Latest / Women Disrupting Tech / How Cyberette Battles Deepfakes So You Can Trust What You See with Julia Jakimenko | Ep. 109

Transcript

How Cyberette Battles Deepfakes So You Can Trust What You See with Julia Jakimenko | Ep. 109 · Women Disrupting Tech

0:00
Let's move back a bit to the self promo and why it is
0:03
difficult instead. Do you think it's difficult for
0:06
female founders more in general because to?
0:09
Build a company. Not to to build a company, but
0:12
also to do promotion for the company.
0:14
Definitely. I feel like from what I've again
0:17
observed from my experience, what I've seen with my female
0:20
colleagues, they feel more pressure to prove that they're
0:25
as capable as their male colleagues, right?
0:28
I, I feel like we have watched a bit more closely how we speak,
0:31
how we dress, how we show up, do we smile enough?
0:34
Are we nice enough how present we are?
0:38
So, you know, we're judged more on the tone, on our appearance,
0:43
on the attitude. And actually, if I've personally
0:46
experienced that because I've received comments, I need to
0:48
smile more before. So, well, actually competence
0:52
alone isn't often enough for women.
0:54
So I think that that adds a layer of pressure as well.
1:10
Welcome to women disrupting tech my name is dear Kanupkas and on
1:14
this podcast I give the women who work in tech the platform
1:17
they need to inspire and empower each other.
1:21
Join me on a journey into the world of entrepreneurship,
1:25
funding biases and role models to discover the strategies for.
1:29
Success in the tech industry. In other words, how women can
1:34
disrupt tech. Are you ready to listen?
1:43
Hey, welcome back to Women Disrupting Tech, and thank you
1:47
for tuning in to episode 109. My guest today is Julia
1:52
Yakimenko. Julia is the founder of Cyber
1:55
Ed, a startup that's building a platform to detect deepfakes
2:00
across video, audio and images. But as you'll hear, she's doing
2:05
much more than that. Julia Yakimenko studied human
2:09
rights law. She shifted into tech because
2:13
she saw the potential of technology to simplify, support,
2:16
and protect people's lives. And now she's on a mission to
2:20
restore trust in what we see and hear online.
2:25
In this episode of Women Disrupting Tech, Julia and I
2:28
talk about what makes deepfakes so dangerous, how you can spot
2:32
them, and why building transparency into AI isn't
2:35
optional. It's urgent.
2:39
What I loved. Most about our conversation is
2:41
that Julia not only goes deep on the tech part, but also shares
2:45
her take on what it means to be a female founder in
2:48
cybersecurity. She opens up about the
2:51
challenges of self promotion, how she's redefined success for
2:55
herself, and we cover the emotional work behind staying
2:59
true to your vision, especially in the space where for female
3:03
founders, competence alone is not enough and where the number
3:07
of role models is still too small.
3:11
So when you're ready to listen and learn how Cyberrat battles
3:15
deepfakes so you can trust what you see, please welcome to women
3:19
disrupting. Tech Julia Yakimenko Welcome to
3:24
Women disrupting. Tech Julia, good to have you
3:28
question. Question I always ask at the
3:32
start is how you got into your current job or current position
3:35
and as the founder of Cyber IT? That is a good question.
3:41
There's a long answer and then there is like a short answer.
3:44
Which one do you prefer? Well, I prefer the loan because
3:48
this is going from content to our.
3:51
OK, so I started completely in a different industry because cyber
3:56
rat is it's a tech company, right?
3:59
It's a cybersecurity tech company.
4:01
But I started my career in law. My first degree was law and I
4:07
was planning to become a lawyer, very passionate about it.
4:10
I was very interested in law and European Union law and actually
4:15
human rights law. That was my favorite subject in
4:19
my bachelor's degree. So and then I moved to England.
4:25
I'd completed my law degree, European Union law degree in
4:29
Poland, moved to England, and then I started working in HR and
4:34
compliance. It's more business oriented, but
4:38
still legal. And I ended up working in a lot
4:43
of tech companies, startups, and that's it.
4:47
OK. There, there was no way back for
4:50
me. The once I've seen the tech once
4:52
I've seen what systems can do, how they can automate, how they
4:58
can simplify our lives. There was no other any other
5:02
industry for me anymore because I was just that's it.
5:05
I found myself. I want to be in tech Yeah.
5:08
And just, I guess I manifested it in a way as well a little
5:13
bit, but then I also sort of tried to shift my career.
5:17
COVID hit and then. So that is a good time to change
5:21
careers into tech. I was already working with
5:24
systems, but I was on the user end rather than actually
5:30
development sites. And then I moved to the
5:34
Netherlands, started working at Rabobank, and yeah, a few years
5:38
later I ended up finding Cyberat.
5:43
OK. And what does cyberat do?
5:46
Good question as well. So we are building pretty much a
5:49
platform to detect deepfakes across video, audio and images.
5:55
Very simple. Well, very simple, but it
5:58
sounds, yeah, it sounds really simple when you pronounce it
6:00
like that. But it sounds also to me like
6:03
really technically complex, because I think I read somewhere
6:07
that 8 out of 10 people cannot detect deep face.
6:12
So how can you? How can you deploy tech to do
6:15
that? Yeah.
6:15
So we actually, you're right, we have quite complex, quite
6:19
advanced technology behind it. So we build our own deep
6:25
learning models and we use multiple combined multiple
6:29
detection approaches. So it's like pattern based fraud
6:32
detection, neural forensics analysis and then provenance
6:37
verification. Many more approaches.
