welcome we are live again this is virtual chat episode 16 and you might be live here too so as usual i welcome you to introduce yourself in the chat my name is fabian westard i'm an entrepreneur and investor with a passion for artificial intelligence um this is also a podcast so you might listen this as a podcast or you watch the replay on youtube so say hello also i have created some polls so if you're live there you can answer the polls today i would like to ask you what are the reasons why b2b integration for ai is lacking behind you know why it's not going that fast as we would like to have it um in preparation for this i thought like okay why are the reasons you know where are we standing and i remember that in 2014 when i went full ai you know deep learning came out and it started to emerge we were all bullish we were all like excited you know singularity is close recruits will everyone says like
hey the eyes are coming now and i would say in 2016 to 2017 everything was all about the eye policy billions of new funding funds a lot of startups hype journalists everyone talked about the eye but 2019 it shifted it's not that it's like an ai winter it's just we were sobering up we were realizing that expectations and what we actually can achieve with it doesn't match we thought it would save the world and it didn't do it yet however that's not the reason the reason is not because ai's not good enough or the research is not good enough it's the opposite it's integration is lacking behind we don't have enough companies actually using ai and therefore we have not enough products out there so this is the topic and we are speaking today about last time we spoke with professor kessling about research and they are doing good and we spoke about how to improve policies with two members of the parliament from germany and with one member of the parliament in europe but
today we speak about b2b integration in the industrial space and my guest there is dr oli waltinger and he is formerly would say the head of machine learning or machine intelligence research at siemens corporate technology but he's also the founder of the siemens ai labs so he started in the machine learning uh area and now he's also head of the ai so he's kind of in charge of everything about ai at siemens and siemens is a huge industrial company globally active um just for his background he grew up in regensburg and i think he wanted to leave so after his abi tour he went to london he went to australia and then he ended up in bielefeld since i'm from the area too i will ask him how he can end up there on how he was able to stay there for a while because he did his master's there plus his phd in his speech he was already in ai about social semantic and machine learning so today we will talk about him ai
integration in the industry in business we talk about frameworks lesson learn uh i say welcome ollie boom fabian thanks thanks very much super stoked what what an introduction right almost awesome right everything i need actually now and i can close down now i can retire right it's uh such a great introduction thanks thanks for having me you know having me on your show yes wonderful so um i'm from australia or some even say uh lipper um it's like a galicia store out there somewhere in germany there is nothing about work there's a lot of industry countryside and nothing else you know if you grow up there it's boring which i enjoyed you know it's nature etc you studied there you went to bilafell tell us a little bit you know billy felt people haven't realized with ai so why what speaks for b no a b a pretty pretty cool actually so i ended also up i didn't know about billy fault you know i actually my two kids are born there right so i'm
heavily still even though now i'm bavarian now in munich right i'm heavily related to the philadelphia area and uh hinterland ai i guess right which you also want to call it right um has a highly uh high affiliation with um ai technology lightness clients neural networks you know cognitive interaction the wax mode right and so these kinds of university and academia right are quite significant and this reflects also to the time where i was there i'm doing my master's doing my phd having my postdoc there um in terms of having the excellent cluster what it's called right the corporate the the government's wide program and and one of these excellence class scitec center of excellence apartment with action was associated also in billable so it it gave me and my family actually a perfect environment to grow to grow up literally on my kid's side but also to grow for me as a as a young researcher um you know having having the research from the man's fundings obviously the boys toys for boys and
girls right um in there but it's also engaging with the scientific community because if you have such a competence of the competent center for robotics azimo you may remember right under the first azimuth robotic center there's also a villa felt located actually on the university and this marvelous interdisciplinary setup of cognitive and computer science was a great a great journey and still feel very much related and i think this is important for most universities we don't only look for computer science ai we look for cognitive science you know biology philosophy all these cross fields have to come together and network and i think the university is good for this then let's jump you went to siemens and you you told me yesterday your career but we would speak about today it is a similar right so i ended up i don't know 2012 i guess right um with a with a um finish on my postdoc and i'm looking for you know i spoke with a small staff in um speaking at the junior professorship
