December 17, 2021 • 41 mins
In this episode, we welcome Jan W. Veldsink, AI Architect and Coach at Rabobank, to explore the vital role of fostering an analytical mindset within large organizations. Jan shares practical insights on how he approaches AI coaching at Rabobank, emphasizing the importance of embedding data literacy and analytical thinking across business units. He discusses how successful data-driven transformation requires more than just technology—it demands cultural alignment, education, and strategic leadership. The conversation delves into:
👤 About the Guest Jan W. Veldsink MSc is a forward-thinking AI strategist, speaker, and educator. With extensive experience in artificial intelligence, cybersecurity, systems thinking, and innovation, Jan has advised top financial institutions and leads AI efforts within compliance and fraud detection at Rabobank. He also serves as Core Faculty at Nyenrode Business Universiteit, where he teaches on AI and cybersecurity in executive MBA programs. Jan’s mission: to create resilient, secure, and high-performing organizations by aligning technology, people, and purpose.
- How to design and scale effective corporate data strategies
- Real-world banking use cases Jan has helped implement
- The promise—and current limitations—of AutoML tools for non-technical teams
👤 About the Guest Jan W. Veldsink MSc is a forward-thinking AI strategist, speaker, and educator. With extensive experience in artificial intelligence, cybersecurity, systems thinking, and innovation, Jan has advised top financial institutions and leads AI efforts within compliance and fraud detection at Rabobank. He also serves as Core Faculty at Nyenrode Business Universiteit, where he teaches on AI and cybersecurity in executive MBA programs. Jan’s mission: to create resilient, secure, and high-performing organizations by aligning technology, people, and purpose.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:06):
Bedroom is an independent data science an aiphirm specialized in
data driven business change. In this podcast, our guests help
us spread knowledge and experience with our listeners.
Speaker 2 (00:28):
Hello, good morning John, How are you doing today?
Speaker 3 (00:32):
Good morning has So I'm doing fine. I'm doing fine.
Good morning to you two.
Speaker 2 (00:36):
So thanks for taking the time. Where are you calling from?
Speaker 3 (00:42):
Oh? I'm calling from my home in the Netherlands, in Hottem,
a small place in the northern part of Netherlands. And
as most of the people in the places are work,
I work from home. So I'm just in my home
situation now doing some work and talking to you.
Speaker 2 (01:01):
Of course, of course sounds good. And tell me what
do you have ahead for the day, what plans or
which enda do you have for allday?
Speaker 3 (01:09):
It's like always, it might be a day full of
full of things to do or new projects to arrive.
And for certain I need to do some preparation because
tomorrow I have a MBA class to give at a
Nayrow University, so I do some preparation for that, and
(01:30):
that course will be on AI and cybersecurity, which is
both our hot items. So looking forward to that too.
Speaker 2 (01:38):
I see, I see so apart from the role that
you have a ravevank. You also collaborate and work at
the university. Right, So I was going to say that
you are the first AI coach because that's the title
that you have in LinkedIn for the role that you
do at a bank. But I'm curious to know what
(02:03):
does it entail to be an AI coach? Is it
more about promoting you know, the the the understanding and
the culture on data science, AA capabilities and so on
and so forth. Is it about, you know, ensuring that
data literacy exists amongst different business units in Reverbank? What
(02:24):
is it all about them? I'm curious to know what's
an AA coach?
Speaker 3 (02:27):
Okay, that's interesting all of the all of the things
and more. I do. I do a lot of things
you mentioned. I do create my own AI machine learning algorithms.
I I'm mostly to be found inside project teams or
inside initiatives within the bank that are willing or going
(02:50):
to use artificial intelligence or AI technologies as I call them,
and I will act as a let's say, a do
you call it an accelerator for them or pivoting point
in creating ideas how to how to use AI in
(03:11):
their problem situation. Or in their question. So I'm busy
with a literacy on AI. What is it? What can
we achieve with it? I'm busy with a kind of
architecture to find the right architecture for the solution. Even
I create my own models and help finding the right
(03:34):
models and the right settings for them to ask answer
the question, because that is a little bit what what
I do. Busy with decision decision support with the use
of artificial intelligence technologies.
Speaker 2 (03:52):
So it's it's a combination of being a consultant for
your business, for the internal business internal a little bit
of data science when it comes to data modeling and
applying analytics, some of the data engineering that you require
or infrastructure work that you require when developing a model.
(04:13):
I also some promoting of this discipline so that people
insteade of the company speak the same language.
Speaker 3 (04:21):
Okay, yeah, and that's and that that's that's not only
on operational level, but also a middle middle middle management
level and top top management level. So it's all across
the stack of business and I T and management and
operations and develops. So all of those.
Speaker 2 (04:42):
Areas many things to do apart from the work that
you do at the university. So I'm curious where you
find the time.
Speaker 3 (04:50):
Oh, that's that's easy. That's the easy part because both
are let's say, energy giving activities. Are you in the
Rabble Bank? I practice what I preach at university. That's
a little bit of thing in the university. It's cool
to be on the edge of the developments and talk
(05:12):
with other professors and researchers on what they what they
do and what they find and and try to build
that bridge between that academic field of working and let's
say the real world inside a company like Rabble Bank.
Speaker 2 (05:30):
I'm curious, and how did you get to the kind
of role that you are performing today? But not only
about the world that you're doing at the bank, but
also that combination of skills that allow you to also
teach and help people to understand what AI like. Where
do you come from in terms of academical knowledge or
(05:53):
academical background, and which roles helped you to move towards
where you are now.
Speaker 3 (06:00):
Oh, that's a that's a journey. That's a that's a
Journey's I started started all at me wanting to become
a teacher in the let's say way back, and I
haven't I have a background in education. I have a
background in computer science and I have my master's in
(06:21):
artificial intelligence, and then I started teaching because I liked it.
Because I didn't, I didn't end up being a teacher.
I ended up being a computer scientist and working in
and with organizations, and I studied some organization dynamics and philosophy,
(06:42):
and I combined those things in the roles I played
over the years. So I started off using my two
hands to program E r P systems, and then I
came into it tea management. I came into education like
the university, and all along the route I gathered all
(07:04):
my knowledge and skills on data, understanding of data, understanding
of business, of course, but also understanding of the AI
technologies and how to apply them in organizations and how
to form organizations around the idea of applying artificial intelligence,
(07:25):
because that's a challenge in its.
Speaker 2 (07:28):
Own plus when you try to embed, build, or design
these capabilities within an organization, it's also hard because of
the popular feeling that if you do it, it's going to
take out some of the jobs that some of the
people that this are performing. When we like to call
(07:50):
it humanized intelligence, because these are mechanisms that are supposed
to empower the individuals, same as you know with robots
that may help you to bring you a beer at
home in such a simple in such a simpler scenario.
