:54:44,622
it's currently doing. And I feel
like it's hard to scale without

920
00:54:44,622 --> 00:54:48,528
an embodiment or without a
financial justification for

921
00:54:48,528 --> 00:54:52,000
scaling. But what are your
thoughts around this?

922
00:54:52,000 --> 00:54:55,399
Kordel France: Yeah, I fully
agree because there's is a

923
00:54:55,399 --> 00:54:58,983
scalability question around the
embodiment and then around

924
00:54:58,983 --> 00:55:02,567
compute to definitely. And how
do we I mean, if we want to

925
00:55:02,567 --> 00:55:06,522
build, you know, true foundation
models for olfaction, we need a

926
00:55:06,522 --> 00:55:10,354
lot of compute, a lot of data.
And, that's I mean, I heard one

927
00:55:10,354 --> 00:55:14,000
metric that GPT four was trained
on, like $100 million. Oh.

928
00:55:14,000 --> 00:55:15,000
Audrow Nash: My God.

929
00:55:14,000 --> 00:55:17,516
Kordel France: Which is like, no
startup in the world can afford

930
00:55:17,516 --> 00:55:20,813
that, right? You literally have
to be a power tech player to

931
00:55:20,813 --> 00:55:24,000
build that type of model. Yeah,
really. So yeah. Go ahead.

932
00:55:24,000 --> 00:55:28,258
Audrow Nash: It's probably
changing for that kind of thing.

933
00:55:28,258 --> 00:55:32,155
I, I, I'm not totally aware I
have been following that

934
00:55:32,155 --> 00:55:36,340
closely, but I think a lot of
open source models have been

935
00:55:36,340 --> 00:55:40,814
trained much, much cheaper. And
that's exciting. So like maybe

936
00:55:40,814 --> 00:55:45,000
getting from 0 to 1 is so
expensive, but getting from like

937
00:55:45,000 --> 00:55:49,474
10 to 11, might only be 100th of
the cost or even less 1000 to

938
00:55:49,474 --> 00:55:53,804
10,000 there. Who knows. But I
wonder if it could be similar

939
00:55:53,804 --> 00:55:58,278
with Santen. But you guys are
probably doing the 0 to 1, which

940
00:55:58,278 --> 00:55:59,000
who knows,

941
00:55:59,000 --> 00:56:02,465
Kordel France: Well, yeah, it's,
Now it's a terrific point. The

942
00:56:02,465 --> 00:56:05,655
pruning and distillation of
these large models is starting

943
00:56:05,655 --> 00:56:09,120
to occur, which is great. And so
the costs. Yeah, you're right.

944
00:56:09,120 --> 00:56:12,365
The costs will come down. The
we're trying to really solve,

945
00:56:12,365 --> 00:56:15,500
with open source, vision
language models, trying to solve

946
00:56:15,500 --> 00:56:18,800
the problem with custom factor
models. And then how do we do

947
00:56:18,800 --> 00:56:22,155
build a custom fusion solution?
So instead of retraining, you

948
00:56:22,155 --> 00:56:25,125
know, these large vision
language models from scratch,

949
00:56:25,125 --> 00:56:28,425
how do we take the state of the
art that's open source so we

950
00:56:28,425 --> 00:56:31,615
don't have to retrain it, use
those weights, and just turn

951
00:56:31,615 --> 00:56:35,025
them, in, in in relation with
our olfaction models. So there's

952
00:56:35,025 --> 00:56:38,490
yeah. That creates, that saves a
lot of money. And I think that

953
00:56:38,490 --> 00:56:41,955
that's probably ultimately the
way to go. But I guess we'll see

954
00:56:41,955 --> 00:56:43,000
as we move forward.

955
00:56:43,000 --> 00:56:46,663
Audrow Nash: Yeah. My, my
impression would be if there's

956
00:56:46,663 --> 00:56:50,653
like an, a low hanging fruit
that provides a lot of value and

957
00:56:50,653 --> 00:56:54,512
lets you get out there, then
that would be awesome, because

958
00:56:54,512 --> 00:56:58,371
I'm thinking also like Tesla
with their self-driving, where

959
00:56:58,371 --> 00:57:02,295
they just have all their users
feeding them like millions or

960
00:57:02,295 --> 00:57:06,285
billions of miles of road data
that then they get to leverage

961
00:57:06,285 --> 00:57:10,275
as opposed to say, like cruise
or other companies that had to

962
00:57:10,275 --> 00:57:14,068
hire people to go drive the
cars. It's like very different

963
00:57:14,068 --> 00:57:18,058
scales of data. And then I think
Tesla benefited tremendously

964
00:57:18,058 --> 00:57:21,917
from all that data. And if there
was some justification for

965
00:57:21,917 --> 00:57:26,169
having lots of those sensors out
there getting lots of data, that

966
00:57:26,169 --> 00:57:28,000
would be absolutely awesome.

967
00:57:28,000 --> 00:57:31,273
Kordel France: I agree, right?
Tesla did a really amazing thing

968
00:57:31,273 --> 00:57:34,545
by being able to have all these
data gatherers on the road that

969
00:57:34,545 --> 00:57:36,000
autonomously just so clever.

970
00:57:36,000 --> 00:57:39,969
Audrow Nash: And you pay for it,
too. Yeah, it's like it was just

971
00:57:39,969 --> 00:57:43,634
brilliant. And I mean, but it
provides real value to people.

972
00:57:43,634 --> 00:57:47,237
But you pay for it to be the
additional thing, sending back

973
00:57:47,237 --> 00:57:50,779
data for them, which is just
like, I don't know, wonderful

974
00:57:50,779 --> 00:57:52,000
and crazy, isn't it?

975
00:57:52,000 --> 00:57:55,204
Kordel France: Yeah. I was like,
if you were to propose this in a

976
00:57:55,204 --> 00:57:58,162
pitch meeting 20 years ago,
people would be like, there's no

977
00:57:58,162 --> 00:57:59,000
way that'll work.

978
00:57:59,000 --> 00:58:01,000
Audrow Nash: But if you go,
yeah, totally.

979
00:58:01,000 --> 00:58:04,917
Kordel France: There's a,
there's a, something you, you

980
00:58:04,917 --> 00:58:09,405
commented on a moment ago made
me think, when you mentioned the

981
00:58:09,405 --> 00:58:13,963
embodiment and include modeling
and when I first started my PhD,

982
00:58:13,963 --> 00:58:18,308
I was trying to build olfactory
navigation for graves, and, I

983
00:58:18,308 --> 00:58:22,724
became I had some success, but
then what became very difficult

984
00:58:22,724 --> 00:58:26,000
to work through was the rotor
wash of the UAV.

985
00:58:26,000 --> 00:58:27,000
Audrow Nash: Oh, yeah.

986
00:58:27,000 --> 00:58:30,285
Kordel France: Because that can
help you because you're kind of

987
00:58:30,285 --> 00:58:33,362
pulling the air in to some
degree, but you're also adding a

988
00:58:33,362 --> 00:58:36,595
bunch of noise to the signal and
drowning out you. It's it's a

989
00:58:36,595 --> 00:58:39,567
it's a it's ultimately a
negative thing. And so the way I

990
00:58:39,567 --> 00:58:42,696
tried to solve it was, actually
putting like, little antenna

991
00:58:42,696 --> 00:58:44,000
like that were a forward.

992
00:58:44,000 --> 00:58:45,000
Audrow Nash: Motion.

993
00:58:45,000 --> 00:58:48,286
Kordel France: Of the UAV
rotors, that were like ahead of

994
00:58:48,286 --> 00:58:51,744
the or ahead of the, the rotor
wash and that seemed to help.