in frankfurt and then directing doubts um you know in munich and doing then landing there um i was always passionate about you know application and this is also what missed you know what i missed the most when you do research um you you explore on you know artificial dangerous hats you you do machine learning or semantics or you're trying to bring you know both worlds together um which i did in my in my my phd um and then you know it's you you're striving for it i want to apply that actually and um so for me the best the best company and the best back is applying and going to a corporate research where you have data computer resources um use cases the challenges but still engaging with the scientific community and that's why you know if you look in the in the map of europe there are not that many corporates back in the days right it's at this kind of fundament and hey you know such a global organization as i'm working now
for um you know has marvelous use cases opportunities did siemens by 2012 already have ai as a term in its research department or was it did you have to create it uh no so um siemens has in corporate research quite quite a long history actually in in ai um so obviously from the naming we call it neural networks we call it experts expert systems right we call it knowledge decision systems right we call it machine learning but uh you know then setting up the umbrella of hype right i i would say i survived two hypes the first was then the big data hype and the second one is is now actually it's still uh still on hippish but we are now moving from the hype on the ai like the umbrella of you know intelligent systems adaptive systems um towers you know how to shipping it to product right shipping it into practice and this is what we see here and this journey right guiding you know a large corporation which has a high footprint
in actually physical systems right controllers turbines energy production you know um ritz and stuff like that um and um guiding that a bit in pieces um along the journey what's the impact of digitalization big work right what's the impact of data driven decisions big work what's the impact of ai in the global organization pushing that how it's an ai first company as you know i guess a couple of people folks would say is possible um ai is an it's a buzzword it's a while it means it's older than computer science uh to build human-like machines i mean we we're all machines in a certain way um there you go nice t-shirt by the way i i know that siemens and others use machine learning and machines and intelligence for decades i mean it was always there but it wasn't classified as ai so you said there were like two hypes two waves what changed in the last eight years inside of siemens and the relevance of ai in terms of resources data like did it
grow exponential or was it always the same just with a different focus a corporate life always exponential right you know we do mod exponential on every case right scalability profits you don't know just kidding so what changed so in in on the one side you have transactions transactions are prosecuting users use cases and that means obviously if you're looking very very back then you see you know the first neural network's deployed 20 years now right um steel industry right optimizing uh the product production right um when i i moved to to um to siemens corp um we deployed the first question answering system for energy systems um so you know asking natural language questions was at the same time as ibm watson right um was pretty hippish right and we in the first task i was asked move to princeton and define with a great team and an architecture and implement the qa system that is able to answer questions randomly about about fleet right building in one side building what is now called i
guess a knowledge graph on you know a digital twin representation of where are our turbines and what status have with services on the other side allowing them to naturally converse and have natural language interfaces so from the application from the application side there has been plethora of that though obviously what changed is that um with the we we are heavily we're always um software centric purely soft authentic it means like a product software centric providing simulations of providing services providing services towards real products with the emergence of digitalization and the digitization of processes and suddenly data arise and suddenly new opportunity arise and suddenly that what is now called the ecosystem changed the way how we do business changes and the corporation adopted on that you know and because everyone completely yes but did it you i mean you you showed me insights into siemens i did not expect so in in first place um what i'm if we are stretching over obviously we have a huge program on big data every i think every
corporate huge you know data warehouses systems and you know tons of consultants um fostering so like in a very early touch base you know how do we leverage the value of data um we call it data to business right how we take with ada and make the business even though it was still back in the in the big data hype so um the mindset was already okay we need to incentivize that we need to be transparent where is our data and what kind of warehouse ems mes systems is it hidden right can we somehow integrate it and the first discussions went on you know how do we manage these kinds of prospects second aspect is how do we generate you know um insights out of that this was the big data and the insights in the corporate world were pretty much uh driven by dashboards to be honest right with dashboards have low spot spotfire dashboards dashboards oh yeah sorry dashboards i'm sorry alright i'm i'm famous for not opening my mouth while i'm talking
