1
00:00:00,000 --> 00:00:10,880
Hey there, welcome to Data Democracy, a podcast where we explore ways to make data and AI

2
00:00:10,880 --> 00:00:12,880
more accessible to everyone.

3
00:00:12,880 --> 00:00:18,000
We do this by interviewing experts across industries, asking them how they think about

4
00:00:18,000 --> 00:00:22,320
data, what are some of the challenges they face when it comes to data, and if they had

5
00:00:22,320 --> 00:00:27,120
a magic wand and time and resources were not constraints, what kind of intelligence and

6
00:00:27,120 --> 00:00:29,560
models would they wish to have.

7
00:00:29,560 --> 00:00:33,320
We're honored to have Karina Namanis with us today.

8
00:00:33,320 --> 00:00:37,960
Karina is an entrepreneur and a thought leader in the world of AI.

9
00:00:37,960 --> 00:00:44,400
She is the CEO of Ukraine and US based computer vision company and a data annotation company

10
00:00:44,400 --> 00:00:46,040
label your data.

11
00:00:46,040 --> 00:00:52,080
Karina and her team are pioneering innovative solutions across industries from enhancing

12
00:00:52,080 --> 00:00:58,000
agricultural practices through computer vision to improving national defense capabilities

13
00:00:58,000 --> 00:01:00,000
with AI-driven projects.

14
00:01:00,000 --> 00:01:04,800
I've been following Karina's work in the field of AI and computer vision and I'm

15
00:01:04,800 --> 00:01:08,000
both inspired and fascinated by her journey.

16
00:01:08,000 --> 00:01:13,400
I'm super excited to learn more from her philosophy on leadership and her vision for

17
00:01:13,400 --> 00:01:15,000
the future of AI.

18
00:01:15,000 --> 00:01:17,000
Karina, welcome.

19
00:01:17,000 --> 00:01:21,640
Yes, thank you very much for such an extensive introduction.

20
00:01:21,640 --> 00:01:23,760
I'm super happy to be here.

21
00:01:23,760 --> 00:01:26,520
Glad we finally made it.

22
00:01:26,520 --> 00:01:34,440
I'm super excited to share my thoughts and always offer a great chat about things that

23
00:01:34,440 --> 00:01:36,440
I'm passionate about.

24
00:01:36,440 --> 00:01:37,440
That's awesome.

25
00:01:37,440 --> 00:01:45,560
We're so glad that we were able to meet and chat about all things data and AI.

26
00:01:45,560 --> 00:01:48,000
Let's start with our journey.

27
00:01:48,000 --> 00:01:49,200
Let's start with your journey.

28
00:01:49,200 --> 00:01:50,200
I'm sorry.

29
00:01:50,200 --> 00:01:54,480
How did you get where you are?

30
00:01:54,480 --> 00:01:56,040
Thank you for the question.

31
00:01:56,040 --> 00:02:03,760
First of all, my career started almost eight years ago in a company called Support Your

32
00:02:03,760 --> 00:02:10,280
App, which now is another company of our company, Label Your Data.

33
00:02:10,280 --> 00:02:12,720
I was only 18 at that time.

34
00:02:12,720 --> 00:02:13,720
That's an easy math.

35
00:02:13,720 --> 00:02:16,240
I'm 25 right now.

36
00:02:16,240 --> 00:02:21,200
I've worked with a customer support consultant at Support Your App for two years, which helped

37
00:02:21,200 --> 00:02:27,640
me to learn how to get along with different people and different levels of different problems.

38
00:02:27,640 --> 00:02:31,080
Then I worked as an onboarding specialist for two years.

39
00:02:31,080 --> 00:02:38,040
Then for the past four years of my career, I've been building Label Your Data.

40
00:02:38,040 --> 00:02:45,240
Label Your Data started off an amazing spin-off idea of R2 co-founders Daria Leschenko and

41
00:02:45,240 --> 00:02:46,240
Peter Bondariewski.

42
00:02:46,240 --> 00:02:53,120
To be honest, I never thought I'd end up in a place like this at this time.

43
00:02:53,120 --> 00:03:02,000
My background is in journalism, and I never thought I'd be the CEO of a tech company ever.

44
00:03:02,000 --> 00:03:07,840
But when I'm usually thinking about this question, I think I got to where I am right

45
00:03:07,840 --> 00:03:11,160
now thanks to two main factors.

46
00:03:11,160 --> 00:03:18,480
The first one is that I was lucky enough to have the right people around me.

47
00:03:18,480 --> 00:03:25,520
I was lucky to have Daria Leschenko as my mentor for the past six years.

48
00:03:25,520 --> 00:03:31,320
She was the one who taught me everything I know about building and managing a company.

49
00:03:31,320 --> 00:03:40,160
But she was also the one who saw the potential in me in my early 20s and was always giving

50
00:03:40,160 --> 00:03:44,800
me the heads up that I can do it.

51
00:03:44,800 --> 00:03:49,720
The second factor I think is my life motto, as I call it.

52
00:03:49,720 --> 00:03:57,800
It stands like this, don't overthink and grab the opportunities presented to you by life.

53
00:03:57,800 --> 00:04:07,360
I was, as I mentioned, 21 when I was offered to join a team of four young and passionate,

54
00:04:07,360 --> 00:04:12,440
amazing professionals and to start building a company.

55
00:04:12,440 --> 00:04:18,400
But then, after some time, I was the only one who made it till the end.

56
00:04:18,400 --> 00:04:24,480
For me, it was definitely a scary experience, full of self-doubt at times.

57
00:04:24,480 --> 00:04:29,960
But nonetheless, it's been super exciting.

58
00:04:29,960 --> 00:04:36,640
Right now, I strongly believe in what we do in my team, in our mission.

59
00:04:36,640 --> 00:04:40,400
This is what gives me the boldness to move on.

60
00:04:40,400 --> 00:04:41,400
That's amazing.

61
00:04:41,400 --> 00:04:44,160
I think that's a short recap of the story.

62
00:04:44,160 --> 00:04:46,160
No, that's amazing.

63
00:04:46,160 --> 00:04:53,000
Yeah, it's a fascinating story and I'm excited to learn more about it.

64
00:04:53,000 --> 00:05:01,160
I love what you're doing at Label Your Data with data annotation and AI.

65
00:05:01,160 --> 00:05:08,800
As AI picks up steam, data annotation becomes more important and will likely be widely adopted.

66
00:05:08,800 --> 00:05:13,640
Can you please tell us more about what data annotation is and what you do at Label Your

67
00:05:13,640 --> 00:05:14,800
Data?

68
00:05:14,800 --> 00:05:18,240
Label Your Data, we'll label your data.

69
00:05:18,240 --> 00:05:26,520
It's a data annotation company and we help AI enthusiasts, evangelists, designers, data

70
00:05:26,520 --> 00:05:32,600
scientists and machine learning engineers around the globe to focus on AI development

71
00:05:32,600 --> 00:05:39,200
while we take care of what's often considered not a sexy part of AI, which is humans in

72
00:05:39,200 --> 00:05:47,760
the loop training, but it is essential for nearly any AI model training.

73
00:05:47,760 --> 00:05:52,680
We work with applications in computer vision and natural language processing.

74
00:05:52,680 --> 00:06:00,200
Over the four years of our operation, we have worked with more than 100 projects around

75
00:06:00,200 --> 00:06:05,240
the globe and our current team is nearly 1,000 people.

76
00:06:05,240 --> 00:06:10,240
But apart from services, we also built products.

77
00:06:10,240 --> 00:06:18,120
One of them, which was launched late 2023, is a destination platform where anyone from

78
00:06:18,120 --> 00:06:24,000
any background of any age, well as long as you're about 18, can register on a platform

79
00:06:24,000 --> 00:06:29,000
and start earning money by training AI by doing simple and interesting data annotation

80
00:06:29,000 --> 00:06:30,000
tasks.

81
00:06:30,000 --> 00:06:36,360
It's super automated, easy to use and I strongly believe in our social mission there too because

82
00:06:36,360 --> 00:06:42,160
it provides so much needed workplaces for people who for some reason cannot work full

83
00:06:42,160 --> 00:06:47,800
time or who just want to earn extra cash for themselves or their family.

84
00:06:47,800 --> 00:06:51,280
And there is another product that we're working on, but it's coming soon, don't want to give

85
00:06:51,280 --> 00:06:56,800
you any spoilers, it's going to be out in a few months and you'll see it.

86
00:06:56,800 --> 00:07:03,560
And then if we talk about data annotation, in simple words, it's the process of using

87
00:07:03,560 --> 00:07:10,640
unstructured data and transforming it into a dataset which then can be used to train

88
00:07:10,640 --> 00:07:12,120
machine learning models.

89
00:07:12,120 --> 00:07:17,400
So for example, the basic example that I usually give is that think of a self-driving

90
00:07:17,400 --> 00:07:20,640
car, imagine you want to build another Tesla.

91
00:07:20,640 --> 00:07:29,640
So you want to teach it to distinguish between crossroads and just road marks or a squirrel

92
00:07:29,640 --> 00:07:32,560
passing by the car or a pedestrian.

93
00:07:32,560 --> 00:07:39,320
And so in order to do that, you need a dataset, you need to show the machine examples of different

94
00:07:39,320 --> 00:07:44,040
situations where a squirrel in the pedestrian can cross the road.

95
00:07:44,040 --> 00:07:52,720
But to machine the computer, the images are just a bunch of colored pixels that don't

96
00:07:52,720 --> 00:07:55,760
really make much sense as they do to us.

97
00:07:55,760 --> 00:08:04,400
So what we do, we build managed teams to go through those images and identify the objects

98
00:08:04,400 --> 00:08:07,600
on them and assign them specific labels.

99
00:08:07,600 --> 00:08:12,320
So say that, hey, this object right here is a pedestrian, this object right there is a

100
00:08:12,320 --> 00:08:18,680
squirrel and then by feeding your model with so many examples of these, it starts to pick

101
00:08:18,680 --> 00:08:22,120
up, you know, what are the similarities.

102
00:08:22,120 --> 00:08:28,000
And then in the end, it was able to identify the pedestrian from a squirrel on its own.

103
00:08:28,000 --> 00:08:30,920
But I'd say that this is the easiest example.

104
00:08:30,920 --> 00:08:38,320
And since, as you said, AI is a buzzword, it's all over everywhere right now, everybody

105
00:08:38,320 --> 00:08:44,040
wants, everybody's chasing AI applications in their work.

106
00:08:44,040 --> 00:08:51,080
We are seeing fewer and fewer of these easy cases.

107
00:08:51,080 --> 00:08:58,960
But we are seeing more and more projects with niche AI requests.

108
00:08:58,960 --> 00:09:05,280
I think we're entering the era of niche AI products where, for example, LLMs are adapted

109
00:09:05,280 --> 00:09:14,800
to oil and gas industry, or when you use radar and AI to navigate and explore the surface

110
00:09:14,800 --> 00:09:16,800
of the ocean.

111
00:09:16,800 --> 00:09:23,920
So yeah, that's in simple words, what data notation is, applications, I can go on and

112
00:09:23,920 --> 00:09:28,000
on about all different kind of cases we have.

113
00:09:28,000 --> 00:09:31,560
But as I said, it's literally everywhere right now.

114
00:09:31,560 --> 00:09:32,560
That's awesome.

115
00:09:32,560 --> 00:09:33,560
Yeah, absolutely.

116
00:09:33,560 --> 00:09:39,800
And then the gen generative AI, I think there's been so much talk about AI and machine learning

117
00:09:39,800 --> 00:09:45,520
that, you know, it's the place to be and the time to be in the AI space.

118
00:09:45,520 --> 00:09:51,880
You're at the right time in the right place.

119
00:09:51,880 --> 00:09:59,360
Can you talk more about, I know you did talk about a few of your projects and examples.

120
00:09:59,360 --> 00:10:06,600
What are some of the real life projects that you're working on now at Label Your Data?

121
00:10:06,600 --> 00:10:13,520
Yeah, well, we work with all kinds of projects where you need to use reinforcement learning

122
00:10:13,520 --> 00:10:17,680
from human feedback, both computer vision and an OP.

123
00:10:17,680 --> 00:10:22,280
So not only computer vision, we work with a lot with text and audio data as well.

124
00:10:22,280 --> 00:10:28,880
And so far, we've been delivering projects in such industries as robotics, agriculture,

125
00:10:28,880 --> 00:10:36,280
manufacturing, automotive, environmental protection, advertisement, entertainment, hyper

126
00:10:36,280 --> 00:10:41,920
automation, civil engineering, fintech, the least goes on and on.