995
00:58:51,744 --> 00:58:54,972
But there's, there's a research
group in, so a technical

996
00:58:54,972 --> 00:58:58,546
university in I think it's Delft
out of the Netherlands with a

997
00:58:58,546 --> 00:59:01,832
partnership with Carnegie
Mellon. They were able to solve

998
00:59:01,832 --> 00:59:05,464
it by modeling the plume. So I
wasn't modeling the plume. I was

999
00:59:05,464 --> 00:59:08,865
just using as low compute as
possible, and figured I'd just

1000
00:59:08,865 --> 00:59:12,381
get ahead of the plume with the
antenna. But they modeled the

1001
00:59:12,381 --> 00:59:15,897
plume. And then on the base of
the quadcopter, they stuck the

1002
00:59:15,897 --> 00:59:19,644
sensor. So like, there's a small
pocket of air that's undisturbed

1003
00:59:19,644 --> 00:59:23,160
on the top of the quadcopter in
the center. And so they stuck

1004
00:59:23,160 --> 00:59:26,446
the sensor there, and they
performed olfactory navigation

1005
00:59:26,446 --> 00:59:29,444
that way. And then they had
some, like, better plume

1006
00:59:29,444 --> 00:59:32,960
modeling of like, this, this
this noise is due to rotor wash.

1007
00:59:32,960 --> 00:59:36,476
This noise is due to this noise
due to whatever. So it was an

1008
00:59:36,476 --> 00:59:39,992
interesting perspective to see,
like the way that they solved

1009
00:59:39,992 --> 00:59:43,163
it. And it's another it's
another tip of the hat to how

1010
00:59:43,163 --> 00:59:46,333
difficult it is to do like
olfactory detection, because

1011
00:59:46,333 --> 00:59:49,504
there are all these different
things that could even be

1012
00:59:49,504 --> 00:59:52,905
imposed by the robot itself on
the factory signal. So, it's

1013
00:59:52,905 --> 00:59:54,000
interesting really.

1014
00:59:54,000 --> 00:59:58,268
Audrow Nash: Yeah. Everything in
the environment, I don't know, I

1015
00:59:58,268 --> 01:00:02,273
just, I so we have a dog, cute
little medium sized white guy,

1016
01:00:02,273 --> 01:00:06,081
like white dog. And he, it's
awesome watching him smell in

1017
01:00:06,081 --> 01:00:10,086
the yard. Like, he'll catch us.
Like, you'll see him tilt his

1018
01:00:10,086 --> 01:00:13,960
head, and he'll have gotten a
sniff of something, and he'll

1019
01:00:13,960 --> 01:00:17,965
kind of follow it like you can
watch him follow a smell. It's

1020
01:00:17,965 --> 01:00:21,641
just it looks effortless for
him. But I imagine having a

1021
01:00:21,641 --> 01:00:25,187
robot move in such a way
because, like, I'm imagining,

1022
01:00:25,187 --> 01:00:28,798
like, a trail of the smell
coming from the source with,

1023
01:00:28,798 --> 01:00:32,934
like, little breaks throughout
that time. But he's following it

1024
01:00:32,934 --> 01:00:36,742
and moving in such a way. It's
like you need sophisticated

1025
01:00:36,742 --> 01:00:40,551
movement. And the trail is
weirdly shaped and all sorts of

1026
01:00:40,551 --> 01:00:44,293
things. Like, it's a it's a
tricky it's a tricky problem,

1027
01:00:44,293 --> 01:00:46,000
probably with all of that.

1028
01:00:46,000 --> 01:00:49,892
Kordel France: Absolutely. Yeah.
Dogs honestly are underestimated

1029
01:00:49,892 --> 01:00:53,664
on how, they can track sense for
kilometers and, yeah, the, the

1030
01:00:53,664 --> 01:00:57,017
sensitivity of their noses.
There's an incredibly higher

1031
01:00:57,017 --> 01:01:00,310
proportion of compute in their
brain attributed to, the

1032
01:01:00,310 --> 01:01:04,082
olfactory bulb than comparison
to humans. And so, I was reading

1033
01:01:04,082 --> 01:01:07,735
a paper that was talking about
how in humans, smell and taste

1034
01:01:07,735 --> 01:01:11,327
are integrated in the brain and
in dogs, smell and sound are

1035
01:01:11,327 --> 01:01:15,159
integrated. And so if you think
about like, I'm not sure if this

1036
01:01:15,159 --> 01:01:18,692
is true, but as I was reading
the paper, I started to think

1037
01:01:18,692 --> 01:01:22,224
about how you can you can hear
chemical like you can detect

1038
01:01:22,224 --> 01:01:25,996
chemical signatures through the
frequencies that they emit at a

1039
01:01:25,996 --> 01:01:29,230
quantum level. And so I'm
wondering if that's how dogs

1040
01:01:29,230 --> 01:01:32,822
ultimately detect things. And
that's why sound and smell are

1041
01:01:32,822 --> 01:01:36,534
integrated in the same part of
their processing. I don't know.

1042
01:01:36,534 --> 01:01:40,247
I don't know if that's true, but
it's just it's interesting to

1043
01:01:40,247 --> 01:01:43,839
think about because, like, why
is it just more compute? They

1044
01:01:43,839 --> 01:01:47,671
have more compute in their brain
for olfactory processing? Or is

1045
01:01:47,671 --> 01:01:51,084
it fundamentally a different
mechanism? I'm not sure. But

1046
01:01:51,084 --> 01:01:53,000
it's interesting to think about.

1047
01:01:53,000 --> 01:01:56,187
Audrow Nash: That is really
interesting. I wonder, it'd be

1048
01:01:56,187 --> 01:01:59,374
cool in like ten years. The
theory is proved correct or 20

1049
01:01:59,374 --> 01:02:02,780
years or something like that.
And it was documented here with,

1050
01:02:02,780 --> 01:02:05,912
like your theory of how that
works for this kind of thing

1051
01:02:05,912 --> 01:02:08,000
that I'm using. That would be so
cool.

1052
01:02:08,000 --> 01:02:10,000
Kordel France: Be cited in all
these research papers.

1053
01:02:10,000 --> 01:02:14,338
Audrow Nash: Yeah. How did you
get interested in smell to begin

1054
01:02:14,338 --> 01:02:18,539
with? Like doing your PhD in it?
I agree with you that it's a

1055
01:02:18,539 --> 01:02:22,808
completely under served sense.
But how did you come to this as

1056
01:02:22,808 --> 01:02:26,665
a big interest? Because it
clearly is something you have

1057
01:02:26,665 --> 01:02:30,934
pursued for many years and are
very interested in. And I think

1058
01:02:30,934 --> 01:02:33,000
there's a lot of potential in.

1059
01:02:33,000 --> 01:02:37,519
Kordel France: Yeah. So
actually, I so, in 2017, my

1060
01:02:37,519 --> 01:02:41,861
co-founder and I started, a
company called Secret

1061
01:02:41,861 --> 01:02:46,646
Technologies, and we were
building we specialized was,

1062
01:02:46,646 --> 01:02:47,000
hey.

1063
01:02:47,000 --> 01:02:48,000
Audrow Nash: That's a great
name.