But sometimes sometimes in the corporate or industrial environments, is
perceived as something negative. That's true, How do you tackle that?
(08:16):
Do you also fight against that with you know, the
the promotion of these technological capabilities within the bank and
at the university.
Speaker 3 (08:26):
Yes, some sort of. The first the first thing is
to try to be as optimistic as possible and be
as realistic as possible. And for the rest, it's all
about having the idea of what the incentive for an
organization is. If the incentive of an organization is that
(08:48):
managers get paid more because they have two hundred staff
instead of two staff, then that incentive needs to be
the subject of your discussion. And in most of the
cases that realm. Of course, a large corporate that has
such incentives, manager with two hundred staff is more important
(09:10):
than somebody with two staff or non staff, So I
promote them. Perhaps I'm more I could be more productive
with one AI than you were two hundred staff. But
that's a difficult conversation. But you always need to start
and at the value. What's the value for the organization,
what's the value for customers, what's the value for society?
(09:34):
And then from that dribbling back, Okay, how can we
then achieve that within our organization knowing the limitations we
have in the knowledge of people of artificial intelligence acceptance,
the knowledge of technology and how to deal with technology.
(09:57):
And I do promote a lot that people really look
into that artificial intelligence and not see it as some
kind of mystic technology that like an oracle, just spits
out some predictions and cannot you cannot question it. So
I'm very much in favor of explainable AI and that
(10:18):
you can ask the model or the AI the question, Okay,
why did you can come to this conclusion? Tell me,
explain to me why you come to this conclusion. And
that's a little bit that the human interface, in human
in the loop, a human centric approach to this technology.
Speaker 2 (10:38):
That's verty much what a lot of the people that
are starting in this discipline need, right, even some of
the stakeholders that sometimes trust a service provider that is
saying that he has a magnificent technology, whether that's a
platform or a custom solution, he needs to understand in
(10:58):
and out what the process is doing, what's happening behind
that algorithm, at least to get a feeling of the
reasoning of what that outcome is coming from, where that
outcome is coming from. I think that this is something
that was lacking in the past. And I believe that
you can call them AI winters, and you can call
(11:20):
them misunderstandings that happened in the past were related to
the feeling of contracting or hiring black boxes, as you know,
a service tool. And I think now now the world
and and and the different teams that are looking to
embed these capabilities within within their teams already know that
(11:41):
they want to know how or what's happening in the background. Okay,
I'm curious to know use cases or platforms or tools
that you've developed internally. I mean at the bank that
you could highlight or tell us about or it's told
me about, because these are interesting. This can be related
(12:04):
to the customer, the end customer of the bank, or
automated operations that you've been able to I don't know,
make more efficient because of the use of a algorithms.
So anything that you would think it's interesting.
Speaker 3 (12:19):
Yeah, that's many things, many many things. So there are
small things, and I do not I do not. I
don't think low of small solutions. So I think that
everywhere in the process where you can see that there
(12:40):
is a decision to be made by a human or
even a machine, and you can optimize that by using
artificial intelligence, are points you can you can tap in.
And for us, I live in the compliance worlds of
frauds and AML and and those kinds of areas, and
so I do a lot of things in monitoring that's
(13:02):
high volume, high velocity stuff. And we do a lot
of work with AI and AI technologies as I call them,
because it's a broad spectrum of not AI is not
for me. It's not just one thing. It's a broad
spectrum of technologies we apply and we use it for
(13:22):
finding fraud, finding money laundering, and finding and determining risks
of behavioral risks of customers. But also a very simple
thing like classifying invoices using r p A a robot
(13:43):
process automation robotic process automation, and in that r p
E there is a machine learned model because it can
perform much much better than a human coded set of codings.
And with with the hundreds of documents we get in
from all kinds of feeds we try we categorize them
(14:08):
into are they relevant for regulatory or not. We have
some invoice classification algorithm, et cetera, so it ranges from
the classical supervised learning models. Okay, if these and these
things are true, then it might be frauds of the
chance of fraud is x y z to a text
(14:34):
mining operation. I have some text in an invoice. Can
you tell me in which bucket this needs to be
put to a normally detection unsupervised learning for kinds of
risk detections or a normally detections in network traffic, et cetera.
(14:54):
So there are a lot of applications in which the
human are and the human mind cannot oversee the vast
amount of data anymore or the number of columns if
you talk in spreadsheets, so find find find the relations
and correlations in two thousand columns spreadsheet. That's impossible to
(15:17):
do for human but a machine and a machine learning
algorithm can put some can give you insight from that data,
so we use it on. We also use machine learning
and AI technologies to gain insight from data. So not
only in the automated way in doing the classifier in
(15:40):
fraud or in a m L or in the in
the classification of documents, but also gaining insight and the
quality of data.
Speaker 2 (15:49):
That is actually a really good point, meaning obviously AI
mechanisms are going to be useful when it comes to
automating stuff, just because some of the human tasks that
a human is supposed to be doing traditionally are not
intended to be for humans because they are too repetitive
(16:11):
and they do not require you know, much reasoning, much thinking,
and the human ideally would be dedicated to some other
tasks where his creativity can be put to use rather
than just hitting in front of a laptop, just clicking
a button. So some of the tasks that you can automate,
I believe AI comes in handy bat at the same
(16:34):
time computational power in the realm of processing data at
a quicker pace and finding as you said, trends or
patterns or insights across massive data sets and a human
brain cannot process that, but an algorithm can within seconds.
(16:54):
I believe that that dual view of how it can
be beneficial is what people need to understand, right, not
substitute in the human, but empowering him to do something
else that's better for him or for her, and at
the same time using the computational power for what's really needed.
Speaker 3 (17:10):
I always like to think on it like a let's say,
AI assistant, So see it as an assistant. My work
is doing the creative stuff and the real decision making,
but I need the machine to do the calculation for me.
So I outsourced some of my capabilities to that machine,
(17:31):
because that's why I said it would be all end
boring and repetitive. And I make mistakes, the machine makes mistakes.
To it, the AI makes mistakes. We have to. We
cannot say that the machine doesn't make mistakes. But so
there needs to be a trade off in that the
assistant paradigm I like to use in this context.
Speaker 2 (17:53):
And at the same time, I think that AI mechanisms
will come in very handy for some of the application
or jobs were safety it's a key feature for you know,
for humans. Sometimes some of the roles or jobs that
humans need to do or are required to do in
their day to day maybe not as safe as you
(18:15):
know ideally they should be. So maybe in the future
robots with the appropriate AI mechanisms embedded in they will
be able to substitute them in here, not only in
the boring stuff, but also in the necessary stuff.