127
00:10:41,920 --> 00:10:48,240
But there are two I'd say most fascinating industries that I probably would want to talk

128
00:10:48,240 --> 00:10:49,240
about.

129
00:10:49,240 --> 00:10:51,600
Number one is universities.

130
00:10:51,600 --> 00:10:54,200
We work a lot with different universities.

131
00:10:54,200 --> 00:10:59,960
We have collaborations with Yale, Carnegie Mellon, Miami universities, universities from

132
00:10:59,960 --> 00:11:01,760
the Middle East.

133
00:11:01,760 --> 00:11:06,720
And because we get the opportunity to work with young researchers on their PhD projects

134
00:11:06,720 --> 00:11:16,360
or their experiments and researches, the variety of these applications where they want to use

135
00:11:16,360 --> 00:11:19,680
AI never seems to surprise me.

136
00:11:19,680 --> 00:11:24,320
So we have, for example, been working on a landfill detection using satellite imagery

137
00:11:24,320 --> 00:11:27,800
with one university.

138
00:11:27,800 --> 00:11:34,440
Well with another one, we've been annotating the drone imagery of ships in Australia, because

139
00:11:34,440 --> 00:11:39,120
apparently the student wanted to develop an autonomous drone shepherd.

140
00:11:39,120 --> 00:11:42,320
So it was a pretty fun project to work with.

141
00:11:42,320 --> 00:11:49,200
So I think that AI is disrupting the industries in general.

142
00:11:49,200 --> 00:11:55,760
But another thing that I also wanted to mention is that so far we have a pretty extensive

143
00:11:55,760 --> 00:12:01,080
experience with miltec industry, which is also booming right now.

144
00:12:01,080 --> 00:12:06,280
So label you that comes originally from Ukraine.

145
00:12:06,280 --> 00:12:14,600
And since the full scale invasion of Russia, which happened more than three years ago,

146
00:12:14,600 --> 00:12:22,600
we have been cooperating closely with the Defense Ministry of Ukraine on multiple projects.

147
00:12:22,600 --> 00:12:32,360
So from annotating drone imagery for the enemy detection to identifying the sounds of flying

148
00:12:32,360 --> 00:12:39,440
missiles to better understand their trajectory and to protect civilians on time, hundreds

149
00:12:39,440 --> 00:12:41,680
and thousands of people.

150
00:12:41,680 --> 00:12:46,320
We got those projects in our portfolio.

151
00:12:46,320 --> 00:12:51,920
So yeah, I think that covers pretty much the main kind of projects that we work with, but

152
00:12:51,920 --> 00:12:54,000
they are not limited to it.

153
00:12:54,000 --> 00:13:00,840
So we have been working with healthcare quite a lot on skin diseases detection with hyper

154
00:13:00,840 --> 00:13:06,840
automation for fintech industries, robotics and manufacturing.

155
00:13:06,840 --> 00:13:08,760
That's a huge thing right now.

156
00:13:08,760 --> 00:13:14,240
And agriculture too, we see more and more of those requests and I'm super happy to see

157
00:13:14,240 --> 00:13:22,600
them because I think these developments will help us to solve some of the world problems.

158
00:13:22,600 --> 00:13:23,600
Absolutely.

159
00:13:23,600 --> 00:13:35,000
You've got a really diverse portfolio in your company and very, very cool to hear that.

160
00:13:35,000 --> 00:13:40,200
You're a thought leader in the field of AI and machine learning specifically, computer

161
00:13:40,200 --> 00:13:45,600
vision and of course data annotation.

162
00:13:45,600 --> 00:13:55,320
We talked about a little bit of how generative AI has impacted your field and also how you've

163
00:13:55,320 --> 00:14:01,920
been helping with national defense capabilities and across different industries.

164
00:14:01,920 --> 00:14:13,920
Are there any public infrastructure projects that you worked on or is how can AI and data

165
00:14:13,920 --> 00:14:18,560
annotation and computer vision help with infrastructure and public services?

166
00:14:18,560 --> 00:14:22,120
I know you've written and talked about it a little bit, so I just wanted to ask you

167
00:14:22,120 --> 00:14:25,120
about it.

168
00:14:25,120 --> 00:14:38,280
Well, speaking of the public services, I think bottom line, I'd start right away that I'm

169
00:14:38,280 --> 00:14:45,640
confident that AI will transform the way public services are now governed.

170
00:14:45,640 --> 00:14:47,400
It will definitely improve the efficiency.

171
00:14:47,400 --> 00:14:50,560
It will definitely improve the satisfaction of the citizens.

172
00:14:50,560 --> 00:14:58,480
So as I mentioned before, we have experience working with things like smart cities projects,

173
00:14:58,480 --> 00:15:05,480
waste management, environmental protection, energy, public health solution and all of

174
00:15:05,480 --> 00:15:14,640
them based on what I hear from our clients, see AI as a tremendous boosting tool towards

175
00:15:14,640 --> 00:15:23,680
bringing solutions at a quicker speed towards the current public sector problems.

176
00:15:23,680 --> 00:15:29,080
And apart from that, we also have experience working with solar panel management, urban

177
00:15:29,080 --> 00:15:36,600
safety, as I mentioned before, skin disease detection and so on, so better diagnosis things.

178
00:15:36,600 --> 00:15:45,040
Moving back to what you mentioned earlier from Gen AI and how it impacted our field,

179
00:15:45,040 --> 00:15:48,840
I probably want to take some time to talk about that as well.

180
00:15:48,840 --> 00:15:55,000
So I think one of the applications that we see in how our clients work with data is that

181
00:15:55,000 --> 00:16:01,640
Gen AI helps them to create data sense they need to train their models faster and at a

182
00:16:01,640 --> 00:16:03,560
higher level.

183
00:16:03,560 --> 00:16:10,840
So then they can use these data sense at an extensive level to train the models, right?

184
00:16:10,840 --> 00:16:19,400
But you have to be very careful there because you might end up working on a very thin ice

185
00:16:19,400 --> 00:16:26,520
by having synthetic data, lots of it and fitting your model with it, you really have to oversee

186
00:16:26,520 --> 00:16:30,520
the quality and the relevance of this data.

187
00:16:30,520 --> 00:16:36,200
Because the annotation may be good, everything may be great, but you might end up with the

188
00:16:36,200 --> 00:16:44,240
biased decisions made by your model or biased outcomes because the data wasn't representative

189
00:16:44,240 --> 00:16:47,360
enough, for example.

190
00:16:47,360 --> 00:16:58,840
Another application that we see is how there is a growing demand for human evaluation of

191
00:16:58,840 --> 00:17:01,360
the data generated by AI.

192
00:17:01,360 --> 00:17:07,360
So when there is a proper feedback needed to be given to the LLM output, for example,

193
00:17:07,360 --> 00:17:12,320
or to a computer vision model, whether this was a good or bad image generated, maybe there

194
00:17:12,320 --> 00:17:19,600
is something lacking or maybe a response to a specific question wasn't correct enough.

195
00:17:19,600 --> 00:17:28,400
And right now we build teams for our clients to improve the efficiency of these existing

196
00:17:28,400 --> 00:17:32,440
JNAI models by providing that feedback.

197
00:17:32,440 --> 00:17:39,200
And as I said before, sometimes it's super niche, sometimes we should onboard the scientists

198
00:17:39,200 --> 00:17:45,400
or people with some very, very specific background like in finance or in legal field because

199
00:17:45,400 --> 00:17:50,880
you really want to train your model with the feedback from people who understand what this

200
00:17:50,880 --> 00:17:52,720
model is going to produce in the end.

201
00:17:52,720 --> 00:17:58,920
So that's a very, very important thing to be careful about when either building an in-house

202
00:17:58,920 --> 00:18:05,440
data annotation team or outsourcing it somewhere is that you really want your team to understand

203
00:18:05,440 --> 00:18:14,360
the end result of your model so that they can shape their answers or annotations or prompts

204
00:18:14,360 --> 00:18:20,400
in the very correct way, understanding the depths of what you want to receive from the

205
00:18:20,400 --> 00:18:22,880
model at the end results.

206
00:18:22,880 --> 00:18:24,040
Yeah, absolutely.

207
00:18:24,040 --> 00:18:30,920
And yeah, I mean, like you said, I think both quality of the data that's coming into the

208
00:18:30,920 --> 00:18:33,200
model is really important.

209
00:18:33,200 --> 00:18:46,120
And also, especially with the JNAI, I think there needs to be a quality check on what the

210
00:18:46,120 --> 00:18:56,960
models are putting out and that there is definitely a need for quality check on the

211
00:18:56,960 --> 00:19:00,040
output end as well.

212
00:19:00,040 --> 00:19:06,360
In our company, yeah, I mean, I work for, I run our data science team in our company

213
00:19:06,360 --> 00:19:08,080
called Legal Shield.

214
00:19:08,080 --> 00:19:17,080
What we've come across is that I think JNAI in my mind is not ready yet for customer-facing

215
00:19:17,080 --> 00:19:26,960
tasks in my mind because of the consistency issues and hallucination issues that we're

216
00:19:26,960 --> 00:19:29,120
facing with the current models.

217
00:19:29,120 --> 00:19:38,960
So what we're doing right now is utilizing JNAI as a productivity tool.

218
00:19:38,960 --> 00:19:45,600
So to help, you know, we brought in GitHub Co-Pilot to help our software engineers and

219
00:19:45,600 --> 00:19:51,360
data engineers get better at coding or get quicker and faster at coding.

220
00:19:51,360 --> 00:20:03,760
We're also planning to bring in JNAI to help write emails and read through documents and

221
00:20:03,760 --> 00:20:04,760
things like that.

222
00:20:04,760 --> 00:20:06,600
So I think that's super helpful.

223
00:20:06,600 --> 00:20:12,040
Yeah, that was a good point that you made on the quality of the data.

224
00:20:12,040 --> 00:20:22,440
What are some of the things that you're doing to improve data quality in outputs of your

225
00:20:22,440 --> 00:20:23,440
models?

226
00:20:23,440 --> 00:20:24,440
Right.

227
00:20:24,440 --> 00:20:31,400
Well, our main goal is to improve the output of the models of our clients by feeding it

228
00:20:31,400 --> 00:20:33,760
with the right data.

229
00:20:33,760 --> 00:20:39,280
It's as simple as following the standard quality assurance procedures that we've established

230
00:20:39,280 --> 00:20:45,600
in the beginning of our work because to us, there are four main qualities in how we do

231
00:20:45,600 --> 00:20:46,920
business and label your data.

232
00:20:46,920 --> 00:20:52,560
Number one is security because oftentimes we deal with very sensitive, very personal

233
00:20:52,560 --> 00:20:57,720
data of our clients, especially those coming from FinTech or healthcare industries.

234
00:20:57,720 --> 00:21:03,080
And you have to be compliant with lots of different things like HIPAA, PCI, DSS, ISO

235
00:21:03,080 --> 00:21:09,000
certifications, CPA compliance, 3DPR, and the list goes on.

236
00:21:09,000 --> 00:21:11,200
So that's our number one value out there.

237
00:21:11,200 --> 00:21:18,120
And number two is quality because having poorly annotated data regardless of how much money

238
00:21:18,120 --> 00:21:22,400
you put into it won't bring you any good view on your model.

239
00:21:22,400 --> 00:21:27,560
So at Label Your Data over the years, we've been implementing a couple of different strategies

240
00:21:27,560 --> 00:21:29,360
and approaches.

241
00:21:29,360 --> 00:21:35,000
Some of them were also things that I personally found when we just started Label Your Data

242
00:21:35,000 --> 00:21:41,240
because at the very beginning of our journey, at 2020, right when the world shut down for

243
00:21:41,240 --> 00:21:45,160
the current team, this is where we were starting a new company.

244
00:21:45,160 --> 00:21:46,840
It worked out great.

245
00:21:46,840 --> 00:21:50,880
We had our first clients in April 2020.

246
00:21:50,880 --> 00:21:55,120
From that time on, we've been growing.

247
00:21:55,120 --> 00:22:02,200
But yeah, at the very first, at the very start of this journey, I was the one first doing

248
00:22:02,200 --> 00:22:03,800
the pilots for our project.

249
00:22:03,800 --> 00:22:08,920
So doing that annotation myself because I really wanted to understand the pain points.

250
00:22:08,920 --> 00:22:13,160
Because I think that, you know, well, data annotation can be perceived in the somewhat

251
00:22:13,160 --> 00:22:19,120
unmanned task, you know, something that you can get easily bored over, but then it can

252
00:22:19,120 --> 00:22:21,360
also impact your productivity.

