7. Frank Block, Head of IT Data Science at Roche, Switzerland

Summary

Der Vortrag von Frank umfasst sein persönliches Engagement in der KI-Branche, beginnend mit seiner Ausbildung als Physiker und seiner Zeit am CERN. Er diskutiert die Bedeutung von Datenqualität, die auch nach Jahrzehnten immer noch eine Herausforderung darstellt, und betont die Notwendigkeit, den Fokus auf die Produktivität und den Nutzen von KI-Anwendungen zu legen. Er reflektiert über die Entwicklung von KI von Expertensystemen über neuronale Netzwerke hin zu heutigen automatisierten Modellen und unterstreicht die Bedeutung von Interdisziplinarität, Soft Skills und der Bedeutung von Qualität und Kundenzentrierung in der KI-Entwicklung.

Takeaways

• 🔬 Datenqualität ist ein immerwährendes Thema, das die Produktivität von AI-Teams und die Skalierung von KI-Projekten beeinträchtigt.
• 🧠 Die Industrielle Nutzung von KI erfordert tiefgreifendes Verständnis der Geschäftslogik und der Wertschöpfung für Stakeholder.
• 💡 Die technische Seite erfordert fortlaufendes Lernen und gute Kenntnisse in Datenwissenschaft, Cloud-Architekturen und fortgeschrittenen Analysewerkzeugen.
• 🛠️ Technologie ist ein Enabler, aber das zentrale Ziel ist es, Probleme zu lösen und nicht nur um der Technologie willen zu arbeiten.
• 🤝 Interdisziplinarität ist entscheidend für erfolgreiche KI-Projekte, da sie die Komplementarität verschiedener Perspektiven und Fähigkeiten nutzt.
• 💼 Soft Skills wie das Fähigkeiten zu netzwerken und die Fähigkeit, Geschichten in der Sprache der Stakeholder zu erzählen, sind für KI-Spezialisten wichtig.
• 👨‍🔬 Die Rolle des Data Scientists hat sich von der reinen Forschung hin zu einer Produktions- und Operations-Orientierung entwickelt.
• 🔍 Die KI-Industrie hat sich von der Ersetzung von Menschen durch KI hin zu einer stärkeren Betonung der Verstärkung menschlicher Fähigkeiten entwickelt.
• 🏢 Organisationsbarriere und mangelnde Bereitschaft der mittleren Management-Ebene, KI zu adoptieren, können die Skalierung von KI-Projekten verlangsamen.
• 📈 Die Bewegung von Daten als Produkt und die Einführung von DevOps-ähnlichen Prinzipien in der KI-Entwicklung sind wichtige Entwicklungen.

Q & A

Wie hat die Erfahrung des Redners bei CERN seine Karriere beeinflusst?

Der Redner ist Physiker ausgebildet und hat bei CERN, dem größten Beschleuniger der Welt, seine Masterarbeit geschrieben. Diese Erfahrung hat ihn stark beeinflusst und ihn für die Arbeit mit großen Datenmengen und der Analyse solcher Daten interessiert gemacht.

Was ist der Large Electron-Positron Collider und wie wichtig war er für die Forschung?

Der Large Electron-Positron Collider (LEP) ist ein 27 Kilometer umfassendes Gerät, das 100 Meter unter der Erdoberfläche liegt. Es war ein gigantisches Instrument, das für die Forschung zum Verständnis der Bedingungen des Urknalls von großer Bedeutung war.

Wie viele Protonen-Proton-Kollisionen pro Sekunde wurden im Large Hadron Collider gemessen?

Im Large Hadron Collider wurden zwei Milliarden Protonen-Proton-Kollisionen pro Sekunde gemessen, was enorme Mengen an Daten erzeugt hat.

Wie hat sich die Einstellung des Redners zur künstlichen Intelligenz (KI) entwickelt?