6:40
OK, so. So there is actually advanced
6:43
tech behind it, yeah. So I won't go into all the all
6:48
the terms that you just mentioned because to to keep
6:51
things simple, but it wouldn't be fair to say that you want to
6:55
bring trust back to the Internet or enable.
6:59
Trust on the Internet with that. Yeah.
7:04
So our goal is not really like to stop on AI, but just to bring
7:09
that confidence back to the Internet, keep people and
7:14
companies confident and what they're seeing online and help
7:20
help to set that boundaries around how AI generated content
7:25
is flagged, identified and responded to.
7:30
So yeah, I guess the goal is to make that verification process
7:34
of the media a default. You know how you have 100% you
7:39
have like utilizing 2 factor authentication or antivirus.
7:44
So what we want, we want to, in an ideal world, make that
7:50
default for the media as well. So any content that you see on
7:54
social media, online or that companies, organizations,
7:58
governments share should be verified and people should see
8:03
that that there should be a level of transparency around it.
8:07
So people can make the distinction between what is real
8:11
and what is deep fake. Exactly in the most approachable
8:14
way. Possible.
8:16
So. So maybe can you clarify for us
8:19
what types of deep fakes there are and what makes it difficult
8:24
for people to, let's say, distinguish them from real
8:28
content? So what is deep fake?
8:30
Deep fake? It's either AI generated or AI
8:34
manipulated or changed by the AI media, which is image.
8:40
It can be voice recording, it can be a video that you see
8:44
online. It can be a phone call.
8:46
So fake video or like a phone call or audio recording.
8:52
It can damage your reputation on like practically overnight, but
8:57
it can also be utilized by scammers to get someone to send
9:02
you money to trick you into sending money.
9:04
They can easily pretend to be someone's daughter or
9:06
grandparent and then ask for help using different techniques
9:11
like push in urgency and then get you to share your personal
9:16
sensitive information and that can be seen online that can be
9:21
used in a phone call. So there are different ways to
9:24
do it. They can clone the voice and
9:26
they only need like 30 seconds of someone's voice to be able to
9:30
clone it. OK.
9:31
So basically what I'm hearing you say is that, for instance,
9:36
the the CEOCFO fraud that would be done via e-mail is now moving
9:42
to phone calls because they can actually clone your voice as
9:47
ACEO or as ACEO. And then if they have a little
9:50
clip or 30 seconds of your voice, they can, well, they
9:54
could of course spoof as they call it, the OR mask the number
10:01
that they're calling from with the number of that particular
10:05
person. And then they call you to say,
10:09
hey, you need to wire that money and you need to wire it now.
10:14
Yeah, correct. It's a form.
10:17
Yeah, exactly. That's one of the forms.
10:19
It's not the only form, but that is one of the forms.
10:22
And what they often do, they try to get you off e-mail or off
10:26
Teams, off Zoom, try to get you into WhatsApp, into phone call,
10:30
because that is easier for them to clone.
10:32
There's still no perfect way to clone video real time to go on
10:39
like a FaceTime call, but on The Voice call they can do it.
10:43
So that's actually one of the probably one of the signals to
10:47
watch out for is they're going to try to get you off the e-mail
10:50
into the phone call or yeah, WhatsApp call saying they're
10:55
busy, they're on the move. Can we, can you give me your
10:58
phone number? I'll call you something like
11:01
that. That is.
11:02
Something that definitely employees should watch out for,
11:05
yeah. OK, that's good to know just as
11:08
a questionnaire because I I was browsing the Internet yesterday
11:12
and I saw that Google has the option to create videos from
11:16
images. Is that also?
11:19
A form of deep fake that is deployed.
11:22
That is synthetically generated deep fake.
11:26
That's one of them. So it's not just changing
11:28
cloning. So we have voice cloning, we
11:32
have image altering. You know, all those Photoshop,
11:35
Photoshop like Pixie changes where you put on face filter to
11:40
prettier or whatever you want to see, you know, in in your
11:45
picture. But yeah, also you have those
11:48
where you can generate really cool videos actually.
11:51
Yeah, I was experimenting yesterday with it, but now that
11:55
I hear from you, I'm like, OK, maybe not such a good idea,
11:59
although it's. But that's fine, right?
12:01
Not everything. The AAI is not only for harm,
12:05
it's also, you know, it's also for fun.
12:08
Yeah, true. Yeah, like every technology, I
12:12
think it's sort of it doesn't care if it's used for wrong or
12:16
right. And let's let's take that, take
12:18
that on a bit bit later. So you talked about a few ways
12:22
to to to spot the deepfake like really taking care or being
12:31
mindful that if people want to take you off e-mail into
12:34
WhatsApp, that's yeah, red flag. Are there any other red?
12:39
Flags to that you can use to to spot deep fakes.
12:45
Definitely. So the pressure to act fast,
12:48
that is probably ultimate signal.
12:51
I would say any security awareness trainings will
12:54
definitely mention that. So, and in any sort of social
12:59
engineering, even by e-mail, if you see someone is trying to
13:02
push you to do something fast, that is the first and the
13:05
biggest rut flag. But if we're talking about the
13:09
video or yeah, if we're talking about video deep fakes, just
13:14
look for any like unnatural blinking or weird light that
13:19
falls on the face, on the skin, any stiff facial expressions,
13:25
any weird poses, robotic tones in the voice.
