Latest / Women Disrupting Tech / Why Your AI Implementation Fails and How to Fix It with Sophia Zitman | Ep. 136

Transcript

Why Your AI Implementation Fails and How to Fix It with Sophia Zitman | Ep. 136 · Women Disrupting Tech

0:00
If I have to come in at an emergency guard department or
0:03
whatever other partner and I need to do like what, how many
0:08
sessions on how to use the AI and like this is how it works
0:12
and this is how you should handle it, then I think you
0:15
designed a wrong AI. It should feel natural to the
0:19
people. What if the biggest risk to your
0:32
AI strategy isn't technology with the people, politics, and
0:37
processes around it? I'm the actual NUKAS host of
0:40
Women Disrupting Tech, the podcast about women shaping how
0:44
technology is really built. In this episode, I talk to
0:47
Sophia Zitman. Sophia Zittman is an AI leader
0:51
who has helped banks, airlines, and hospitals work with AI, and
0:55
we talk about why so many AI projects fail, how to design AI
1:00
people actually work with, and how to deal with fear and
1:03
resistance. If you're a founder working with
1:07
AI or an investor, this episode will help you avoid building or
1:11
backing AI tools that look impressive in the slide deck but
1:15
never get used to in reality. So when you're ready to listen
1:19
and learn, here's my conversation with Sophia Zitman.
1:23
Sophia Zitman, welcome to women disrupting tech.
1:27
Good to have you. Yeah.
1:30
We met at The NEXT WEB after your keynote.
1:33
Why your AI strategy is failing and how to fix it.
1:37
And one of the things I remember from that discussion is that you
1:40
told me that you get asked on stage not because you're a
1:48
woman, but you're just very good at what you do.
1:52
So I was wondering if you could share a bit of the back story,
1:56
how you got onto the the stage at The Next Web and what you do
2:01
at Kickstart AI? Yeah, of course.
2:04
And maybe to get into the first part, I mean, I of course hope
2:09
and sincerely wish that that's the reason.
2:11
And I think that can definitely be the reason that that's why
2:14
why I was having a keynote on the on the Next web.
2:17
But I think, and that also ties in with the things we're
2:19
probably going to be discussed later on.
2:22
Maybe that is not always the case.
2:23
And that's really a double edged sword, but we'll dive into that
2:27
one later, I guess. Indeed.
2:30
Sofia Sipman, director of AI projects at Kickstart AI.
2:34
And we as Kickstart AI are a nonprofit founded by corporates
2:39
in the Netherlands. KLM, Aldo has NS and ING are
2:42
behind it with the mission to accelerate the adoption of AI.
2:47
And we've done that, tried doing that and still doing that in a
2:51
variety of ways. And one of those ways is really
2:55
AI implementation and adoption projects from A to C.
2:59
So from like the idea of maybe something can become more
3:02
efficient to having it fully operational in those partner
3:06
companies or other organizations that we work with.
3:09
And that is where I come in as well.
3:12
I'm responsible for the success, the business success or the
3:16
societal success of all these projects, which I think is
3:20
really fun. And I joined Kickstart AI two
3:23
years ago. I think by now and before that
3:26
I've always been in consultancy on the AI things, used to be AI
3:30
engineer myself and myself written my fair share of go to,
3:35
but also did a whole lot of projects and account management
3:38
in that setting. And I think that makes make me
3:40
very well suited to set up this part of Kickstart AI as well.
3:45
Cool. Couple of questions.
3:48
Do you think it's prerequisite that you can actually do do a
3:52
bit of coding or at least understand how AI works in order
3:56
to be able to do consultancy and to lead projects on?
4:02
AII think it's a very big asset if you know how to.
4:06
I think there might be a couple of people out there who don't
4:09
and still do an amazing job. But I really see myself and they
4:15
also to work with the with my colleagues that we're doing,
4:18
we're kind of bridging the tech and the human world and making
4:22
it productive together. And if you want to be this, this
4:25
bridge or this, this translator as some people call it, you need
4:29
to be able to speak both languages.
4:31
And therefore I do think you need to know a bit about the
4:35
inner workings of AI, but also, of course, about your
4:39
organizations and the people that are going to use it.
4:41
So yeah, I'd say so. OK.
4:44
And the other question that was in my mind is you mentioned that
4:49
you built of course for organizational value, probably
4:53
for customer value, but also bid for societal value.
4:58
How do you define that? How do you define if it's value
5:02
for society? Yeah, that's that's a good one
5:05
because in in a sense of this is bit of a fake one.
5:07
There's not one metric that immediately comes to mind when
5:11
of course you look at business value, you can one way or
5:13
another always translate it to money when it comes to society.
5:17
Sometimes you you can I think in maybe what you put back into the
5:21
economy or savings should generate for an organization
5:25
with a really big societal cause like another nonprofit, for
5:28
example. But sometimes we also look at
5:32
different ways and an example there as we wanted a project
5:35
with umbrella, which is the logistics side of of bowl.
5:40
And there we our business case was actually in CO2 reductions.
5:44
So emission reductions, which I think has a very big societal
5:49
benefit if one does. And I I did like that a whole
5:53
lot because then you can make it really tangible, you can
5:55
calculate, you can really see, hey, did we achieve what we
5:57
wanted to achieve or not? And really, yeah, in that way.
6:01
Look at what's of what you want to achieve.
6:04
Yeah. So I read a piece this morning
6:07
about AI sort of having impact on society at large and not
6:14
necessarily positive and the need for regulation.
6:19
I was wondering how you look at it at that topic also because
6:24
you're building AI also with us in part.
6:28
Sometimes a societal at least a societal impact, especially if
6:32
you work for NS and ING, it will sort of impact society at some
6:38
point. Well, I think in general, I
6:43
would like I always approach this from a bit of a different
6:46
angle, because sometimes I feel in media or in the conversations
6:50
people having, it's kind of saying either you have the
6:52
people who like do AI and then you have the people who want to
6:54
regulate it and they're kind of opposites to 1 each other.
6:57
At least they're they're facing each other.
6:59
I don't think I don't see it like that because I mean, I'm of
7:03
course the one also like building the AI.
7:05
That's what we do with Kickstarter.
7:06
We want to have it like implemented and and do a
7:08
beautiful things. And I think I do want the same
7:12
as the people who want to regulate.
7:14
I do want the same as regulators.
7:16
I wanted to have a net positive for all of us.
7:19
Why otherwise do it? So I don't think in what we want
7:24
to achieve were different whatsoever.
7:27
Only the angle of which we sometimes approach it is
7:30
different. And there are sometimes you can
7:32
get a bit of friction or potential misalignment.
7:35
But in general, I think regulations like they're good.
7:41
I mean, I, I self regulate when I do projects, right, because I
7:44
just don't want to do something for whatever.
7:46
I, you know, I, I have my own ethics and desires and like will
7:52
for it for it to become a positive thing.
7:56
So that's kind of my take on it. And what, but what I do think is
7:59
important when you look at regulations is that you should
8:02
not limit the people who are actually doing it and like have
8:06
those same books environments. I would like to see regulations
8:09
actually stimulate the use of AI, but in a safe and sound
8:13
environment and not say not do AI.