Speaker 3 (18:31):
And also in locations where humans cannot operate. And also
a colleague, professor of Mine, Chris the Ivand is researching
on evolutionary robotics, where he says, Okay, there are two
things in evolution we need to take into account. That's
(18:52):
reproduction and selection. And if we can can do that
with a quite quite autonomous robot, we can place them
on Mars and let them if evolve into a being
that is suitable for that context that we cannot even
imagine from our own standpoint.
Speaker 2 (19:14):
I'm curious just to know how does the bank where
you're working at structure the different teams in regards to
data science, the data capabilities AI. Do you have a
team that you work with, Do you have a ceo
vertical and all of the when I mean cdo I
(19:35):
mean Chief Data Officer, shift, THETA officer and everything in
regards to day that depend from it. What's the strategical
sort of organizational structure that you have in place.
Speaker 3 (19:45):
Oh, that's that's a good question. That's a super good question.
But we're still in the forming of that, and there
are different schools within the Bank, and of course as
a large corporate, we have a data officer and we
have a data science department. All of that, we have
a competent center, etc. That's all being centralized. And on
(20:07):
the other side, we have a lot of data science
or decision engineering capabilities in the teams, in the business teams.
Because if you only see it from a data if
you only perceive artificial intelligence as a data science aspect,
that's only for me. There's only one aspect of artificial intelligence,
(20:30):
and using those technologies, most of the things are coming
from the business knowledge and the domain knowledge and the
data domain data knowledge that resides within the business teams
within the Bank. So what I see is a little
bit the duality we have. Of course, we are forming
(20:51):
a data lake, and that's all being centralized, centralized archase.
Everybody can use the data that is allowed to that's
the centralized part. But creating the real functional artificial intelligence
technologies to support the business in making their decisions. That's
why I caught decision engineering instead of data science, and
(21:12):
that decision engineering is for me, a combination of data science,
computer science, psychology, managerial science, decision science, because you need
to be in that area where it belongs, because you
need also to have capable people there to maintain the
(21:37):
artificial intelligence. It's not something like we thought in the
beginning of software. I created an application and that will
last forever. And when you go to models and artificial intelligence,
you need to think on do I need to retrain it?
And can I interpret the outcome in business in the
(21:58):
form of business value? And does it still perform business
value for me? Those are typically other questions. Then if
you see a data scientist that says, okay, the accuracy
is ninety percent, but that the ninety percent has no
meaning in the business context if you cannot place real
(22:19):
value against that. And I think in the future there
will be of course centralized functions like data science and
data management and data governance, et cetera. But the satellites
in the organization and in the business teams are the
ones who are going to create the real, functional and
(22:42):
useful and valuable AI for the business.
Speaker 2 (22:45):
I guess it all relates to having the required training
at least to perform some basic functions. So I guess
that the data science AA team or different roles that
you know would belong to the teams that you're talking
about within be able to support all of the business functions.
If the business functions don't don't have at least a
(23:07):
few key people that can communicate with the other ones,
you know, and that as the receivers and the well
informed receivers of these projects right.
Speaker 3 (23:19):
Now, but I want to go one step further. I
think they should be able to create their own models
if the if the conditions are being set up in
such a way that you can say, okay, but my
ideal team for a for artificial intelligence project is a
(23:41):
business domain specialist, a business domain data specialist, and and
one guy that that knows artificial intelligence and machine learning
and knows the statistics around measures and and evaluation criteria,
et cetera. That's my ideal team and not three. But
(24:03):
it can be more, and every function can be more
depending on the size of the of the problem. But
the ownership also should lie inside the business and not
in data science teams that that in our case is
located within a DTO or in an I kind of organization.
Speaker 2 (24:21):
I guess that this takes me to my next question.
So if we if we or any team on the
data engineering architectural side of things is able to set
up an infrastructure that supports all of the use cases
that you or that someone aims to launch. Then a
(24:43):
useful tool STAG will make use of platforms like data robot,
H two, assure, machine Learning, bigger MAL. These are the
kind of tools that if the foundational steps were appropriately performed,
then someone who isn't extremely technical, can I still use
(25:04):
these interfaces or these platforms with a UX to build
and deploy models at least to give them a try,
and to even operationalize some of the use cases that
the business is thinking of.
Speaker 3 (25:18):
Right, I use what we have. Of course, we are
a large organization. We have different stacks. We have of
course a quiet in stack, and we have some people
working in R and mad lab, and we have some
people working in NIME, and I think data equal. But
I predominantly use bakem L because and of course we
(25:39):
have agile mL and stuff like that, but I predominantly
use BAKEML to just do that to give people the
opportunity to okay, I can just start with the model
and they're not being able to put it in production
because that's one step beyond. But some of the models
we use and we create being put in production because
(26:02):
it's in an automated it's been automated. But it started
all off with creating a model in big L and
say okay, this is a good model, this is the data.
It's a good model. Let's put it in the production
and without recoding it. Because that's that's the most horrorsome
scenario you can find. And at my in my classes,
(26:24):
at my MBA classes, I teach I teach my students
how to use tools like big L because if you
are on a management position, because MBA classes are about
management positions, I think you need to understand the rationale
of machine learning, and you need to understand the rationale
(26:47):
of data. And I just finished my round of assignments,
and one assignments I created for them is, okay, here
do you have a data set having have to deal
with churn. Just build me a model that that can
prevent and predict churn in in this data set. And
(27:08):
it's it's amazing that people who are not predominantly technical
can from that business perspective, with the data set and
and basic understanding of the business problem, have have created
models that work. And I think that sounds very simple,
but in the in the in the early days of automation,
(27:29):
that was limited to a few people in the I
T department that could program. So I think we're really
at that level of democratizing software creation. That that that
that are companies doing things like a big mel or Novolo,
that the non coding platforms where we as business can
(27:55):
build our functionality and more and more. And you touched
a little bit upon on that regulators and governments will
ask you as an organization to be in control of
the AI you are going to use. So having said that,
you you will have you will have the need for
(28:15):
capabilities inside your organization to build, maintain, and define artificial
intelligence technologies because if you if you get them from
pre defined if you use pre cooked or pre defined
models from vendors, then I think in the end you
will come into a problem because you need to understand
(28:37):
what that model exactly does. So then that does make
it as easy as possible to create a model and
be in control on the hyper parameters and the goal
and the data data yourself as an organization and with
with tools like like the platforms you mentioned, it is
(28:59):
it is easier and it's rather cheap to create a
model and if it doesn't work, just straight away make
a new model and try out as many as many
as you want. And that's a different notion than we
had probably ten fifteen years ago, where it was a
super science to create a simple model, and that's been
(29:20):
taking away from us luckily for the most part.