13:29
Also a background noises that do not match the picture that you
13:33
see if you've seen someone at the office but then you see some
13:37
car in the background. Like I don't know some
13:40
construction or actually construction noises is
13:43
realistic. But if I don't know, if you see
13:45
some completely opposite noise to what it should seem like,
13:51
then it's better to ask or question that.
13:55
And that's already like a small flag.
13:57
Why is that noise? It's completely opposite to
14:01
yeah, what I'm seeing. So I think it's just like a
14:03
combination of multiple red flags.
14:07
Then that should raise questions in your head.
14:11
Yeah. We call it zero trust in
14:13
cybersecurity, zero trust approach, zero trust.
14:17
OK, so I'm still of the generation of trust but verify.
14:21
But then your approach is more like don't.
14:24
Yeah, I think millennials are zero trust.
14:27
Bit more skeptic, yeah. Yeah, our our generation is.
14:31
Just to add to that, because if you look at the AI and also at
14:35
video for that matter, the platform that we're recording
14:40
this on Riverside does actually does a pretty good job in
14:44
creating clips, etcetera, and help you with putting together
14:49
really nice looking video clips. So it, it, it, it works really
14:56
fast or it progresses really fast, I should say.
14:59
And also already described how difficult it is to, to spot
15:05
deepfakes, especially if you're not paying attention.
15:07
How can how can we ensure also taking that zero, maybe taking
15:14
that zero trust approach, how can we ensure that we keep using
15:18
AI for good? Well, I would say bringing that
15:22
transparency around the AI because if now we go into social
15:26
media or read some news, the only reason we can't suspect
15:32
that the source is verified is just by looking, OK, this is
15:36
Bloomberg. It must be real, right?
15:38
Or I don't know, some new news daughter.
15:43
Now that must be real, right? But we don't know because they
15:45
also getting their media, you know, from some witnesses or
15:54
journalists and how did they know that what they're sharing,
15:59
it's 100% true unless they've taken that picture themselves or
16:03
you know, taken that video. So I think bringing that
16:05
transparency to the next level, because people should always
16:11
know what's like when content is generated and when it is real,
16:15
especially when it, when, when it comes to politics or
16:19
journalists or something very serious, even general, even
16:23
regulations is not enough. We have AI act now in US.
16:27
We have some bills passing through, I think in California
16:30
around deepfakes. But then in Asia, there's still
16:34
nothing, right? And we read Asian news, right?
16:37
So we still, we Earth, Earth is not that big.
16:40
We're still all, you know, open to, you know, absorbing new
16:45
sources from all over. And so I think having
16:48
standardized regulation globally, that would really
16:52
help. And of course, clear boundary
16:54
boundaries, part of that regulation, how AI can be used
16:58
to manipulate everyone and deceive people.
17:02
I mean, these days, personal data is not personal anymore.
17:06
Any like social media platforms, they just have access to
17:09
anything you signed up on, like, you know, that's how we all
17:12
know. But now also we have AI models
17:14
that can just with a tiny bit of input, they can guess things
17:20
like your location, your income, your gender, and yeah, so we all
17:26
watched Black Mirror. So we actually entering the era
17:31
where AI can learn human behavior, predict and influence
17:37
the decision making process. So yeah, I think the focus
17:39
should be. How do you look?
17:41
At that yeah so it's I wouldn't I wouldn't say we should panic
17:48
and be scary we should definitely utilize the potential
17:51
of AI for good because there's so much good it can bring you
17:54
know in Med tech and in longevity as well helping people
18:00
you know especially like in medical fields.
18:02
And then we're talking about data, we're talking about
18:04
genetics. So it has a lot of views, right?
18:07
It can also help us progress and all kind of developments.
18:11
That's also I would say. So the way I look at it that we
18:16
just need we should, you know, benefit and enjoy the potential
18:20
that AI can bring while also, you know, bringing that
18:23
transparency regulation and enforcing those regulations
18:28
where there's clear boundaries between what can we can do and
18:33
what we cannot do. So I think just that openness,
18:37
transparency and enforcing the boundaries would definitely
18:41
help. So the focus should not be on
18:43
like building hype, you know, but also to protect people,
18:48
support people and bring that long term stability.
18:51
Yeah, and teach people how to use it.
18:53
And that as well, yeah. And that that's.
18:57
Probably at the front of it all. Yeah, I think that's, that's
19:00
really important to to especially teach people how
19:04
things work, what they see, how to detect deep fakes, but also
19:08
how to deploy AI for really good causes because they are, as you
19:14
said, a lot of them. So is cyber and AI.
19:18
Are those still very male dominated fields?
19:21
Work fields? Yeah, still quite male dominated
19:26
industry, I'd say. Yeah.
19:28
OK. So would it then also be good to
19:33
have more women in those fields? Also to make sure that we're
19:38
mindful of the dangers, that we just not go ahead blindly
19:42
because it's good. But then also really think about
19:46
the ethics part of it, the impact part of it.
19:49
Definitely. I mean we all know what gender
19:52
diversity brings so and how it helps to build systems.
20:00
We need diverse, you know, points of view, diverse
20:05
backgrounds to build the best, most efficient systems.
20:08
So of course. Of course, gender diversity,
20:12
cultural diversity, it definitely helps to diversify
20:17
and bring that aspect into cybersecurity 100%.
20:23
But I, I, I wouldn't just see it as cybersecurity, I would see it
20:26
as tech in general. Tech itself, as an industry,
20:31
it's expanding. We see that digitalization in
20:35
more small aspects of our everyday life.