8:17
And that is, I think where one should have the discussion, if
8:20
that makes sense. Yeah, just let's double click on
8:23
that. So how do you make it safe
8:25
without putting limitations on it?
8:28
Well, I think sometimes you can have like limitations, of
8:32
course, like like I said, I do, I do self regulation as well.
8:38
But what I would like to see and what I see a lot in like
8:41
companies, sometimes they don't even start with AI or keep the
8:44
scope very small because they're afraid of either like big fines
8:49
or having headlines in newspapers and saying they're
8:52
doing something wrong. And there's this huge fear which
8:56
is blocking people and organizations from actually
8:59
using it. And what I would love to see for
9:01
regulations is saying like, hey, we of course want to do it
9:05
properly. So if you have this regulatory
9:07
body say, hey, this is the aims that we have.
9:10
And I can, I'm, I'm very sure that like 99.9 and 9% of of all
9:15
organizations align with those values behind it.
9:18
And let's be done together with this regulatory body can say,
9:21
OK, let's find a way on how we then can actually do it in this
9:24
safe and sound way like like a sandbox or like a collaboration.
9:30
And then let's find out a way on how to do it and not just say,
9:33
hey, if you fuck up, we're going to slash you, but say, hey, we
9:37
both want to get the benefits out of it.
9:40
We both have the same values backing it.
9:43
Let's find a way to do it. And I think that's a completely
9:45
different way of approaching it and a very different mindset.
9:49
And then all of a sudden it's not a limitation, but it becomes
9:52
an enabler. Yeah, yeah, that's true.
9:57
And and I agree that that collaboration usually works
10:01
better. I also say that coming from a
10:04
more regulatory background in the sense that I used to do risk
10:09
and compliance. So but I, I still think to your
10:13
point, indeed collaboration works best.
10:17
On the other hand, I do feel that you you always need both a
10:22
carrot and a stick because if you've messed up because you're,
10:26
you were just not intending to do well or because you just had
10:34
a very bad design and still implemented it knowingly, then I
10:38
think there's yeah, you need to have some, some punishment of.
10:44
Sort, yeah, you know, of course, but that's of course the thing
10:46
when you did do it knowingly so. So in a way you don't say, hey,
10:48
I don't align with these values of creating something with a net
10:52
positive. And then, yeah.
10:55
OK, let's switch bits. Well, keep it a bit into the AI
11:00
domain, of course. So your keynote at The Next Web
11:05
was about why AI strategies are failing?
11:10
Yeah. Why do they fail?
11:13
Well, I mentioned a whole bunch of reasons in that keynote, but
11:17
I think the common denominator in all those reasons are
11:21
actually humans. OK, oddly enough, and I think
11:26
what I mean by that is when I implement and develop AI in like
11:33
the projects that that that we do the technique or the AI, like
11:39
developing the eye itself is never really the problem.
11:42
Of course it's a puzzle and you need a whole bunch of clever
11:45
minds and you need to do it like carefully and thoroughly and all
11:51
that. But if you look at the time
11:53
spent and maybe if you start counting your mind aside, like
11:57
size on like, Oh, no, it's, it's, you know, it's difficult,
12:00
then I think for the development, it's actually
12:03
rather limited. But when you look at then
12:07
actually putting it into an interface that works for the
12:11
people who need to start working with the AI or can start
12:13
benefiting from the AI, that's that's, that's already way more
12:18
difficult. But even before that, at the
12:21
start of a project, really finding a project that can
12:24
deliver a whole lot of value is sometimes really difficult
12:28
because it's so often approached from this innovative tech push
12:33
kind of angle that people kind of forget the organizational
12:37
context that they live in. For example, when when we start
12:42
up a project, more often than not, I, I get a call and they'd
12:45
be like, oh, Sophia, like Gen. AI or agentic AI was just kind
12:49
of taking over the hype. Now we need it.
12:51
It's great. And I, and then I always say,
12:53
yeah, indeed, it's, it's great. It can do great things.
12:56
But before I don't want to go into an an organization and just
13:00
start looking for a random place to put in some agentic AI.
13:04
I want to start at the company at like their long term goals
13:07
and their vision and the targets that they have and the struggles
13:11
that they now have to meet those targets and get into direction
13:15
of that vision. Because that's where the real
13:17
business problem is. And that's then also where I say
13:20
you can deliver most value. And then you can say, OK, what
13:24
then does the tech needs to look like to actually solve that
13:26
problem? And I think you see all sorts of
13:30
things across all a lot of organizations, which just kind
13:36
of makes us a bit inefficient sometimes.
13:40
Because they don't start with organizational goals in mind or
13:44
because they just want to have agentic AI for the purposes of
13:48
having agentic AI and director being able over drinks to say
13:53
hey, but we have an AI agent. Yeah, no, exactly right.
13:57
Because that that's kind of the thing because.
13:58
And they're like, Oh yeah, I know it's cool, innovative.
14:00
I can talk about it during like the Friday afternoon drinks or
14:04
like the Christmas holiday party with the family.
14:06
And I'm not saying I never do those things.
14:09
I do those things too. It's cool to talk about it.
14:11
But when you say hey, we really want to deliver value, I just
14:16
say there is a more efficient place to start looking.
14:18
Yeah. So how do you solve for this?
14:22
What what is the? Let's say how do you turn this
14:25
around because you get these calls and then you tell them
14:31
basically, OK, well, if you let's look at your goals, let's
14:35
look at probably also look at the tech stack that you have and
14:40
the people you have. How do you trans transform that
14:45
into something that is going has a good chance of being
14:49
successful? Yeah, it's a bit of a multi
14:53
stage process and I think you mentioned a lot already there.
14:57
I always start indeed on the organizational side and I try to
15:02
go through all layers. So indeed, on the most like
15:04
abstract layer, what's the division for the organization
15:09
that, that we're doing this project with?
15:11
Because you, you want to have, you want to build something that
15:14
is also durable within the organization.
15:16
You don't want to solve a problem that they don't expect
15:18
to have any more in like 3 years or whatever.
15:21
So that's where you go. And then you'll go a bit like
15:24
lower down in a pyramid and say, OK, what are your targets for
15:28
like upper and upper middle management?
15:31
Because and that always gives a really good insight in like the
15:34
metrics and how you eventually going to evaluate your projects
15:38
because that's they're the company basically already tells
15:41
you, hey, this is what we find important and this is what fits
15:44
in our system and our way of working.
15:49
And then I go way deeper, then I start mapping processes that
15:54
then affect these targets or the targets we potentially want to
15:58
help them with. And that is really where it
16:01
becomes also really operational is you can, there is always a
16:05
process, people having tasks, team doing things, delivering
16:09
things. And then within that process you
16:11
can start looking for, OK, then where is the bottleneck or where
16:16
is the blocker or how can we enable people to do things
16:20
better, more joyful, you name it.
16:22
So then you have a really clearview of the organization
16:26
from like all different layers and you can really check
16:28
yourself, hey, am I working with the right problem?