Speaker 2 (29:25):
I very much agree. And if I was in a
situation of, you know, building out some capabilities within a
smaller bank in regards to, you know, setting up some
automation for processes, living a side the more traditional modeling
such as client scoring and things like that, some fairly
(29:48):
new stuff which requires me to i don't know, hire
some data scientists and data engineers and think of the
deck stack that I'm going to use. So living aside
these tools that we've mentioned for like low code or
no code mel yeah, I'm also talking about languages, about
cloud infrastructure, about a little bit here and there. Probably
(30:11):
you've you've been through this road, and you've been through
this path, and you've probably learned from mistakes. So if
you were to provide a very high level advice to
whoever is you know, going through this thought process at
the moment, what would you say at the moment of
time where we are now in terms of the technology,
(30:32):
the discipline, everything, it's a very problem.
Speaker 3 (30:35):
I can I can give it a try. I can
give try. I can give it a try. If I
was a very small business, I would start off with
a platform like Blemel or a big mel or something
like that. That is easy, that is low hanging, that
that's low, that's that's cheap. Because if I would, if
(30:57):
I would be started, I would want the cheap, fast
and available. I would think, on okay, what's the data
that I'm going to provide that algorithm, and I always
would want to be able to get a model out
of there that I would be independent of that platform
(31:20):
where I learned the model for execution. That's one I
would if in my case, I have let's say relative
sensitive data, then I chose to put that on prem
on a few machines because I think it's it's in
(31:41):
the world I live. Very very important to be very
secure and to be very precise on where your data resides.
And with all the say Betriet Act and Save Harbor
stuff in the US, you need to be very very
precise on what you are going to put in the
cloud or not what my learning is. Do not start
(32:03):
with an AI project. Just start with a process in
which you want to automate some decision steps and go
look for those decisions in your process and see if
there is an example or if there are algorithms that
will help you support that decision in that process. So
(32:27):
I would always start from that point. So look at
your processes, look at your most valuable processes. Where is
your value created and where can you add that decision
automated decision capability. Yeah, do not start with large teams.
So I was contacted by a large bank and they
(32:51):
had hired four hundred data scientists and they didn't get
one model life in not a year. So the idea
is to start with the let's say the real practical
practical problems and also start with the start with the
(33:13):
execution of the model in mind, where do you want
to execute? Is an app? Is it ap API capable?
Do I have to write Java or make a note
GS or do Python or whatever. So look at the
place where you're going to apply the artificial intelligence. What
(33:35):
touch point do you have of technology? And start building backwards.
So if I need a Java binding or I need
a no GS binding, do I have my technology stech
that creates the model ready to create that note GS
component or model that so it can run into my process.
(33:58):
So what I see a lot is that people building
beautiful models but not having thought about how to implement
it into the reality. And always provide and always look
for feedback loops. Can you can you get feedback from
the real world on your suggested decisions?
Speaker 2 (34:19):
Very interesting? You were mentioning a large bank that was
thinking of hiring four hundred data SI data scientists. Wouldn't
it make sense that this happens progressively and during that
time apart from you know, thinking of technology that could
(34:40):
be a low hanging fruit for finding insights or for
visulating visulation some data sets that they also think about
feeling or being supported by specific boutique companies that are
a top notch in providing X data science services. Or
would you think that this needs to happen once the
(35:04):
team is mature enough on once there is a roadmap,
or do you think that some strategical advice from the
outside could also be beneficial at the start of this
building of teams? What do you think?
Speaker 3 (35:17):
I see the latter It can be productive to do
that and you always need to be in charge. So
somebody can advise you, but you have to make the
choice because you have to live it. And most of
the time the consultants can advise you and help you
and make your first steps, but be sure that after
(35:40):
they are gone, you can make the same steps and
you can progress, progress over and over so that you
do not depend on get addicted what I call addicted
to those consultancy services. Build your own capabilities also should
be should be the aim of when when hiring consultants
(36:03):
for this, the aim should be building your own capabilities.
Speaker 2 (36:07):
No, I completely agree with that. I was just asking
because it's I'd say that at the start, it is
pretty hard to already on board so many individuals. Having
sometimes an outside view with a broader context can also
make you think, Okay, do I really need these kind
(36:28):
of profiles long term? Maybe giving it a try with
uh embedding some people from the outside allow you to conclude, Okay,
I do need this, I do not need it, or
this has worked well, let's continue to grow in this
area now with my own workforce.
Speaker 3 (36:49):
Short of say, and because we always I hear a
lot around me that those people, those data scientists are
scarce and naming it scarce means that everybody wants to
gather them and have them because they are scarce. But
you need to start off the point where do I
(37:10):
want to create value and what do I need to
create that value? And perhaps you do not need not
need four on the data scientists, you only need ten
and good supported support for your business.
Speaker 2 (37:21):
Teams obviously and always having that business sense. So as
you said before, it's about decision engineering or business science
and not only thinking about data science, right. I mean
data science here is a tool, as I mean, to
an end that is supposed to be solving business questions.
But it isn't about doing the best performing model only,
(37:46):
but about making sure that that model is useful for
something and not creating it in isolation to the stakeholders
that may need the output of that of that model.
So I really.
Speaker 3 (37:57):
Agree with that.
Speaker 2 (37:57):
Okay, John, thanks very much for the time. I only
have one question pending here that I always ask is
if you were to tell me or give me one
name of someone that I should be calling or that
I should be speaking about aileada science applied to business
(38:20):
real cases or that operates in the research industry in
regards to data or A. I would you have a
name for me?
Speaker 3 (38:30):
Yes, I would. I would have a name for you.
I even would have two names for you, but you
only ask one.
Speaker 2 (38:37):
You can give me two on fine with that, okay.
Speaker 3 (38:40):
The first one I would speak to is Mark Kuckelbert
Mark Coockulbert is a is A is a scientist working
on the ethics of AI and and working on let's
say the academic and really academic view on this topic.
So very interesting guy to speak to. The second guy
(39:02):
I would speak to Corbiaus core Bias is a leading
architect in the Netherlands using artificial intelligence, but also an
inventor of a new new ways of thinking on how
to architecture and how to model applications in a distributed
(39:23):
and AI driven way. So that's both of the guys
would be very interesting for you to talk to.
Speaker 2 (39:31):
I do need some help with the surname of the
first one, you said, Mark Kulger. How do you spell that.
Speaker 3 (39:41):
C oh ce O E c k E l b
E r G H not.
Speaker 2 (39:53):
But and I was supposed to guess that that was
a difficult one. Okay, but both both knowed and what
you told me about them already sounds really interesting and appealing.