about it there's a dashboard that means um we enter a phase where data-driven decisions um by distribution of some kind of data right in a visual manner um and that means it leads to incentivizing training skilling on you know how do we gain this descriptive um aspects based on our data um and then ai picked up and that means we we shift the towers we don't only want to have prescript descriptive statistics data informed decisions but we also want to predict them so that means we want to have you know more adaptive systems more um process optimization right and gaining more efficiency right and this is always what you're targeting and guess what one of the toolboxes that are pretty much cruel for that is machine learning systems right so that and this is the hype so identifying then on the management board that this is key individualization if we moon over towers you know somehow the the digital service and the digital and the area also on a lot of cooperation means um we
need to set programs right and programs have been set from top down and um you gave us a very kind introduction of right head of the ai and stuff like that siemens organized from top down will and bush actually initiated a program which is called company core technologies that means they defined a set of technologies which which identifies like these are keys and these shall be programs obviously corporates run programs right one of this program is called analytics and ai right which incorporates either around 200 230 pure researchers plus a plethora of business colleagues working on ai and organizing themselves as an eye so also a systematic program so this is very corporate style and we'll ask you later how this can be adapted for middlestand for smaller companies who don't have this structure but so i understand first it was all about data sure you need data for everything else and then it was about dashboards who provide you insights but they were descriptive and when we turned with machining to prediction or even
optimization um this changed too and machine learning got more important so this is a trend and siemens was playing along with this trend because it has a lot of software out there i mean a lot of factories use machines by siemens which run them within operating systems which is from siemens too um did it change on the management level it it sounded a little bit like before it was not you told me yesterday it's almost ceo level today this topic so is it is it increase it is sure it is uh c level right it has c level attention and with c level attention it's it's a program right which is it's lab by mathemati and cooper you know cct all the appropriation and corporate world like the program you're assigning your portfolios and with portfolios you'll send projects and you know the typical a bit of also hierarchical thing how you organize and how you you know let's say um control and how you actually communicate stuff the second one is um launching a
training and a corporate-wide trading and i mean like top-down training it's called abb applying digitalization to our business um you know i know it's crazy but the content is right contents right so it means um you know the the notion is is pretty short right um siemens uh sea lava says like you know if ai is and they analytics and cyber security and cloud are those essential concepts we want that everybody in the organizations um understands and speaks the same language and looks behind two or three steps behind the buzzword not everybody needs to be a neural network right um you know engineer right or data scientist but they need to understand the profound impact of these technologies and and this is what we have done and my team and me also you know rolled out worldwide we have a lot of manager believe me in there you know what is hands on actually what is machine learning how do you build now computer vision what is the cnn doing what is an activation function
and what is back propagation right and okay so that makes fully sense to educate i wish all politicians would take the same course too so they're on the same level um did you only educate other scientists or opera developers did you educate managers too no it's a top so you designed a workshop training program for manager and siemens to bring them all on the same level yes where it is you know it's called it was i guess they called the a team right we have a bunch of you know uh yeah and 15 people um you know designing individual classes um on my side on the ai side on the other side cyber security on the other one are connectivity um designed hands-on classes um on on one side this is important understanding inspiration is important understanding technology is important but also a third part is how do you ship that in products how do you ship that and make the converge because managers do like you know playing toys understanding concepts and they're really
smart they really catch up very easily but obviously they need to see the opportunity and this is this is the challenge also you know how do you make it bridge over to you know um make it in the product so so i summary it's top level so the ceo or the board of directors said ai is a core technology that's first step for everyone out there your ceo needs to believe in it then you educate your management level the first let's say 10 about the core technology how to use it and why and then you give them tools at the hand to execute and to make money with this and now we come to this you know this part is how did you change those people to make build products to make money and to execute on it yeah so that that's a it's a tough one isn't it right so it's not solved that's why we talked today and you were sharing yeah exactly so um you know it's it's um it's actually you