253
00:22:21,360 --> 00:22:28,880
So we take care about that a lot, about how people see this job and what we can do from

254
00:22:28,880 --> 00:22:34,480
our end to make it better for them and to ensure more productivity.

255
00:22:34,480 --> 00:22:38,960
And then after I was doing the annotation, at some point of time, I was also the quality

256
00:22:38,960 --> 00:22:39,960
assurance specialist.

257
00:22:39,960 --> 00:22:44,560
So because I really wanted to make sure the data that our clients receive, it is of a good

258
00:22:44,560 --> 00:22:50,120
quality because they're first clients, a reputation, and it still goes on like this, right?

259
00:22:50,120 --> 00:22:55,480
So every project that we work on, it goes through the obligatory QA.

260
00:22:55,480 --> 00:22:59,960
So there are two levels, the annotation level and then the quality assurance level, which

261
00:22:59,960 --> 00:23:02,120
is all held on our end.

262
00:23:02,120 --> 00:23:08,880
And only then the client can go on and, you know, check the data for the quality and maybe

263
00:23:08,880 --> 00:23:15,240
return some things, you know, if there are some edge cases or gray areas, so to say.

264
00:23:15,240 --> 00:23:20,960
So the data goes through the obligatory, another second pair of eyes from our end from the

265
00:23:20,960 --> 00:23:25,560
quality assurance team, but we also offer hybrid approaches.

266
00:23:25,560 --> 00:23:31,200
So sometimes if we are talking about, well, it's more applicable to NLP than computer

267
00:23:31,200 --> 00:23:36,200
vision to be honest, because in NLP you receive more subjective data, right?

268
00:23:36,200 --> 00:23:43,600
So the text can be perceived with certain emotions or sometimes the words out there really stand

269
00:23:43,600 --> 00:23:45,760
for something different, right?

270
00:23:45,760 --> 00:23:48,200
If you don't understand the context or something.

271
00:23:48,200 --> 00:23:53,040
Or the reviews of the restaurants, for example, or the hotels or different places, they can

272
00:23:53,040 --> 00:23:56,240
mean different things depending on how you read them.

273
00:23:56,240 --> 00:24:00,040
So to ensure high quality of that kind of data and to make sure the client receives what

274
00:24:00,040 --> 00:24:04,560
they want, we can sometimes implement cross-reference quality assurance method.

275
00:24:04,560 --> 00:24:09,240
It's when the same piece of data is annotated simultaneously by different people, sometimes

276
00:24:09,240 --> 00:24:14,080
from different cultural backgrounds, because we recently had this interesting project where

277
00:24:14,080 --> 00:24:22,640
we had to listen to the audio recording of people, but the annotator didn't have to know

278
00:24:22,640 --> 00:24:28,000
the language spoken, they just had to grab the emotions.

279
00:24:28,000 --> 00:24:36,320
But it was super fun because people speak differently already, depending on their culture,

280
00:24:36,320 --> 00:24:38,280
depending on their mood.

281
00:24:38,280 --> 00:24:42,900
So not knowing the language and then detecting the emotion of the person speaking on the

282
00:24:42,900 --> 00:24:48,440
other side of the phone is very, very interesting and challenging.

283
00:24:48,440 --> 00:24:54,720
And then we can also implement things like embedding a golden data set into the data,

284
00:24:54,720 --> 00:24:56,560
the work of the data annotator.

285
00:24:56,560 --> 00:25:03,840
So it's just imagine the annotator annotates 1000 images, 10% of which have already been

286
00:25:03,840 --> 00:25:05,960
annotated by us and we know they are right.

287
00:25:05,960 --> 00:25:11,400
So we did that random spot check whether the annotator does a good job or not.

288
00:25:11,400 --> 00:25:15,880
And then the data never knows whether it's a new piece of data or it's part of that golden

289
00:25:15,880 --> 00:25:17,240
data set.

290
00:25:17,240 --> 00:25:25,200
So we are also always open to some new techniques and sometimes we even use elements to do the

291
00:25:25,200 --> 00:25:29,800
annotation and to compare it against what our annotators produce.

292
00:25:29,800 --> 00:25:35,000
So it acts like the third or a fourth or a fifth annotator.

293
00:25:35,000 --> 00:25:41,880
So yeah, we're always trying to bottom line is that there is always a quality assurance

294
00:25:41,880 --> 00:25:42,880
in our end.

295
00:25:42,880 --> 00:25:48,400
And then based on our experience and our use cases and as we learn the specific request

296
00:25:48,400 --> 00:25:54,360
of the client, we can then offer the better way of ensuring the quality so that the client

297
00:25:54,360 --> 00:26:01,560
doesn't have to go over however many thousands of images or data points they sent our way.

298
00:26:01,560 --> 00:26:05,480
Yeah, that's very interesting.

299
00:26:05,480 --> 00:26:11,640
Like you said, I'm glad that you have all these techniques and redundancies in place.

300
00:26:11,640 --> 00:26:15,840
So your quality is maintained, especially in data annotation.

301
00:26:15,840 --> 00:26:24,680
I think you'll have to think about both human errors and also machine errors, which you're

302
00:26:24,680 --> 00:26:25,680
already doing.

303
00:26:25,680 --> 00:26:27,320
That's super interesting.

304
00:26:27,320 --> 00:26:32,440
Well, apart from that, we sometimes also offer a hybrid approach.

305
00:26:32,440 --> 00:26:38,920
So it's not always right now, it's more and more not about classic data annotation.

306
00:26:38,920 --> 00:26:43,800
As we think of it, right, you just get a data set you annotate it, you feed your model with

307
00:26:43,800 --> 00:26:45,640
it and there you go.

308
00:26:45,640 --> 00:26:51,120
But more and more, we are being onboarded by our clients to the existing models that

309
00:26:51,120 --> 00:26:52,520
they have.

310
00:26:52,520 --> 00:26:56,560
And so they want that humans in the loop team to always be there to evaluate the results

311
00:26:56,560 --> 00:27:03,760
produced by their model, fix things here and there, or just review if it was done correctly,

312
00:27:03,760 --> 00:27:05,080
for example.

313
00:27:05,080 --> 00:27:10,880
So more and more, it's that hybrid model of cooperation, humans in the loop team working

314
00:27:10,880 --> 00:27:14,760
together with the machine, with the model.

315
00:27:14,760 --> 00:27:21,640
Yeah, it sounds like you have clients from all over the world and across different industries.

316
00:27:21,640 --> 00:27:25,600
So that's definitely fascinating, exciting, and also challenging.

317
00:27:25,600 --> 00:27:35,240
So that was a great example of, you know, label annotating data, annotating conversations,

318
00:27:35,240 --> 00:27:38,080
emotions of the conversations without knowing the language.

319
00:27:38,080 --> 00:27:40,920
That is fascinating, right?

320
00:27:40,920 --> 00:27:45,760
So, yeah, it's super exciting.

321
00:27:45,760 --> 00:27:51,320
You recently wrote an article in Europe's conference of AI rules.

322
00:27:51,320 --> 00:27:57,760
Can you please tell us what they are and how can other countries across the world learn

323
00:27:57,760 --> 00:27:58,760
from it?

324
00:27:58,760 --> 00:28:02,760
Well, that's an interesting thing, right?

325
00:28:02,760 --> 00:28:10,960
Well, in simple words, as I said, in simple words, as I said, this European Parliament

326
00:28:10,960 --> 00:28:21,400
AI Act, the main goal of it is to prevent or manage rather than prevent the risks of violating

327
00:28:21,400 --> 00:28:27,440
human rights with the existing emerging AI technologies, while at the same time trying

328
00:28:27,440 --> 00:28:33,520
not to limit the development of the technology.

329
00:28:33,520 --> 00:28:39,360
Because I don't think AI can be perceived as a good or bad, like in a black and white.

330
00:28:39,360 --> 00:28:45,400
It's a trajectory in a black and white perspective.

331
00:28:45,400 --> 00:28:49,760
I think it's both, right, depending on how you use it.

332
00:28:49,760 --> 00:28:56,520
And overall, I'm happy to see the notion, because I think this is the kind of the step

333
00:28:56,520 --> 00:29:03,960
that brings us closer towards preventing the many concerns people have about AI technology.

334
00:29:03,960 --> 00:29:12,800
And not only concerns about, not only concerns, but actual potential threats.

335
00:29:12,800 --> 00:29:19,600
So I think it's important to categorize risks and put the policies in place to prohibit

336
00:29:19,600 --> 00:29:27,120
things like what the Act prohibits rights of social scoring, emotional recognition,

337
00:29:27,120 --> 00:29:35,280
violating human behavior, or categorization by biometrics.

338
00:29:35,280 --> 00:29:39,240
I think it's undoubtedly something we should only continue to work on.

339
00:29:39,240 --> 00:29:40,600
And I'm excited to see what's next.

340
00:29:40,600 --> 00:29:45,920
I think it's a very important first break.

341
00:29:45,920 --> 00:29:51,920
But I also believe that it's not only about passing a law, but it's also about a strong

342
00:29:51,920 --> 00:29:59,800
cooperation between tech businesses and government, because one of the risks, I think, lies in

343
00:29:59,800 --> 00:30:08,880
the fact that the laws are sometimes passed by people who lack understanding of how tech

344
00:30:08,880 --> 00:30:14,160
and AI actually works.

345
00:30:14,160 --> 00:30:21,880
And so this is where the competence and expertise of AI leaders and businesses can really help

346
00:30:21,880 --> 00:30:30,800
to bridge that gap and make sure these laws and governmental decisions are put in place

347
00:30:30,800 --> 00:30:31,800
thoroughly.

348
00:30:31,800 --> 00:30:32,800
Yeah, absolutely.

349
00:30:32,800 --> 00:30:38,680
There's a lot of ethical considerations.

350
00:30:38,680 --> 00:30:45,120
And the policies are always catching up to technology and more so with AI, I mean, with

351
00:30:45,120 --> 00:30:47,240
the pace of development.

352
00:30:47,240 --> 00:30:52,200
It's really hard for governments to really catch up.

353
00:30:52,200 --> 00:30:57,640
So I think, yeah, definitely there needs to be a lot more collaboration with tech companies

354
00:30:57,640 --> 00:31:01,880
and policymakers across the world.

355
00:31:01,880 --> 00:31:10,720
And also, I think between countries, leaders of countries, between governments of different

356
00:31:10,720 --> 00:31:14,680
countries as well.

357
00:31:14,680 --> 00:31:22,720
Could you discuss the potential for international collaboration in AI projects?

358
00:31:22,720 --> 00:31:29,240
You've got plans from all across the globe and a lot of different countries.

359
00:31:29,240 --> 00:31:33,680
What are the potential for international collaborations in AI projects?

360
00:31:33,680 --> 00:31:41,160
Particularly those aimed at common global challenges like climate change or anything.

361
00:31:41,160 --> 00:31:43,680
Yes, of course.

362
00:31:43,680 --> 00:31:49,360
Well, it's hard for me to give my general opinion on it.

363
00:31:49,360 --> 00:31:54,160
I'll be talking only through my perspective as I can see it.

364
00:31:54,160 --> 00:32:01,880
I think we really need success cases, which will then be implemented.

365
00:32:01,880 --> 00:32:08,400
Because you can't just take something out there and, okay, let's do it.

366
00:32:08,400 --> 00:32:13,960
It's important to see something that, for example, worked for one country and then scale

367
00:32:13,960 --> 00:32:14,960
it up.

368
00:32:14,960 --> 00:32:21,160
I can talk, for example, in Ukraine, we have this amazing initiative called DIGIA, which

369
00:32:21,160 --> 00:32:26,520
is basically a digital country in Europe.

370
00:32:26,520 --> 00:32:32,520
We do our European vision voting through it.

371
00:32:32,520 --> 00:32:36,160
European vision voting through it.

372
00:32:36,160 --> 00:32:37,160
We have our passport.

373
00:32:37,160 --> 00:32:38,480
We have our language.

374
00:32:38,480 --> 00:32:43,680
Recently, we were having our diplomas in it, we're paying our fines through it.

375
00:32:43,680 --> 00:32:48,480
So that's a huge success case, which is now being implemented globally.

376
00:32:48,480 --> 00:32:55,160
But it wasn't the thing that naturally said, okay, we need to have the digital country

377
00:32:55,160 --> 00:33:00,520
in a phone that's cooperating and do it globally.

378
00:33:00,520 --> 00:33:04,120
It comes from those local successes.