Der Redner begann mit der Arbeit an Expertensystemen und wechselte dann zu neuronalen Netzwerken, nachdem bewiesen wurde, dass nicht-lineare Neuronen fast alles modellieren können, was zu einer Renaissance der KI-Forschung führte.

Was ist die Bedeutung von Datenqualität für KI-Anwendungen?

Datenqualität ist ein konstanter Faktor, der die Skalierung von KI beeinträchtigt. Ohne hochwertige Daten ist es schwierig, KI-Teams produktiv zu gestalten und die volle Potenzial von KI zu nutzen.

Wie wichtig sind Soft Skills im Bereich der künstlichen Intelligenz?

Soft Skills sind sehr wichtig, da sie das Verständnis und die Kommunikation zwischen verschiedenen Stakeholdern und Teams erleichtern. Dazu gehören Fähigkeiten wie das Aufbauen von Netzwerken, das Erzählen von Erfolgsgeschichten und das Anpassen des Tempos der Innovation an die Bedürfnisse anderer.

Was sind die Herausforderungen beim Übergang von Prototypen zu Produktion in der KI-Entwicklung?

Ein Hauptherausforderung ist die Verkürzung der Zeit von der Erstellung eines Prototypen bis zur Produktion. Es ist wichtig, Plattformen zu nutzen, die einen nahtlosen Übergang ermöglichen, um die Zeit bis zur Produktion zu reduzieren.

Wie kann die Lücke zwischen verfügbaren Data Scientists und Marktbedarf überbrückt werden?

Durch die Förderung von Self-Service-Analytik und Citizen Data Scientists kann die Lücke schrittweise überbrückt werden. Dies erfordert jedoch eine hohe Datenqualität, um sicherzustellen, dass die Ergebnisse der Analysen verlässlich sind.

Wie wichtig ist die Interdisziplinarität in KI-Projekten?

Interdisziplinarität ist sehr wichtig, da sie die Zusammenarbeit verschiedener Fachbereiche wie Business, Data Science und Engineering ermöglicht, um komplexe Probleme zu lösen und umfassende KI-Lösungen zu entwickeln.

Was sind die aktuellen Entwicklungen, die der Redner in der KI-Branche beobachtet hat?

Der Redner beobachtet eine Verschiebung von KI als Ersatzmöglichkeit für Menschen hin zu KI, die unsere Fähigkeiten erweitert und unsere Arbeit interessanter macht. Er betont auch die Notwendigkeit, Datenqualität und Governance zu verbessern, um KI erfolgreich zu skalieren.

Outlines

00:00

🔬 Einführung in die Welt der Physik und KI

Der Sprecher, ein Physiker, erzählt von seiner Karriere und wie er durch seine Ausbildung und seine Zeit am CERN, dem größten Teilchenbeschleuniger der Welt, beeinflusst wurde. Er beschreibt seine Erfahrungen mit riesigen Datenmengen, die durch den Betrieb des Large Hadron Colliders generiert wurden, und wie er zu der Überzeugung kam, dass KI und KI-Systeme für die Analyse solcher Daten unerlässlich sind. Er spricht auch über seine frühen Erfahrungen mit Expertensystemen und der Entwicklung zu neuronalen Netzwerken, die er später in der Teilchenphysik anwenden konnte.

05:03

🔧 Automatisches Modellmanagement und AI-Produkte

Der Sprecher beschreibt seine frühen Arbeiten mit automatischem Modellmanagement und der Entwicklung von KI-Produkten. Er betont die Bedeutung der Datenqualität und der kontinuierlichen Verbesserung von Modellen, um den Wert von KI-Anwendungen zu maximieren. Er diskutiert auch die Herausforderungen der Skalierung von KI in Unternehmen, die oft durch organisatorische Barrieren und die Notwendigkeit, den Wert von KI nachzuweisen, begrenzt wird. Der Sprecher hebt hervor, wie wichtig es ist, den Fokus auf die Qualität und den Kunden zu legen, um erfolgreiche KI-Produkte zu entwickeln.