know you you don't say you know you you have to have to do these and these and these right because we're talking about launch organization and you talk about actually about leadership and about mindset you know i'm accepting that we live in a world of of scale we live in a world of continuous uncertainty and the continuous uncertainty um needs to be somehow tamed or at least embraced in a positive manner and um so it this is a lot of about our leadership you can all already obviously mail it down so like okay top use cases are predictive maintenance for uk's income supply chain then autonomous manufacturing and you know um resource optimization so you can nail down you know the opportunities with the markets use explosives but this doesn't help on the job profile you need to give them blueprint and best practice on their hand how do you leverage and identify opportunities in your business together with your customer right and these we do do that too for that right it's called design
thinking right about value proposition uh leveraging pains and gains about customer journeys about personas stepping in the foot of a customer identifying what kind of pains do they have because it is not like now we have this ai this does you know garbage in gold out thing but you know this machine learning system works effectively um if you define the objective function very effective you'd if if if you define this objective function identify the function and it's also right i always say it's like start from the deployed object and work backwards then you know you can identify these cases which may scale in your business okay so i have the understanding and i get a process to identify opportunities together with my customers and then to build products um first there how do they do it do they all have their own ai team or do they book the central ai team second you know do they build their own knowledge stack or do they consult you and second is did you give them like
a pdf and this is the secret sauce for z men's to to be ai ready or did did this happen a lot of iterations are you like internal consultants now yeah it's a good one right you need to have somehow the sizzling right it will be a you know a central function basically um and and therefore our aim is to push innovation um to our colleagues and with our colleagues and with our colleagues to the external customer world right so so yes we have somehow central function which does that as well what um what also sea level did and say like okay if they don't fix an ai if that of course then we need to put the teams and the lab structures onto the regions and into the in the sections um where the customers are and where the verticals are so where they're actually businesses so a uh they're called application centers right so put an application center for mobility ai and analytics mobility close to the to the mobility right to in
allah in the sales right um energy management civilian manchester water right our customer service right airport dubai mobility and city singapore right so putting competences teams and know-how in those regions those aspects where the business and where the customer is so we can work together with the customers because the change the significant changes normally normally as a corporate you would go back you would design a product and say like here it is yes all the features you have or you don't have right and this is not the work we currently it's in right we need to embrace this uncertainty like actually we develop features and maybe also the product along with the product journey together with the customer and this is a significant mindset i know that's not ai geeky right but this is how you develop that's very that's very good because this first worlds with every technology and i think this is a very needed infrastructure in a corporation to implement the fast impact of new technologies very fast with customers i think
this is a structural thing which is needed i'm with the comments here yeah michael says that the strategy is often lacking and it's an issue for operation leaders in the uk so you know you have the strategy you have the education then you build the competence centers out there that's super good let's take one example for now and take and at least something industrial because i like everything which is manufacturing where how did it work like one piece where we can understand how the process work and where you brought in ai into the value chain um yes there's plethora right so there's a plus or so if you start from from industry industry you know you you talk to define okay um you um in industry is is pretty much um still on what is called the attractions teeth so depth of value the depth of value so um we'll hopefully talk about the breadth of value in a second but let's talk for the first place of adaptive value so depth of value um