I'm sure they will both make great guests of the
data stand up Shosh.
Speaker 3 (40:13):
Suddenly suddenly I know for certain.
Speaker 2 (40:17):
Again, John, thanks for taking the time. Thanks for a
very interesting call. I've learned lots of what you're doing
at the bank and also really aspiring to know that
apart from the day to day job that you're running
at the bank, you're also doing some work at the university,
which to me it's it's, you know, a motivation because
(40:38):
you need to find the time and I'm sure that's
rewarding to you and in the future I would love
to do that as well.
Speaker 3 (40:43):
So thanks, thank you.
Speaker 2 (40:47):
I hope you'll have a nice day.
Speaker 3 (40:48):
By Jack AMBERSTSTSTSTRUCTU
Bedroom is an independent data science an aiphirm specialized in
data driven business change. In this podcast, our guests help
us spread knowledge and experience with our listeners.
Speaker 2 (00:28):
Hello, good morning John, How are you doing today?
Speaker 3 (00:32):
Good morning has So I'm doing fine. I'm doing fine.
Good morning to you two.
Speaker 2 (00:36):
So thanks for taking the time. Where are you calling from?
Speaker 3 (00:42):
Oh? I'm calling from my home in the Netherlands, in Hottem,
a small place in the northern part of Netherlands. And
as most of the people in the places are work,
I work from home. So I'm just in my home
situation now doing some work and talking to you.
Speaker 2 (01:01):
Of course, of course sounds good. And tell me what
do you have ahead for the day, what plans or
which enda do you have for allday?
Speaker 3 (01:09):
It's like always, it might be a day full of
full of things to do or new projects to arrive.
And for certain I need to do some preparation because
tomorrow I have a MBA class to give at a
Nayrow University, so I do some preparation for that, and
(01:30):
that course will be on AI and cybersecurity, which is
both our hot items. So looking forward to that too.
Speaker 2 (01:38):
I see, I see so apart from the role that
you have a ravevank. You also collaborate and work at
the university. Right, So I was going to say that
you are the first AI coach because that's the title
that you have in LinkedIn for the role that you
do at a bank. But I'm curious to know what
(02:03):
does it entail to be an AI coach? Is it
more about promoting you know, the the the understanding and
the culture on data science, AA capabilities and so on
and so forth. Is it about, you know, ensuring that
data literacy exists amongst different business units in Reverbank? What
(02:24):
is it all about them? I'm curious to know what's
an AA coach?
Speaker 3 (02:27):
Okay, that's interesting all of the all of the things
and more. I do. I do a lot of things
you mentioned. I do create my own AI machine learning algorithms.
I I'm mostly to be found inside project teams or
inside initiatives within the bank that are willing or going
(02:50):
to use artificial intelligence or AI technologies as I call them,
and I will act as a let's say, a do
you call it an accelerator for them or pivoting point
in creating ideas how to how to use AI in
(03:11):
their problem situation. Or in their question. So I'm busy
with a literacy on AI. What is it? What can
we achieve with it? I'm busy with a kind of
architecture to find the right architecture for the solution. Even
I create my own models and help finding the right
(03:34):
models and the right settings for them to ask answer
the question, because that is a little bit what what
I do. Busy with decision decision support with the use
of artificial intelligence technologies.
Speaker 2 (03:52):
So it's it's a combination of being a consultant for
your business, for the internal business internal a little bit
of data science when it comes to data modeling and
applying analytics, some of the data engineering that you require
or infrastructure work that you require when developing a model.
(04:13):
I also some promoting of this discipline so that people
insteade of the company speak the same language.
Speaker 3 (04:21):
Okay, yeah, and that's and that that's that's not only
on operational level, but also a middle middle middle management
level and top top management level. So it's all across
the stack of business and I T and management and
operations and develops. So all of those.
Speaker 2 (04:42):
Areas many things to do apart from the work that
you do at the university. So I'm curious where you
find the time.
Speaker 3 (04:50):
Oh, that's that's easy. That's the easy part because both
are let's say, energy giving activities. Are you in the
Rabble Bank? I practice what I preach at university. That's
a little bit of thing in the university. It's cool
to be on the edge of the developments and talk
(05:12):
with other professors and researchers on what they what they
do and what they find and and try to build
that bridge between that academic field of working and let's
say the real world inside a company like Rabble Bank.
Speaker 2 (05:30):
I'm curious, and how did you get to the kind
of role that you are performing today? But not only
about the world that you're doing at the bank, but
also that combination of skills that allow you to also
teach and help people to understand what AI like. Where
do you come from in terms of academical knowledge or
(05:53):
academical background, and which roles helped you to move towards
where you are now.
Speaker 3 (06:00):
Oh, that's a that's a journey. That's a that's a
Journey's I started started all at me wanting to become
a teacher in the let's say way back, and I
haven't I have a background in education. I have a
background in computer science and I have my master's in
(06:21):
artificial intelligence, and then I started teaching because I liked it.
Because I didn't, I didn't end up being a teacher.
I ended up being a computer scientist and working in
and with organizations, and I studied some organization dynamics and philosophy,
(06:42):
and I combined those things in the roles I played
over the years. So I started off using my two
hands to program E r P systems, and then I
came into it tea management. I came into education like
the university, and all along the route I gathered all
(07:04):
my knowledge and skills on data, understanding of data, understanding
of business, of course, but also understanding of the AI
technologies and how to apply them in organizations and how
to form organizations around the idea of applying artificial intelligence,
(07:25):
because that's a challenge in its.
Speaker 2 (07:28):
Own plus when you try to embed, build, or design
these capabilities within an organization, it's also hard because of
the popular feeling that if you do it, it's going to
take out some of the jobs that some of the
people that this are performing. When we like to call
(07:50):
it humanized intelligence, because these are mechanisms that are supposed
to empower the individuals, same as you know with robots
that may help you to bring you a beer at
home in such a simple in such a simpler scenario.
But sometimes sometimes in the corporate or industrial environments, is
perceived as something negative. That's true, How do you tackle that?
(08:16):
Do you also fight against that with you know, the
the promotion of these technological capabilities within the bank and
at the university.
Speaker 3 (08:26):
Yes, some sort of. The first the first thing is
to try to be as optimistic as possible and be
as realistic as possible. And for the rest, it's all
about having the idea of what the incentive for an
organization is. If the incentive of an organization is that
(08:48):
managers get paid more because they have two hundred staff
instead of two staff, then that incentive needs to be
the subject of your discussion. And in most of the
cases that realm. Of course, a large corporate that has
such incentives, manager with two hundred staff is more important
(09:10):
than somebody with two staff or non staff, So I
promote them. Perhaps I'm more I could be more productive
with one AI than you were two hundred staff. But
that's a difficult conversation. But you always need to start
and at the value. What's the value for the organization,
what's the value for customers, what's the value for society?