16:32
And especially that process layer where you actually map the
16:36
tasks of people. If you know that and you know
16:40
where you want to increase some sort of efficiency or whatever,
16:44
then you already have a whole lot of requirements on what the
16:47
solution needs to look like as you know who it needs to work
16:50
for. And then you can really start
16:52
designing, OK, what does in an ideal world does the solution
16:56
look like? And if you have that dot on the
16:59
horizon, it's, it's a matter of mapping what you need to do to
17:03
get there and start building. So that is, that's what we do.
17:07
And it takes quite, it takes quite some time.
17:09
It usually takes a couple months before you have this entire
17:13
clear picture with like interviews or shadow runs or
17:18
like shifts that people do. For example, a project where
17:22
I've been involved myself a whole lot was for the Hospice
17:25
MC, the, the hospital in Rotterdam.
17:28
We did a project for the the emergency care and I did a
17:32
couple of shifts in like full on nurse suit myself to really
17:35
understand what the tasks of the people at the emergency care
17:38
look like to really check that we're designing something that's
17:42
works for them. And that is what my colleagues
17:46
also do throughout the project to really get a feel of, yeah,
17:52
we are doing it for yeah. Out of curiosity, how do you
17:55
make it future proof? Because if you're measuring the
17:58
situation as, as it is right now, you can project a bit,
18:03
let's say, modernize it a bit. Probably there's quite a bit of
18:07
inefficiency to be resolved. But how do you make it?
18:12
How do you make that technology stick in the minds of people so
18:17
they actually keep using it? Because otherwise there's the
18:20
next thing or the, let's say tech, not so positive person who
18:28
is like, yeah, but we can also do it, make it work without.
18:32
How? How do?
18:33
You avoid that. How do you make it successful or
18:37
so long term? Yeah, I think that is really
18:41
part of. I think if you do that process
18:43
mapping and although shadow runs successfully, you should come up
18:47
with a design that that works because in the end you want to
18:53
make something that benefits. And sometimes that that can be a
18:58
customer or a patient when you look at for example the Hospice
19:01
project, but also the people like the nurses and the doctors
19:05
working on the emergency guards should work for them as well.
19:08
It should work everywhere, should be a benefit everywhere.
19:11
And if you do ones work properly, there is this benefit.
19:17
So people have this natural incentive to use it.
19:20
And that is I think also really comes in with, I think a general
19:24
philosophy we have about tech, like tech should work for people
19:28
and not the other way around. Like the tech should bend around
19:31
people. If I have to come in at an
19:33
emergency guard department of or whatever other partner and I
19:37
need to do like what, how many sessions on how to use the AI
19:42
and like this is how it works and this is how you should
19:45
handle it, then I think you designed a wrong AI.
19:49
It should feel natural to the people, like it should aid the
19:53
people and not the other way around.
19:55
And if you do that, it feels natural to people, There is a
19:59
benefit for people. So why wouldn't you use it, so
20:02
to say. And I think within that
20:04
philosophy, you can also add that it actually being useful is
20:10
the best example on why this can definitely be a positive change
20:15
for for society and all of us. Absolutely.
20:19
So just stepping back a bit, in the interview, you spoke about
20:26
tech leaders who sort of rush into tech before reflecting on
20:33
on the real business challenge that they have to solve.
20:37
Is that is that a particular AI problem or do you see that in a
20:42
broader context? And people rush to.
20:46
Tech rather than as a sort of. I've heard that it works, so it
20:51
must work for me. Well, I definitely see it a
20:54
whole lot in AI and I, I can assume it works in other fields
20:58
as well. I don't think I can comment on
21:01
that a whole lot because my work is within AI.
21:05
But yeah, I think it's, it's a mindset within especially like
21:12
the, the larger corporates I I also work for.
21:14
So I wouldn't be surprised if you've other have had other
21:17
people on the podcast or get other people on the podcast that
21:20
work in a different like side of tech and, and really see this
21:23
happening too. Yeah, yeah.
21:26
Yeah, it was more broadly that I spoke to Anu Klikman a couple of
21:30
weeks back and she said, well, in order to have an inclusion
21:34
inclusive organization, you need to be able to understand
21:39
yourself to to connect with yourself as a person first.
21:42
Because otherwise you a, you don't know what inclusion could
21:48
look like and B, you rush into building an inclusive
21:55
organization for the purposes of just checking off some DEI
22:01
checklist. Instead of that it you do it
22:04
because it makes real business sense.
22:07
Yeah, but. Yeah, I, I think, yeah, I
22:09
wouldn't be surprised if there's a very similar or if not the
22:11
same principle at play. You know, you can want to go on
22:17
like or like have efforts on being more diverse because
22:20
indeed you want to have that checklist.
22:22
I think that's very similar in saying, Oh yeah, I read about
22:25
agentic AI, so now I need to do something with my innovation
22:28
department as well, so I can also have a check in the books
22:31
there. But then, of course, you're very
22:35
likely not to have the beneficial outcomes that it
22:38
actually can have if you would have done it properly to begin
22:41
with. Yeah, yeah.
22:44
So I think it was a quote of yours that you believe that
22:50
people, not algorithms, can make or break AI initiatives.
22:57
Can you share an example where this happens or where?
23:03
Yeah, essentially where people made or broke the initiative.
23:07
Yeah, I'll, I'll. I'll stay a bit fake in this
23:10
descriptive. Maybe let's start with the with
23:12
the breaking one that's a bit more juicy.
23:13
And then we can go into the the endless of like positive
23:16
examples. We we of course have as well.
23:18
Yeah, we wanted a project also for for one of our partner
23:24
companies and we we landed on this topic which was very core
23:31
and things they they did with it within the business and they
23:35
already had some efforts going on on this particular topic.
23:40
So that was great. And we, we take the long on a
23:43
different side and we brought a lot of speed, a lot of
23:45
attention. We brought a whole lot of people
23:48
together because we really had this focus on actually getting
23:51
it implemented and not just staying within this experimental
23:55
research phase, which of course is really important you of
23:58
course also need to do. But we really wanted to be more
24:01
iterative, so bring what we already got through production
24:04
and then later grow on that. But then at some point, I think
24:10
we went a bit a bit too fast because people really sold this
24:15
as a really difficult, difficult topic.
24:19
And then we showed a whole lot of progress real quick, and that
24:25
did not necessarily in line with the political incentives of
24:29
other people. And then you see them kind of
24:34
like, you know, then you start seeing a bit of pushback and all
24:37
that kind of stuff. And then at some point, such a
24:39
project becomes a bit of a, a pawn in a, in a bigger political
24:44
play. And I think the irony here is
24:48
that because the tech was so great and we got to that point
24:52
so, so fast, that there was a real irony there that it still
24:56
then got slowed down by by quite a lot because it got this like,
25:03
yeah, political nature to feel over.
25:08
It Yeah. Yeah, so.
25:10
What's that? A classic case of of what we
25:13
call in Dutch as an ex sister in Underham.
25:17
So. So basically being going
25:20
bankrupt based on its own success, Yeah.
25:24
Yeah, I think so. I think so.
25:26
And I mean, there's still a bit of of life to it.