in ai and the analytics let's say or machine learning analytics is about gaining more throughput gaining more efficiency right optimizing yeah optimizing towers efficiency this is what the main aim is and then you look at paint what is currently in the production process you know one of the factories um what kind of uh you know pains do we have and then you identify okay we we want to increase with machine learning the throughput of the production of the pcb boards right this is you know moving out and we need some manually checking the pcb board it's like oh yeah here right out here rework here out here rework so can we somehow give this process right what if this process digit is already digitized what is not digitized you know can we leverage the power of machine learning in this case visual recognition you know um can we power to inject and step in here you know augment this decision process by means of machine learning so then capturing obviously a lot of data right
and not just raw hey you i will i want all everything data but very targeted how it's objective function we want to uh you know analyze um the quality of certain uh fonts in that based on the quality criteria and then um deploy it observe it improve it right and this is this is the step right um but you need to nail it down to that it sounds easy for us too because it's like on the on the board we have one comment here that mostly it it never gets over the proof of concept of mvp so how do you make sure that a small mvp is really deployed at the end um yeah that's that's the you know um there's a nice you know nice stats actually you know cooper is gartner and stuff like that right 80 percent of you know of things end up at the poc level um i think us in term guys you may not change the number of adc 80 will end up at the plc level to
be honest right because certain pieces needs to be tested and agreed then this is not doable or that this doesn't provide the value we wanted to expect it right the natural reflex prior was obviously a manager says show me the business case if the business case is calculated it will do it right so but for for many pieces in the corporation you can't calculate precisely the business case because it changes and that makes it a bit flexible one aspect i always see is if you don't have sea level attention on your use case will never see a light so it can be the marvelous case as well but if no c-level attention is there right if your manager doesn't support that right um ain't nothing second one is people tend to uh not include the operational from the beginning on if somebody needs to be productivized these people who productivity thing needs to be from the very beginning on there right a third one have the right competences in there and the right competences means
also customer involved there if you're not developing and putting the customers from the beginning in the process up there you ain't gonna end up at a product the customer would love and like right so including that and it means bridging the world of management stakeholder customer journey in their um data science expertise no question about it but operational engineers and and these are the full pillars if you combine them from the beginning on there then you have from from my perspective higher chances to gather it on the road as you would explore just like you know what i have a working student who tried it out but it never worked right you know what i mean in most of the cases i see working students are driving our digital transformation to be honest right oh i agree i i told you our intern set up the whole obs system here i mean he was fixing something no one was able to fix i mean the smartest people are is often the youngest and the freshers
because they have less politics in their mind and the manager needs to protect them and you know leverage it and you said 80 of your cases were ended at a poc or mvp level which i think is fine because it's like a startup it's a test if it doesn't work you don't scale but 20 means you rolled out do you know how many thousands or dozens of a projects you have at siemens like like do we speak about dozens or about thousands which are in the global corporation yeah so um obviously um since it's a program right what does a program do it somehow um you know um communicates and i tries to aggregate aggregating of a company of 380 000 people do that that's crazy important you can you can somehow estimate and obviously we do hundreds hundreds of of pocs um and and software development and projects some more data analytics so more ai some deep down in in reinforcement learning some deep down in computer vision um not everything gets reported right
who wants to be just on the reporting structure right you can imagine that this number is quite high to be honest right we have a good overview and i wanna you know call out not now a precise number because somebody nails me down and that my major comes with like what but it's believe me it's it's quite quite a mass what what is being tested right and to give you just maybe a round reel right we're talking about global services right global services double shared services right speech recognition you know um translation uh nlp extracting from documents from travel costs from you know you name it did ocrs and treble cos reimbursement right because i use case i guess we're most corporates out there right we talked about hr uh we're talking about you know how do we how do we interface with employee conversation ai you know carl is such a virtual assistant right which uh interfaces with with the colleagues the monthly search how do we access the right information can i ask