20:39
So it's not just tech, it's just integrated in our daily life.
20:43
And I mean, what, roughly 50% of population is women, so it's 50%
20:50
of customer. Yeah.
20:52
But do you also see that because you've been working in tech for
20:56
quite a quite some time, do you also see that change?
21:00
So do you see more women come into tech, for instance as a
21:04
gabo or? Definitely, if organizations are
21:09
more conscious of diversity and inclusion and actually in their
21:14
conscious and integrating that into the whole workflow into
21:19
completely changing everything you know from the start.
21:22
And yeah, if if diversity and inclusion is not just a policy
21:27
and a box they took, but actually they are generally
21:31
interested in it and they see a benefit of it.
21:34
So yeah, then I definitely have observed that change.
21:37
Let's say Rabobank, certainly they put a lot of effort into
21:41
diversity and inclusion. I think it's like one of the few
21:45
employers that I work for where I've really seen, you know,
21:50
genuine interest, genuine dedication to bringing that
21:55
diversity and inclusion into the company, not just on a policy
21:59
level. Yeah.
22:00
Out of curiosity, how does that show?
22:02
Well, can you give an example? Of course, it's just it's
22:07
enabling initiatives. So for example, like they have
22:11
women in tech community and setting that up.
22:15
I wouldn't say it was difficult because everyone was very
22:18
excited about it, passionate. A lot of people are willing to
22:22
join and just do the work that needs to be done.
22:26
Where what I've heard from some of female colleagues at
22:30
different organizations where they face more hurdles, you
22:34
know, setting up a community from from above HR trying to
22:39
sort of manage them a bit and slow them down.
22:43
Where I have not experienced any of that at the Rabobank,
22:46
everyone was actually enabling, pushing, given space.
22:49
And so that that was just the most basic example that I can.
22:55
So you you received a lot of support also in practical terms.
22:59
100%, yeah. OK.
23:02
Do you? Experience the same as as a
23:04
founder of Cyber Ed. As a founder, you're bit more
23:06
isolated, so I had to consciously build a community of
23:10
people around me or join communities.
23:14
I, I see a lot of, I see a huge gap, gender gap in, in
23:19
entrepreneurship, in tech entrepreneurship.
23:21
And actually in the last four or five years, there's been a drop
23:28
in female founders versus before 2020.
23:32
I've actually spent a whole day researching the other day and I
23:35
was a bit concerned. So actually before 2020 has been
23:41
a bit higher number of female entrepreneurs and in Europe
23:44
especially and that number dropped.
23:45
And I started looking more into the data and I've seen the
23:50
reason. The reason behind it is that
23:53
women receive a less funding at a startup stage.
23:57
It's only up to 2% globally. Regions like Europe, I think
24:01
it's a one point half percent if I'm not mistaken, but it's up to
24:04
2% globally. Of all VC funds, you know of all
24:07
VC money women receive, which is.
24:10
Actually, well, now less than 1% globally and less than half a
24:14
percent in all lands. Wow.
24:18
It has dropped and there are there are several factors that
24:21
play a role in this. So funding is one of the causes
24:24
that there are less female founders.
24:27
What are other causes? I think another reason is just
24:32
because there's much less role models, so women struggle to
24:37
push through in between start up into scale up.
24:41
So that's where I've seen the gap.
24:44
So when women move from start up into scale up, there's a huge
24:48
drop in female founders that's make it through and that's.
24:52
So that make it to the second stage.
24:55
Yeah, make it to the second stage.
24:57
Yeah, exactly. So that's where the the issue
25:02
because, yeah, you get the first initial investment, maybe
25:05
friends and family maybe, yeah, maybe not VC itself.
25:09
It doesn't always make sense to get VC at the very early stage.
25:12
Maybe you get some grants, but then at the next stage, actually
25:17
women then that their women struggle to get funding, get
25:21
support, Yeah, there's not enough role models.
25:24
Maybe they feel isolated and I have to say every now and then I
25:29
do feel isolated as well. But I'm so consciously not wired
25:33
that I know OK, I have my best friend, she's very successful
25:37
entrepreneur. So I can text her, call her
25:39
anytime, 24/7. And then I'd say most of my
25:43
close friends are entrepreneurs or either ex entrepreneurs or
25:47
current entrepreneurs. Maybe not all of them are tech
25:51
entrepreneurs, but they all entrepreneurs.
25:53
So I'm just surrounded. But then even even then, I feel
25:57
every now and then isolated. There are communities.
26:01
So I think that is very important to be part of some
26:05
community and you know, go to the meet ups and have those
26:10
people around you that support you for sure.
26:14
How How difficult is self promotion?
26:17
That is challenging. And actually, funny thing, the
26:19
other day I went to the amazing event and there was a dinner and
26:24
someone was sitting across the table, someone that works at, I
26:28
think an investment firm. And he of course found out I'm a
26:33
founder. And he asked me the first
26:34
question. So tell me why your product is
26:37
better. And then I started giving them
26:39
him reasoning like, OK, so it's not a straightforward answer.
26:44
And I started going into like a bit of statistics and explaining
26:48
the product and he's like, you know, only females give me that
26:52
kind of answer. When I asked the same question,
26:54
when I asked a male the same question, they go straight away
26:57
into, oh, we're the best here, we're the best there, we're the
27:00
best there. And only female tries to be not
27:04
just more analytical, but tries to give that, you know, ego less
27:09
answer. And that made me think it's just
27:11
on such a core level where I don't want to go like as a woman
27:18
right into self promotion. I want to try to stay humble and
27:21
also just give benefits of the products and also give the whole
27:25
background, you know, help him understand the whole industry
27:29
products, so then he can understand how that looks.