1064
01:02:48,000 --> 01:02:52,337
Kordel France: Yeah. So yeah, we
were building edge applications

1065
01:02:52,337 --> 01:02:56,607
for AI. So back then it was much
harder to pack neural networks

1066
01:02:56,607 --> 01:03:00,674
onto like, a mobile phone or
like, a Ros application. And so

1067
01:03:00,674 --> 01:03:04,740
we specialized in condensing
these big neural nets down into

1068
01:03:04,740 --> 01:03:08,942
something that could be used in
more edge applications. And we

1069
01:03:08,942 --> 01:03:12,602
were contracted out by, a
company, called So Tech that

1070
01:03:12,602 --> 01:03:16,126
wanted to build these AI
applications for or this AI

1071
01:03:16,126 --> 01:03:20,057
model for their sensors that
could detect, coronavirus and

1072
01:03:20,057 --> 01:03:23,649
pneumonia. So, we were
originally just contracted out

1073
01:03:23,649 --> 01:03:27,918
by then, but we aced the job so.
Well, and our AI become such a

1074
01:03:27,918 --> 01:03:31,849
foundational part of their
product that they ended up like

1075
01:03:31,849 --> 01:03:35,848
acquiring us in. So kind of
became, technical team, of that

1076
01:03:35,848 --> 01:03:37,000
startup and then.

1077
01:03:37,000 --> 01:03:38,000
Audrow Nash: So cool. Congrats.

1078
01:03:38,000 --> 01:03:42,186
Kordel France: Yeah, yeah. And
then we, we, the, the original

1079
01:03:42,186 --> 01:03:45,960
intent was, medical and
industrial. So we were supposed

1080
01:03:45,960 --> 01:03:50,008
to focus on kind of, industrial
applications like explosive

1081
01:03:50,008 --> 01:03:54,125
detection and when the pandemic
hit that, our investors just

1082
01:03:54,125 --> 01:03:57,899
went all into the medical
domain, and they never really

1083
01:03:57,899 --> 01:04:01,811
came out of that. And so, I
ended up leaving that company

1084
01:04:01,811 --> 01:04:05,379
just because, it's very
difficult to innovate in the

1085
01:04:05,379 --> 01:04:09,702
health care field. But, that's
really how we got started as we,

1086
01:04:09,702 --> 01:04:14,093
we had no any observability into
the world of olfaction until we

1087
01:04:14,093 --> 01:04:18,279
ended up taking that job. And
then from that point forward, I

1088
01:04:18,279 --> 01:04:22,602
was like, I was hooked. I was so
fascinated by it because I saw

1089
01:04:22,602 --> 01:04:26,651
so much more than what we were
doing by just looking at us.

1090
01:04:26,651 --> 01:04:30,287
And, I just a lot of the
problems that I assumed were

1091
01:04:30,287 --> 01:04:34,610
solved had not been solved. So I
just thought the field was way

1092
01:04:34,610 --> 01:04:38,796
more mature than it is. And it's
I don't think it's a lack of

1093
01:04:38,796 --> 01:04:42,982
talent or lack of of knowledge.
It's just a lack of awareness

1094
01:04:42,982 --> 01:04:47,305
about what you can do with the
sense of, of action and how much

1095
01:04:47,305 --> 01:04:51,422
data there is and the breadth
around us. So, yeah, from that

1096
01:04:51,422 --> 01:04:53,000
point on, I was hooked.

1097
01:04:53,000 --> 01:04:57,694
Audrow Nash: Oh, yeah. Yeah. It
does seem like something. It's

1098
01:04:57,694 --> 01:05:02,464
like once you see it, you can't
unsee it for this kind of thing

1099
01:05:02,464 --> 01:05:06,780
about how powerful it is to
understand, send around us. I

1100
01:05:06,780 --> 01:05:11,474
wonder, what would you think? So
how would so like all of this

1101
01:05:11,474 --> 01:05:16,320
talk about all of this? Like, I
want to start building something

1102
01:05:16,320 --> 01:05:20,257
that has some sort of scent
model sensing for smell,

1103
01:05:20,257 --> 01:05:24,800
modality. How would people get
started with. I know you guys

1104
01:05:24,800 --> 01:05:29,267
at, Scentience are building
sensors for this kind of thing,

1105
01:05:29,267 --> 01:05:33,810
but, like, what? What is the
path to building something with

1106
01:05:33,810 --> 01:05:38,731
scent look like? And what is the
timeline look like? For you guys

1107
01:05:38,731 --> 01:05:43,426
with your devices and I guess,
how does it work for makers who

1108
01:05:43,426 --> 01:05:46,000
want to do stuff with all of
this?

1109
01:05:46,000 --> 01:05:49,808
Kordel France: Honestly, I think
it comes down to the first step,

1110
01:05:49,808 --> 01:05:53,441
the most fundamental step that
we have to do for anyone trying

1111
01:05:53,441 --> 01:05:56,839
to work in olfaction is to
standardize the data medium. So

1112
01:05:56,839 --> 01:06:00,413
how do we make sure that, like
vision is PNG and Jpeg? How do

1113
01:06:00,413 --> 01:06:03,929
we build that type of standard
for our data and get a common

1114
01:06:03,929 --> 01:06:07,034
consensus from the entire
community that this is what

1115
01:06:07,034 --> 01:06:10,256
we're going to do moving
forward? And I think if we can

1116
01:06:10,256 --> 01:06:13,772
do that, then everything else,
is enabled because Scentience

1117
01:06:13,772 --> 01:06:17,404
itself is at risk of going in
the wrong direction if we choose

1118
01:06:17,404 --> 01:06:20,627
the wrong data standard in
someone with more authority,

1119
01:06:20,627 --> 01:06:24,377
more money and more power comes
in and says, now we're not going

1120
01:06:24,377 --> 01:06:27,599
that we're going this way. And
all of our processing is

1121
01:06:27,599 --> 01:06:31,349
optimized for one data standard,
or we build our, hardware to be

1122
01:06:31,349 --> 01:06:34,923
optimized for a specific, you
know, like olfactory processing

1123
01:06:34,923 --> 01:06:38,087
technique. Then, you know,
there's there's always that

1124
01:06:38,087 --> 01:06:41,368
risk. So I think if you could,
if we can standardize the

1125
01:06:41,368 --> 01:06:45,117
sensible action, then that's the
first step. And then the second

1126
01:06:45,117 --> 01:06:48,574
step, I think, would be to just
get extremely fluent in the

1127
01:06:48,574 --> 01:06:52,148
state of the art and not like
within the last five years, but

1128
01:06:52,148 --> 01:06:55,488
what's been done in the last
like 40 or 50, because there

1129
01:06:55,488 --> 01:06:58,769
have been some incredible
problems that have been solved

1130
01:06:58,769 --> 01:07:02,284
that maybe were introduced in
like the 70s or 80s, but never

1131
01:07:02,284 --> 01:07:06,034
went anywhere because there just
wasn't a strong enough interest

1132
01:07:06,034 --> 01:07:09,491
or maybe there wasn't enough
compute. And so if you throw a

1133
01:07:09,491 --> 01:07:12,772
bunch more compute at the
problems that they were trying

1134
01:07:12,772 --> 01:07:16,170
to solving, you can really you
get either, a more scalable

1135
01:07:16,170 --> 01:07:19,744
solution or you get a solution
that's much more accurate. And

1136
01:07:19,744 --> 01:07:23,201
so olfaction is literally I
mean, senses itself is built on

1137
01:07:23,201 --> 01:07:27,009
the shoulders of giants. And all
the work that's been done in the

1138
01:07:27,009 --> 01:07:30,349
last few years, the GCMs that
the gas chromatography mass

1139
01:07:30,349 --> 01:07:34,099
spectrometry machine is telling
you about, a machine the size of

1140
01:07:34,099 --> 01:07:37,849
the fridge, very accurate. Takes
very long to, to get, a result,

1141
01:07:37,849 --> 01:07:41,423
that's been around for like 50,
60 years and it's still being

1142
01:07:41,423 --> 01:07:44,587
used because it works. But
there's this part of it is,