379
00:33:04,120 --> 00:33:07,200
And this is the way it should be.

380
00:33:07,200 --> 00:33:12,600
You have to test it small scale, see it works, and then implement it on a global scale.

381
00:33:12,600 --> 00:33:20,280
But maybe, and not maybe, of course, adapted to specific country needs or culture needs

382
00:33:20,280 --> 00:33:25,480
or geographic needs, because not always something that worked for us, for example, will work

383
00:33:25,480 --> 00:33:27,440
for some other countries.

384
00:33:27,440 --> 00:33:32,880
Assessing the threats, cybersecurity threats, environmental threats.

385
00:33:32,880 --> 00:33:34,320
So it's a long process.

386
00:33:34,320 --> 00:33:36,480
It's not an easy thing to do.

387
00:33:36,480 --> 00:33:43,720
But I think learning from success cases and then implementing it maybe slowly but steadily

388
00:33:43,720 --> 00:33:49,800
is a way to go for the international cooperation for the DEI development.

389
00:33:49,800 --> 00:33:57,520
Yeah, that's a really good point on identifying local successes and then scaling it up across

390
00:33:57,520 --> 00:33:59,000
other countries.

391
00:33:59,000 --> 00:34:07,160
Because I'm originally from India and I live in the US now and hadn't been to India for

392
00:34:07,160 --> 00:34:08,160
about five years.

393
00:34:08,160 --> 00:34:15,280
And then when I went back last year and I was so surprised by the adoption of digital

394
00:34:15,280 --> 00:34:19,600
payment systems in India.

395
00:34:19,600 --> 00:34:27,520
So it's PayTM and Google Pay are the kings of the, you know, kings of the road.

396
00:34:27,520 --> 00:34:36,160
So basically, there's, I pulled out my credit card and gave it to them at a salon.

397
00:34:36,160 --> 00:34:41,800
Literally they had to pull out the card reader and then dust it out because nobody has used

398
00:34:41,800 --> 00:34:42,800
it in months.

399
00:34:42,800 --> 00:34:48,040
So everybody just pays through their, you know, mobile phones or smartphones.

400
00:34:48,040 --> 00:34:57,400
So it's fascinating how I think South Asia and Africa are kind of leading in the digital

401
00:34:57,400 --> 00:34:59,120
payments systems.

402
00:34:59,120 --> 00:35:06,920
And you gave a great example of how Ukraine is leading the charge in terms of, you know,

403
00:35:06,920 --> 00:35:13,160
it doesn't, you don't have to carry your documentation, your, you know, every, I mean, basically

404
00:35:13,160 --> 00:35:20,800
it sounds like the whole anything that you need from a documentation government standpoint

405
00:35:20,800 --> 00:35:24,040
can be done digitally.

406
00:35:24,040 --> 00:35:25,960
That's really cool.

407
00:35:25,960 --> 00:35:30,520
And what I also love is the openness of the creators of the government about it, right?

408
00:35:30,520 --> 00:35:34,880
So we are ready to share it with people with those countries who want it.

409
00:35:34,880 --> 00:35:41,080
What's great as well is that, you know, the government is open to, to cooperating with

410
00:35:41,080 --> 00:35:43,080
the tech minds, right?

411
00:35:43,080 --> 00:35:48,760
So, okay, if you think we have any vulnerabilities, okay, go ahead and tell us, you know, we'll

412
00:35:48,760 --> 00:35:52,920
pay you the check for it or something like this, you know, to make sure we all work together

413
00:35:52,920 --> 00:35:53,920
on improving it.

414
00:35:53,920 --> 00:35:57,800
And it's important to be that open minded, open for sharing.

415
00:35:57,800 --> 00:35:59,800
So, yeah.

416
00:35:59,800 --> 00:36:01,800
Yep, yep, absolutely.

417
00:36:01,800 --> 00:36:12,200
Yeah, that's, that's, you know, that's kind of the tech.

418
00:36:12,200 --> 00:36:13,520
That's the beauty of tech, right?

419
00:36:13,520 --> 00:36:21,240
It's always been open source as idealized in the tech industry.

420
00:36:21,240 --> 00:36:27,520
And hopefully it will continue in the AI space as, you know, it started off with science

421
00:36:27,520 --> 00:36:32,600
and then, you know, the tech industry adopted the openness and then hopefully, again, AI

422
00:36:32,600 --> 00:36:35,560
and other engineering fields do that as well.

423
00:36:35,560 --> 00:36:36,560
Of course.

424
00:36:36,560 --> 00:36:43,240
Well, another great example I can give you will give it my, my origin is in the middle

425
00:36:43,240 --> 00:36:44,240
tech industry.

426
00:36:44,240 --> 00:36:49,640
So, as I mentioned before, since the full scale invasion, terrific events, terrific

427
00:36:49,640 --> 00:36:54,760
invasion of Russia and Ukraine, we have been working with on different middle tech projects.

428
00:36:54,760 --> 00:37:04,000
And one of them was about removing that manual analytics of drone imagery and so, you know,

429
00:37:04,000 --> 00:37:09,440
switching it towards more automated things, whether it's enemy or target detection, whether

430
00:37:09,440 --> 00:37:12,480
it's observation of certain points.

431
00:37:12,480 --> 00:37:15,040
We have worked together on it.

432
00:37:15,040 --> 00:37:20,560
And it is the technology that is now presented in NATO used for the defense purposes.

433
00:37:20,560 --> 00:37:27,360
And so I think that, you know, we will, there will be a lot to share even in this year because

434
00:37:27,360 --> 00:37:30,240
this is just one of the projects we've been working on.

435
00:37:30,240 --> 00:37:36,080
So I know another super successful commercial project, which started off also as on a volunteering

436
00:37:36,080 --> 00:37:41,800
basis here in Ukraine, where we together with another amazing Ukrainian startup and a couple

437
00:37:41,800 --> 00:37:47,200
more amazing tech people from our industry have been working on building a system that

438
00:37:47,200 --> 00:37:52,400
will use, they will analyze the sounds of flying missiles around the, around Ukraine.

439
00:37:52,400 --> 00:37:53,400
Ukraine is huge.

440
00:37:53,400 --> 00:37:59,720
So it'll detect the sound, it'll tell whether it's a wind or an animal or a helicopter

441
00:37:59,720 --> 00:38:01,720
or a missile.

442
00:38:01,720 --> 00:38:08,440
And these small stations located around the country will be able to track it, track it,

443
00:38:08,440 --> 00:38:13,560
locally detect it and understand, okay, what this missile is heading, how we can prevent

444
00:38:13,560 --> 00:38:18,120
it from hitting the potentially civilian building or something.

445
00:38:18,120 --> 00:38:22,040
And it has also become a potential commercial project.

446
00:38:22,040 --> 00:38:26,880
So I think it's a great technology that will help to protect, philosophically tonight to

447
00:38:26,880 --> 00:38:35,440
protect civilian population against some horrific terror attacks.

448
00:38:35,440 --> 00:38:44,600
Yeah, like you said previously, AI is a tool and it can be used for good and bad and you've

449
00:38:44,600 --> 00:38:49,520
given great examples of how it can be life saving, right?

450
00:38:49,520 --> 00:38:56,320
So it's definitely fascinating.

451
00:38:56,320 --> 00:39:02,080
Although I've never really grown anything myself, coming from the US and having lived

452
00:39:02,080 --> 00:39:08,360
in small towns in the US, coming from India and live, sorry, having lived in small towns

453
00:39:08,360 --> 00:39:17,640
in the US, I know friends with ranches and farms who are into agriculture.

454
00:39:17,640 --> 00:39:22,920
You wrote a great piece on the role of computer vision and agriculture and you've worked

455
00:39:22,920 --> 00:39:25,160
on a lot of projects in that.

456
00:39:25,160 --> 00:39:31,120
Can you please talk to us about those projects a little deeper and how can AI and computer

457
00:39:31,120 --> 00:39:35,240
vision or data annotation help farmers in agriculture?

458
00:39:35,240 --> 00:39:37,240
Yeah, of course.

459
00:39:37,240 --> 00:39:40,400
I mean, it's a growing industry.

460
00:39:40,400 --> 00:39:49,280
First thing first, I think it's aiming at solving a very global food production problem.

461
00:39:49,280 --> 00:39:53,480
In terms of our expertise at Labelgear data, we have been working both with startups and

462
00:39:53,480 --> 00:39:55,600
enterprises in this field.

463
00:39:55,600 --> 00:40:04,680
They were mostly about developing a robotic solution to either detect some diseases or

464
00:40:04,680 --> 00:40:10,600
automating the crop selection, right?

465
00:40:10,600 --> 00:40:17,280
Where you have, I don't know, strawberries or some other plants that you are growing

466
00:40:17,280 --> 00:40:21,480
to basically gather your harvest.

467
00:40:21,480 --> 00:40:26,960
This is a super fun project for the editors to work on, so that they always enjoy it.

468
00:40:26,960 --> 00:40:32,680
It's that kind of project where you're just doing the work and you're really enjoying

469
00:40:32,680 --> 00:40:33,680
it.

470
00:40:33,680 --> 00:40:41,160
But it's not limited only to the cases that I've mentioned.

471
00:40:41,160 --> 00:40:44,960
So you look back to the piece that was written.

472
00:40:44,960 --> 00:40:50,720
Yes, there are so many applications of computer vision in agriculture, not only robotics,

473
00:40:50,720 --> 00:40:58,200
not only things like disease detection, but also things like water management, better

474
00:40:58,200 --> 00:41:02,960
soil management, better prediction, right?

475
00:41:02,960 --> 00:41:12,000
So I think overall, we will be seeing a huge improvement in the final metrics of the industry.

476
00:41:12,000 --> 00:41:14,080
It's really disruptive.

477
00:41:14,080 --> 00:41:16,800
It's amazing to see these things.

478
00:41:16,800 --> 00:41:23,080
So yeah, I'm also excited to see if this prediction really come to life.

479
00:41:23,080 --> 00:41:29,640
And I strongly believe that projects like AI and agriculture will help us to, might sound

480
00:41:29,640 --> 00:41:33,560
sentimental, but make the world a better place.

481
00:41:33,560 --> 00:41:35,400
No, absolutely.

482
00:41:35,400 --> 00:41:41,200
I mean, it's one of the fundamental industries that's never gone away.

483
00:41:41,200 --> 00:41:45,520
So more productive we can be, the better.

484
00:41:45,520 --> 00:41:52,960
And on a smaller scale, just an off-top story, I've been hunted by this Instagram ad of an

485
00:41:52,960 --> 00:41:57,960
app that helps you to detect the diseases of your plants.

486
00:41:57,960 --> 00:42:05,680
And me, like, well, I've grown in a small village, so we had a lot of, you know, planting

487
00:42:05,680 --> 00:42:09,200
potatoes and cucumbers and tomatoes going on.

488
00:42:09,200 --> 00:42:12,360
And for my whole childhood, I said, I hate it.

489
00:42:12,360 --> 00:42:17,760
I'm never going to do it because it was, you know, you have to do things as a child and

490
00:42:17,760 --> 00:42:19,080
you have to help your parents.

491
00:42:19,080 --> 00:42:21,480
And it's not always the best thing.

492
00:42:21,480 --> 00:42:26,240
It's not always the thing you dream of doing on your summer vacation, right?

493
00:42:26,240 --> 00:42:33,160
But then recently, and I've never had, as an adult, I've never had like home plants

494
00:42:33,160 --> 00:42:34,480
in the apartment.

495
00:42:34,480 --> 00:42:39,160
But recently, I got five plants.

496
00:42:39,160 --> 00:42:43,480
I bought five plants, almost five in a row, because for some reason I just had this, you

497
00:42:43,480 --> 00:42:48,600
know, excitement about having some home plants in the apartment.

498
00:42:48,600 --> 00:42:55,360
And I was so scared that I'm not going to manage it, that I will fail, that they will

499
00:42:55,360 --> 00:42:56,840
die.

500
00:42:56,840 --> 00:43:01,440
And then, you know, I thought, well, I knew things, right?

501
00:43:01,440 --> 00:43:03,400
I know how to run a company.

502
00:43:03,400 --> 00:43:06,120
I know how to manage people.

503
00:43:06,120 --> 00:43:14,160
But I was still so, you know, petrified about managing my five small plants in my kitchen.

504
00:43:14,160 --> 00:43:21,000
So yeah, well, things like even having those small apps helping you to take, to give a

505
00:43:21,000 --> 00:43:25,400
better care to your plants, I think it's also a good thing.