27:33
Yeah. So that that was very
27:35
interesting. So 2 questions.
27:37
So. So if I understand you
27:38
correctly, what he basically told you is give me less less
27:42
context. I think, yeah, what he was
27:44
trying to tell is don't. He mentioned that.
27:48
Don't be afraid to put on a bit of ego.
27:52
That is OK. OK, so how did that make you
27:55
feel? Oh, that is a great question as
27:58
well. So I agreed with him partially,
28:03
and then partially it just made me question like, why that was
28:07
not my first reaction. Why did I not right away started
28:11
giving the benefits? And I thought, and then my next
28:15
thought was, well, because something that I'm building is
28:18
bigger than me, is bigger than you or him, You know, it's for
28:24
the world. So I want him to see that, you
28:27
know, from my standpoint rather than just, you know, giving a
28:32
sales pitch. You know, all our accuracy is so
28:34
high. We have descendants approaches
28:39
that no one else has, detection approaches that no one else has.
28:42
So I think, yeah, I think it is just having that mission above
28:47
anything else. And that's how I'd see that.
28:50
And maybe not everyone is like that.
28:54
Yeah, true, but let's do some shameless self promo then.
28:59
So why is why is Cyberrat better?
29:02
What makes it better? What makes it unique?
29:04
Let me put it that way. We have explored multiple
29:09
detection approaches that I've mentioned earlier.
29:12
I think So. What we're trying to build is
29:17
all in one defect detection platform.
29:22
So we don't just take one architecture and that's what I
29:27
know quite a few competitors do. They have one architecture that
29:32
they use and a couple of detection approaches.
29:35
We explore multiple detection approaches and our accent is on
29:40
forensics, forensics detection and giving that forensic
29:45
reporting, forensic explanation and comprehensive overview of
29:52
what is real, what is not. So it's not as simple as OK, the
29:56
image, the video is real or not. It's giving the full
30:00
explanation, the background behind it.
30:02
And we're going as deep as giving the metadata information
30:07
where, for example, the media was created, maybe even details
30:12
like date, like locations. So it's not high level insights,
30:17
but we focus on deep level insights while also ensuring
30:23
that we diversify our detection methods because Gen.
30:27
AI is improving. And with I would say the light
30:33
speeds, every day I open my Instagram and then I see some
30:38
new Chinese, you know, Gen. AI.
30:41
So we build our own data sets from the scratch.
30:45
So we have a whole team of people that are sitting and
30:47
building data sets and we make sure that we our data sets
30:51
quality is impeccable. So that is also a huge part in
30:56
in building a great product. Your data sets have to be
30:58
impeccable. And we have privacy first
31:02
approach as well. So there are multiple factors
31:06
Going back to that question that you asked, it's not about why we
31:10
are the best, it's about how we build a great product that is
31:15
very good quality product and it's, you know, meets the needs
31:19
of our customers. OK.
31:21
So it's more holistic in terms of the detection, but also in
31:27
terms of approach, it gives more analytical data to users.
31:34
So they can actually can they actually use that?
31:37
How can they use that data? Because if you give more data,
31:40
there's probably use case for that data with customers.
31:44
Exactly. So those kind of reports,
31:48
they're very helpful in forensics investigation.
31:51
It's aimed at the government's law enforcement, public sector
31:57
defence customer who really want to see, you know, the nitty
32:01
gritty details. I know what we offer on our our
32:04
dashboard as well is multiple views.
32:07
So whether you want to see more technical details or if you want
32:11
to test different approaches, we give that option that you can
32:14
see how that how different detection approaches look like
32:19
actually visually because I am very visual person.
32:22
So I believe and actually testing everything with my own
32:27
eyes. Yeah.
32:29
And the last thing I heard you say is that you focus on
32:33
building like an impeccable data set.
32:36
Maybe let's dive a bit deeper into why that is important, both
32:42
from a detection point of view, but also from a yeah, from a
32:45
customer point of view. Well, from the technical point
32:49
of view, actually now if we just make a little step back, having
32:55
a great data set is probably about 80% of your win.
32:59
Well, maybe I'm being a bit yeah, generic, but I would say
33:04
it's 70 to 80% of win in building a great detection
33:09
product, at least in our industry.
33:12
Because the better your data sets are, the better you can
33:16
train your models to, you know, differentiate and spot those
33:21
differences between real and what is real, what is vague and
33:25
spot those anomalies and patterns.
33:29
So actually those data sets, well, I foundation 2 great
33:34
models. So that that is, that is one of
33:36
the reasons. And of course our job is to
33:41
maintain them and update them on a regular basis.
33:45
So it's a lot of work, but it's a fun work.
33:48
We enjoy it. OK, well, that's important to
33:52
enjoy the the work that you're doing.
33:54
At least in my experience, it keeps you going.
33:57
Definitely, yeah. And of course, from the customer
34:00
perspective, we want to deliver clean, high quality product and
34:05
we want to make sure that our product is privacy first as
34:08
well, hence, you know. So why is that important?
34:11
Because we, we, we, you just highlighted.
34:14
So your customers are mainly law enforcement, law firms, forensic
34:20
firms who want to pick part. Let's say we need more data to
34:26
to detect deepfakes or is it is your product or that's.