1143
01:07:44,587 --> 01:07:48,102
there hasn't been a whole lot of
optimization on it. I mean,

1144
01:07:48,102 --> 01:07:51,676
there has been a little bit, but
not at the rate of something

1145
01:07:51,676 --> 01:07:54,840
like, the rest of the
computation industry. So I think

1146
01:07:54,840 --> 01:07:58,472
a thorough understanding of the
state of the art is, can, will

1147
01:07:58,472 --> 01:08:01,695
like, be hugely valuable for
anyone that wants to enter

1148
01:08:01,695 --> 01:08:05,445
olfaction because it'll save you
from spending a lot of money in

1149
01:08:05,445 --> 01:08:08,843
the wrong direction. And, I
guess ingesting a lot of noise

1150
01:08:08,843 --> 01:08:12,534
that really doesn't have a lot
of signal as far as, like what's

1151
01:08:12,534 --> 01:08:15,756
important for, moving the
industry forward and which AI

1152
01:08:15,756 --> 01:08:19,155
model should be used, etc.. I
think those are probably the

1153
01:08:19,155 --> 01:08:22,494
two. Those would be the two
enablers. I wish I would have

1154
01:08:22,494 --> 01:08:25,951
paid more attention to what has
been done in the past for a

1155
01:08:25,951 --> 01:08:29,701
faction, because I saw I started
solving a lot of or I assumed a

1156
01:08:29,701 --> 01:08:33,099
lot of problems had been solved
that hadn't. And I started

1157
01:08:33,099 --> 01:08:36,439
solving problems that were
already solved, which a lot of

1158
01:08:36,439 --> 01:08:40,071
which made me spin my wheels.
You know, a lot of the first few

1159
01:08:40,071 --> 01:08:43,763
years. And so I started reading
a lot more and stopped, stopped

1160
01:08:43,763 --> 01:08:47,161
building and just started
observing and listening for, you

1161
01:08:47,161 --> 01:08:50,676
know, the next probably year
after that, to just get an idea

1162
01:08:50,676 --> 01:08:54,367
of what has been happening and
what's been going on. And it was

1163
01:08:54,367 --> 01:08:57,824
I wish I would have done that
from Deloitte. Usually I open

1164
01:08:57,824 --> 01:08:58,000
it.

1165
01:08:58,000 --> 01:09:02,219
Audrow Nash: Yeah. But do you
have like, like I think that's

1166
01:09:02,219 --> 01:09:05,945
quite cool. From your
perspective, it seems, is that,

1167
01:09:05,945 --> 01:09:09,883
or from like your experience, it
seems like sharing your

1168
01:09:09,883 --> 01:09:14,383
knowledge on this kind of thing,
could be very valuable to guide

1169
01:09:14,383 --> 01:09:18,461
people down this, because I
think there's going to be some

1170
01:09:18,461 --> 01:09:23,031
people that listen to this, that
want to go start a scent related

1171
01:09:23,031 --> 01:09:27,039
startup now or in the near
future. And, the path as clear

1172
01:09:27,039 --> 01:09:31,398
as you can make it will benefit
them for this kind of thing. I

1173
01:09:31,398 --> 01:09:35,195
don't know if you guys are
intending to do like a blog

1174
01:09:35,195 --> 01:09:39,555
post, blogs, or any community
outreach kind of thing, but that

1175
01:09:39,555 --> 01:09:43,914
could be really cool to kind of
educate an army of people that

1176
01:09:43,914 --> 01:09:47,922
are going to go do this,
startups for this kind of thing.

1177
01:09:47,922 --> 01:09:52,000
That seems like a very valuable
thing from my perspective.

1178
01:09:52,000 --> 01:09:55,452
Kordel France: Yeah, I fully
agree. The so we will we have a

1179
01:09:55,452 --> 01:09:59,077
blog that's in development right
now. And so that was something

1180
01:09:59,077 --> 01:10:02,759
to look forward to. What we hope
is that, I mean, Scentience, we

1181
01:10:02,759 --> 01:10:06,499
really don't want to become like
an application based company. We

1182
01:10:06,499 --> 01:10:09,778
don't want to focus on one
industry in particular. If you

1183
01:10:09,778 --> 01:10:13,518
imagine, like Nvidia is the GPU,
they build GPUs. They don't care

1184
01:10:13,518 --> 01:10:16,970
what you use it on, what you
train it AI model, self-driving

1185
01:10:16,970 --> 01:10:20,077
car or video game industry. So
yeah, we want to be the

1186
01:10:20,077 --> 01:10:23,299
olfactory processing unit
company and just build the set

1187
01:10:23,299 --> 01:10:26,866
of APIs and the hardware that
enable all the developers moving

1188
01:10:26,866 --> 01:10:30,318
forward. Because if we try to
attack every application right

1189
01:10:30,318 --> 01:10:33,367
now wouldn't work, we'll dive
into indigestion versus

1190
01:10:33,367 --> 01:10:34,000
starvation.

1191
01:10:34,000 --> 01:10:36,000
Audrow Nash: So, yeah, it's a
good way to put it.

1192
01:10:36,000 --> 01:10:39,955
Kordel France: Yeah. So we don't
want we don't want to try to, to

1193
01:10:39,955 --> 01:10:43,301
go we want to try to go
horizontal, but if we can build

1194
01:10:43,301 --> 01:10:47,073
a very, very cutting edge and a
very accurate set of hardware,

1195
01:10:47,073 --> 01:10:50,541
set of sensors, or factory
processing units, and the next

1196
01:10:50,541 --> 01:10:54,252
set of APIs that are so easy to
use that developers just it's

1197
01:10:54,252 --> 01:10:57,963
just cake for them to be able to
integrate with other sensing

1198
01:10:57,963 --> 01:11:01,310
modalities. Then there are so
many startups that can be

1199
01:11:01,310 --> 01:11:04,960
started. Oh yeah. And totally
just based off of that. Right.

1200
01:11:04,960 --> 01:11:08,428
Because the key enabler right
now is like, if you want to

1201
01:11:08,428 --> 01:11:12,018
build, if you want to build an
olfaction startup, that, you

1202
01:11:12,018 --> 01:11:15,546
know, that does quality control
on food or agriculture or,

1203
01:11:15,546 --> 01:11:19,501
detects organic compounds on the
surface of Mars, you need better

1204
01:11:19,501 --> 01:11:23,212
sensors. And so it starts with
better sensors. And, yeah, the

1205
01:11:23,212 --> 01:11:26,985
folks and the people that use
sentence products to build those

1206
01:11:26,985 --> 01:11:30,574
applications, I feel like we'll
be extremely wealthy in the

1207
01:11:30,574 --> 01:11:31,000
future.