506
00:43:25,400 --> 00:43:26,400
Yeah, absolutely.

507
00:43:26,400 --> 00:43:32,320
I mean, I've never been a plant person, but I'm slowly trying to be now.

508
00:43:32,320 --> 00:43:37,200
I think this plant is kind of going, I need that AI.

509
00:43:37,200 --> 00:43:41,120
Oh my God, I saw there was a background and I was just about to make a joke about it in

510
00:43:41,120 --> 00:43:44,400
the beginning saying you didn't plant anything, but here you go.

511
00:43:44,400 --> 00:43:45,680
I thought that was a background.

512
00:43:45,680 --> 00:43:46,680
Oh, thank you.

513
00:43:46,680 --> 00:43:47,680
That's great.

514
00:43:47,680 --> 00:43:49,080
I'll take that as a compliment.

515
00:43:49,080 --> 00:43:53,960
But I do need that app you're talking about to see if this plant's going bad.

516
00:43:53,960 --> 00:43:56,080
But yeah, it is definitely fascinating.

517
00:43:56,080 --> 00:43:57,600
You're taking good care of it.

518
00:43:57,600 --> 00:43:58,600
So.

519
00:43:58,600 --> 00:43:59,600
Oh, thank you.

520
00:43:59,600 --> 00:44:00,600
Yeah, absolutely.

521
00:44:00,600 --> 00:44:11,000
And also, I think as you were talking about applications of AI in agriculture, I think

522
00:44:11,000 --> 00:44:20,200
since the world is rapidly evolving in adoption of AI across industries, I think we'll see

523
00:44:20,200 --> 00:44:28,880
a lot of cross-pollination and innovations in other industries helping something else.

524
00:44:28,880 --> 00:44:37,960
Say for example, if we get really better, even better than right now at predicting weather,

525
00:44:37,960 --> 00:44:47,480
maybe that'll help the agricultural industry, the rains and predicting exact quantities and

526
00:44:47,480 --> 00:44:49,160
timelines of the rains and everything.

527
00:44:49,160 --> 00:44:58,040
So I think we will see a lot of cross-pollination of innovations and applications of AI from

528
00:44:58,040 --> 00:45:02,560
other industries of doubt in across different ones.

529
00:45:02,560 --> 00:45:06,320
Well, I heard this fun story recently.

530
00:45:06,320 --> 00:45:08,120
Our marketing lead shared it with me.

531
00:45:08,120 --> 00:45:13,920
Well, not quite sure it's true, but apparently in Japan, I think it was Japan, they were

532
00:45:13,920 --> 00:45:17,960
developing an AI computer vision model.

533
00:45:17,960 --> 00:45:24,280
They've been doing it for five or so years to be able to detect between 200 different

534
00:45:24,280 --> 00:45:31,720
bakeries, bakery products, because it's a lot of manual work and they wanted to automate

535
00:45:31,720 --> 00:45:37,560
it, but it was so difficult to do because these 200 different types of bakery products,

536
00:45:37,560 --> 00:45:40,040
they almost looked alike.

537
00:45:40,040 --> 00:45:41,920
And then they finally developed it.

538
00:45:41,920 --> 00:45:47,240
And then there was this guy who came in and said, hey, you know what, actually on a larger

539
00:45:47,240 --> 00:45:53,480
scale, these bakery products, they look similar to some cancer cells or something like this.

540
00:45:53,480 --> 00:45:59,360
So they have used this model to, and obviously they have developed it over time, but they've

541
00:45:59,360 --> 00:46:05,040
used it as a foundation to develop the model to detect some cells, you know, to improve

542
00:46:05,040 --> 00:46:07,800
in the healthcare industry.

543
00:46:07,800 --> 00:46:10,200
So that's another example.

544
00:46:10,200 --> 00:46:11,200
Well, not sure.

545
00:46:11,200 --> 00:46:16,000
I'm pretty sure it's a real one, but it sounds like a made-up story, but it proves the point,

546
00:46:16,000 --> 00:46:17,000
right?

547
00:46:17,000 --> 00:46:18,720
Yeah, no, absolutely.

548
00:46:18,720 --> 00:46:22,800
I wouldn't be surprised if that's true, right?

549
00:46:22,800 --> 00:46:30,280
There's a lot of things that we don't think about as humans, and as AI gets better and

550
00:46:30,280 --> 00:46:34,480
you know, it's the knowledge base of the world more so than the internet.

551
00:46:34,480 --> 00:46:40,120
I think it'll be able to make those connections that we've never made before, I think.

552
00:46:40,120 --> 00:46:47,080
So I'm super excited about, you know, what AI brings us in that space.

553
00:46:47,080 --> 00:46:50,000
Yep, for sure.

554
00:46:50,000 --> 00:46:59,640
Yeah, you've led your company and leading your company through enormous crises in a pandemic

555
00:46:59,640 --> 00:47:01,920
and war.

556
00:47:01,920 --> 00:47:07,440
What can you teach us about leadership and leading a team in crisis?

557
00:47:07,440 --> 00:47:10,720
What can we learn from your experience?

558
00:47:10,720 --> 00:47:17,640
I love talking about this because I'm just sharing my experience.

559
00:47:17,640 --> 00:47:24,600
I don't think, I don't think you can pass the leadership course or management course,

560
00:47:24,600 --> 00:47:27,120
and then all of a sudden become a great leader, right?

561
00:47:27,120 --> 00:47:29,680
It's all about practice.

562
00:47:29,680 --> 00:47:40,200
I had my mistakes and had my good things that I just learned from and learned from my lessons,

563
00:47:40,200 --> 00:47:41,360
but I love this.

564
00:47:41,360 --> 00:47:47,160
I recently listened to, I think it was Simon Altman's podcast with Lex Friedman, and he

565
00:47:47,160 --> 00:47:54,280
was saying about the Hactic Weekend that happened last November, and he was talking about Mira

566
00:47:54,280 --> 00:48:00,040
who took over, you know, temporarily the open AI.

567
00:48:00,040 --> 00:48:10,520
And he said this thing that I think that just got, you know, deeply in my mind that we usually,

568
00:48:10,520 --> 00:48:21,400
we usually evaluate the great leaders by how they act in crises or in stressful situations

569
00:48:21,400 --> 00:48:25,680
or just with those one-time moments.

570
00:48:25,680 --> 00:48:33,440
But being a good leader, I think, is, as Sam said, and I also stand by that, is in those

571
00:48:33,440 --> 00:48:39,640
day-to-day things, right, in how you show up to the meetings and how you make your decisions

572
00:48:39,640 --> 00:48:44,320
and how you solve challenges on a day-to-day basis.

573
00:48:44,320 --> 00:48:51,000
And it's what about, it is about what you do on a regular day, regular morning, regular

574
00:48:51,000 --> 00:48:53,440
afternoon or evening.

575
00:48:53,440 --> 00:49:03,200
So I think genuinely what helped me a lot is that I strongly believe in what we do.

576
00:49:03,200 --> 00:49:05,440
I strongly believe in our social mission.

577
00:49:05,440 --> 00:49:09,440
I strongly believe in our mission to table your data, which is go building AI-driven

578
00:49:09,440 --> 00:49:11,000
economy.

579
00:49:11,000 --> 00:49:17,040
And this is the thing that I always tell my people, my executive teams and the leaders

580
00:49:17,040 --> 00:49:23,560
of the teams about we hold strategic sessions two times a year and communication sessions

581
00:49:23,560 --> 00:49:31,600
with the whole team, including tag team, marketing, HR, service delivery, business development.

582
00:49:31,600 --> 00:49:35,920
And what we always look back to, we start every report with our mission.

583
00:49:35,920 --> 00:49:39,360
So everybody told you, reminded where we are going.

584
00:49:39,360 --> 00:49:43,680
And from that mission, we start to build our tactical steps.

585
00:49:43,680 --> 00:49:47,960
And we always communicate to every single person.

586
00:49:47,960 --> 00:49:50,040
We spend a lot of time brainstorming it.

587
00:49:50,040 --> 00:49:59,000
And we always communicate how every single team member can help the company achieve its

588
00:49:59,000 --> 00:50:06,720
goals by doing their day-to-day things, whether you're a content writer, whether you're a

589
00:50:06,720 --> 00:50:12,240
software developer, whether you're an SDR or anything.

590
00:50:12,240 --> 00:50:19,480
So it's all about, I think, for me, it has always been about vision, keeping in mind

591
00:50:19,480 --> 00:50:20,800
regain and focus.

592
00:50:20,800 --> 00:50:26,680
My day-to-day routine, by the way, is always about talking to people, whether it's potential

593
00:50:26,680 --> 00:50:32,360
clients or our existing clients or team members or partners.

594
00:50:32,360 --> 00:50:41,600
I do talking a lot, but then I learned to dedicate one day a week to strategic tasks,

595
00:50:41,600 --> 00:50:44,040
leaving everything else behind.

596
00:50:44,040 --> 00:50:48,600
Because I think it was easy to do so when you were relatively small, but as we scale

597
00:50:48,600 --> 00:50:52,320
up, you get a lot of operational stuff on your shoulders.

598
00:50:52,320 --> 00:50:56,800
And there are lots of things you get to deal with, including finances or interviewing

599
00:50:56,800 --> 00:51:05,120
new team members or maybe just spending lots of time fighting a battle, the Google spreadsheet,

600
00:51:05,120 --> 00:51:09,360
trying to get that formula right.

601
00:51:09,360 --> 00:51:15,720
But it is important to stop there and to just rise above and gain that, well, you can call

602
00:51:15,720 --> 00:51:19,640
it a helicopter vision or a 360 vision.

603
00:51:19,640 --> 00:51:27,480
And why am I here, what value can I bring as the CEO to the company, regain in focus,

604
00:51:27,480 --> 00:51:32,840
looking back to the very beginning and understanding that you're here to drive strategy.

605
00:51:32,840 --> 00:51:37,920
And my main goal is to make sure that our strategic plans are being executed correctly

606
00:51:37,920 --> 00:51:41,280
throughout a certain period of time.

607
00:51:41,280 --> 00:51:44,040
So definitely vision, definitely about being transparent.

608
00:51:44,040 --> 00:51:46,600
I think there's another key thing.

609
00:51:46,600 --> 00:51:49,960
And there is a crisis, you say there is a crisis, right?

610
00:51:49,960 --> 00:51:55,640
But I'm here, we're all in this together, I'm here to give you all the resources, all

611
00:51:55,640 --> 00:52:01,800
the means to make sure you stay on board because I know I value you and I want you to be, you

612
00:52:01,800 --> 00:52:08,040
know, to stay in the team and we're going to go through it, every crisis ends.

613
00:52:08,040 --> 00:52:15,640
So you know, it's those who survived the crisis will make it after that.

614
00:52:15,640 --> 00:52:21,920
So yeah, I think those key things to sum up vision, regain in focus on strategic things

615
00:52:21,920 --> 00:52:29,240
and being transparent with your team and with yourself too, is what I learned from these

616
00:52:29,240 --> 00:52:32,080
four years of building label your data.

617
00:52:32,080 --> 00:52:33,080
That's awesome.

618
00:52:33,080 --> 00:52:37,920
I mean, a couple of things really stood up to me, like you talked about everyday leadership,

619
00:52:37,920 --> 00:52:38,920
right?

620
00:52:38,920 --> 00:52:45,520
And it's not just in the, you know, points of crisis, but in how you operate every day

621
00:52:45,520 --> 00:52:47,720
is really important as a leader.

622
00:52:47,720 --> 00:52:51,160
That's super important.

623
00:52:51,160 --> 00:52:56,600
And also you talked about the importance of repetition and consistency, you know, of clarifying

624
00:52:56,600 --> 00:53:02,120
your messages as a leader, you know, in terms of what your vision is and what your, you

625
00:53:02,120 --> 00:53:08,400
know, projects are, you know, being transparent with your team, especially on the repetition

626
00:53:08,400 --> 00:53:09,400
and consistency.

627
00:53:09,400 --> 00:53:19,480
I did speak to Jim James, a PR guru on this podcast a couple of weeks ago, and he said

628
00:53:19,480 --> 00:53:29,120
the same thing that the most important thing for a leader is from a PR standpoint is repeating

629
00:53:29,120 --> 00:53:31,160
and staying consistent with the message.

630
00:53:31,160 --> 00:53:33,920
So I think you're doing the right things already.

631
00:53:33,920 --> 00:53:35,920
Yeah, absolutely.

632
00:53:35,920 --> 00:53:41,760
And then communicating them, right, you may think, I mean, you may stay focused on it,

633
00:53:41,760 --> 00:53:46,400
you know, on your own, but it's also important to communicate it to the team.