34:29
Not our main customer, but that is a customer for whom we build.
34:35
We're building the forensics dashboard for sure, but we have
34:40
a more user friendly versions for media or maybe a bit less
34:45
techie type of customer. So it's not all very technical,
34:49
but we have an option for it to be more advanced.
34:52
OK. Because I was the like the more
34:56
technologically advanced people usually care a lot about today
35:02
getting as much data as possible and maybe less about privacy
35:08
because especially in law enforcement, you want to enforce
35:11
law. Correct.
35:12
Yes, but also. There's the balance with
35:16
privacy. DRS is different from, let's
35:20
say, if you're a regular company where you at least have to want
35:24
to report privacy rights of your employees and customers.
35:28
Exactly that is the case, yes. But also what I also tried to
35:33
say with the privacy first approach is it's really
35:36
important to me make sure that the way you collect data to
35:40
train your models, because we all know, you know, we put the
35:42
personal data into ChatGPT and that's not your data anymore.
35:47
That belongs now to everyone except you.
35:50
So we do not retain the data. So that is one of the probably
35:55
also very important bits that I'd say everyone in our industry
36:01
should be very conscious of that we do not retain any data from
36:05
the customer without permission. It's, you know, privacy first
36:10
built at its core and the way the data sets are built as well.
36:15
We do not, you know, utilize public data that does not belong
36:20
to us and we do not have, you know, if you, we do not have a
36:23
permission to use it, we're not going to use it.
36:25
So, so then we had to go through a few, you know, rounds of
36:29
really manually collecting and gathering the data and putting
36:33
quite a bit of effort from the team members who, you know,
36:37
created those data sets. So that's what I'm also
36:41
mentioning when it comes to collecting data, you have also
36:44
have to be very privacy first minded like it has to be at the
36:50
core of the mindset. Is that something you support in
36:53
in mil Co founders as well that they have this privacy first
36:57
approach? I think our industry is still
36:59
very new. It's not very sadly not very
37:05
transparent right now around, you know, defect detection
37:09
products. The companies are not are not as
37:12
transparent as I think we should be around, you know, knowledge
37:17
sharing and maybe having one united platform for benchmarking
37:23
and also for, you know, also regulating who's regulating us.
37:29
This whole AI regulating AIAI companies is still, you know, up
37:34
and coming those regulations. So I think until that is set, we
37:40
actually don't know who is doing what.
37:44
We can only trust their word. There's no organization that
37:47
can, you know, come and audit our data sets.
37:50
Let's say that doesn't, well, it exists, but not to the point
37:54
where it's widely used and widely utilized and widely
38:00
known. And I think it's, you know,
38:03
should not be the case because you have common standard audits
38:07
and compliancy and that should be applicable for AI companies,
38:12
you know, companies that build models and it's up and coming,
38:17
but we're not fully there yet. So we can.
38:20
Only trust the word of what the people say, right?
38:24
And you know, anyone can say anything really.
38:27
Yeah. But then still I would think
38:29
that they would have to comply with ISO type of standards.
38:35
Of course, yeah. Those are the sort of.
38:37
Starts are are really the basics about at least the organ, how to
38:42
build up the organization. Of course, those are the basics.
38:45
Yeah, ISO. Yeah, so let's move back a bit
38:50
to to self promo and why it is difficult for is that do you
38:54
think it's difficult for female founders more in general because
38:58
you. Build a company.
39:00
Not to to build a company, but also to do promotion for the
39:03
company. Definitely.
39:05
I feel like from what I've again observed from my experience,
39:08
what I've seen with my female colleagues, they feel more
39:11
pressure to prove that they're as capable as their male
39:16
colleagues, right? I feel like we have watched a
39:19
bit more closely how we speak, how we dress, how we show up, do
39:23
we smile enough? Are we nice enough how present
39:26
we are? So, you know, we judge more on
39:30
the tone, on our appearance and the attitudes.
39:34
And actually, if I've personally experienced that because I've
39:37
received comments, I need to smile more.
39:39
So while actually competence alone isn't often enough for
39:43
women. So I think that that adds a
39:46
layer of pressure as well for us, you know, all the all the
39:53
norms, older social standards. Again, when we start, you know,
40:00
going to school and studying what is, you know, normal for a
40:04
girl to be like for a guy, that where we end up, you know, a lot
40:08
of females end up not being very driven about building their own
40:15
business in the end because it requires a lot of masculine
40:18
energy to push through. OK, I'm a bit silent now because
40:24
what? Why does it require more
40:27
masculine energy? Because it's competitive.
40:31
Let me, let me rephrase this. There's always going to going to
40:34
be competition of sorts. So that's, that's I think a
40:39
masculine value that is really difficult to to evade.
40:45
But other than that, is there a way for a woman to build a
40:48
business on her own terms? There is, but I think we need to
40:52
understand what it takes to build a business.
40:54
First. It it takes a lot of self belief
40:59
if you don't especially come from a family of successful
41:02
entrepreneurs or parents who are financially well off to support
41:06
you in any journey that you take.
41:10
Where if you don't come from that money or those role models,
41:14
you have to have the blind belief that you are capable.
41:19
And also your upbringing will be a, you know, very core factor on
41:27
how on yourself image of yourself.
41:30
And if you grow up in an environment where you bullied a
41:32
lot, you judged a lot, you you're questioned a lot,
41:35
everything you say is used against you, then yourself
41:39
belief is going to be somewhere at the rock bottom.