10:05

🤝 Interdisziplinäre Zusammenarbeit und Soft Skills

Der Sprecher betont die Bedeutung der Zusammenarbeit zwischen verschiedenen Disziplinen, wie Business, Data Science und Engineering, um erfolgreiche KI-Projekte umzusetzen. Er spricht über die Notwendigkeit, die Geschwindigkeit der Innovationen anzupassen, um sicherzustellen, dass alle Beteiligten mithalten können. Er erwähnt Soft Skills wie das Fähigkeit, ein Netzwerk aufzubauen, die Fähigkeit, Geschichten zu erzählen und die Bedeutung von Führung, um KI-Projekte voranzubringen. Der Sprecher diskutiert auch die Herausforderungen, Talente zu finden und zu halten und wie man effektive Teams aufbaut.

15:07

🌐 Aktuelle Entwicklungen in der KI-Industrie

Der Sprecher reflektiert über aktuelle Trends und Entwicklungen in der KI-Branche, wie die Verwendung von KI zur Steigerung der Fähigkeiten der Menschen und zur Verbesserung ihrer Arbeitsbedingungen. Er spricht über die Bedeutung von Datenqualität und Governance, die weiterhin die Skalierung von KI-Lösungen behindern können. Er diskutiert auch die Notwendigkeit, AI-Technologien zu standardisieren und zu industrialisieren, um sie effizienter und skalierbarer zu machen. Der Sprecher erwähnt auch die Bedeutung von Agile-Methoden und DevOps in der KI-Entwicklung.

20:08

📊 Datenqualität und KI-Implementierung

Der Sprecher beantwortet Fragen zu den Herausforderungen der KI-Implementierung, insbesondere in Bezug auf Datenqualität und Governance. Er diskutiert die Notwendigkeit, ein Gleichgewicht zwischen der Bereinigung von Daten und der schnellen Implementierung von KI-Lösungen zu finden. Der Sprecher betont, dass KI-Projekte nicht auf perfekte Daten warten sollten, sondern mit den verfügbaren Daten arbeiten und kontinuierlich verbessern sollten. Er spricht auch über seine Erfahrungen beim Übergang von Prototypen zu Produktion und die Bedeutung einer effizienten Plattform für diesen Prozess.

25:08

⏱️ Zeit zur Produktion und KI-Lernungen

Der Sprecher teilt seine Erkenntnisse über den Übergang von Prototypen zu Produktion in der KI-Entwicklung. Er betont die Bedeutung, die Zeit von der Erstellung eines Prototypen bis zur Produktion zu verkürzen und wie man Plattformen nutzt, um diesen Prozess zu beschleunigen. Der Sprecher diskutiert auch die Herausforderungen, die mit der Skalierung von KI-Lösungen einhergehen und wie man diese effektiv bewältigen kann.

Keywords

💡Datenqualität

Datenqualität bezieht sich auf die Genauigkeit, Vollständigkeit und Verlässlichkeit von Daten. Im Kontext des Videos ist dies ein zentrales Thema, da der Sprecher betont, dass er 90% seiner Zeit damit verbringt, Datenprobleme zu lösen. Dies zeigt, wie wichtig die Datenqualität für die Entwicklung und den Einsatz von KI-Anwendungen ist und wie sie die Produktivität von AI-Teams beeinflusst.

💡Industrielleisierung

Industrielleisierung bezieht sich auf den Prozess, bei dem Technologien und Verfahren standardisiert und automatisiert werden, um die Effizienz und Skalierbarkeit zu erhöhen. Im Video wird dies erwähnt, um die Herausforderungen zu beschreiben, die mit der Skalierung von KI-Lösungen verbunden sind, und wie sie die Produktivität und den Wert, den KI schaffen kann, beeinflusst.