634
00:53:46,400 --> 00:53:52,560
And oftentimes we have some starting conversations with my team and they, they, they, they, I

635
00:53:52,560 --> 00:53:57,560
think it's important to cultivate the safe space for the team, right?

636
00:53:57,560 --> 00:54:02,200
When they can openly share ideas as well as concerns they have.

637
00:54:02,200 --> 00:54:07,040
And then when you understand that you need to maybe sometimes take time to just listen

638
00:54:07,040 --> 00:54:09,720
and understand, okay, I understand your concern.

639
00:54:09,720 --> 00:54:11,480
I understand your doubt.

640
00:54:11,480 --> 00:54:17,000
Well, maybe even share your personal story that you wouldn't have shared otherwise.

641
00:54:17,000 --> 00:54:25,080
Is it any other, you know, any other in any other situation, but maybe, you know, just

642
00:54:25,080 --> 00:54:30,240
being transparent and also telling people sometimes about your weaknesses and how you

643
00:54:30,240 --> 00:54:32,600
acted in those moments of self-doubt.

644
00:54:32,600 --> 00:54:40,520
It also brings a lot of trust towards you as a leader and as a person as well.

645
00:54:40,520 --> 00:54:45,360
Those are very intimate moments, especially if we're talking about building the right

646
00:54:45,360 --> 00:54:50,800
executive team, core team, people who then go and communicate the message to their team

647
00:54:50,800 --> 00:54:51,800
members.

648
00:54:51,800 --> 00:54:56,880
It's in those sometimes intimate moments, not about, you know, I'm the manager here,

649
00:54:56,880 --> 00:54:58,480
I'm the leader here, I'm the CEO.

650
00:54:58,480 --> 00:55:01,040
You do everything I tell you to do.

651
00:55:01,040 --> 00:55:08,640
It's about sharing that experience which is sometimes painful, sometimes exciting, right?

652
00:55:08,640 --> 00:55:11,000
And sharing the emotions.

653
00:55:11,000 --> 00:55:17,920
Because then I think in the end we all work, it's all about people, right?

654
00:55:17,920 --> 00:55:21,320
I do business development, I do sales, sometimes they buy your data.

655
00:55:21,320 --> 00:55:25,080
And I know it's all about people, it's all about genuine interest in people rather than

656
00:55:25,080 --> 00:55:28,920
make closing a successful deal or going to a conference, right?

657
00:55:28,920 --> 00:55:30,800
And trying to get as many leaders as possible.

658
00:55:30,800 --> 00:55:35,840
No, it's always about talking to people and showing your genuine appreciation, showing

659
00:55:35,840 --> 00:55:41,880
your genuine interest in who they are and learning how they got there.

660
00:55:41,880 --> 00:55:45,360
So I think that's what is value in communication.

661
00:55:45,360 --> 00:55:48,160
Yeah, no, absolutely.

662
00:55:48,160 --> 00:55:55,040
And as in my leadership journey, I've learned that I don't have all the answers at all

663
00:55:55,040 --> 00:55:56,520
times, right?

664
00:55:56,520 --> 00:56:04,320
So just being transparent and also sharing your own journey and self-doubts, like you

665
00:56:04,320 --> 00:56:13,920
said, it makes you a more relatable leader and not somebody who just reads a book and

666
00:56:13,920 --> 00:56:17,520
just talks just about that.

667
00:56:17,520 --> 00:56:23,080
It's about personal journeys, it's about we're all people, we're all, you know.

668
00:56:23,080 --> 00:56:26,840
Yeah, sure, I think being humble is important.

669
00:56:26,840 --> 00:56:33,920
So for us, well, just to give you another example from a recent experience, I'm not a

670
00:56:33,920 --> 00:56:38,200
marketing person, I don't have a lot of marketing experience.

671
00:56:38,200 --> 00:56:47,120
But when we were onboarding a new marketing lead, I knew I have to make sure he does a

672
00:56:47,120 --> 00:56:48,800
good job, right?

673
00:56:48,800 --> 00:56:52,400
But then how can I do that if I don't know marketing?

674
00:56:52,400 --> 00:56:57,920
And that in the end, it all came down to, for example, him showing me a report, which

675
00:56:57,920 --> 00:57:02,120
I, for example, there were lots of numbers, lots of concepts, right?

676
00:57:02,120 --> 00:57:03,760
I might not know.

677
00:57:03,760 --> 00:57:08,240
But I'm showing my willingness to learn and to explain this to me because I really want

678
00:57:08,240 --> 00:57:09,840
to know what this means.

679
00:57:09,840 --> 00:57:12,040
I really want to be on the same page with you.

680
00:57:12,040 --> 00:57:16,200
Well, someone can say it, my doctor is okay, you don't know things, right?

681
00:57:16,200 --> 00:57:17,200
Why are you in here?

682
00:57:17,200 --> 00:57:19,160
Why are you my manager?

683
00:57:19,160 --> 00:57:24,040
But no, I see it, it does a completely opposite thing.

684
00:57:24,040 --> 00:57:25,760
I'm genuinely interested in what you do.

685
00:57:25,760 --> 00:57:30,000
I'm genuinely interested in how you want to bring me that result.

686
00:57:30,000 --> 00:57:32,160
And in your experience, so please share it with me.

687
00:57:32,160 --> 00:57:35,840
At any other convenient time, I'll be sharing my experience with you, maybe because maybe

688
00:57:35,840 --> 00:57:40,200
there are some things that you are not, you are lacking expertise in where I can fill

689
00:57:40,200 --> 00:57:43,400
in and vice versa.

690
00:57:43,400 --> 00:57:49,040
And my team knows that they are always learning something, they're always passing some course

691
00:57:49,040 --> 00:57:53,920
courses, watching things, and we're always learning from each other.

692
00:57:53,920 --> 00:57:58,800
They come and say, I passed this course, I heard this and that.

693
00:57:58,800 --> 00:58:02,760
Here is what we can implement in the company and say, okay, sure, looks like a good idea.

694
00:58:02,760 --> 00:58:09,720
And then when I hear something or I pass some course or do some education, I go and I talk

695
00:58:09,720 --> 00:58:12,520
to them about it and this is how we grow together.

696
00:58:12,520 --> 00:58:18,560
And this creates, as I said, that safe idea lab rather than eco chamber when you can exchange

697
00:58:18,560 --> 00:58:25,840
ideas and open for criticism, open to discussion, open to the out of the box ideas are not afraid

698
00:58:25,840 --> 00:58:30,720
to be humble and to share what you really think.

699
00:58:30,720 --> 00:58:36,840
Because in that, you see how you contribute to our common goal or common success.

700
00:58:36,840 --> 00:58:37,840
Yeah, absolutely.

701
00:58:37,840 --> 00:58:48,160
I mean, one of my philosophies is don't be an asshole, be a jerk, maybe.

702
00:58:48,160 --> 00:58:55,360
And so, yeah, I mean, there's a lot of things you can learn from anybody and everybody, right?

703
00:58:55,360 --> 00:59:03,960
So one of the things I keep a journal off or keep notes off is what did I like about

704
00:59:03,960 --> 00:59:11,040
people I talked to in my in the podcast or at my work or any place we meet.

705
00:59:11,040 --> 00:59:16,880
So how can I emulate some things I like from each person I meet?

706
00:59:16,880 --> 00:59:23,840
You know, that that has helped me be more positive and trust people more and also be

707
00:59:23,840 --> 00:59:28,160
a better leader on the whole.

708
00:59:28,160 --> 00:59:31,440
But I think it all ends up to that one motto of yours.

709
00:59:31,440 --> 00:59:33,440
Don't be a jerk.

710
00:59:33,440 --> 00:59:34,440
Exactly.

711
00:59:34,440 --> 00:59:37,600
Yeah, humility is super important, right?

712
00:59:37,600 --> 00:59:51,080
So as we see more and more that being a problematic leader brings a lot of eyeballs, but I think

713
00:59:51,080 --> 01:00:02,640
in my own little world, I think I'm championing a humble leader and importance of being a

714
01:00:02,640 --> 01:00:04,640
humble leader.

715
01:00:04,640 --> 01:00:11,520
Yeah.

716
01:00:11,520 --> 01:00:18,480
You wrote an article asking if we should be afraid of AI and a lot of people are like

717
01:00:18,480 --> 01:00:21,520
we talked about.

718
01:00:21,520 --> 01:00:23,280
And for some good reasons, right?

719
01:00:23,280 --> 01:00:26,680
There are the genuine concerns that you mentioned.

720
01:00:26,680 --> 01:00:31,200
Can you please give us your take on it for our audience?

721
01:00:31,200 --> 01:00:39,080
Well, that's a never ending argument, isn't it?

722
01:00:39,080 --> 01:00:46,640
There is no doubt regarding how powerful AI techniques can be and impact in both good

723
01:00:46,640 --> 01:00:50,680
and bad ways the humanity.

724
01:00:50,680 --> 01:00:58,920
But my personal philosophical take on it is simple.

725
01:00:58,920 --> 01:01:08,480
I think that AI cannot possess bigger threats for humanity than humans themselves.

726
01:01:08,480 --> 01:01:20,640
That we are managing, though not always successfully, to prevent our worst fears from happening.

727
01:01:20,640 --> 01:01:28,320
It somehow, the world somehow ends up being, again, somewhat balanced, right?

728
01:01:28,320 --> 01:01:33,320
One party develops AI, another one regulates it.

729
01:01:33,320 --> 01:01:39,400
So I think that one, we have to admit we can stop it.

730
01:01:39,400 --> 01:01:47,680
I think the case with the letter to pause the development for six months really, you

731
01:01:47,680 --> 01:01:53,880
know, shown that no one can stop it from happening.

732
01:01:53,880 --> 01:01:56,560
It's already there.

733
01:01:56,560 --> 01:02:02,320
But let's do it in a smart and sustainable way.

734
01:02:02,320 --> 01:02:03,320
That's it.

735
01:02:03,320 --> 01:02:05,440
Yeah, no, absolutely.

736
01:02:05,440 --> 01:02:14,520
There's no, you made a good point of, you know, the world, humanity finds a way to make

737
01:02:14,520 --> 01:02:15,840
things okay.

738
01:02:15,840 --> 01:02:20,000
We saw this with the nuclear warfare and the ADs.

739
01:02:20,000 --> 01:02:22,320
I mean, we've heard about it.

740
01:02:22,320 --> 01:02:23,960
You and I weren't there.

741
01:02:23,960 --> 01:02:34,800
But so I think I am positive as well that we will be okay at the end.

742
01:02:34,800 --> 01:02:43,240
We will see some bad actors using AI or any kind of technology in bad faith.

743
01:02:43,240 --> 01:02:47,360
But I think overall, over-arching Lee will be okay.

744
01:02:47,360 --> 01:02:51,240
I think overall, it challenges us tremendously, right?

745
01:02:51,240 --> 01:02:59,000
It challenges our ability to have power.

746
01:02:59,000 --> 01:03:02,960
It changes in some ways the human nature.

747
01:03:02,960 --> 01:03:07,800
It challenges how we think about ourselves.

748
01:03:07,800 --> 01:03:10,480
It challenges how we think about other people, right?

749
01:03:10,480 --> 01:03:15,720
Because one thing I think about is, okay, what if it ends up in the hands of bad people,

750
01:03:15,720 --> 01:03:16,720
right?

751
01:03:16,720 --> 01:03:23,960
It challenges lots of foundational things about human nature, about the current state

752
01:03:23,960 --> 01:03:27,720
of things, about our future.

753
01:03:27,720 --> 01:03:36,760
And I think that because we are, I don't know, we are going with such a huge speed at it

754
01:03:36,760 --> 01:03:43,120
that our inability to predict things is what scares us the most, right?

755
01:03:43,120 --> 01:03:51,320
So we would, even with the regulation of AI, we oftentimes hear these thoughts, okay, we

756
01:03:51,320 --> 01:03:58,840
are, we are nearly late to implement the regulation because it's all over, you know, it's full

757
01:03:58,840 --> 01:04:02,360
speed development.

758
01:04:02,360 --> 01:04:08,920
So I think it's that inability to predict things at this current pace is what is scary.

759
01:04:08,920 --> 01:04:10,640
It's okay, right?

760
01:04:10,640 --> 01:04:11,800
It's the time it should be.

761
01:04:11,800 --> 01:04:17,480
You don't know what's out there that makes you feel scared about it.