41:42
So it takes a lot of self preservance and self belief and
41:47
sometimes a little bit of delusion.
41:50
But I'm capable, you know, of doing this and maybe a little
41:56
bit of ego as well, while healthy ego.
42:00
So how do you do that for yourself?
42:03
Yeah. How do you keep yourself?
42:04
I wouldn't say delusional, but how do how do you keep your
42:07
spirits up, especially when times get hard?
42:11
For me it took a lot of inner work, so I don't come from
42:15
money, I come from a very normal family of working class parents.
42:24
I also went through, like most of you know, I think females in
42:30
the 90s where it's been still, you know, post Soviet as well,
42:35
environments where you know as a woman you are more taught to be
42:41
compliant, be nice, be. To smile.
42:47
To smile to, you know, be a pleaser rather than, you know,
42:55
standing up for yourself and having that healthy rage maybe
43:00
or healthy, you know, boundaries, boundaries, what is
43:04
boundaries? So that's something that I had
43:06
to learn. So doing a lot of self work
43:09
definitely helped me to stop trying to fit into who I thought
43:14
I am supposed to be and actually helped me shape the idea of what
43:20
success looks like. Because in my head that was
43:22
something else. And then I would say after
43:26
really taken a few years to focus on my mental health, I've
43:31
come to my Zen. Then I realized, OK, this is not
43:37
what I see success as. And that helped me to step into
43:40
my power, into my potential, accept that the inner potential,
43:45
and then just, you know, act from it, speak from it.
43:49
So it's a journey. Yeah, so how has your definition
43:53
of success changed? It definitely changed before I I
43:57
thought it's, you know, having a stable job and, you know, having
44:01
your own house, maybe, you know, having like a family and
44:06
settling in one place and that that is definition of success
44:10
now changed. Now, for me, success is
44:12
something more holistic, something that comes from
44:16
inside. It's all those external values,
44:19
you know, I cancel them and transitioned into more, you
44:23
know, finding my own peace, my happiness that, you know,
44:27
doesn't rely on that portrait image of what success is like.
44:31
It's for me. It's just, you know, acting from
44:34
my inner voice. It's trusting my intuition, it's
44:37
feeling. Every day I wake up and I'm
44:39
grateful for being alive. That is what success for me, you
44:42
know, being happy to be where I am and it doesn't matter where
44:46
it is, it can't be anywhere. So you said you also worked or
44:53
lived and worked in different, let's say, cultures because you
44:58
come from a Soviet post Soviet background, Ukraine I think.
45:03
Yeah, that's correct. Yeah, I was born in Ukraine.
45:05
OK. And you lived in London and now
45:08
you live in the Netherlands. How does that help you as a
45:12
founder or as a human? That definitely helps me to be a
45:17
bit more adaptive. So where I had to, you know,
45:21
switch between different cultures, different systems and
45:25
that definitely helped me learn to accept people maybe more the
45:31
way they are, except differences and you know, feel at home
45:38
anywhere really. So that is probably one of the
45:43
things that's definitely helped me and that adaptability I bring
45:48
with me to, you know, to business where you all, you
45:52
know, start up environment. It's all about being agile.
45:55
And that is in me. I strive in change.
45:59
I love change. We moved around a lot as well
46:02
when I was a child. So that's it's completely fine
46:05
for me. I'm very minimalistic.
46:07
I don't need much. I need one suitcase to be happy
46:11
and have my things. It fits in the one suitcase and
46:16
that is enough. And I think it also helped me to
46:19
be a bit more perceptive. So because I've met so many
46:22
people in my life and travelling, I've been to 38
46:25
countries, I've lived in like over 7 countries.
46:28
So yeah, I think being perceptive and just seeing
46:34
through, you know, the surface level or performative dynamics
46:38
and that again really helps in career and business.
46:42
Also psychology degree helps with that as well.
46:46
Yeah. That I can imagine, because that
46:50
I think you're then also taught to observe.
46:55
Definitely made observe and make decisions really fast.
47:00
And I think as an entrepreneur, that's something that is a must
47:05
because the more decisions we'll make, the more mistakes were
47:09
made make and then, you know, we can improve faster.
47:15
So I, I think it's, yeah, from my experience, it's, that's
47:19
definitely helped. Just don't overthink.
47:21
Sometimes. Sometimes I also end up falling
47:23
into a pattern where I overthink and then I'm like, wait a
47:28
minute, why does that matter? You don't know?
47:31
So just make a decision and and that really helps, you know.
47:36
So basically also being able to intercept your head running
47:41
away. Exactly.
47:44
So I want to move back to to success for a moment and I want
47:48
to move forward in time to 2032 and let's imagine that cyber AT
47:54
is really successful. I was wondering what is if
48:00
you're really successful building a business and and
48:04
selling your products or giving yeah enabling people to work
48:09
with your product and detect deep fakes.
48:12
What has changed for companies and the people working for them?
48:16
Well, first of all, people no longer have to waste their time
48:22
questioning whether content is fake or real, so they can spend
48:27
their time on doing something more productive.
48:30
And that means that that verification process is fully
48:34
embedded into devices, work flows like 2 factor
48:40
authentication, you know, antivirus, it's just there so it
48:46
becomes a standard. I think that's the first thing
48:48
that really comes to my mind. And then people don't have to
48:52
then second guess. Or am I looking at a real
48:55
message? Is that it or like real, you
48:57
know, image video, is that a, is that a deep fake or am I
49:01
receiving call from my mother or is that someone pretending to be
49:05
her? So it should be embedded into
49:09
systems, into devices. Secondly, I would say global
49:14
standards and those providers of global regulations around media
49:20
authenticity like C2 PA or there's a global united body
49:25
that regulates AI and those regulations are in force and
49:30
companies are held accountable for the content that they share.