questions oh i need to convert now um combining conversation i have a question here is my here's my leaflet please proceed again so rpas robotic process automation right we're talking about siemens financials so financial industry risk assessment banking assessment right um we're talking about industry obviously you know predictive maintenance production visual recognition visual identification uh we're talking about energy uh control optimization reinforcement learning deployed on in in various turbines right i'm not not that many people that don't do any recovery rates do that right we're talking about additive manufacturing observing while printing you know a camera touch to that to see at the quality inspection to reflect on when the printer is actually ready that the quality is turned off it sounds like siemens is an ai company no one knows outside of siemens in your lab maybe yeah might be right i i think yeah so i if i think this is also one approach you asked me hey dude you're just are you you know you have this plethora of talks you're just
talking right no actually you know we're doing i work i'm delighted to work with awesome great team right my core team is like 20 25 people and we have the lab um as an incubator for exactly facilitating these kinds of journeys right um and yet we think that this and ai narratives are really important because i think from my perspective my personal perspective europe has just an image problem in terms of you know if you draw the line of you know of hey you asked advanced and agent um you know we need to make transparent hey where we have an impact and how we work together with customers right we can continue right i haven't touched mobility right right during autonomous driving right you know cyber security we observe right and also probably in our lab right you serve every every you know security is such an important and obviously cyber security comes from outside with ai and deep learning so we want to tackle it with insight with deep learning health and ears because
patterns right so you know there's quite a lot actually since is so huge 380 000 people you said there are many many many different units and therefore i mean it's like thousands of startups and one or thousands or dozens of companies in one uh but i i summary so you have the c-level attention you have educated the management you have giving tools frameworks aa you call the competence centers or you know application centers for all the sub-units so they can execute then you do mvps you measure them then you roll out them with your teams with the client with the seed management support now you have a lot of learnings and we talked about this after this we came to a point like the feedback loop which is keyword explaining the like trustworthy eye i mean you were working with data you said it's like used in recruiting it's used with finances what are your learnings a best practice like integration what you learn and what's key so everyone here in the talk can take
out but second learning's what occurred how do i do trustworthy ai because this was not a topic five years ago yeah it wasn't uh yeah um so in first place i think you know you you initially you said like you know what's the difference between i guess you know shipping machine learning in the consumer space and shipping machine learning in the business space especially maybe industry space right let's start with data right so data in is somewhere entirely different right in the b2b area because um in in the consumer area this the data is produced and deployed within the map we call it bad data right instagrams sales transactions right so they ask they stay in you know the data is actually produced outside the web in the digital area right and needs to convert uh in the nuclear power plant you know someone purchased the product by siemens and they produce the data exactly and that means they own the data right we ain't gonna touch that right um so and that means uh
we have you have these two words and um i would say you know um even though we're talking highly about you know ai's augmenting our tasks and augmenting our lives if you look in the b2b world it's actually caved to be honest it's it's we approaching the augmentation more and more it's it's pervasive and we are on the roadmap of pervasive by actually casing it and this means right we have robotics which are closed down the systems right but they are clearly separated to be honest still clearly separated between human interaction and machine interaction you know let me just similarly say take bamo take memo right key fencing right right more or less team fencing right exactly and it's we actually we are not in the last three years we're not dominating by augmenting yet right but actually separating is like this is where you know robotics and adaptive systems of autonomous systems run and this is where the human acts and we are now converting emerging collaborative robotics converging you know digital spice and
process and and and also um given looking at a process see what's digitized and what's not digitized and then placing the weakest signal this is how you gain efficiency right and that demands trust and this trust is the second differentiator because people were used to make this task to optimize to take the decision because of the a massive domain of how an experience that now this machine should start because otherwise we break something right and now suddenly data-driven decisions not only as informed decision diagnostic but also actually activations like actions are taken that demands a high different level of trust so just before the machine the human set stopped the machine because it was separated human and machine and now we come to collaborative machines um where the machine can make its own stop decision you know where you give more i would call you open the world model for the for the machine it understands human exists and considers also the human in its own decision process exactly and this demands trust all cloud