762
01:04:17,480 --> 01:04:19,000
But yeah, I also tend to be optimistic.

763
01:04:19,000 --> 01:04:24,880
I think it's also a lot to do with education, but with educating people on what it really

764
01:04:24,880 --> 01:04:25,880
is.

765
01:04:25,880 --> 01:04:31,760
As I said, with the governmental policies, right, it's sometimes people should then know

766
01:04:31,760 --> 01:04:32,760
the capabilities.

767
01:04:32,760 --> 01:04:38,280
They don't know what AI in essence is and they perceive it as some, I don't know, either

768
01:04:38,280 --> 01:04:42,040
a magical pill, which is also a bad thing, right?

769
01:04:42,040 --> 01:04:47,800
Some people just want to use AI to solve every single problem out there, which is not true.

770
01:04:47,800 --> 01:04:51,200
It's not right, at least at this point of time.

771
01:04:51,200 --> 01:04:58,240
But then the other camp sees it as some terrific, you know, weapon that's going to kill us all

772
01:04:58,240 --> 01:05:04,200
and going to rebel against humanity and kill everyone.

773
01:05:04,200 --> 01:05:11,240
Without educating people so that they can realistically assess risks like in the act

774
01:05:11,240 --> 01:05:12,240
that was released.

775
01:05:12,240 --> 01:05:17,880
It's not about some transformers, you know, robots coming and killing everyone.

776
01:05:17,880 --> 01:05:20,520
It's about things like manipulating.

777
01:05:20,520 --> 01:05:25,840
And so you have to be cautious because even without AI, a lot of people are manipulated

778
01:05:25,840 --> 01:05:29,640
with the media, with social media and stuff.

779
01:05:29,640 --> 01:05:35,240
So yeah, I think that's my take on it, optimistic, but definitely there is a lot of things we

780
01:05:35,240 --> 01:05:42,560
should do about it, to regulate it, to do it in a smart, to do it in a sustainable way.

781
01:05:42,560 --> 01:05:52,760
And be cautious that it will definitely change the way things are, not only, you know, in

782
01:05:52,760 --> 01:05:59,800
the reality, but philosophically to who we are and what are our capabilities.

783
01:05:59,800 --> 01:06:00,800
So yeah.

784
01:06:00,800 --> 01:06:05,560
Yeah, it's basically enhancing our intelligence.

785
01:06:05,560 --> 01:06:13,240
We've augmented our ability to walk and run with automobiles.

786
01:06:13,240 --> 01:06:25,720
We've augmented our ability to, you know, say, shout or communicate with digital technology,

787
01:06:25,720 --> 01:06:28,600
microphones and speakers and anything like that.

788
01:06:28,600 --> 01:06:33,960
And AI is more of an augmentation of your intelligence, right?

789
01:06:33,960 --> 01:06:36,360
No, of course, it's artificial intelligence.

790
01:06:36,360 --> 01:06:43,160
But yeah, I mean, definitely no one knows where we're going to go from here, learn to

791
01:06:43,160 --> 01:06:49,320
your point, regulation and keeping ourselves in check is super important.

792
01:06:49,320 --> 01:06:56,280
But again, maybe 10 years from now, we'll just use AI to create tech talks and random

793
01:06:56,280 --> 01:06:57,520
emojis, who knows.

794
01:06:57,520 --> 01:07:00,000
No, I'm just kidding.

795
01:07:00,000 --> 01:07:02,120
I think we're already doing it.

796
01:07:02,120 --> 01:07:03,120
We are.

797
01:07:03,120 --> 01:07:04,120
We are.

798
01:07:04,120 --> 01:07:05,120
I mean, 10 years from now, that's all we'll be doing.

799
01:07:05,120 --> 01:07:07,320
We're creating tech talks.

800
01:07:07,320 --> 01:07:09,240
But yeah, it's there.

801
01:07:09,240 --> 01:07:18,840
There's a lot of ways this will go and a lot of industries are being disrupted, right?

802
01:07:18,840 --> 01:07:27,720
We talked about the role of regulators and the role of tech leaders to be involved.

803
01:07:27,720 --> 01:07:34,160
We also talked about ethical considerations while building AI and the data sets that go

804
01:07:34,160 --> 01:07:37,000
into AI.

805
01:07:37,000 --> 01:07:50,280
There is a lot of talk about tech leaders and engineers having way too much power because

806
01:07:50,280 --> 01:07:53,720
AI is taken over.

807
01:07:53,720 --> 01:08:03,520
How can, what are some of your tips to engineers or building these models and AI technology

808
01:08:03,520 --> 01:08:12,800
to keep their own biases in check and bring in different voices or not make sure that

809
01:08:12,800 --> 01:08:15,080
it's equitable to everybody?

810
01:08:15,080 --> 01:08:20,840
Well, I think you got it there in your question.

811
01:08:20,840 --> 01:08:24,520
But I don't think I'm in a position to give some sort of advice.

812
01:08:24,520 --> 01:08:25,520
That's just my perspective.

813
01:08:25,520 --> 01:08:27,520
Again, I'm trying to be humble.

814
01:08:27,520 --> 01:08:28,920
I don't know everything.

815
01:08:28,920 --> 01:08:40,720
But I think on a basic level, AI leaders should be a visioner and look above the short-term

816
01:08:40,720 --> 01:08:45,800
revenue goals or self-centric goals.

817
01:08:45,800 --> 01:08:51,560
But focus really on understanding a mission and vision of the company or not only the

818
01:08:51,560 --> 01:08:58,720
company but the product or technology or a solution you're building on a global level.

819
01:08:58,720 --> 01:09:04,880
And it is a good thing, I think, to look back to on a day-to-day basis.

820
01:09:04,880 --> 01:09:08,320
And as you said it there, I think you got it completely right.

821
01:09:08,320 --> 01:09:12,720
It's developing that habit of self-regulation.

822
01:09:12,720 --> 01:09:17,880
Not waiting for the government to pass the law and prohibit what you're doing or someone

823
01:09:17,880 --> 01:09:22,760
to sign the letter to pause AI development.

824
01:09:22,760 --> 01:09:31,560
But it's important to challenge yourself on what risks and opportunities my creation is

825
01:09:31,560 --> 01:09:32,960
going to produce.

826
01:09:32,960 --> 01:09:40,880
It's about self-check, self-assessment, self-regulation, the motives.

827
01:09:40,880 --> 01:09:47,200
But also trying to look above that, what would another person or the different motive be

828
01:09:47,200 --> 01:09:51,120
able to do with what I'm creating.

829
01:09:51,120 --> 01:09:57,040
I think that's the basic thing that I would think of if I were to design a unique AI

830
01:09:57,040 --> 01:09:59,960
patent, AI solution, for example.

831
01:09:59,960 --> 01:10:06,040
Yeah, that's a really important point that you made that it's important for anybody

832
01:10:06,040 --> 01:10:12,840
creating a technology to have a foresight of how else can this technology be used.

833
01:10:12,840 --> 01:10:22,560
Because it's baffling that a lot of times scientists and engineers are so focused on

834
01:10:22,560 --> 01:10:28,160
that their technology is answered to a specific set of problems.

835
01:10:28,160 --> 01:10:35,360
We don't think about how else, if this gets in the hands of a bad actor, how can this

836
01:10:35,360 --> 01:10:36,640
be used?

837
01:10:36,640 --> 01:10:44,760
So I think adding those guardrails or thinking about it at least will be a really good practice

838
01:10:44,760 --> 01:10:46,760
for engineers.

839
01:10:46,760 --> 01:10:49,320
For sure.

840
01:10:49,320 --> 01:10:53,440
Sweet.

841
01:10:53,440 --> 01:10:56,560
What excites you about current advancements in AI?

842
01:10:56,560 --> 01:11:03,800
I'm sure there's a lot of things that you're chatting to people about and what are you

843
01:11:03,800 --> 01:11:04,800
excited about?

844
01:11:04,800 --> 01:11:07,880
Gosh, well, lots of things, as you said.

845
01:11:07,880 --> 01:11:15,920
How can it not excite someone in one domain or another?

846
01:11:15,920 --> 01:11:20,720
Oftentimes, putting the work aside, the things that we are seeing on a day-to-day basis at

847
01:11:20,720 --> 01:11:26,120
label your data, these tremendously interesting AI applications and things that people are

848
01:11:26,120 --> 01:11:30,680
doing around the globe, I think in the end I'm getting more and more excited about the

849
01:11:30,680 --> 01:11:32,720
day-to-day life.

850
01:11:32,720 --> 01:11:37,080
That's how it challenges the current state of things and the human nature.

851
01:11:37,080 --> 01:11:39,360
So just yesterday I had a chat with this woman.

852
01:11:39,360 --> 01:11:48,080
I know she was talking about her 10-year-old son who is created an Instagram page for their

853
01:11:48,080 --> 01:11:56,040
dog and this 10-year-old boy uses chatGPT to write the text to the post, to compose

854
01:11:56,040 --> 01:12:05,320
the content plan for the next couple of months and then to ask, and then he asks, chatGPT

855
01:12:05,320 --> 01:12:11,840
what kind of pictures I should take of my dog to make it an interesting looking profile.

856
01:12:11,840 --> 01:12:13,400
And I was listening to it.

857
01:12:13,400 --> 01:12:19,680
Well, I know that it sounds kind of common right now, but I could do, I probably wouldn't

858
01:12:19,680 --> 01:12:25,960
be able to do as good, and she showed me that Instagram page and I thought, okay, I'm

859
01:12:25,960 --> 01:12:31,280
not going to be able to do that as good as he does it.

860
01:12:31,280 --> 01:12:38,320
So that one thing that I was just a little bit shocked to hear about, realize, right,

861
01:12:38,320 --> 01:12:46,520
because I hear these stories, lots of things, but when you realize it, you know, it's different.

862
01:12:46,520 --> 01:12:54,680
Another thing I think is that what excites me is how not specific developments in AI

863
01:12:54,680 --> 01:13:03,160
or, you know, like models that are deployed or presented to the world, but I'm always

864
01:13:03,160 --> 01:13:08,000
thinking about how this can challenge the way things are right now.

865
01:13:08,000 --> 01:13:15,360
So for example, challenge of workplaces, but not only like the number of workplaces, right,

866
01:13:15,360 --> 01:13:22,440
but how we will be changed, how our agility will be changed.

867
01:13:22,440 --> 01:13:30,040
We hear it a lot from Yvonne Noaharari, right, that people will have to stay flexible and

868
01:13:30,040 --> 01:13:40,200
agile at ages of 40, 50, 60 to learn new skills, because again, we're moving at such a high

869
01:13:40,200 --> 01:13:47,800
speed that no one really knows what you're going to need in terms of the skills in the

870
01:13:47,800 --> 01:13:48,800
future.

871
01:13:48,800 --> 01:13:53,480
And then sometimes I'm even thinking about my future kids, right, I don't have kids right

872
01:13:53,480 --> 01:13:58,520
now, but let's imagine four years ago when I was thinking about it, I thought, okay,

873
01:13:58,520 --> 01:14:05,680
maybe my kids would go and study mathematics and, you know, like computer science, so that,

874
01:14:05,680 --> 01:14:12,320
you know, my imaginary is done, so that by, you know, because I know that it's kind of

875
01:14:12,320 --> 01:14:16,560
a thing right now, so he will be successful in his life with basic thoughts, right.

876
01:14:16,560 --> 01:14:20,480
Of course, there is a lot of other things involved, but still basic things.

877
01:14:20,480 --> 01:14:25,240
I'm thinking about what kind of skills were needed at that time.

878
01:14:25,240 --> 01:14:32,440
But right now, like four years later, I'm pretty convinced it's going to be about creative

879
01:14:32,440 --> 01:14:38,720
skills rather than some hard skills, design thinking, creative problem solving.

880
01:14:38,720 --> 01:14:43,920
And now I'm not sure, you know, what in the next four years time spent, I think would

881
01:14:43,920 --> 01:14:46,240
be a necessary skill.

882
01:14:46,240 --> 01:14:56,360
So it's that challenge to really make us more agile as people, as humans.

883
01:14:56,360 --> 01:15:02,160
It's something that I think, you know, challenge of philosophical questions, the basic question

884
01:15:02,160 --> 01:15:08,880
gets me all excited about what I have about AI in general.

885
01:15:08,880 --> 01:15:13,440
I'm putting aside all of the amazing things that are currently happening.