(09:34):
And then from that dribbling back, Okay, how can we
then achieve that within our organization knowing the limitations we
have in the knowledge of people of artificial intelligence acceptance,
the knowledge of technology and how to deal with technology.
(09:57):
And I do promote a lot that people really look
into that artificial intelligence and not see it as some
kind of mystic technology that like an oracle, just spits
out some predictions and cannot you cannot question it. So
I'm very much in favor of explainable AI and that
(10:18):
you can ask the model or the AI the question, Okay,
why did you can come to this conclusion? Tell me,
explain to me why you come to this conclusion. And
that's a little bit that the human interface, in human
in the loop, a human centric approach to this technology.
Speaker 2 (10:38):
That's verty much what a lot of the people that
are starting in this discipline need, right, even some of
the stakeholders that sometimes trust a service provider that is
saying that he has a magnificent technology, whether that's a
platform or a custom solution, he needs to understand in
(10:58):
and out what the process is doing, what's happening behind
that algorithm, at least to get a feeling of the
reasoning of what that outcome is coming from, where that
outcome is coming from. I think that this is something
that was lacking in the past. And I believe that
you can call them AI winters, and you can call
(11:20):
them misunderstandings that happened in the past were related to
the feeling of contracting or hiring black boxes, as you know,
a service tool. And I think now now the world
and and and the different teams that are looking to
embed these capabilities within within their teams already know that
(11:41):
they want to know how or what's happening in the background. Okay,
I'm curious to know use cases or platforms or tools
that you've developed internally. I mean at the bank that
you could highlight or tell us about or it's told
me about, because these are interesting. This can be related
(12:04):
to the customer, the end customer of the bank, or
automated operations that you've been able to I don't know,
make more efficient because of the use of a algorithms.
So anything that you would think it's interesting.
Speaker 3 (12:19):
Yeah, that's many things, many many things. So there are
small things, and I do not I do not. I
don't think low of small solutions. So I think that
everywhere in the process where you can see that there
(12:40):
is a decision to be made by a human or
even a machine, and you can optimize that by using
artificial intelligence, are points you can you can tap in.
And for us, I live in the compliance worlds of
frauds and AML and and those kinds of areas, and
so I do a lot of things in monitoring that's
(13:02):
high volume, high velocity stuff. And we do a lot
of work with AI and AI technologies as I call them,
because it's a broad spectrum of not AI is not
for me. It's not just one thing. It's a broad
spectrum of technologies we apply and we use it for
(13:22):
finding fraud, finding money laundering, and finding and determining risks
of behavioral risks of customers. But also a very simple
thing like classifying invoices using r p A a robot
(13:43):
process automation robotic process automation, and in that r p
E there is a machine learned model because it can
perform much much better than a human coded set of codings.
And with with the hundreds of documents we get in
from all kinds of feeds we try we categorize them
(14:08):
into are they relevant for regulatory or not. We have
some invoice classification algorithm, et cetera, so it ranges from
the classical supervised learning models. Okay, if these and these
things are true, then it might be frauds of the
chance of fraud is x y z to a text
(14:34):
mining operation. I have some text in an invoice. Can
you tell me in which bucket this needs to be
put to a normally detection unsupervised learning for kinds of
risk detections or a normally detections in network traffic, et cetera.
(14:54):
So there are a lot of applications in which the
human are and the human mind cannot oversee the vast
amount of data anymore or the number of columns if
you talk in spreadsheets, so find find find the relations
and correlations in two thousand columns spreadsheet. That's impossible to
(15:17):
do for human but a machine and a machine learning
algorithm can put some can give you insight from that data,
so we use it on. We also use machine learning
and AI technologies to gain insight from data. So not
only in the automated way in doing the classifier in
(15:40):
fraud or in a m L or in the in
the classification of documents, but also gaining insight and the
quality of data.
Speaker 2 (15:49):
That is actually a really good point, meaning obviously AI
mechanisms are going to be useful when it comes to
automating stuff, just because some of the human tasks that
a human is supposed to be doing traditionally are not
intended to be for humans because they are too repetitive
(16:11):
and they do not require you know, much reasoning, much thinking,
and the human ideally would be dedicated to some other
tasks where his creativity can be put to use rather
than just hitting in front of a laptop, just clicking
a button. So some of the tasks that you can automate,
I believe AI comes in handy bat at the same
(16:34):
time computational power in the realm of processing data at
a quicker pace and finding as you said, trends or
patterns or insights across massive data sets and a human
brain cannot process that, but an algorithm can within seconds.
(16:54):
I believe that that dual view of how it can
be beneficial is what people need to understand, right, not
substitute in the human, but empowering him to do something
else that's better for him or for her, and at
the same time using the computational power for what's really needed.
Speaker 3 (17:10):
I always like to think on it like a let's say,
AI assistant, So see it as an assistant. My work
is doing the creative stuff and the real decision making,
but I need the machine to do the calculation for me.
So I outsourced some of my capabilities to that machine,
(17:31):
because that's why I said it would be all end
boring and repetitive. And I make mistakes, the machine makes mistakes.
To it, the AI makes mistakes. We have to. We
cannot say that the machine doesn't make mistakes. But so
there needs to be a trade off in that the
assistant paradigm I like to use in this context.
Speaker 2 (17:53):
And at the same time, I think that AI mechanisms
will come in very handy for some of the application
or jobs were safety it's a key feature for you know,
for humans. Sometimes some of the roles or jobs that
humans need to do or are required to do in
their day to day maybe not as safe as you
(18:15):
know ideally they should be. So maybe in the future
robots with the appropriate AI mechanisms embedded in they will
be able to substitute them in here, not only in
the boring stuff, but also in the necessary stuff.
Speaker 3 (18:31):
And also in locations where humans cannot operate. And also
a colleague, professor of Mine, Chris the Ivand is researching
on evolutionary robotics, where he says, Okay, there are two
things in evolution we need to take into account. That's
(18:52):
reproduction and selection. And if we can can do that
with a quite quite autonomous robot, we can place them
on Mars and let them if evolve into a being
that is suitable for that context that we cannot even
imagine from our own standpoint.
Speaker 2 (19:14):
I'm curious just to know how does the bank where
you're working at structure the different teams in regards to
data science, the data capabilities AI. Do you have a
team that you work with, Do you have a ceo
vertical and all of the when I mean cdo I
(19:35):
mean Chief Data Officer, shift, THETA officer and everything in
regards to day that depend from it. What's the strategical
sort of organizational structure that you have in place.