💡Expertensysteme

Expertensysteme sind KI-Anwendungen, die Entscheidungsunterstützung bieten, indem sie auf einem Wissen basieren, das von Fachleuten in einem bestimmten Bereich bereitgestellt wird. Im Video erwähnt der Sprecher, dass er an Expertensystemen gearbeitet hat, was auf seine frühe Beteiligung an der KI-Entwicklung hinweist und die Entwicklung von KI von symbolischer KI hin zu neuronalen Netzen zeigt.

💡Neuronale Netze

Neuronale Netze sind ein Typ von KI-Systemen, der sich durch die Verwendung von Verbindungen und Abstufungen arbeitet, die an die Funktionsweise des menschlichen Gehirns erinnern. Der Sprecher erwähnt, dass er zu neuronalen Netzen übergegangen ist, nachdem bewiesen wurde, dass sie komplexere Probleme lösen können, was auf ihre Bedeutung für die KI-Entwicklung und -Anwendungen hinweist.

💡Datenanalyse

Datenanalyse ist der Prozess der Untersuchung von Daten, um Informationen und Erkenntnisse zu gewinnen. Im Video wird die Begeisterung des Sprechers für die Datenanalyse und die Analyse großer Datenmengen hervorgehoben, was zeigt, wie wichtig diese Fähigkeit für die Arbeit mit KI und der Erforschung von Phänomenen wie dem Urknall ist.

💡CERN

CERN ist das Europäische Organisation für Kernforschung und ist bekannt für seine Partikelbeschleuniger und Experimente zur Teilchenphysik. Im Video erwähnt der Sprecher, dass er bei CERN war, als der Large Hadron Collider in Betrieb genommen wurde, was die enorme Skala der Daten und die Herausforderungen bei der Datenanalyse im wissenschaftlichen Forschungskontext verdeutlicht.

💡Datenwissenschaft

Datenwissenschaft ist ein interdisziplinäres Fachgebiet, das sich mit der Analyse und dem Verständnis von Daten aus verschiedenen Quellen und in verschiedenen Formen beschäftigt. Im Video wird die Notwendigkeit von Datenwissenschaftskenntnissen betont, um KI-Lösungen zu entwickeln und zu implementieren, und wie diese Fähigkeiten zur Problemlösung beitragen.

💡Künstliche Intelligenz (KI)

Künstliche Intelligenz bezieht sich auf die Technologie und die Forschung, die darauf abzielt, kognitive Funktionen wie das Lernen, Problemlösen und Entscheidungsfindung in Computerprogrammen und Robotern zu simulieren. Im Video wird KI als zentrales Thema behandelt, und der Sprecher teilt seine Erfahrungen und Einsichten über die Entwicklung und den Einsatz von KI in verschiedenen Anwendungsbereichen.

💡Automatisierung

Automatisierung bezieht sich auf den Prozess der Ersetzung manueller oder menschlicher Tätigkeiten durch Maschinen oder KI-Systeme. Im Video erwähnt der Sprecher frühe Arbeiten zur Automatisierung des Modellbetriebs und der Modellersetzung durch neue Modelle, was auf die Bedeutung der Automatisierung für die Effizienz und Skalierung von KI-Lösungen hinweist.

💡Interdisziplinarität

Interdisziplinarität beschreibt die Zusammenarbeit und den Austausch zwischen verschiedenen Disziplinen oder Fachbereichen. Im Video wird betont, wie wichtig es ist, dass verschiedene Perspektiven und Expertisen, wie die von Geschäftsleuten, Datenwissenschaftlern und Ingenieuren, zusammenarbeiten, um KI-Projekte erfolgreich umzusetzen.

Highlights

Constant importance of data quality over time despite advancements in AI.

Observation of recent developments and trends in AI and data analysis.

Background as a physicist influenced the speaker's approach to data and AI.

Experience at CERN working with large-scale data from particle accelerators.

The evolution from expert systems to neural networks in AI applications.

Importance of understanding business models and value generation in AI projects.

The necessity for continuous learning and staying updated with AI advancements.

The shift from automating model production to proving the value of AI applications.