1208
01:11:31,000 --> 01:11:35,113
Audrow Nash: Probably true. Do
you? So one thing with you

1209
01:11:35,113 --> 01:11:39,659
talking about the standards, and
nailing the standard first and

1210
01:11:39,659 --> 01:11:44,061
related things, one thing that
it makes me think of is do you

1211
01:11:44,061 --> 01:11:48,679
ever read there was an essay on
Lisp, called worse is Better. Do

1212
01:11:48,679 --> 01:11:53,009
you know it? Yes, it makes me
think of that. And so the idea

1213
01:11:53,009 --> 01:11:57,627
of that was c b lisp because it
was easy to implement on lots of

1214
01:11:57,627 --> 01:12:01,956
different machines. And so it
wasn't necessary. Like Lisp is

1215
01:12:01,956 --> 01:12:06,358
far better and I like I love
programing in Lisp, but if there

1216
01:12:06,358 --> 01:12:10,760
was a good type list, I typed
list that had a good ecosystem,

1217
01:12:10,760 --> 01:12:15,306
I would be programing it all the
time. But C won out and became

1218
01:12:15,306 --> 01:12:19,419
like Linux and all sorts of
operating systems use it, and

1219
01:12:19,419 --> 01:12:23,604
all sorts of software was
written on it. Much, much bigger

1220
01:12:23,604 --> 01:12:27,934
than Lisp in my knowledge. I
wonder if there's like, a first

1221
01:12:27,934 --> 01:12:32,624
mover advantage for this kind of
thing. If it's a portable format

1222
01:12:32,624 --> 01:12:36,881
and you just have data already,
because it would be like if

1223
01:12:36,881 --> 01:12:41,211
someone interested in this area
can just bootstrap something

1224
01:12:41,211 --> 01:12:45,107
with what you've already
provided. Like if you look at

1225
01:12:45,107 --> 01:12:49,076
like open CV or a lot of
libraries that are like OpenCV

1226
01:12:49,076 --> 01:12:53,117
still gets a lot of community
use and their, software is

1227
01:12:53,117 --> 01:12:57,519
pretty painful to use. In my
experience. But it's just no one

1228
01:12:57,519 --> 01:13:01,343
else has taken the time to
implement these algorithms

1229
01:13:01,343 --> 01:13:04,879
because OpenCV already did. And
so it's not a big

1230
01:13:04,879 --> 01:13:09,209
differentiator. I know they're
like sci Pi doing some of the

1231
01:13:09,209 --> 01:13:13,682
stuff, but I wonder if, I wonder
if there's value in like or I

1232
01:13:13,682 --> 01:13:18,228
guess, what do you think of that
kind of a what do you think of

1233
01:13:18,228 --> 01:13:22,630
that kind of comment where it's
perhaps there's just value in

1234
01:13:22,630 --> 01:13:27,176
getting something out there to
everyone so that it gets adopted

1235
01:13:27,176 --> 01:13:31,289
and providing some data versus
nailing it down perfectly.

1236
01:13:31,289 --> 01:13:35,474
Because getting something out
there gets by, in and by and

1237
01:13:35,474 --> 01:13:38,000
makes it sticky. What do you
think?

1238
01:13:38,000 --> 01:13:41,595
Kordel France: I fully agree,
and that's the direction we've

1239
01:13:41,595 --> 01:13:45,249
taken with Scentience. Right? So
we've got those two products

1240
01:13:45,249 --> 01:13:48,724
that we pushed out. And yeah,
while they're in close sale,

1241
01:13:48,724 --> 01:13:52,498
they're still for sale. And so
we have customers that are using

1242
01:13:52,498 --> 01:13:56,093
them and we're trying to gather
as much data as possible, as

1243
01:13:56,093 --> 01:13:59,867
quickly as possible so that we
can, we can start to build out a

1244
01:13:59,867 --> 01:14:02,863
large library of chemical
compounds. And sure, the

1245
01:14:02,863 --> 01:14:06,398
products can be optimized even
further, but they might take

1246
01:14:06,398 --> 01:14:10,112
more resources, more money. And
if we can just gather feedback

1247
01:14:10,112 --> 01:14:13,467
to improve that product
development cycle, we can mature

1248
01:14:13,467 --> 01:14:17,122
the product much faster and so
love it. Yeah, we're trying to

1249
01:14:17,122 --> 01:14:21,016
make data driven decisions with,
I would say, I would say subpar,

1250
01:14:21,016 --> 01:14:24,490
products, but products that
probably could be optimized in

1251
01:14:24,490 --> 01:14:27,845
iterating. Yeah, yeah, yeah,
curating. There you go. And

1252
01:14:27,845 --> 01:14:31,560
with, with hopes that we can get
sensors out there that people

1253
01:14:31,560 --> 01:14:34,975
then can use to build out all
the other applications that

1254
01:14:34,975 --> 01:14:38,629
they're thinking of. But yeah,
I, I'm in full embrace of kind

1255
01:14:38,629 --> 01:14:42,284
of pushing things out before
they're perfect to try to get as

1256
01:14:42,284 --> 01:14:45,878
much data because I feel like a
if you if you wait till it's

1257
01:14:45,878 --> 01:14:49,533
perfect, you can get overcome by
competitors. Or you can that

1258
01:14:49,533 --> 01:14:53,068
last 10%, which is always the
hardest part, could be in the

1259
01:14:53,068 --> 01:14:56,602
wrong direction. And I'm a huge
believer of like talking to

1260
01:14:56,602 --> 01:15:00,437
customers and letting customers
drive, some of the decisions you

1261
01:15:00,437 --> 01:15:04,151
make, not all of them, but some
of them to inform, like what's

1262
01:15:04,151 --> 01:15:07,746
going to sell faster. And so
basically, if you get the right

1263
01:15:07,746 --> 01:15:11,460
customer, that is going to give
you feedback on how things can

1264
01:15:11,460 --> 01:15:14,935
be improved. Then you can use
that to build a product that

1265
01:15:14,935 --> 01:15:18,649
they love, that they will go,
you know, by and spread the word

1266
01:15:18,649 --> 01:15:22,304
on and become kind of your own
marketing team to some degree.

1267
01:15:22,304 --> 01:15:25,000
So I fully agree with your
thoughts. Oh yeah.

1268
01:15:25,000 --> 01:15:29,212
Audrow Nash: Okay. That's really
cool. I really like that you're

1269
01:15:29,212 --> 01:15:33,160
heading in that direction. Tell
me about, you mentioned data

1270
01:15:33,160 --> 01:15:37,175
privacy for customers that are
using your products to kind of

1271
01:15:37,175 --> 01:15:41,452
help you get data. Tell me a bit
about that whole pipeline of how

1272
01:15:41,452 --> 01:15:45,269
you are absorbing data from
other people using it. And, it

1273
01:15:45,269 --> 01:15:49,152
sounds very valuable. Once you
have the data. And actually,

1274
01:15:49,152 --> 01:15:53,298
it's it's also interesting of,
like, how would you label it for

1275
01:15:53,298 --> 01:15:57,180
everything to understand the
context of what it is, because

1276
01:15:57,180 --> 01:16:00,668
otherwise it's just kind of
arbitrary numbers in your

1277
01:16:00,668 --> 01:16:04,485
schema, of RGB like things for
smell. But tell me, tell me

1278
01:16:04,485 --> 01:16:08,170
about your whole data harvesting
from customers. Or data

1279
01:16:08,170 --> 01:16:11,000
harvesting from devices that are
out there.

1280
01:16:11,000 --> 01:16:14,909
Kordel France: I suppose so,
well, the there's the Scentience

1281
01:16:14,909 --> 01:16:18,817
app that connects to the devices
to stream data and, give you

1282
01:16:18,817 --> 01:16:22,854
more insight into what's going
on and what the device observes.

1283
01:16:22,854 --> 01:16:26,827
And on that app, there are two
models. There's the edge model,

1284
01:16:26,827 --> 01:16:30,608
which allows you to have no
internet connection. There's no

1285
01:16:30,608 --> 01:16:34,324
talk to the database. It's in
our agreement. So there's no

1286
01:16:34,324 --> 01:16:38,425
saving anything. Right. Like the
intent of that edge model is to

1287
01:16:38,425 --> 01:16:41,757
basically gather trusted
customers to say, we're not

1288
01:16:41,757 --> 01:16:45,474
harvesting your data. We're not,
we're not we're not doing

1289
01:16:45,474 --> 01:16:49,190
anything that, might be that
those edges. Right? Yeah. But

1290
01:16:49,190 --> 01:16:53,227
there is a there's a way to opt
in to have a more sophisticated

1291
01:16:53,227 --> 01:16:56,944
model, more sophisticated
olfaction vision language model.