Speaker 3 (19:45):
Oh, that's that's a good question. That's a super good question.
But we're still in the forming of that, and there
are different schools within the Bank, and of course as
a large corporate, we have a data officer and we
have a data science department. All of that, we have
a competent center, etc. That's all being centralized. And on
(20:07):
the other side, we have a lot of data science
or decision engineering capabilities in the teams, in the business teams.
Because if you only see it from a data if
you only perceive artificial intelligence as a data science aspect,
that's only for me. There's only one aspect of artificial intelligence,
(20:30):
and using those technologies, most of the things are coming
from the business knowledge and the domain knowledge and the
data domain data knowledge that resides within the business teams
within the Bank. So what I see is a little
bit the duality we have. Of course, we are forming
(20:51):
a data lake, and that's all being centralized, centralized archase.
Everybody can use the data that is allowed to that's
the centralized part. But creating the real functional artificial intelligence
technologies to support the business in making their decisions. That's
why I caught decision engineering instead of data science, and
(21:12):
that decision engineering is for me, a combination of data science,
computer science, psychology, managerial science, decision science, because you need
to be in that area where it belongs, because you
need also to have capable people there to maintain the
(21:37):
artificial intelligence. It's not something like we thought in the
beginning of software. I created an application and that will
last forever. And when you go to models and artificial intelligence,
you need to think on do I need to retrain it?
And can I interpret the outcome in business in the
(21:58):
form of business value? And does it still perform business
value for me? Those are typically other questions. Then if
you see a data scientist that says, okay, the accuracy
is ninety percent, but that the ninety percent has no
meaning in the business context if you cannot place real
(22:19):
value against that. And I think in the future there
will be of course centralized functions like data science and
data management and data governance, et cetera. But the satellites
in the organization and in the business teams are the
ones who are going to create the real, functional and
(22:42):
useful and valuable AI for the business.
Speaker 2 (22:45):
I guess it all relates to having the required training
at least to perform some basic functions. So I guess
that the data science AA team or different roles that
you know would belong to the teams that you're talking
about within be able to support all of the business functions.
If the business functions don't don't have at least a
(23:07):
few key people that can communicate with the other ones,
you know, and that as the receivers and the well
informed receivers of these projects right.
Speaker 3 (23:19):
Now, but I want to go one step further. I
think they should be able to create their own models
if the if the conditions are being set up in
such a way that you can say, okay, but my
ideal team for a for artificial intelligence project is a
(23:41):
business domain specialist, a business domain data specialist, and and
one guy that that knows artificial intelligence and machine learning
and knows the statistics around measures and and evaluation criteria,
et cetera. That's my ideal team and not three. But
(24:03):
it can be more, and every function can be more
depending on the size of the of the problem. But
the ownership also should lie inside the business and not
in data science teams that that in our case is
located within a DTO or in an I kind of organization.
Speaker 2 (24:21):
I guess that this takes me to my next question.
So if we if we or any team on the
data engineering architectural side of things is able to set
up an infrastructure that supports all of the use cases
that you or that someone aims to launch. Then a
(24:43):
useful tool STAG will make use of platforms like data robot,
H two, assure, machine Learning, bigger MAL. These are the
kind of tools that if the foundational steps were appropriately performed,
then someone who isn't extremely technical, can I still use
(25:04):
these interfaces or these platforms with a UX to build
and deploy models at least to give them a try,
and to even operationalize some of the use cases that
the business is thinking of.
Speaker 3 (25:18):
Right, I use what we have. Of course, we are
a large organization. We have different stacks. We have of
course a quiet in stack, and we have some people
working in R and mad lab, and we have some
people working in NIME, and I think data equal. But
I predominantly use bakem L because and of course we
(25:39):
have agile mL and stuff like that, but I predominantly
use BAKEML to just do that to give people the
opportunity to okay, I can just start with the model
and they're not being able to put it in production
because that's one step beyond. But some of the models
we use and we create being put in production because
(26:02):
it's in an automated it's been automated. But it started
all off with creating a model in big L and
say okay, this is a good model, this is the data.
It's a good model. Let's put it in the production
and without recoding it. Because that's that's the most horrorsome
scenario you can find. And at my in my classes,
(26:24):
at my MBA classes, I teach I teach my students
how to use tools like big L because if you
are on a management position, because MBA classes are about
management positions, I think you need to understand the rationale
of machine learning, and you need to understand the rationale
(26:47):
of data. And I just finished my round of assignments,
and one assignments I created for them is, okay, here
do you have a data set having have to deal
with churn. Just build me a model that that can
prevent and predict churn in in this data set. And
(27:08):
it's it's amazing that people who are not predominantly technical
can from that business perspective, with the data set and
and basic understanding of the business problem, have have created
models that work. And I think that sounds very simple,
but in the in the in the early days of automation,
(27:29):
that was limited to a few people in the I
T department that could program. So I think we're really
at that level of democratizing software creation. That that that
that are companies doing things like a big mel or Novolo,
that the non coding platforms where we as business can
(27:55):
build our functionality and more and more. And you touched
a little bit upon on that regulators and governments will
ask you as an organization to be in control of
the AI you are going to use. So having said that,
you you will have you will have the need for
(28:15):
capabilities inside your organization to build, maintain, and define artificial
intelligence technologies because if you if you get them from
pre defined if you use pre cooked or pre defined
models from vendors, then I think in the end you
will come into a problem because you need to understand
(28:37):
what that model exactly does. So then that does make
it as easy as possible to create a model and
be in control on the hyper parameters and the goal
and the data data yourself as an organization and with
with tools like like the platforms you mentioned, it is
(28:59):
it is easier and it's rather cheap to create a
model and if it doesn't work, just straight away make
a new model and try out as many as many
as you want. And that's a different notion than we
had probably ten fifteen years ago, where it was a
super science to create a simple model, and that's been
(29:20):
taking away from us luckily for the most part.
Speaker 2 (29:25):
I very much agree. And if I was in a
situation of, you know, building out some capabilities within a
smaller bank in regards to, you know, setting up some
automation for processes, living a side the more traditional modeling
such as client scoring and things like that, some fairly
(29:48):
new stuff which requires me to i don't know, hire
some data scientists and data engineers and think of the
deck stack that I'm going to use. So living aside
these tools that we've mentioned for like low code or
no code mel yeah, I'm also talking about languages, about
cloud infrastructure, about a little bit here and there. Probably
(30:11):
you've you've been through this road, and you've been through
this path, and you've probably learned from mistakes. So if
you were to provide a very high level advice to
whoever is you know, going through this thought process at
the moment, what would you say at the moment of
time where we are now in terms of the technology,
(30:32):
the discipline, everything, it's a very problem.