Emphasis on the product focus and customer-centric approach in AI development.

The challenge of data quality as a barrier for scaling AI and its impact on productivity.

The need for interdisciplinary teams in AI projects for a comprehensive approach.

The importance of soft skills in AI, such as the ability to network and communicate effectively.

The shift in AI focus from replacing human work to augmenting human capabilities.

The role of organizational barriers in the scaling of AI and the need for data-driven decision-making.

The potential of self-service analytics to address the shortage of data scientists.

Balancing data quality and governance with the need for timely insights in data-driven businesses.

Learnings from moving AI projects from prototyping to production and the importance of time to production.

Transcripts

of constant over time and that may be

interesting um to to mention as well as

a few observations um more recent

developments and and trends uh that i

think i'm i'm seeing here on on the

horizon

um and

so if if i continue here let me see oops

now the controls are back so um

where do i come from and um so

one second i'm still having some issues

here with my back and forth so now it's

good um so where did i come from i'm i'm

a physicist by training and that that

influenced me a lot so let me tell you a

little bit

why i like data so much and um

analysis of big amounts of data so when

i when i started

studying towards the part when you do

your master thesis so the last two years

of my studies i was lucky and could go

to cern cern is in geneva it's the

biggest accelerator you will find on

earth

and as a matter of fact it was just

about the time when this huge machine

was was being launched uh into operation

back then under the name of a large

electron positron collider and this is a

huge 27 kilometer circumference device

about 100 meters below surface

so everything for me as a young student

there was was gigantic right so starting

with this huge machine then going on and

here i'm showing now the upgrade that

that it received like 10 years after it

it started

working as a lap it was called large

hadron collider and there were huge

experiments that are placed on the earth

under the surface and these huge

machines you see a little bit here

perhaps

the dimensions compared to to us humans

so these are huge machines think about

it as microscopes

and and and they are there to collect

tons of data

and that data is then what will be

analyzed later on and um so

what is then of course also gigantic is

the amount of data that's generated in

such an environment

um and and so you have something like

two billion

of these proton proton collisions head

on collisions per second and that just

generates crazy amounts of data right so

um yeah that was huge

of course then all the computing you

need is huge

um

i don't know a few of you may remember

that name of the craze super computer

today you would probably put a iphone

and it would do probably the same job

but that was was a big computer back

then so that was a certain approach um

in an architecture that was invoked

and cern was the first place where you

would find that in europe and it was

also the first place where it would be

decommissioned because um they

definitely went uh in into other

architectures like farms of um you know

of unix machines and then the servers

and so on and so on and of course

overwhelming the physics that's why i

was there so understanding more about um

the big bang right so what were the

conditions

we had a model the standard model and so

on and so on so this is all the

the story and i really just wanted to

give you that kind of background to see

you know what has motivated me to really

work

very early on in this ai space so i

started off

working on expert systems so that was

kind of a paradigm

kind of from symbolic ai that was

on vogue back then

after earlier on marvin minsky had shown

that

using simple what they call perceptron

so not yet

you know the neural networks as we know

them but these were simpler

architectures they couldn't even solve

simple problems such as the xor problem

so

that got you know that kind of neural

network research into a longer period of

darkness or standby so i was happy

working on on these um at cern you know

doing diagnostics of certain falls of

complex machines so that was an

interesting approach to learn and then

of course very quickly after

i switched to neural networks which came

in vogue again after someone else proved

that well if you have non-linear neurons

you can model almost anything you know

you just have to have enough enough

neurons in your in your

deep learning model

so

there were many applications i could

start then playing with you know be it

for particle track reconstruction

using different setups or identifying

certain fundamental particles

so all of that

was you know back then you had to write

your own code

um we didn't have to open open source

libraries which came a bit later so

that takes a bit of time until you get

something deployed as i moved out of

academia when i finished my phd

it was pretty much about

already then automating

um you know the whole operation of

model production and operation

and replacement of models by new models

and massive amounts of models hundreds

of different models um that would be