25:29
You know, it's it's no waste of what what is what has been done,
25:32
the work that has been done. But sometimes, sometimes you can
25:38
also move a bit too quick, I guess.
25:40
Yeah. So how did?
25:41
How did that make you feel that it sort of was so successful
25:45
that it sort of torpedoed itself?
25:50
Well, that sucked. I I did feel bad.
25:53
Yeah. No, I mean also with the team
25:55
and also with like a lot of people from from from that
25:58
company. We we spend, we spend a fair
26:01
number of like hours on it. We we really liked it.
26:05
We we did a whole lot for it. It yeah, no, it was people
26:11
definitely had a passion for that project And and that showed
26:14
also in in a success, of course. So and it then gets torpedoed
26:18
like that, that that sucks. So that was kind of my first
26:25
wave and the second wave was also kind of like, should I have
26:28
managed things better? Should I have played way more of
26:31
a political game? You know, you get a bit of at
26:34
least that's what I what I always get.
26:35
Then I start kind of like doubting myself and being like,
26:39
what if I engineered the world's completely different?
26:41
What is what we know? What if I make sure there was
26:44
was another beer at the Friday afternoon drinks and like sent
26:48
that e-mail like a day out earlier or later, whatever I
26:50
said, you know, it could have changed and I, I think not, but
26:57
you know, it sucks. And then then, then you move on
26:59
and then you start looking for other ways.
27:01
And sometimes we, we pause it for a little bit and I focus our
27:04
times and energy on other beautiful things and come back
27:07
to it when the time is there. Did you do?
27:11
That in this particular case, yeah.
27:15
And did it help to take your foot off the gas and then later
27:19
come back to it when mines were more, let's say sometimes I
27:24
under I, I hear what you say sometimes also dependent on the,
27:30
let's say the political agenda of people.
27:33
Yeah, no, yeah, for sure. And I think my, my default state
27:37
is always keeping my foot on the gas.
27:39
That's kind of how we break barriers, how we move mountains.
27:43
It's really how we also kick start identify ourselves as the
27:46
doers. But every once in a while, you
27:49
need to take your foot off the gas, let it kind of drive, just
27:53
let it go on for a bit. And then for whatever reason, it
27:56
will always find momentum because it was so good to begin
28:00
with. And then you can just pick it
28:02
up. Yeah.
28:03
Do you have a certain way of dealing with these political
28:08
forces or let's say, situations where you're doubting yourself?
28:14
Should I've sent this e-mail a little bit earlier or later or
28:18
maybe had an altar beer on the Friday drinks?
28:22
How do you deal with, how do you deal with that as a person?
28:26
Does this stay in your mind long or do you, is it a fault that
28:29
comes up that you and that you then sort of push that you're
28:35
able to push out? I think there are two kind of
28:39
like separate questions in there.
28:40
I think because how I deal with it more from like an
28:43
organizational work perspective, of course you deal with it
28:47
because you know, when you want to implement AI from like AD and
28:52
do it, like I said in the beginning, the organizational
28:55
context and where you want to go and aligning people and
28:59
incentives and bringing departments together.
29:01
It's a, it's a political game you play.
29:04
And I mean that in a good sense of the word.
29:07
It's a political game. Like there, there's no negative
29:10
connotation for me there. So of course it's a big part of
29:15
our projects when we do those mappings, especially in the
29:17
beginning and like looking at targets of people and
29:21
departments to have that all mapped out to make sure that
29:24
we're actually building something that the organization
29:27
wants, that people want that fit into their own political agendas
29:32
within the company. So from from the get go, you
29:35
start navigating that, you start mitigating that because you know
29:39
you need you need a whole different bunch and different
29:42
people throughout such an organization.
29:44
So that is a key success factor in in, in what we do.
29:49
And I also think we, we do that well, if you look at the results
29:52
we have with our projects and how some how fast we can get
29:55
things live or get very sensitive topics moving to begin
30:00
with, doesn't mean it like always works like this
30:04
occasional example that I gave you.
30:05
Sometimes you know, it, it doesn't.
30:09
And that's, and that's part of it.
30:10
But I think that's part of everybody's work.
30:14
And I'm very proud of kind of the success rate that we have
30:17
in, in the in the projects that we do.
30:20
So that is the the work answer of the things.
30:23
And then on a more personal level, yeah, that that pops up
30:31
and I can feel that for like half a day.
30:34
And then it kind of like I get a little bit of like frustration,
30:42
a little bit of sadness. But for me personally, that
30:46
really turns to fuel quite quickly because in a, in a bit
30:51
of old way, it's a, it's a challenge to then mitigate or
30:54
navigate that and then kind of try to see like the puzzle in
31:01
it. And my mind starts spinning on
31:04
like how how to best go about it.
31:07
OK. So you move to more
31:10
dissolutionising mode pretty quickly.
31:14
Yeah, yeah, for sure. OK, So that's an example where
31:18
initially at least, the AI initiative was sort of broken up
31:24
in order to, well, you at least at you to take your foot off the
31:28
gas. Can you share an example where
31:32
the initiative was wildly successful, but then in a
31:36
positive way? Yeah, yeah, of.
31:38
Course, I can. I can even share a couple if I,
31:43
if I, if I may, I think the one I already mentioned before on
31:49
para. I think what was a big success
31:51
there is that we really went from like kind of the inception
31:55
of an idea to having something fully operational in their
31:59
business processes in about half a year.
32:02
So that's like rapid speed, like from the moment I picked up that
32:05
that phone for more like the innovation people being like,
32:08
oh, let's do something cool to actually having something
32:12
running. If you see for example, if you
32:14
see a bull busher like a fan picking up parcels, whatever
32:20
shipping them, most likely the estimates for the volume of
32:25
those parcels and the planning process, we did that together
32:29
with them. So that's there's been a
32:30
collaboration with Kickstart, which I think is very cool.
32:34
When you look for example at the Osmos, the hospital project, we
32:38
did it there in nine months. So that's also incredibly fast
32:44
and not just and what? What was that project about
32:49
again? Was that about planning the the
32:51
need for emergency care? Yes, something like that.
32:56
It was indeed like they they kind of they came to us with the
32:59
the the problem or the the use case saying we still go on like
33:04
code red as they call it too much.
33:06
So basically they have this this coloring system saying if
33:09
they're green, they can they still have plenty of capacity
33:12
that they can take in people and whatever kind of emergencies.
33:15
But when they go on code red, as they say, they basically say,
33:19
hey, we're, we're full capacity. We actually cannot take in
33:23
people anymore. Like we'll still do if if shit
33:25
hits the fan, but it's it's anything far from ideal.
33:29
And this has been a huge also been part of the public debate
33:33
in, in times of of COVID where a lot of people had to get go
33:39
through the emergency care. And they basically said, well,
33:42
it's kind of moved out of the public eye, but to us, it's
33:45
still very much a problem because we're still on very
33:48
similar, uh, frequencies of going on red.
33:52
Can you find a way for us to like, make this a little bit
33:56
better and potentially solve it? And that, that of course, was,
34:00
was, it was, it was a really beautiful, lovely case.
34:02
And also from a very societal, uh, point of view by, by nature.