886
01:15:13,440 --> 01:15:21,360
So things like Sora, Jennie and artists and, you know, implementation of AI in industries

887
01:15:21,360 --> 01:15:30,200
previously where we couldn't think, like just recently I had a call with Oil and Gas Company,

888
01:15:30,200 --> 01:15:37,640
they are implementing AI, they're implementing to predict it or to analyze the data that

889
01:15:37,640 --> 01:15:41,680
they have to create business intelligence tools.

890
01:15:41,680 --> 01:15:44,960
So data that they have been gathering for the past 10 or 20 years.

891
01:15:44,960 --> 01:15:49,520
This is the kind of the industry I thought it's just, you know, a couple of years ahead

892
01:15:49,520 --> 01:15:52,880
there, no, it's right there.

893
01:15:52,880 --> 01:16:00,320
As you mentioned it early, disrupts industries, and as I said, it disrupts human nature.

894
01:16:00,320 --> 01:16:08,520
So you know, it gives me chills, but also super excited to see this exciting time to

895
01:16:08,520 --> 01:16:09,520
live in.

896
01:16:09,520 --> 01:16:14,840
So exciting that sometimes you think I'd love to live in some boring times because no, it's

897
01:16:14,840 --> 01:16:19,800
too much exciting when the boring time is coming because I want to just, you know, chill

898
01:16:19,800 --> 01:16:20,800
down a bit.

899
01:16:20,800 --> 01:16:21,800
I know.

900
01:16:21,800 --> 01:16:25,240
I always talk about having a week where nothing happens.

901
01:16:25,240 --> 01:16:29,840
Maybe, you know, I need a boring week, but no, never happens.

902
01:16:29,840 --> 01:16:36,120
But I mean, you made some great points there.

903
01:16:36,120 --> 01:16:43,680
Maybe with a will to do something or, you know, that wants to create something can do

904
01:16:43,680 --> 01:16:49,840
it, can do that wants to complete a project, can do it like you gave a great example of

905
01:16:49,840 --> 01:16:56,360
a 10 year old, you know, using generative AI to do a lot of things.

906
01:16:56,360 --> 01:17:03,320
And we also have, I was talking to Robert Brill, a media guru on the podcast a few weeks

907
01:17:03,320 --> 01:17:05,600
ago as well.

908
01:17:05,600 --> 01:17:17,920
He talked about that he's excited about seeing a single person company that's evaluated at

909
01:17:17,920 --> 01:17:18,920
a billion dollar.

910
01:17:18,920 --> 01:17:22,040
Yeah, that's not true.

911
01:17:22,040 --> 01:17:28,760
Because it's democratized and the productivity gain is so much that there will be a tipping

912
01:17:28,760 --> 01:17:35,760
point where there will be, you know, a solo entrepreneur will create a billion dollar

913
01:17:35,760 --> 01:17:37,280
business.

914
01:17:37,280 --> 01:17:40,240
And what's more exciting, it can be anyone, right?

915
01:17:40,240 --> 01:17:41,240
Exactly.

916
01:17:41,240 --> 01:17:46,960
Because you don't have to, you know, be born in a particular country or go to a particular

917
01:17:46,960 --> 01:17:49,960
university or being surrounded by a particular set of people.

918
01:17:49,960 --> 01:17:51,480
Of course, everything is important.

919
01:17:51,480 --> 01:17:53,000
All of this stuff is important.

920
01:17:53,000 --> 01:17:58,720
But yeah, as I said, democratization of these things and it could be anyone, which means

921
01:17:58,720 --> 01:18:05,560
that it exposes us to such a huge talent pool out there, which we previously hadn't seen

922
01:18:05,560 --> 01:18:13,680
into so many other perspectives to the world problems, challenges and other situations.

923
01:18:13,680 --> 01:18:19,960
Yeah, I think as I think about it, like, like, like you said, hard skills, like coding, I

924
01:18:19,960 --> 01:18:26,320
mean, I can talk to chat GPT or these gen AI tools and create code.

925
01:18:26,320 --> 01:18:35,040
But as I've been thinking about this, one of the things that became just more important

926
01:18:35,040 --> 01:18:43,640
in the world is your English speaking or writing skills, right?

927
01:18:43,640 --> 01:18:51,360
Because a lot of these AI tools are being developed in English, more so in English than

928
01:18:51,360 --> 01:18:52,840
in other languages, right?

929
01:18:52,840 --> 01:18:58,920
So because I've come from India where there's a ton of different languages.

930
01:18:58,920 --> 01:19:08,640
So I'm acutely aware of importance of language and importance of English as a global language.

931
01:19:08,640 --> 01:19:17,160
And I think we're in the middle of the time in history where English just became a little

932
01:19:17,160 --> 01:19:19,960
more important than it already was.

933
01:19:19,960 --> 01:19:28,760
True, but don't you think it in a way became less important with the development of the

934
01:19:28,760 --> 01:19:31,720
technologies that we have, right?

935
01:19:31,720 --> 01:19:35,560
Do you sometimes you don't necessarily have to know?

936
01:19:35,560 --> 01:19:40,720
Well, I think for the display right now, maybe not so much, right?

937
01:19:40,720 --> 01:19:49,640
But it's going to a way where, I don't know, you just, my mom moved to the US almost a

938
01:19:49,640 --> 01:19:57,960
year ago with zero knowledge of English and how to speak English.

939
01:19:57,960 --> 01:20:00,120
She lived there successfully for a year and a half.

940
01:20:00,120 --> 01:20:05,720
She goes to a bank, she puts up her phone, translate her and she solves her things perfectly.

941
01:20:05,720 --> 01:20:10,360
So it's just a matter that you have to pull it up and to say something and then to wait.

942
01:20:10,360 --> 01:20:16,680
But I know there are so many technologies that kind of bridge that gap already.

943
01:20:16,680 --> 01:20:19,800
So yeah, well, that's an interesting topic to discuss as well.

944
01:20:19,800 --> 01:20:21,600
Yeah, that's a good point that you made.

945
01:20:21,600 --> 01:20:28,280
So it's definitely going to be more accessible to people who doesn't know English as an end

946
01:20:28,280 --> 01:20:36,880
user, but as a developer of technology and data, maybe English is more, I need to think

947
01:20:36,880 --> 01:20:42,640
about it a little more, but yeah, that's a really well made point.

948
01:20:42,640 --> 01:20:43,840
Cool.

949
01:20:43,840 --> 01:20:52,720
And this is one of our key staple questions on this podcast.

950
01:20:52,720 --> 01:21:01,280
If you had a magic wand and time and resources were not constrained, what is the AI model

951
01:21:01,280 --> 01:21:06,200
and data models that you would build and have access to?

952
01:21:06,200 --> 01:21:12,800
Oh gosh, when I was thinking about that question, I spent a decent amount of time thinking about

953
01:21:12,800 --> 01:21:13,800
it.

954
01:21:13,800 --> 01:21:21,240
Spoiler alert, haven't come up with some fascinating answer or something unique, but I thought

955
01:21:21,240 --> 01:21:23,640
about it in two different perspectives.

956
01:21:23,640 --> 01:21:30,920
One, self-centric, I think that I'd be really excited about.

957
01:21:30,920 --> 01:21:32,320
I'm not a scientist myself.

958
01:21:32,320 --> 01:21:39,920
I'm not a physicist or an astrophysicist myself, but I think I'd love to do something

959
01:21:39,920 --> 01:21:45,400
related to prediction or exploring the outer world out there because I think as much as

960
01:21:45,400 --> 01:21:50,640
AI challenges us today, that will challenge us too.

961
01:21:50,640 --> 01:21:59,120
We have been challenged by what we've seen in terms of how small we are, how really we

962
01:21:59,120 --> 01:22:03,880
are there because of a coincidence of lots of different things and how we continue to

963
01:22:03,880 --> 01:22:07,440
live on because for some reason things are the way they are.

964
01:22:07,440 --> 01:22:19,280
So I think the development in going above what we currently see and know is exciting.

965
01:22:19,280 --> 01:22:25,000
And I oftentimes like to think, I'm not a conspiracy theorist myself, but I often like

966
01:22:25,000 --> 01:22:33,880
to think that people know something that is so grand that it might be just a little bit

967
01:22:33,880 --> 01:22:38,120
too much to think about as a single person not knowing certain concepts.

968
01:22:38,120 --> 01:22:44,280
So given that I have no constraints and resources and money, that would be something I'd explore.

969
01:22:44,280 --> 01:22:50,640
But other than that, I mean, I think I cannot pick one problem I'd solve.

970
01:22:50,640 --> 01:22:59,840
There are so many problems out there in the world that need immediate solutions, including

971
01:22:59,840 --> 01:23:06,640
wars, right, that's something that is my pain point right now, including environmental

972
01:23:06,640 --> 01:23:10,200
changes, famine and all these things.

973
01:23:10,200 --> 01:23:18,560
So it wouldn't definitely be one thing, but making the world a better place in any way

974
01:23:18,560 --> 01:23:26,480
I can would be definitely something I'd strive for, given that I have access to all the amazing

975
01:23:26,480 --> 01:23:30,000
technology and all the money around the globe.

976
01:23:30,000 --> 01:23:35,280
Well, that's some sentimental answer, but that's one thing I really like to the core

977
01:23:35,280 --> 01:23:36,760
of it.

978
01:23:36,760 --> 01:23:38,560
I think that's what I'd do.

979
01:23:38,560 --> 01:23:40,160
No, absolutely.

980
01:23:40,160 --> 01:23:42,360
And you're already doing a lot of it, right?

981
01:23:42,360 --> 01:23:48,000
So you are running a company and making the world a better place.

982
01:23:48,000 --> 01:23:49,400
It's really fascinating.

983
01:23:49,400 --> 01:23:59,400
I mean, I do share your sentiment on space, Neil deGrasse Tyson, the director of Hayden

984
01:23:59,400 --> 01:24:11,680
Planetarium in New York, he always talks about how if you look at things from a cosmic perspective,

985
01:24:11,680 --> 01:24:20,120
problems of our world become so tiny or you have a different perspective, we don't need

986
01:24:20,120 --> 01:24:21,560
to be fighting.

987
01:24:21,560 --> 01:24:26,640
We don't need to be just because the colors of our skin are different.

988
01:24:26,640 --> 01:24:28,160
We're still the same species.

989
01:24:28,160 --> 01:24:29,680
We're still humans.

990
01:24:29,680 --> 01:24:36,040
You know, everything about we have more things in common than not.

991
01:24:36,040 --> 01:24:43,320
If an alien looks at both of us, maybe you're almost the same to me.

992
01:24:43,320 --> 01:24:49,080
So I think the differences and the wars and everything kind of comes puny.

993
01:24:49,080 --> 01:24:56,520
So I think developing the space consciousness and how AI can help us get there is super

994
01:24:56,520 --> 01:24:58,720
fascinating to watch.

995
01:24:58,720 --> 01:25:00,720
I agree.

996
01:25:00,720 --> 01:25:01,720
Awesome.

997
01:25:01,720 --> 01:25:05,560
No, thank you so much, Karina.

998
01:25:05,560 --> 01:25:06,560
It's been a pleasure.

999
01:25:06,560 --> 01:25:10,760
I appreciate your time and I learned a lot.

1000
01:25:10,760 --> 01:25:12,120
Thank you very much as well.

1001
01:25:12,120 --> 01:25:14,080
The pleasure is all mine.

1002
01:25:14,080 --> 01:25:16,920
As I said, I was super excited about the chat.

1003
01:25:16,920 --> 01:25:21,760
I'm super excited right now because I got a chance to talk about things that I'm passionate

1004
01:25:21,760 --> 01:25:22,760
about.

1005
01:25:22,760 --> 01:25:26,720
And I think it ended in a very good note.

1006
01:25:26,720 --> 01:25:29,640
In the end, we're all humans.

1007
01:25:29,640 --> 01:25:36,680
There are things that matter, right?

1008
01:25:36,680 --> 01:25:37,680
We have to remain humans.

1009
01:25:37,680 --> 01:25:38,760
We have to remain humble.

1010
01:25:38,760 --> 01:25:43,240
We have to remain kind to each other and together make the world a better place.

1011
01:25:43,240 --> 01:25:46,880
So that's a good ending, I think, of the conversation.

1012
01:25:46,880 --> 01:25:47,880
Absolutely.

1013
01:25:47,880 --> 01:25:50,440
Hopefully AI technologies can help us with that.

1014
01:25:50,440 --> 01:25:51,440
Absolutely.

1015
01:25:51,440 --> 01:25:53,560
Let's end on a positive note.

1016
01:25:53,560 --> 01:25:55,440
Thank you so much.

1017
01:25:55,440 --> 01:26:01,560
Thank you too.