49:35
And then, of course, employees are trained to work with and
49:38
their fights good than the everified privacy for systems.
49:43
And people understand AI. They know how to work with AI.
49:47
They're not afraid of AI. Yeah.
49:48
So in a nutshell, you know. Is that also the ultimately what
49:53
you want to achieve is make people less afraid of AI because
49:58
AI is used for good? Definitely.
50:01
Because you know what? We are usually afraid of
50:03
something that we don't understand, but if we
50:07
understand, we know how to utilize it and there is
50:10
boundaries around it that's safe to protect us.
50:13
There's digital, you know, digital safety.
50:17
I think people will be less scared of the AI and its
50:21
potential if it's regulated in a transparent, you know, globally
50:25
enforceable way. Digital safety becomes a default
50:29
for our society, not something that we have to ask for, Send an
50:33
e-mail to cancel my, I don't know, don't use my data.
50:37
But it's, it's a default. You know, systems have to be
50:41
able to guarantee that safety by default.
50:44
And, you know, governments should really support that.
50:48
Yeah, fully agree. I'd say that's the very end
50:52
goal. Bigger mission?
50:55
Yeah, cool. So my bigger mission is to make
50:58
inclusion in tech to new normal by the end of 2032.
51:03
What is one thing that we need to do to get there that I would
51:07
need to do to get there? Wow.
51:11
I think they're already doing so much right.
51:13
The role models visibility promoting that and I think we
51:16
really need more role models. I mean, I've mentioned that
51:20
already a few times. If you don't have those role
51:23
models, you know, growing up, how can you know that someone
51:26
that looks like you can be successful?
51:28
So I think that is essential, you know, people who break the
51:31
mold and show that there isn't just one way to succeed in tech.
51:36
We need to accept everyone for who they are.
51:39
And I think that comes to my second point is stop forcing
51:43
everyone to fit in the same box. Like let people lead and build
51:48
and show up as who they actually really are, and that gives space
51:53
for everyone you know to show their real talents and step up
51:58
as their full potential without being scared to show it.
52:01
Yeah, SO also gives back trust. Yeah.
52:04
So inclusion can't just, you know, tick a box.
52:09
It has to be integrated into into everything, into our minds,
52:15
how we think, into products, how they designed, the way teams are
52:20
formed and the way decisions are made.
52:22
It has to be a default as well. Sounds like a great place to
52:27
close. Julia Yakimenko, thank you very
52:31
much for an amazing conversation.
52:32
If people want to learn more about you and Cyber Ed, where
52:36
would you like to send them? They can check our website, it's
52:40
www.cyberrat.ai or my LinkedIn or LinkedIn of Cyberrat.
52:48
OK, cool. And we'll include those links in
52:50
the show notes, of course. So once again, thanks very much
52:54
for a great conversation, really insightful, learned a lot about
52:59
deepfakes and how to detect, but also what detection does in the
53:03
bigger scheme of things. And this time, thank you very
53:06
much for listening. And until the next episode, keep
53:09
being awesome. Well, that's a wrap for episode
53:16
109 of Women Disrupting Tech with Julia Yakimenko on how
53:21
Cyberat battles deepfakes. So you can trust what you see,
53:26
Julia. And I hope that you found your
53:28
conversation as insightful and inspiring as we intended it to
53:32
be. And if you have any questions
53:34
for Julia or for me, please leave them in the comments on
53:37
Spotify, Apple, or YouTube. I do read all the comments and
53:42
I'll make sure they get answered.
53:45
If you want to learn more about Julia or Cyber Ed, please check
53:48
out the links in show Notes. And as usual, you can find show
53:52
notes at Women Disrupting Tech dot Blog.
53:55
You'll also find a link in the description.
53:58
If you've enjoyed this episode, please consider leaving a rating
54:02
or a review on Apple or Spotify. If you leave a rating or a
54:07
review, this helps listeners find podcast and it supports my
54:11
guests by giving them a bigger platform.
54:14
If you're a regular listener, make sure to follow Women
54:18
Disrupting Tech on Spotify, Apple, Amazon or YouTube.
54:22
If you follow us, it's a simple 0 cost way to support Julia and
54:27
my older guest in making inclusion in tech the new normal
54:30
by the end of 2032. For more about Women disrupting
54:35
tech, please follow the podcast on social media, on LinkedIn.
54:38
You can find us at the True Story Company and on Instagram,
54:43
YouTube and Sub Stack. We're at Women's disrupting tech
54:48
on Sub Stack. I also shared in depth stories
54:51
in behind the scenes, insights from every episode, plus trends
54:56
that I've spotted across 100 plus episodes so far.
55:00
So don't forget to subscribe to my sub stack for more women
55:04
disrupting tech. Finally, if you'd like to
55:08
connect with me directly, you can do so via e-mail, dear Kyon
55:12
at women disrupting tech dot blog or you can connect on
55:16
LinkedIn. On LinkedIn, simply search for
55:19
my name, dear Kyon, and you'll find me.
55:22
With that, I wish you a wonderful day.
55:24
I hope to see you in the next episode and until then, keep
55:28
being awesome.