34:08
And then we started looking into them like, OK, how can we
34:11
actually change some of the processes that you go on?
34:15
Let read less often. And one of the things we saw is
34:20
that a lot of the work they do and how they like assign people
34:25
or the frequency, like the speed in which they take in people or
34:29
they send people home or to different departments of the
34:32
hospital is very reactive on what is happening at that
34:35
moment. And we can ease that, like flow
34:39
away more if we can help them be more proactive in making those
34:44
choices. But to be proactive, you need to
34:47
have a bit of a look into the future on how much people will
34:52
actually be coming in. And that is then what we made
34:56
together with them and this little window into the future
35:00
and with predictions on how many people would be coming in, which
35:05
then allowed them to be more proactive in the things they
35:09
already did. Yeah.
35:11
So actually sort of createspace proactively on the on the ICU,
35:20
yeah. Or sorry.
35:21
On the on the emergency room, Yeah.
35:24
What's on the ICU? But.
35:26
That that, for example, was a thing and like from that initial
35:29
start and just like, hey, we still have this problem, but not
35:32
knowing where to exactly change an aid or like help people out
35:37
to actually having this, this little window into the future
35:41
that the nurses and doctors can use to start acting more
35:44
proactively that there was a nine, nine month project.
35:47
And like having both the, the, the AI that that can give this
35:52
little look into the future, but also having the entire technical
35:55
infrastructure behind it. Like this thing is running 24/7,
36:00
like every hour it's looking once once again into the future
36:03
and updating itself. But then of course also finding
36:06
a way that this information can be very easily and naturally
36:11
kind of fat to the, to the nurses and doctors at the
36:15
emergency care. That it's that is intuitive that
36:17
in just a glance that you can take in and digest that
36:20
information and then do something with it.
36:22
It had so many, so many angles. And that also then in this ever
36:27
changing complex organization that a hospital is.
36:30
I think to make this work also from a technical point of view,
36:33
we had to collaborate across 8 different teams and.
36:39
That is. That is quite a thing.
36:41
It's a challenge we took on together and it and that and
36:44
that worked. And that's also really because
36:46
of the people, because people do want it.
36:49
They do want to find a way. They were basically like, hey,
36:52
we all have our own set of expertise or our own knowledge,
36:56
but we're still kind of finding a way within the hospital
36:59
ourselves to bring this together in the most efficient way.
37:03
And really let's use this case, this collaborative project also
37:07
as this send books or first experiment and say, let's let's
37:12
pick something and see how that works.
37:13
And then of course, and that's also what they're doing now,
37:16
reflecting on that entire process of that project and
37:20
saying, hey, what did work really well for us?
37:22
And we see that they're already copying some of that in their
37:25
own way of working or in their own tech stack.
37:27
And with other things they're saying, hey, we learned from
37:29
this and now we're going to change it a bit and we're going
37:31
to do a similar thing ourselves. And that is purely mindset for
37:37
people. And I think, I think that's
37:39
beautiful. Yeah, that's.
37:41
Absolutely is. I was wondering about one thing.
37:46
What I have not heard you say outright is that people were
37:51
resisting the tech. So there were the the obstacles
37:57
that you had on the way were more around, let's say, the
38:02
politics, the organizational politics and how to get it
38:07
implemented from a, from a, let's say, a tech perspective.
38:13
Have you ever ran into cases where a project ran into
38:19
headwinds because it was AI itself?
38:23
So the the resistance against AI as a technology?
38:27
Well, it's. I think almost every project I
38:31
get, I get there's somebody who mentions it for for sure.
38:37
I think we can navigate that pretty well though, because
38:42
usually people bring it up from some sort, some sort of like
38:46
fear or fear of the unknown or like this public debate.
38:49
And they're a bit like, oh, but those people, if you then start
38:55
explaining what it is and that we're there with the same moral
39:00
values, the same goals as them, as making also their life more
39:06
pleasant. And for some people, indeed,
39:08
work can change a bit like some tasks, you know, and the AI can
39:12
either help them with or maybe completely deal, but of course
39:15
that gives them room to do then other things within those
39:17
companies. Or when we say, hey, I think
39:23
what's also always a thing, people kind of automatically
39:26
assume that it is the end of the world if an AI makes a single
39:32
mistake. And in some cases that's, you
39:35
know, you can, you can definitely make that argument,
39:37
but in a whole lot, I also don't think it is because humans make
39:41
plenty mistakes too. And in the end, you want to have
39:44
this increase in efficiency, productivity, quality, you name
39:49
it. And so as long as you generate
39:52
that improvements without any substantial loss or cost and
39:56
whatever somewhere else, then you're still creating that that
40:01
benefits and having people understand that and just
40:06
basically taking the time to listen and talk.
40:10
Them through and have those people involved in designing the
40:14
solution or at least requirements for the solution
40:18
takes that out of the way So what I'm.
40:20
Hearing you say is is especially also involve them.
40:25
So don't take them out, but take them in rather into your project
40:30
and make them sort of productive if they're giving you headwinds
40:35
because they may actually have, let's say, legitimate concerns.
40:40
Yeah. For sure.
40:41
It's like it's it's almost always like a legitimate thing,
40:44
like a concern, a question, whatever.
40:48
And I think a whole lot, sometimes a whole many things
40:50
sound like a concern or some sort of anti sentiment at first
40:56
are just a question and that of course you can answer.
41:00
And in all honesty, it's usually those people because it's the
41:03
people who have a heart for the company they work for or they
41:07
really enjoy their job. And it's also usually the people
41:10
who understand the organization by the law, otherwise they
41:14
wouldn't bring it up. So those people are, I'd say,
41:18
immensely valuable for really getting that proper
41:20
understanding and make something that work is if you can get, if
41:25
you can make something that works for those people, it will
41:27
work for almost everybody. Yeah, that's.
41:33
Absolutely true. If you if you can make that work
41:36
for for for them, it's usually a takes away a lot of the
41:40
resistance that you you would otherwise have had.
41:44
I think because usually those the people voicing resistance
41:49
are who are voicing concerns are the people who are, let's say
41:57
the focal about it. But there's usually my
42:00
experience at least with implementing new tech, is that
42:04
there are a lot more people who have the same concerns but are
42:08
not focal about it. So there are it.
42:11
It reduces your chance of silently killing your project,
42:16
yeah. Yeah, of course.
42:18
And what I sometimes I think in, in quite a lot of the projects
42:21
actually see is that those people, they, it becomes they,
42:25
they, they're very likely to become some sort of early
42:27
adopter or a champion of the project as well.
42:31
And that's, that's, that's just great to see.
42:34
Yeah. That's true.
42:35
That is true as well. Indeed, it's sort of they, they
42:41
move from, as I used to call it, backbenchers to, to, to fund
42:47
women or to fund people and then being your ambassadors by by
42:53
telling their own crowd that this is this is just really
42:57
great stuff. Yeah, yeah.
42:59
Exactly. So many of the the listeners who
43:04
are listening to this podcast are either female founders or
43:08
their investors. And I want to see if we can get
43:13
free, free sort of learnings out of this for cases where they use
43:20
or start to use AI in their companies.