providers do have to trust currently and they stay in the backboard even though you know exactly what you know the facebook realms the instagrams and twitter google do with their data but you still have a high layer of trust because you stay in the action and action of your you know it's it's not shutting down or closing the door it's not opening up your oven it's not shutting down your kitchen right yet proactively but we're talking in industry we're talking about you know real money value and real physical harm right yes and for that you know that's a different trust sleep you have to take right and this is also something but first one but let me keep this for a second for everyone out there this trust issue is very very important interesting because i have a portfolio company with asgard and it's called microbes industries and they build software for collaborative robots so at the end they have taught robots to learn to adapt to the environment to optimize on their own you
know to learn their humans and interact with them it's a lot of trust and they have to build by the way the trust first with the purchasing department and with the client let's say abb for example to to use their software it's it's interesting so you have to tell them trust the software in a certain extent yeah the the fourth thing is um so one leading use case predictive maintenance what is predictive maintenance doing predictive maintenance like hey i want to identify a failure you know something corrupt you know in your process or production process before it happens um guess what these not happening a billion trillion times right so if you collect data you end up with very rare evidences right when you may get broken a a a plastic may be broken process needs to be fixed right that means we're not talking about trillion of cats in the web that we're talking about rare events in the north set of sensory information that makes it really tough right and it ends up
at unfancy sounding animally detection outlier detection in these kinds of cases right um and the fourth pillar fifth pillar is then you know you need to deploy it in the product itself again and this is also something um you know which which i guess is is pretty much you know crazy first of all a collection of data how do we connect the assets many of the partners and customers are still not connected they're still in their face of cloud first right they start with cloud first and even though we're shifting now to the edge paradigm that means that currently i would say is eighty percent of data you know crunched in in the cloud twenty percent and we shift in the next years eighty percent on edge twenty percent the cloud right so are a huge pusher of this edge technology which at the means you don't have it it's the mixture between cloud and on-premise server it's at the end accessible cloud but close to the machine to reduce latency for example yeah exactly
because if you if you if you need to change and adapt to real-time processes you want to be close to the edge and you want to use the capabilities close to the edge sometimes data should not even leave the factory just for for certain conditions you want to use still obviously cloud capabilities because you want to have asset monitoring asset indexing conditional monitoring overload so this is you want to train in the cloud and deploy an edge and these kinds of errors might be interesting michael in the chat is here commenting uh hacking machines a serious issue so we speak about trust but then we can speak about servers too so let's let's see how do you defend your systems especially if an edge computer is now a gateway to control a whole machine process yes yeah that's that's that will be uh if it is and and will continue a huge challenge right and yes obviously machine learning um that's a great um great value add as well right it starts from all kind
of interaction the biggest biggest weakness i guess is not necessarily maybe the physical interaction of injecting but rather the man in the middle you know putting the usb stick the hacked usb stick in the computer right um and and opening up clicking on the map click on the links with this you know fusion of cloud accessibility and network excel is really these are um you know heavily heavily fragile to be honest as well right um there are processes right pen tests and client slides um yeah i would i would from my perspective and that means it's also clear and this is also from management top attention um which said like alone we will not make it there is no chance by ourselves to have a holistic uh holistic security and there will be no hundred percent security i'm sure right there will be because that's always there and this is what um what i know it's a bit again corporate things right but it's called charter for us i mean these kinds of things organizing
different corporates together it's like hey let's work together and embrace and also make transparent this is very important again make transparent if you're hacked if you have certain incidents report it right commit that you report it so we know what's what's on right there and this this is not a user process for corporates isn't it right but this is this is you know being transparent working together and accepting there is no hundred percent security um and working together to make sure together we are stronger i think this is a good message first make a transparent admit mistakes and second penetrations as you call a pinterest but i when i did a security offering two years ago i invested a lot into billy fed company who did like two days full penetration testing you know trying to get into the cloud databases wordpress system etc it's expensive but it is critical important so we have roughly 15 minutes left um i would talk about two more things how to build trust and how a machine can