Speaker 3 (30:35):
I can I can give it a try. I can
give try. I can give it a try. If I
was a very small business, I would start off with
a platform like Blemel or a big mel or something
like that. That is easy, that is low hanging, that
that's low, that's that's cheap. Because if I would, if
(30:57):
I would be started, I would want the cheap, fast
and available. I would think, on okay, what's the data
that I'm going to provide that algorithm, and I always
would want to be able to get a model out
of there that I would be independent of that platform
(31:20):
where I learned the model for execution. That's one I
would if in my case, I have let's say relative
sensitive data, then I chose to put that on prem
on a few machines because I think it's it's in
(31:41):
the world I live. Very very important to be very
secure and to be very precise on where your data resides.
And with all the say Betriet Act and Save Harbor
stuff in the US, you need to be very very
precise on what you are going to put in the
cloud or not what my learning is. Do not start
(32:03):
with an AI project. Just start with a process in
which you want to automate some decision steps and go
look for those decisions in your process and see if
there is an example or if there are algorithms that
will help you support that decision in that process. So
(32:27):
I would always start from that point. So look at
your processes, look at your most valuable processes. Where is
your value created and where can you add that decision
automated decision capability. Yeah, do not start with large teams.
So I was contacted by a large bank and they
(32:51):
had hired four hundred data scientists and they didn't get
one model life in not a year. So the idea
is to start with the let's say the real practical
practical problems and also start with the start with the
(33:13):
execution of the model in mind, where do you want
to execute? Is an app? Is it ap API capable?
Do I have to write Java or make a note
GS or do Python or whatever. So look at the
place where you're going to apply the artificial intelligence. What
(33:35):
touch point do you have of technology? And start building backwards.
So if I need a Java binding or I need
a no GS binding, do I have my technology stech
that creates the model ready to create that note GS
component or model that so it can run into my process.
(33:58):
So what I see a lot is that people building
beautiful models but not having thought about how to implement
it into the reality. And always provide and always look
for feedback loops. Can you can you get feedback from
the real world on your suggested decisions?
Speaker 2 (34:19):
Very interesting? You were mentioning a large bank that was
thinking of hiring four hundred data SI data scientists. Wouldn't
it make sense that this happens progressively and during that
time apart from you know, thinking of technology that could
(34:40):
be a low hanging fruit for finding insights or for
visulating visulation some data sets that they also think about
feeling or being supported by specific boutique companies that are
a top notch in providing X data science services. Or
would you think that this needs to happen once the
(35:04):
team is mature enough on once there is a roadmap,
or do you think that some strategical advice from the
outside could also be beneficial at the start of this
building of teams? What do you think?
Speaker 3 (35:17):
I see the latter It can be productive to do
that and you always need to be in charge. So
somebody can advise you, but you have to make the
choice because you have to live it. And most of
the time the consultants can advise you and help you
and make your first steps, but be sure that after
(35:40):
they are gone, you can make the same steps and
you can progress, progress over and over so that you
do not depend on get addicted what I call addicted
to those consultancy services. Build your own capabilities also should
be should be the aim of when when hiring consultants
(36:03):
for this, the aim should be building your own capabilities.
Speaker 2 (36:07):
No, I completely agree with that. I was just asking
because it's I'd say that at the start, it is
pretty hard to already on board so many individuals. Having
sometimes an outside view with a broader context can also
make you think, Okay, do I really need these kind
(36:28):
of profiles long term? Maybe giving it a try with
uh embedding some people from the outside allow you to conclude, Okay,
I do need this, I do not need it, or
this has worked well, let's continue to grow in this
area now with my own workforce.
Speaker 3 (36:49):
Short of say, and because we always I hear a
lot around me that those people, those data scientists are
scarce and naming it scarce means that everybody wants to
gather them and have them because they are scarce. But
you need to start off the point where do I
(37:10):
want to create value and what do I need to
create that value? And perhaps you do not need not
need four on the data scientists, you only need ten
and good supported support for your business.
Speaker 2 (37:21):
Teams obviously and always having that business sense. So as
you said before, it's about decision engineering or business science
and not only thinking about data science, right. I mean
data science here is a tool, as I mean, to
an end that is supposed to be solving business questions.
But it isn't about doing the best performing model only,
(37:46):
but about making sure that that model is useful for
something and not creating it in isolation to the stakeholders
that may need the output of that of that model.
So I really.
Speaker 3 (37:57):
Agree with that.
Speaker 2 (37:57):
Okay, John, thanks very much for the time. I only
have one question pending here that I always ask is
if you were to tell me or give me one
name of someone that I should be calling or that
I should be speaking about aileada science applied to business
(38:20):
real cases or that operates in the research industry in
regards to data or A. I would you have a
name for me?
Speaker 3 (38:30):
Yes, I would. I would have a name for you.
I even would have two names for you, but you
only ask one.
Speaker 2 (38:37):
You can give me two on fine with that, okay.
Speaker 3 (38:40):
The first one I would speak to is Mark Kuckelbert
Mark Coockulbert is a is A is a scientist working
on the ethics of AI and and working on let's
say the academic and really academic view on this topic.
So very interesting guy to speak to. The second guy
(39:02):
I would speak to Corbiaus core Bias is a leading
architect in the Netherlands using artificial intelligence, but also an
inventor of a new new ways of thinking on how
to architecture and how to model applications in a distributed
(39:23):
and AI driven way. So that's both of the guys
would be very interesting for you to talk to.
Speaker 2 (39:31):
I do need some help with the surname of the
first one, you said, Mark Kulger. How do you spell that.
Speaker 3 (39:41):
C oh ce O E c k E l b
E r G H not.
Speaker 2 (39:53):
But and I was supposed to guess that that was
a difficult one. Okay, but both both knowed and what
you told me about them already sounds really interesting and appealing.
I'm sure they will both make great guests of the
data stand up Shosh.
Speaker 3 (40:13):
Suddenly suddenly I know for certain.
Speaker 2 (40:17):
Again, John, thanks for taking the time. Thanks for a
very interesting call. I've learned lots of what you're doing
at the bank and also really aspiring to know that
apart from the day to day job that you're running
at the bank, you're also doing some work at the university,
which to me it's it's, you know, a motivation because
(40:38):
you need to find the time and I'm sure that's
rewarding to you and in the future I would love
to do that as well.
Speaker 3 (40:43):
So thanks, thank you.
Speaker 2 (40:47):
I hope you'll have a nice day.
Speaker 3 (40:48):
By Jack AMBERSTSTSTSTRUCTU
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy, Jess Hilarious, And Charlamagne Tha God!