1292
01:16:56,944 --> 01:17:01,045
Now, that model is larger. So it
lives on the cloud and it makes

1293
01:17:01,045 --> 01:17:05,017
requests. And that model does
relay data back to a database to

1294
01:17:05,017 --> 01:17:08,862
be able to try to predict and
detect, what's going on around

1295
01:17:08,862 --> 01:17:12,579
it. So we're trying to give
customers, ability to, to work

1296
01:17:12,579 --> 01:17:16,423
both ways. Now, the one that
lies on the, on the device gets

1297
01:17:16,423 --> 01:17:20,588
slightly personalized over time.
So we have to constrain it a bit

1298
01:17:20,588 --> 01:17:24,305
because as it learns, we don't
want it to become a runaway

1299
01:17:24,305 --> 01:17:28,085
train. But in order to ensure
the promise made to customers

1300
01:17:28,085 --> 01:17:31,994
that none of their data will
talk to our databases, we're not

1301
01:17:31,994 --> 01:17:36,031
saving anything from it. So, for
example, like if if the device

1302
01:17:36,031 --> 01:17:39,812
does not detect a whole lot of
benzene and for several days

1303
01:17:39,812 --> 01:17:43,656
there's no benzene detected,
maybe we can adjust the model a

1304
01:17:43,656 --> 01:17:47,757
bit to like, maybe put more, or
adjust the weights to focus more

1305
01:17:47,757 --> 01:17:50,000
on other compounds besides
benzene.

1306
01:17:50,000 --> 01:17:52,878
Audrow Nash: So prunes itself in
a sense. Right. Like these

1307
01:17:52,878 --> 01:17:55,756
aren't that important. We won't
worry about that that much.

1308
01:17:55,756 --> 01:17:56,000
Okay.

1309
01:17:56,000 --> 01:17:59,562
Kordel France: They're they're
still recognizable, but their

1310
01:17:59,562 --> 01:18:03,065
sensitivity goes down because
the model now kind of adjusts

1311
01:18:03,065 --> 01:18:06,686
self to not be true to that.
However, with the one that lives

1312
01:18:06,686 --> 01:18:10,367
in the cloud, that is taking a
bunch of data and it's learning

1313
01:18:10,367 --> 01:18:13,870
from everyone's so everyone's
doing it to try to make this,

1314
01:18:13,870 --> 01:18:17,254
universal model that can, that
can that can, detect every

1315
01:18:17,254 --> 01:18:20,638
compound in every condition. So,
yeah, there's, there are

1316
01:18:20,638 --> 01:18:24,378
different ways of handling it.
We'll see which one ends up kind

1317
01:18:24,378 --> 01:18:27,703
of, varying faster. Or I guess
gathers trusted customers

1318
01:18:27,703 --> 01:18:31,265
faster. Maybe people don't care,
but, we're saving olfaction

1319
01:18:31,265 --> 01:18:34,709
data. It's a new standard. I
would think that maybe, maybe

1320
01:18:34,709 --> 01:18:38,449
you don't care, but perhaps at
the end of the day, if you do, I

1321
01:18:38,449 --> 01:18:42,130
mean, dogs authenticate humans
by scent, right? So, yeah, they

1322
01:18:42,130 --> 01:18:45,989
they first interpret that you're
their owner by smelling you. And

1323
01:18:45,989 --> 01:18:49,195
so. Okay. Well, that in the
future may become personal

1324
01:18:49,195 --> 01:18:52,935
identity, personal identifiable
information. Right. And so it's

1325
01:18:52,935 --> 01:18:56,616
like is that something we'll run
into? The future is like your

1326
01:18:56,616 --> 01:19:00,000
olfactory signature is now PII.
Maybe. I don't know, but,

1327
01:19:00,000 --> 01:19:02,000
Audrow Nash: We'll cross that
bridge later.

1328
01:19:02,000 --> 01:19:04,716
Kordel France: Yeah. We're
trying to put the necessary,

1329
01:19:04,716 --> 01:19:07,679
guardrails in place to to be
ahead of that. If in, you know,

1330
01:19:07,679 --> 01:19:10,593
the future does so people look
back and say, wow, like, oh,

1331
01:19:10,593 --> 01:19:13,605
they were thinking about from
the very beginning, even though

1332
01:19:13,605 --> 01:19:16,667
it's not a law yet or it's not a
standard like Scentience was.

1333
01:19:16,667 --> 01:19:18,000
Thinking about our privacy.

1334
01:19:18,000 --> 01:19:22,318
Audrow Nash: I like that. Hell
yeah. Okay. Do you so how do you

1335
01:19:22,318 --> 01:19:26,498
see your gathering, the data?
How do you label it? How do you

1336
01:19:26,498 --> 01:19:30,953
get how do you make sense out of
it? Like, it strikes me that you

1337
01:19:30,953 --> 01:19:35,065
probably need a good bit of
context about what's going on to

1338
01:19:35,065 --> 01:19:39,246
make it useful. Or do they tell
you how how does that work? I

1339
01:19:39,246 --> 01:19:40,000
suppose so.

1340
01:19:40,000 --> 01:19:43,239
Kordel France: With the vision
capabilities of the app. If

1341
01:19:43,239 --> 01:19:46,702
you're if there's no camera tie
to the API. So like the device

1342
01:19:46,702 --> 01:19:50,164
here interacts with, you know,
different audio and vision APIs

1343
01:19:50,164 --> 01:19:53,571
for different cameras and and
microphones, if there's no data

1344
01:19:53,571 --> 01:19:57,033
attached to that, then the app,
you have the option to turn on

1345
01:19:57,033 --> 01:20:00,440
the camera, right? So and audio.
So it, we're contextualizing

1346
01:20:00,440 --> 01:20:04,014
based off of the scene that the
camera sees and the audio around

1347
01:20:04,014 --> 01:20:07,476
it. Now it kind of makes more
sense why that data might not be

1348
01:20:07,476 --> 01:20:10,939
saved. Right? We don't want to
film people's homes and and say

1349
01:20:10,939 --> 01:20:12,000
that data wherever.

1350
01:20:12,000 --> 01:20:14,000
Audrow Nash: They're using it.
Yeah.

1351
01:20:14,000 --> 01:20:17,542
Kordel France: Yeah. So it's
contextualized based on what it

1352
01:20:17,542 --> 01:20:21,203
sees in the image and tries to
relate. Okay. Well, I see this.

1353
01:20:21,203 --> 01:20:24,863
I compounds how do I map them
back to the objects in the image

1354
01:20:24,863 --> 01:20:28,583
with high enough confidence. And
then those mappings are saved.

1355
01:20:28,583 --> 01:20:32,125
If it's if it's not the edge
model, those mappings are saved

1356
01:20:32,125 --> 01:20:35,727
to the database where we can
then analyze and say, okay, like

1357
01:20:35,727 --> 01:20:39,269
we was this, was this correct?
Like, did we predict that the

1358
01:20:39,269 --> 01:20:42,989
alcohol like, scent was coming
from, the series of fruit across

1359
01:20:42,989 --> 01:20:46,590
the room or the carbon monoxide
was in probably in an outside

1360
01:20:46,590 --> 01:20:50,428
environment where there's, a lot
of vehicle traffic. Right? Like,

1361
01:20:50,428 --> 01:20:53,203
does it make sense?
Contextualized based on the

1362
01:20:53,203 --> 01:20:56,982
image in the text that was sent
with the with the model. So that

1363
01:20:56,982 --> 01:21:00,701
process would be fine tuned over
time? The way we build that is

1364
01:21:00,701 --> 01:21:04,421
we try to make it as intelligent
as possible, and then we prune

1365
01:21:04,421 --> 01:21:08,022
it down, that becomes an edge
model. And then, now that model

1366
01:21:08,022 --> 01:21:11,269
lives on the device, it's
intelligent enough. Yeah. And

1367
01:21:11,269 --> 01:21:14,694
that a cycle recurs. Right. And
we build another big model

1368
01:21:14,694 --> 01:21:18,000
primitive, prune it down, and it
just updates over time.