live

um and

automated so this was something that we

started very early on in in the 2000s

and i think this is still you know today

a topic under

you know automl and and these are kind

of the the tags that we have today but

in fact that was a work uh

on my side that started very early on

and then it moved more and more into

measuring also the value really proving

the value of of ai of a application so

you can prove it to a b testing

depending on the applications where you

come from that may be more difficult in

a b2b context for instance and then

today i think very important this this

product focus everything is a product

ai product and that brings automatically

the customer

um the user into the focus and quality

with it so i think we we get more and

more that um very good focus on

quality

so

now

summarizing looking back

from from my experience at least what

are the things that i think stayed

constant over time and these may be

interesting for you here and there to

you know to have a look into

um so one of the

unfortunate

evergreens and constants i found at

least is data quality

uh which is a pity because we 20 30

years ago i would have told you yeah i

waste 90 of my time

resolving these data issues and today

the answer is more or less the same

so i think we still haven't gotten out

of that

remove that big barrier for

scaling up

ai

this

of course limits the productivity of our

ai teams our data science teams

analytics teams industrialization is

slowed down or even impossible and uh of

course the value finally that we can get

out of

ai is greatly limited by that

so i'm happy to hear about also others

uh here in the audience you know what

what your experience is then on the

skill side yes um if we want to create

um

advanced applications that contain some

ai components we need to understand what

what is that business of course that

we're working for so how does that work

how does it generate value

um what are the

pain points uh not always easy to

identify

how are the stakeholders and

incentivized and

what are the value metrics that we

should be defining and measuring

um and i think also not to be

underestimated is of course the

organizational change that that it

implies it may often be a barrier

for adopting ai is just the change that

it would

imply

and usually

in you know

adopting ai-based innovation means

change of the current way of working

that's

very much true in most cases

on the technical side of skills

of course we expect that um you know

the the stakeholder or the the people

who develop those solution solutions

that they have very good and deep data

science skills depending on the area

you're working on

um so we need to have some continuous

learning we need to be up to date with

what's going on on the ai front which is

developing very quickly but i think also

you need that kind of experimental

scientific approach to to cut bigger

problems into pieces formulate

hypothesis and so on and so on

um and visualization definitely another

important ingredient

more on the i call it i.t in quotes

let's say skills um there's all the

other technical um skills we need

you need to have some some good

knowledge of programming languages today

more and more cloud architectures and so

on uh be you know being capable of

working with different um advanced

analytics tools um but i think it should

not um

make us blind to

you know in the sense of it's not

all about orbitating around technology

technology is the enabler but what we're

really trying to do is solve problems

and that is at the center and not the

technology

from my perspective

one thing we may um

or i've seen several times is of course

that when we

you know we built these models we almost

fall in love with them and we want them

to be perfect

that most likely will never happen um

so sometimes it's really better to

have a reasonable model that works

reasonably well let's get it out there

it starts generating some value and

let's keep optimizing it afterwards

so that that would be a recommendation

then the other

topic um i've been um of course

observing

um is

interdisciplinarity and here i'm just

showing a few

uh

profiles um this

may change according to the area you're

working in of course you will have

different personas participating in this

but you know to give an example so we

have the business side represented

usually

wants to minimize the risk while

maximizing the return so that's

the balance act that they're usually

working on

on the more like data scientists slash

ai side well we do all these experiments

we try models you know that very well

things can go wrong anytime but that's

part of the game

and we keep generating knowledge

which then

is usually put into practice

and production by more of the

engineering

people that that contribute uh to this

whole

project or or activity

that you may be working on so um i think

this this complementarity is very

important so each have each of these

have a different way of working and

together that really makes a lot of

sense

some words about

soft skills

definitely one topic that i have

observed

is um you know if you are so

um

you just see that the innovation is

working greatly there is many things you

can do you can you can go very far ahead

and by doing that you may

forget

perhaps that other people will not