43:22
I think 1 is to at least is to listen to concerns of other
43:29
people and to take them into the, to the project.
43:35
What? What other things do you see
43:38
wrong? Let's say from a technical
43:39
perspective, and this could be fixable, but.
43:45
Yeah, there, there can be plenty of things.
43:49
I think when I reflect on a project so that we deal with
43:53
kickstarts, I think it's often underestimated how much time and
43:58
effort we actually need to get into the infrastructure side of
44:01
things. So not just building the model,
44:03
because that's what most people, you know, talk about.
44:06
It's kind of the most sexy topic of it all, but having just like
44:12
the computer databases, all the like programs talking to one
44:16
another, doing that in a resilient way, like from a
44:19
technical perspective, resilient way, having these like life
44:22
cycles, the monitoring, you name it, that that comes into play.
44:27
Definitely don't underestimate that.
44:31
So that is I think from a technical point of point of
44:34
fuel, what I say taken into account.
44:37
But at the same time, I would like to notice also, don't let
44:43
yourself be limited by it. I think definitely for more
44:48
startups or scale UPS or new initiatives, I would love it.
44:53
And I think it's definitely possible if AI can become this
44:56
great equalizer because, you know, it can start to code for
45:00
you as well. It can do a whole lot of these
45:03
things for you too. So don't, don't be limited by
45:08
it. Absolutely not like, like use it
45:10
to your advantage because there's so much more you can now
45:13
do do yourself, especially in those those early phases.
45:18
Yeah. So just just do it.
45:20
Yeah. Yeah.
45:21
Do you think? That new organisations or
45:24
smaller organisations have a benefit compared to bigger
45:29
organisations when it comes to implementing AI.
45:34
Because they are. Just more and more.
45:36
Let's say agile, much more. Absolutely.
45:40
A lot less legacy when it comes to tech.
45:44
Well, yeah. For sure.
45:45
I think it definitely from a mindset perspective as well,
45:47
because usually you see people that are a bit more, you know,
45:50
that are more agile themselves or more innovative themselves
45:53
in, in, in certain ways are a bit more fast moving.
45:57
And I think definitely with the tools that AI can offer you
46:00
today, that is becoming so much of an advantage for, for one,
46:06
you don't need like necessarily like a bunch of like an entire
46:09
department or whatever of developers to get the smallest
46:12
idea of the ground. You can play with it yourself so
46:15
easily, right? Then of course, when things
46:17
start to get really big and whatever, at some point it's
46:20
good to get some expert matters, like some some experts involved
46:24
in making sure everything is sound.
46:27
Also all the edge cases and so on.
46:29
But that's so, so much in the later stage.
46:32
OK, Is there? A sweet spot when you start
46:34
doing this or let's say red flags when you have to start
46:38
thinking like, hey, we need to get people on board.
46:44
Yeah, there might be. I don't know if I'm necessarily
46:47
the one to pinpoint that exactly.
46:52
I think as in if you just want to test an idea, if you want to
46:59
like play like have a toy, have a shiny object, and then there's
47:04
so much you can like do yourself.
47:10
And usually in a start up 1 is never truly alone.
47:13
You have a least a small cloud of people that surround you.
47:16
When you start to become more of in a skilled face that that
47:19
obviously gets a whole lot more. I think maybe what's even more
47:25
important is that the people you bring on board still share that
47:28
mindset of moving fast. And I think also when I also
47:34
look at developers and also with the team we we have now at
47:37
Kickstart and if we bring on new people, of course, we test them
47:42
on the, the technical skills and you know, how how technically
47:45
sounded their work is. But next to that, it's also very
47:49
much the mindsets that they have.
47:51
Do they have this like or can build, can do can find a way
47:55
mentality instead of them developers who just get a ticket
47:59
and do it just as the ticket describes and don't consider the
48:03
broader context or don't consider like the broader scope
48:08
or targets of the project, kind of think ahead.
48:14
So if if anything there is, you know, bring on people whenever
48:17
you feel you need to bring on people, but the people you bring
48:21
on, make sure you have that bit of an entrepreneurial techie as
48:24
well. Yeah.
48:26
I think that's that's very good suggestion.
48:29
How do you do that for yourself? How?
48:31
What do you look for? Well, what kind of?
48:35
The words maybe even were things that that that pop that pop into
48:43
your mind like OK, this is when I am getting enthusiastic about
48:49
a person sitting across of me. Oh, that's, that's a good one.
48:57
Well, we just, we just hired a bunch of people.
48:59
We're still in the process of, of hiring people.
49:02
So like in that sense it is really relevant.
49:05
And like I said, of course there is a certain technical baseline
49:08
1 needs to meet, but I do very much like it if besides they, of
49:17
they, of course they like the tech part of it.
49:20
They enjoy, enjoy those technical puzzles and they,
49:23
they, they want to, they want to play with it.
49:25
They enjoy that part of their, their work.
49:27
But I think when I look at another the common denominator
49:31
of, of, of people within the in the team, they all have in one
49:36
way or another, this drive to do something that actually makes it
49:40
to production, that makes it into real life.
49:43
And some of the people on the team really come from this
49:45
society or environmental angle where they say, hey, I spent, I
49:50
don't know, a couple of years in my career just kind of being
49:54
part of a really big organization in a silo and kind
49:57
of just being a really small part of the process.
50:00
And I want to be part of like way more of that process.
50:03
I want to be way more in touch with the thing I actually build
50:05
and see it come to life. And that is something that of
50:09
course that we do. So yeah, that is I think that
50:15
that spirit that that kind of needs for that.
50:19
That's very, it's kind of having, having a purpose or
50:22
seeing what you do really materialize in front of your
50:25
eyes. People who enjoy that, who, who
50:28
kind of need that in their job. I, I love it.
50:31
I have that myself as well. And I think it works great
50:34
within our team. OK.
50:37
So out of curiosity, how do you keep up to date?
50:41
Because I know you like the lunch table as a source of
50:44
information, but how do you keep up to date otherwise?
50:49
Are there specific blogs or YouTube channels or?
50:55
Whatever you. Follow I.
50:58
Don't necessarily follow a whole lot of these myself because like
51:01
I said, like at their actual lunch table at the office, like
51:04
a lot of people from the team, they do it and we discussed that
51:08
a whole lot. So I think indirectly I, I get a
51:11
lot, I get a lot of that insight already, but I think the best
51:15
way to stay up to date is actually use or play with it as
51:20
much as you can yourself. And I think the easiest way for
51:25
plenty of people is, you know, there's this all the LLMS that
51:28
they're out there, like the ChatGPT Gemini group, whatever,
51:32
always make sure the toy with it.
51:34
And The funny thing is actually start developing preference for
51:38
certain kind of models for certain tasks oneself.
51:42
And when you have that really hand on experience, it's
51:47
something and you allow yourself to reflect on it and think on it
51:51
for a little bit. Having all those other
51:53
conversations on like the actual benefits of AI or what it can
51:58
do, what it's not yet can do, or where the dangers actually lie
52:03
that become so much more tangible, that become so much
52:05
more concrete. And then you kind of move out of
52:09
this hypothetical space that I think still dominates a lot of
52:14
the the conversations around AI in the public domain.