1369
01:21:18,000 --> 01:21:21,412
Audrow Nash: Yeah. Hell, yeah.
Do you have companies or the

1370
01:21:21,412 --> 01:21:25,170
customers that you have already?
Are they startups that are using

1371
01:21:25,170 --> 01:21:28,524
this or like are or are they
individuals or companies that

1372
01:21:28,524 --> 01:21:31,935
are trying to optimize some
process? I'm just imagining how

1373
01:21:31,935 --> 01:21:35,463
valuable it would be to just
have like your sensor on a stick

1374
01:21:35,463 --> 01:21:38,932
with a camera and someone goes
and bops all the fruit in the

1375
01:21:38,932 --> 01:21:42,575
grocery store kind of thing, or
hold it over the fruit for like

1376
01:21:42,575 --> 01:21:45,813
two seconds, five seconds or
whatever. I don't know that

1377
01:21:45,813 --> 01:21:49,224
based on the refresh rate, but
like, I don't know what what

1378
01:21:49,224 --> 01:21:52,000
kinds of uses are they doing
with it? I suppose.

1379
01:21:52,000 --> 01:21:55,358
Kordel France: So for the the
smaller device, there are a lot

1380
01:21:55,358 --> 01:21:58,661
of agricultural companies, most
of the agricultural that are

1381
01:21:58,661 --> 01:22:01,468
using it with some, but
surprising, but makes total

1382
01:22:01,468 --> 01:22:04,000
sense now that I see a lot of
their use cases.

1383
01:22:04,000 --> 01:22:06,000
Audrow Nash: So one I would have
guessed to. Yeah.

1384
01:22:06,000 --> 01:22:09,548
Kordel France: Yeah. And then
for the coffee maker, size one,

1385
01:22:09,548 --> 01:22:12,572
research institutions are
expected to be the biggest

1386
01:22:12,572 --> 01:22:16,004
customers because, so far with
the conversations we've had,

1387
01:22:16,004 --> 01:22:19,726
they want to be able to profile
the air for various experiments.

1388
01:22:19,726 --> 01:22:23,158
Right. And while that's not
necessarily that application is

1389
01:22:23,158 --> 01:22:26,647
not necessarily helpful for
science itself, the data we get,

1390
01:22:26,647 --> 01:22:30,021
I mean, research institutions
have chemical like all these

1391
01:22:30,021 --> 01:22:33,627
different chemical compounds,
which you have licenses for. And

1392
01:22:33,627 --> 01:22:37,000
it can be very difficult to get
unless you have, you know.

1393
01:22:37,000 --> 01:22:39,000
Audrow Nash: Super clearances of
some sort.

1394
01:22:39,000 --> 01:22:42,352
Kordel France: Exactly. So, we
get a bunch of data from them

1395
01:22:42,352 --> 01:22:45,480
with, with, with agreement that
allows us to improve our

1396
01:22:45,480 --> 01:22:48,832
olfaction models over time. So
we would never be able to get

1397
01:22:48,832 --> 01:22:51,849
like, like, for example,
fentanyl or explosive, right.

1398
01:22:51,849 --> 01:22:55,257
Like, how are we going to get
that anywhere else other than a

1399
01:22:55,257 --> 01:22:58,665
research institution, unless you
have really great friends, I

1400
01:22:58,665 --> 01:22:59,000
guess.

1401
01:22:59,000 --> 01:23:02,000
Audrow Nash: But, what a thing.

1402
01:23:02,000 --> 01:23:05,246
Kordel France: Yeah. So that
that helps us. I mean, if we

1403
01:23:05,246 --> 01:23:08,719
want to be able to detect, you
know, be like border security,

1404
01:23:08,719 --> 01:23:12,021
drug detection, we've got to
have that target somehow. And

1405
01:23:12,021 --> 01:23:15,609
so, yeah, being able to require
that stimulus from the research

1406
01:23:15,609 --> 01:23:18,000
institutions is going to be
super helpful.

1407
01:23:18,000 --> 01:23:21,499
Audrow Nash: That is super cool,
Yeah. What a neat thing. I'm

1408
01:23:21,499 --> 01:23:24,999
just I really like the idea of
just having a stick that has a

1409
01:23:24,999 --> 01:23:28,556
camera on it like it has, and
you just, like, go in, you like,

1410
01:23:28,556 --> 01:23:32,227
put it up on everything and you
see what it says. And then after

1411
01:23:32,227 --> 01:23:35,784
a bit, it learns what the thing
is. So you can, like, identify

1412
01:23:35,784 --> 01:23:39,169
the same object or something
like that. Like something like

1413
01:23:39,169 --> 01:23:42,496
that would be so, so cool
because one of the needed one of

1414
01:23:42,496 --> 01:23:45,651
the things that was most
enjoyable about working with a

1415
01:23:45,651 --> 01:23:48,979
thermal camera at some point was
I had one of those little

1416
01:23:48,979 --> 01:23:52,535
handheld things with a camera on
top. So just hold it, and I'd

1417
01:23:52,535 --> 01:23:56,035
walk around looking through it,
and having some sort of thing

1418
01:23:56,035 --> 01:23:58,961
like that for scent would be
amazing. Like, I would

1419
01:23:58,961 --> 01:24:02,345
absolutely love that. If you
make one, please. I would love

1420
01:24:02,345 --> 01:24:05,845
it. Yeah, you can. I'll spend a
bunch of time walking around,

1421
01:24:05,845 --> 01:24:09,000
bop and stuff with it and trying
to see what it thinks.

1422
01:24:09,000 --> 01:24:11,325
Kordel France: If you put I
guess if you put your your

1423
01:24:11,325 --> 01:24:13,823
there's a sentence app, right?
If you put your iPhone on a

1424
01:24:13,823 --> 01:24:16,364
selfie stick and you have the
senses close, I guess you can

1425
01:24:16,364 --> 01:24:18,000
kind of like replicate the same
thing.

1426
01:24:18,000 --> 01:24:20,000
Audrow Nash: Yeah, yeah. So
cool.

1427
01:24:20,000 --> 01:24:22,328
Kordel France: And if you
imagine, like from a robot

1428
01:24:22,328 --> 01:24:25,104
perspective, like these humanoid
robots, right. If they, while

1429
01:24:25,104 --> 01:24:27,791
their eyes aren't on a stick,
the cameras are on their head.

1430
01:24:27,791 --> 01:24:30,657
And as they walk around and they
have a sense of smell, they can

1431
01:24:30,657 --> 01:24:32,000
start to learn, like, pick it.

1432
01:24:32,000 --> 01:24:33,000
Audrow Nash: Up and.

1433
01:24:33,000 --> 01:24:35,650
Kordel France: Yeah,
associations between objects and

1434
01:24:35,650 --> 01:24:38,300
chemicals and everything. And,
yeah, there's a lot of

1435
01:24:38,300 --> 01:24:40,000
opportunity they're excited
about.