follow you at the same speed

and it may be worth slowing down a bit

the pace of innovation because otherwise

you will find yourself alone running

ahead and nobody behind you

so um

the other thing um i i put here ability

to network so i think it's very

important that

um you know we as data scientists as ai

specialist that we

we know what's going on we we make

ourselves known but that we also

understand

a network

you know with all different areas of

the companies of the organizations

in a little bit a role of an internal

consultant

always looking for opportunities usually

and

trying to identify the right or the real

pain points by asking the right

questions and that is not always easy

it's very often i find even close almost

to an art

because the real problems

from my experience they're behind many

layers of apparent problems

until you get to them

and then finally if you tell a story a

success story about

ai being employed here and there

uh definitely you need to do that in in

your stakeholders language and not in a

very technical

language which is usually not not very

much

appreciated

then

of course you will have to to find the

people

that have the skills to create your

your ai applications solutions

that will always be a challenge i guess

there will never be enough data

scientists

and among those you still have to find

the ones that are the the right talents

for for what you're out to do

and then of course the next thing is

once you have hired a certain

you know the talents that you were

looking for how do you retain them

because that's the next challenge um

usually the market being very hot

you know what is it that is really

retaining them and

from my perspective it's very much about

the interesting challenges you can

provide

to be resolved so if these are really

challenging high value adding i think

then you have a good chance of retaining

them

and um then the next thing is how do you

make that team efficient right so that

we don't just do research in any kind of

direction but really pointing towards

generational value

interdisciplinary

teams i mentioned that before

the last point here i would make is

about the kind of leadership

that i've seen over the years also

myself as i learned

i see much more benefits in a kind of a

pool leadership which is more kind of a

servant

manager kind of

approach than pushing

into the teams

what you think they should be doing i

think rather better that they follow you

because you have a certain proposition

to make that sounds interesting

um coming towards the end um a few notes

on recent developments um that i've been

observing and i'm sure

you see them as well

um one of the things

over the years

that i find interesting

and absolutely correct also the

development that we

i think in the beginning it was very

much about um ai machine learning

simply replacing people in what they're

doing and i think that focus is now

shifting more and more

definitely in the area where i am in

into augmenting our capabilities

and also making our jobs more

interesting

by the help of of ai that that empowers

us

um

another topic that i i really

stumbled across over the last years is

really the point sometimes we ask so why

can't we really scale up ai in a bigger

way data quality certain one but the

others i think the purely organizational

barriers and

i think part of that is also middle

management which is largely not yet

ready to adopt

ai on a large scale so you may end up

doing

many great projects you can show value

but in the end nothing will change

the the new ways of working will not be

adopted

and that i see as a little bit of a

barrier

so we need to enable many more people to

really start using data themselves and

become more data driven

um

i think the tools are around

the the reservation i would have is

still

um the data quality side as long as we

have those issues i guess you get more

trouble than benefits if you have

everyone doing auto ml on any kind of um

suspicious data so this must go hand in

hand

you know

before we can unlock the value of data

really

and regulations i guess they they

continue to increase so we're seeing a

lot um coming up already out there

that will certainly continue growing

so i think we need to be

ready and embrace it um

and industrialization standardization

automation of ai definitely as well

and to close

um

you've seen some of that certainly um

the envelopes concepts you know that we

augment or increase the

the well-known devops approach to also

contain the

the continuous exploration using uh ml

and ai methods

um

are part of this and we also have some

architectures

frameworks that are being

described and proposed just an example

here from from google which i find quite

useful of course now without going into

any any details here

um in the ways of working the agile and

the scaled agile i think also good

movements here

that um you know have a

product centric um and and team

empowering

kind of focus

so i think all in all

very desirable

developments

and and finally on the data front i i

see some hope there so there is

the fair data movement that you may have

seen already

i think that is already addressing some

data quality dimensions hopefully also

putting data quality more into the

center of detention and another movement

that i'm observing that is creating that