52:18
But when you actually work hands on with it yourself, and even if
52:22
you're a little bit ballsy, I don't know, develop a bit of an
52:25
AI agent yourself, you know, go through one of these models,
52:28
spend a day in a weekends, build a building 1 go.
52:32
That quick. If it's, if it's a simple one,
52:35
yeah. OK, because that's that's
52:38
something really cool that people can actually do and also
52:41
get started with AI to get over the, let's say, out of the fear
52:47
zone and into the possibility zone.
52:50
Yeah. Yeah, for sure, for sure.
52:52
And also like if if it doesn't work within a day, also don't
52:55
beat yourself up over it. And I think a lot of people
52:59
already kind of use it sometimes to summarize documents or to
53:03
write an e-mail. And I think I think that's
53:05
great. But actually building something
53:08
with it yourself or if you have another task that is way more
53:11
related to the physical rules. I sometimes use it for cooking,
53:16
for example, which I think is, is kind of funny.
53:18
And like in the midst of it, I'm kind of saying like, oh, what
53:21
would have actually happened if I did like XYZ, kind of like
53:24
experimental, like random, random things.
53:30
So, so yeah, those kind of things were for hobbies.
53:35
I, I have a big passion for, for lipsticks and of every random
53:38
side note. But it's very fun.
53:40
It's very fun to talk with AI about those things.
53:44
So in, in whatever way, like just just just have fun with it,
53:48
play with it, use it. And if, if you feel comfortable,
53:51
maybe, I don't know, build something with it.
53:55
Yeah, I think. That's very good.
53:57
At least experiment with it, get to know the tool and and
54:01
understand what you're talking about.
54:04
Yeah, for sure. And of course AI is way bigger
54:06
than those LLMS. But for example, I don't know if
54:11
people, I don't know if you drive or use public transfer
54:14
doesn't even matter. But if you say hey I want to
54:16
know more about how those best routes are calculated when I
54:20
when I open Google Maps. Can you tell me?
54:22
How that's done, you know, they, they can tell you.
54:25
And then if you say, oh, if I were to make such a program
54:27
myself, like what does that look like or what are the components
54:30
in it? I think AI can be such a great
54:33
tool for, for education at some point as well.
54:36
And it's still very early phases.
54:37
And I'm not saying, you know, I'm not saying take the kids out
54:40
to school and just give them GFT.
54:41
Absolutely not. But it's it's just so much fun
54:46
when you when you toy with it, do you think?
54:48
People or yeah, children should be taught how to deal with AI in
54:54
school. Yeah, I think.
54:56
So I mean, it's a it's a tool, it's an innovation just like any
55:01
other. Like back in the days there was
55:05
this discussion on whether schools should have computers
55:08
for kids. And to me, this this feels kind
55:11
of like a repeat of that. And my answer is absolutely.
55:17
Does that mean it's a free for all and there are no risks and
55:21
whatsoever? No, of course not.
55:22
But I think it's way better to do it and to do it mindfully and
55:26
maybe sometimes when you realise, oh, that's not the best
55:29
way to teach kids about it and change a little bit than to not
55:31
do it at all. So yeah, definitely so looking.
55:36
10 years ahead, how do you hope that AI will change how we build
55:42
businesses? Education society, Yeah.
55:48
I I very much hope that it will be the great equalizer of
55:54
things. In what way?
55:56
Equalizer. When you, for example, look at
56:00
building companies and startups at that point, like AI can do so
56:06
much for you. So if you have this
56:08
entrepreneurial mindset and a good idea, the the cost of like
56:13
setting that up and actually trying it, I think should be so
56:17
much lower than it is now. Because you don't need this
56:20
entire army of developers to get something off the ground or
56:26
maybe other things and navigating the law or you name
56:30
it things, things that are limitations right now and
56:34
actually making a product. So in that sense, I think it can
56:39
be a great equalizer as like the cost of starting up, can they
56:44
become a whole lot smaller. And also when it comes to
56:48
education, education is expensive.
56:53
I think in the Netherlands we're doing like, you know, where I
56:55
think on your side point of view, especially compared to the
56:57
rest of the world, we're doing really well.
56:59
But in other parts of the world you see vastly different like
57:03
levels of education and quality of education, like also within
57:06
countries. And it all has to do with access
57:09
mostly of like the state of sort of money the parents have for
57:13
their kids. But if so much of that becomes
57:17
accessible and available online, then once again, I like to see
57:20
that cost shrink there. And you're kind of like
57:23
levelling some sort of playing field or at least level it way
57:26
more than it is now. And I truly believe AI has has
57:32
the potential to do that. Also, for example, when it comes
57:35
to businesses in the Netherlands.
57:37
Doesn't mean it will automatically happen to say that
57:40
because also when it comes to startups, you need to hold up
57:43
more than AI to have a positive startup climate when it comes to
57:48
like rules on like, I don't know, paying people in in stocks
57:52
and getting people on board and all.
57:54
Like there's how much more to do.
57:55
So as a society like us, the companies, the nonprofits, but
58:00
also definitely the the government there, like we still
58:03
need to do a whole lot to get the most out of it to harvest
58:07
all that potential. But I I do believe it's there.
58:10
And that's also where where I very much hope it goes and also
58:14
what we escape starts are yeah, working on that it that it goes
58:19
there. Yeah, do.
58:21
You also see it as a as a sort of equalizer in terms of
58:25
opportunities of women, yes. For sure.
58:29
I think like for everybody, like whatever spectrum you want to,
58:34
you want to you want to get whether whether it's sex,
58:36
whether it's like access to to money or connections or
58:42
whatever. So I think yes.
58:44
So I'm building an AI tool at the moment, one that connects
58:50
female founders to investors to to bridge the funding gap.
58:53
It does use AI and it does also to to connect people to to each
59:00
other that otherwise may or may not have met each other.
59:06
So. From your perspective, what is
59:08
the most important ingredient to make sure that AI actually
59:13
supports better decisions instead of reinforcing biases?
59:19
Well, I think. That once again kind of relates
59:22
back to how you start like start up a project and why you do it.
59:28
But I can give a really like technical answer to this because
59:30
within the space of like AI and machine learning, you have
59:33
techniques and ways of measuring bias and all that stuff and
59:40
people, if, if you want, we can, we can talk a bit about those
59:43
things. But I think what is way more
59:47
interesting maybe in the context of the things we've discussed
59:51
prior and also kind of linking back to previous questions.
59:55
If you just say, hey, I have this good cause of, you know,
59:59
bringing these people together and creating a benefit there and
1:00:03
bridging this this funding gap. And then if you just say, hey, I
1:00:06
want to use the eye because it it sounds nice, sort of checking
1:00:09
the books. Yeah, then.
1:00:11
You have very little incentive or you're probably overlooking
1:00:15
more bias, whatever may get introduced or like unwanted
1:00:18
bias, because some sometimes you, you also do want bias, but
1:00:22
that that's a whole different story.
1:00:23
Maybe because maybe in a way you want to have bias the other way
1:00:28
around because you're moving towards the start.