1436
01:24:40,000 --> 01:24:43,929
Audrow Nash: Yeah. Yeah. It
seems like a super exciting

1437
01:24:43,929 --> 01:24:48,071
space. What do you so with
looking out in the future, what

1438
01:24:48,071 --> 01:24:52,286
do you expect the timeline of
things to be? We said kind of

1439
01:24:52,286 --> 01:24:56,643
it'll be bigger in five years
and it'll asymptote at reaching

1440
01:24:56,643 --> 01:25:00,786
everyone. Basically in ten
years. What do you predict like

1441
01:25:00,786 --> 01:25:04,857
years one, two, three, four,
five? Like, how do you think

1442
01:25:04,857 --> 01:25:08,929
everything will go? What do you
think for Scentience just

1443
01:25:08,929 --> 01:25:12,000
looking out in the future. What
do you see?

1444
01:25:12,000 --> 01:25:15,668
Kordel France: I think we're
going to have equal approach or

1445
01:25:15,668 --> 01:25:19,397
converge on equal adoption for
olfactory sensors and a factor

1446
01:25:19,397 --> 01:25:22,698
data, in a similar manner that
we have for cameras and

1447
01:25:22,698 --> 01:25:26,182
microphones, etc., and audio
data. So like in the future,

1448
01:25:26,182 --> 01:25:29,300
instead of seeing like ChatGPT
with vision language

1449
01:25:29,300 --> 01:25:32,906
capabilities, there will be
factory capabilities that allow

1450
01:25:32,906 --> 01:25:36,880
us how cool some sort of sensor,
right? My ultimate goal and what

1451
01:25:36,880 --> 01:25:40,670
I think will happen is, well,
those sensors will be integrated

1452
01:25:40,670 --> 01:25:44,215
into like wearables and our
smartphones. Right? So like we

1453
01:25:44,215 --> 01:25:47,639
have an audio sensor for a
microphone and there are four

1454
01:25:47,639 --> 01:25:51,367
cameras on every smartphone. Now
we'll have olfaction sensors

1455
01:25:51,367 --> 01:25:54,913
that provide real time air
quality monitoring on the phone

1456
01:25:54,913 --> 01:25:58,520
itself. So cool air within
wearables. The interesting thing

1457
01:25:58,520 --> 01:26:02,249
with wearables is so there's
there's air itself, there's your

1458
01:26:02,249 --> 01:26:05,427
breath, but with the same
technology you can look at

1459
01:26:05,427 --> 01:26:09,217
sweat, and all these other data
points on your body. Right. So

1460
01:26:09,217 --> 01:26:13,068
chemicals that are emitted from
your skin or the chemicals that

1461
01:26:13,068 --> 01:26:16,858
are made from your sweat, your
saliva, there are I feel like a

1462
01:26:16,858 --> 01:26:20,710
lot of the, like, smartwatches
and everything in the future are

1463
01:26:20,710 --> 01:26:24,011
going to have chemical sensors
that give a much better

1464
01:26:24,011 --> 01:26:27,801
indication about what's going on
in your body. And so I really

1465
01:26:27,801 --> 01:26:31,407
hope with breath that someone
makes it through the clinical

1466
01:26:31,407 --> 01:26:34,892
trial process and, those, a lot
of breathalyzers that can

1467
01:26:34,892 --> 01:26:38,804
analyze your, your, the data in
your breath in real time because

1468
01:26:38,804 --> 01:26:41,799
there are so, so many
applications with that. But

1469
01:26:41,799 --> 01:26:44,000
clinical trials take a lot of
money.

1470
01:26:44,000 --> 01:26:48,836
Audrow Nash: Yeah. And medical
device companies take a long

1471
01:26:48,836 --> 01:26:53,672
time. Yeah. To it sounds so
exciting. What do you hope that

1472
01:26:53,672 --> 01:26:57,934
our listeners and watchers take
away from this whole

1473
01:26:57,934 --> 01:26:59,000
conversation?

1474
01:26:59,000 --> 01:27:03,280
Kordel France: I hope there's a
lot more awareness about all the

1475
01:27:03,280 --> 01:27:07,359
applications that olfaction can
be attributed to, and all the

1476
01:27:07,359 --> 01:27:11,439
possibilities that we can do by
utilizing the data in the air

1477
01:27:11,439 --> 01:27:15,652
around us. And if there are, is
there a there's a motivation to

1478
01:27:15,652 --> 01:27:19,263
move forward? I hope that, that
they'll consider using

1479
01:27:19,263 --> 01:27:23,409
Scentience, data and sensors to
springboard their applications

1480
01:27:23,409 --> 01:27:27,488
into whatever industry that they
might decide or whatever use

1481
01:27:27,488 --> 01:27:31,568
case they might decide to use.
And I hope that there's just a

1482
01:27:31,568 --> 01:27:35,246
much bigger embrace for,
olfaction AI moving forward so

1483
01:27:35,246 --> 01:27:38,924
that we can really build it on
par with the rest of the

1484
01:27:38,924 --> 01:27:42,000
modalities and give robots the
sense of smell.

1485
01:27:42,000 --> 01:27:46,557
Audrow Nash: That's awesome.
Let's see. And then are there

1486
01:27:46,557 --> 01:27:50,800
any links or any other
information you'd like to share

1487
01:27:50,800 --> 01:27:53,000
with listeners and watchers.

1488
01:27:53,000 --> 01:27:57,155
Kordel France: Yes. So the app,
where is going to be available

1489
01:27:57,155 --> 01:28:01,444
on the App Store next week. The
our website training Scentience.

1490
01:28:01,444 --> 01:28:05,667
I Scentience with SCA like the
sense of, scent olfaction scent.

1491
01:28:05,667 --> 01:28:09,889
There's not a whole lot of the
website right now because we are

1492
01:28:09,889 --> 01:28:14,178
still under stealth, by argument
from our investors. But, within

1493
01:28:14,178 --> 01:28:18,400
the next few months, you'll see
a lot more traffic. And there's

1494
01:28:18,400 --> 01:28:22,287
an option to sign up for our
newsletter online as well, so

1495
01:28:22,287 --> 01:28:25,705
that you can get, a lot more
insight into olfactory

1496
01:28:25,705 --> 01:28:29,056
applications and, just
everything you can do with,

1497
01:28:29,056 --> 01:28:31,000
Scentience olfaction sensors.

1498
01:28:31,000 --> 01:28:34,440
Audrow Nash: Oh, yeah. Awesome.
Well, this has been super

1499
01:28:34,440 --> 01:28:38,060
interesting. This is something I
don't consider often, and I

1500
01:28:38,060 --> 01:28:41,319
think there is a lot of
potential around this. I can't

1501
01:28:41,319 --> 01:28:45,000
wait to see the giant wave of
startups that come out of this.

1502
01:28:45,000 --> 01:28:48,735
Kordel France: Likewise. Yeah.
Thank you so much for your time.

1503
01:28:48,735 --> 01:28:52,352
And thank you to your audience.
A big fan of the podcast and,

1504
01:28:52,352 --> 01:28:55,968
big fan of, all the guests you
agree on. So, thanks for being

1505
01:28:55,968 --> 01:28:59,170
the, thanks for being an
inadvertent teacher to me for

1506
01:28:59,170 --> 01:29:00,000
every episode.

1507
01:29:00,000 --> 01:29:03,829
Audrow Nash: Oh, yeah. Thank you
very much. All right. Bye,

1508
01:29:03,829 --> 01:29:07,723
everyone. You made it. What an
interesting idea to use smell

1509
01:29:07,723 --> 01:29:11,681
for a whole number of things.
What killer applications do you

1510
01:29:11,681 --> 01:29:15,186
think smell could be used for?
I'd love to know in the

1511
01:29:15,186 --> 01:29:18,950
comments. I'm going to be
thinking about this for a while.

1512
01:29:18,950 --> 01:29:22,000
Anyways, that's all for now. See
you next time.