kind of attention on data is this kind

of data as a product approach that we

see in the data mesh

context since 2019 more or less so i see

some adoption of those principles

i've just brought here you know what are

the kind of

requirements or product properties for

these data products and many of them of

course they are they are related to data

quality so let's hope that this

brings the desired effect and good data

plenty of good data so that we could

generate plenty of good

applications using ai in future thanks

for your attention this is all i i had

for you today thank you happy to take

some questions

thanks a lot thank you so much for

sharing the presentation thank you now

just we have time for q a uh we have

given

uh possibility of microphone camera on

to all participants before launch so

anybody who has a question please just

switch on the camera

so i we have just a feedback from niraj

uh

ibsen pharmaceuticals excellent

presentation frank thanks a lot we

agreed

hi frank thank you very much for this

interesting insights and based on your

long-term experience you mentioned one

point that data scientists is a shortage

they're not enough on the market to

fulfill all the demand of companies and

how do you see the possibilities of

self-service analytics to fill up a

little bit this gap and

let's say get

the

the demand of

knowledge workers closer to the

situation of missing data scientists

yeah now thanks for the question marcus

and this is uh spot on right

i mean we have those talks about the

citizen data scientist um i think the

idea is great

the only dependence i would see or

restriction is the data quality

so i mean you have great tools you know

tools like yours and others that that

provide those kind of easy

access to

data preparation and data analysis and

ml i think the tools are there

now we need to make sure that the data

is also there and good

because otherwise you will have many

people

that start using the data and they will

get contradictory results and then you

get a lot of discussion everywhere

and then in the end they will even say

it's the tool right um which is not of

course the case so that is my only

restriction um you know as soon as you

have areas of data where you say

this is safe we have you know we can

guarantee the quality of that i think

then you can open it up for

you know business users

to widely start working on that

thank you very much frank well thank you

thanks a lot and now there was somebody

else hannah took on if you have any

question you can switch on microphone as

well yes

yes thanks frank uh interesting talk

because i think i want to ask a relevant

question is like

because what we are facing is we are

facing a huge amount of data and of

course data quality and governance is

important

but then

and some people will emphasize too much

on this and

this will slow down the let's say ins

driving the insights

so we're in this dilemma is like because

actually

the huge amount of data what type of

insights we should focus on what type of

insights comes first

or

let's clean up all the data but this we

don't know this cleaning up the data is

a huge tedious

very time consuming work and the inside

trailing is of course more welcomed from

the business and

i mean what is the balance there

thanks um for the question i think

that's exactly you know the the

complicated situation we're in and

you know we as data scientists

well we work in this environment and we

would never say let's wait until all the

data is fixed and then we start working

that's just not feasible so very often

also our work i mean our productivity

goes down because we have to fight

against these data quality uh problems

at the same same time

as we do this work we also

understand certain um you know we get

data intelligence which can then be used

again to nicely model it subsequently

into

a data mesh data product a data lake

whatever but we could we also provide

help to those people who then you know

make data nice right in the end but very

often we're at the forefront of this

right so yeah we we cannot wait we need

to move on but

wouldn't it be great if we could move on

like three times faster right

thanks a lot thank you uh

jack

yeah um hi frank nice to see you again

thank you

you mentioned at the end moving into

production so what what do you have kind

of top two learnings when you move from

this prototyping mvps to production and

then what do you need to take into

account and and what's your learnings

in that journey

well great great question and and

you know what we are working um

very much on is is to shorten that time

you know we're very good in in

prototyping um

and shorten the time from prototype to

production or to product right and this

it could still be that after three four

months we have a prototype but then it

takes another nine months or so to get

the product out and it's too long and so

we are now also trying to work on

certain

platforms which then would allow us more

of a seamless path from idea to

prototype to production so you know many

products out there and definitely we

looked into some data science

workbenches

and we are getting some benefits from

there so this is something that we're

really uh ramping up strongly now so i

think time to production is

is key um

yeah i would say that's the major

learning part