1:00:30
So once again saying like, Hey, I have this, I have this goal, I
1:00:34
have this purpose, I have this vision and to, to get there, I
1:00:38
have these processes and this is where I can still optimize for
1:00:41
these processes. Well, I think if you followed,
1:00:44
you have that way of thinking and that that flow and also what
1:00:49
you mentioned previously is like staying connected to, to
1:00:51
yourself, then you have all the right incentives and the reasons
1:00:57
to do it properly and not have those unwanted biases that
1:01:01
you're trying to mitigate. So once again, it starts really
1:01:05
from not being this, this tech Bush or this check box that you
1:01:10
want to check off, but really by you're doing what you're doing
1:01:14
in a broader context in the 1st place.
1:01:17
Yeah. And seeing AISA tool, yeah.
1:01:21
Yeah. So just taking you up on the on
1:01:25
the biases part, what are ways to measure biases in?
1:01:30
AI, I think there are there are plenty.
1:01:33
I must admit, I do not know all the names by heart anymore.
1:01:38
Back way back when, when I was still in uni and doing my my
1:01:43
thesis, my master's thesis, I did a whole lot and explainable
1:01:46
AI. And then I also preached out
1:01:48
into to think what they back then, maybe still the same name,
1:01:51
but I called fair, fair AI in the in the scientific world.
1:01:56
And you have you have ways of measuring that in a sense that,
1:02:02
for example, the classic, the classic loan example, right?
1:02:07
You go to bank, you want to apply for a loan or mortgage and
1:02:11
you either get accepted or rejected.
1:02:13
And then of course, you can test your own AI and saying like,
1:02:17
hey, for for example, the access on like on sex.
1:02:21
Does that differ if you have everywhere else just same
1:02:24
profile but still different acceptance rates?
1:02:26
Well, there is, there is a bias there.
1:02:28
And I think we can all agree it's an, it's an unwanted bias
1:02:32
where the same goes with like no applications or whatever.
1:02:37
And then there are a variety of techniques to kind of mitigate
1:02:41
that, especially in like the classical AI models.
1:02:46
I think if you have a quick search on like fair AI, there's
1:02:50
there's this really good library.
1:02:51
I think also, I believe done from research at the to
1:02:55
Eindhoven, they they have a whole website and this that
1:03:00
approaches it from a technical perspective.
1:03:03
And so there are definitely ways you can go about it.
1:03:06
But in the end, it's still as a human, how you then interpret it
1:03:09
and how you how you choose to go about it.
1:03:12
And then also when it comes to wanted and unwanted biases, like
1:03:15
an AI is not going to give you that answer.
1:03:17
You yourself have to say, hey, this is the bias I want or do
1:03:21
not want in my particular case. So it's not the AI kind of
1:03:27
making that that trade off. It's the human that that
1:03:29
implements it in the end sounds. Like a great place to move to my
1:03:35
final 2 questions. So.
1:03:38
Let's take it beyond AI. What would a more inclusive
1:03:43
world look like to you and to other women who want to lead in
1:03:47
the AI space, were in the tech space.
1:03:50
What advice would you give them? I think.
1:03:53
What a more inclusive world would look like is is in a way a
1:03:57
world where this question isn't relevant anymore or you just do
1:04:00
it that that. I think that'd be a great.
1:04:03
That'd be a great indicator how that then exactly looks like.
1:04:08
I'm myself, I'm not like the biggest fan of all those, those
1:04:11
quotes unnecessarily over needing everything to be like
1:04:14
5050 or whatever. But once again, I'd like to
1:04:19
approach it. But then from a societal point
1:04:21
of view, like, hey, what is it that we want to achieve like as
1:04:24
a society and what is the most efficient way to get there?
1:04:29
And of course, there is not like a single answer to that.
1:04:33
People have tons of different opinions and things are known
1:04:36
where we as a country or as as as a continent need to go.
1:04:42
But I think when I think there's, there's this clear sign
1:04:49
out that, you know, we can benefit from from more women in
1:04:53
the space to everybody I'd like, but also also some because of
1:05:01
course, there's so many more action like axes on where you
1:05:04
can have this diversity. Conclusion is my advice would be
1:05:09
to just to just do it to just go there, right?
1:05:13
I'm, I'm doing it too. And of course I, you know, like
1:05:17
to think that's it's so exceptional when yadda, yadda,
1:05:20
yadda. But in the end, it's not.
1:05:21
It shouldn't be. Yeah.
1:05:25
It shouldn't be. And don't get me wrong, like
1:05:29
I've, I've walked into meetings where, where I saw like, you
1:05:33
know, some people looking at me like, oh, who's this little girl
1:05:35
coming in? But in all honesty, I've also
1:05:38
walked into meetings and I and I thought, who is this ugly dude
1:05:42
here, right? I mean, I'm not, I'm not saying
1:05:45
I'm, I'm a perfect human being, right?
1:05:46
I have that too. Like the world is unfair.
1:05:49
We should move to a fair world. And I every day try to work
1:05:53
towards that, like put in, put in my effort.
1:05:56
But in the end, to me, it feels like, do you then let yourself
1:06:01
be limited by it because you do have agency and you do have
1:06:05
wiggle room? Is it sometimes harder?
1:06:07
Is it sometimes unfair? Yes.
1:06:10
But there's not a reason to not do it.
1:06:12
Absolutely not. And I would love to be and I
1:06:16
think, oh, the other people you have on the podcast, please let
1:06:20
them be most beautiful examples that it that is possible and
1:06:25
that you can just that you can just do it.
1:06:29
That's the only way for it. Just do it.
1:06:31
Thank you very. Much really nice quote to end
1:06:35
the episode on Sophia's Hitman. If people want to learn more
1:06:39
about you, connect with you, or learn more about Kickstart AI
1:06:43
and your work there. Where would you like to send
1:06:45
them? Well, you can always send me a
1:06:47
direct message on LinkedIn, just Sophia, Clare Zipman or go to
1:06:52
the kickstart dot AI website and you can find us there as well.
1:06:56
And if you ever want to meet in person, we host plenty of meet
1:06:59
ups, so join one. Cool.
1:07:02
And we'll, we'll include those links in the show notes.
1:07:06
Sophia Zipman, thanks very much for an amazing conversation.
1:07:10
Listeners, Thank you very much for listening.
1:07:12
And until the next episode, keep being awesome.
1:07:21
Well, that's it. Thank you for listening to
1:07:23
episode 136 of Women Disrupting Tech titled Why Your AI
1:07:28
Implementation Fails and How to Fix It with Sophia Zitman.
1:07:33
If you want to connect with Sophia or learn more about her
1:07:36
work, please use the links in the show notes.
1:07:38
And as usual, you can find the show notes at Women Disrupting
1:07:41
Tech Talk blog via the link in the caption.
1:07:45
If you found this episode useful, please take a moment to
1:07:48
rate and review the show on Spotify, Apple Podcasts, or
1:07:52
YouTube. And if you haven't already,
1:07:54
please hit the subscribe button so you never miss a future
1:07:57
episode. Once again, thank you for
1:08:00
listening. I hope to see you in the next
1:08:02
episode. And until then, as always, keep
1:08:05
being awesome.