provide trust to humans which is explained in biliary and second how siemens is cooperating or integrating opening up to the ecosystem because you said we can't do it alone so let's talk about again about trustworthy ai technical how do you build trust okay so um there's a nice saying trust is not only about uh technology but it's also about interaction right there are certain pieces in the paradigm and formula is you know i trust my wife i have no clue what she's currently doing about so transparency maybe you know something not not always you know explainably the eye once was called right you're visualizing a neuron maybe not the right decision though obviously if you compare again b2b and b2c so it's like when we talk about the b2b area we're talking mostly and also the research applied research it's about human tasks it's about gene translation it's about speech recognition it's about the ranking of products stuff like that um we don't expect hundred percent nope i'm i'm okay it's just perfect google exactly
i'm i'm fully okay and the people are fully okay with not hundred percent right if you are in a system in a system area right robotics operation in the health yeah but it's hard right but it's you you want to have it robust yes but the car the car has to drive 100 i don't expect that you know well this is the best i can do based on the data today but tomorrow it's raining you want to have a robust in first place and you want to have at least at least explainable and somehow transparent so it says like now please take over the steering i'm not confident anymore right has search ever ended up on your side dude i actually don't know i just give you uh a thousands of results i don't have to do whether this is right now never right because you don't expect it because you accept failure and this is different in systematic and that means it's robust so you need a measure for robustness second one it needs
to be explainable so it needs some kind of aspects which does that so let's talk about robustness robustness is in industry machine learning is uh pretty much you know driven by domain adaptation that means we we are still living in an age where we take a data and we think this data represents the real world we throw it in the machine learning algorithm whatever you know representation you want to learn right and then you deploy it and say like i hope i hope the real world doesn't change and so then you you know new application arises or these systems are you know embedded in different domains suddenly it's dark it's wide it's winter and spring it's a different camera perspective in factory stuff like that suddenly these systems um are still doing a high accuracy in terms of what they output but actually they're not robust anymore right and for that we need not only a measure of you know of probability but also on robustness and this is what what we do right if
you i don't know how close you are on neural networks what ends up the most guy is doing is at the very end layer of neural networks we add the soft max layer to classifier right the distribution of probabilities there and um this is nice but they also have the disadvantage um they will always give you some high confidence on certain aspects even though they're entirely in a different domain it's because it's like um how do you say the algorithm is just fulfilling your basic needs like your first there's this term for that the algorithm is optimizing for what you asked for not what you really need for exactly like if you train the system aren't overfitting was it yes somehow overfitting whether um it's if you're trained uh on cats and dogs you will always try to identify cats and dogs whether it's elephant and and muscles right it doesn't matter right it's trying to predict that and it means we need to identify um and we do that by it's called temperature scaling
right what is it you want to have two feedbacks back one is the probability distribution and the other one is the likelihood that the domain is still existing where you are living it right and this this is one aspect where you send you know making things more robust um in your application and we extend that by means of you can pass systems by uploading data to our system to the system what we deploy and it adopts the weights uh towards the new domain so you don't need to retrain it because you just have you know different lighting conditions or different aspects but it adjusts towards the more robust solutions so you want to have an indication am i still in the domain the main adaptation is key here and the second one is having it somehow adaptable and fast deployable and technical do you have a dashboard for this everyone can use or everyone has to build it on its own or do we have to look on the code for this like visual how
can a visual get this very quick measurement to know okay this system seems to be robust and seems to be trustworthy um yeah so you um you get a score back right score a score right so similar you're probably building 80 percent card yeah it's somehow somehow right it's it's it's not a product that is sold outside but it's um you know this is um we we contribute also to scientific community right first but let's check out the papers and second one is we apply to you know to internal from robotic handling from pieces handling currently customizability on a software second one is graphical models such as you know bayesian networks which you know it gives us a bit of entering towards you know graphs and basic moods um which are a bit more explainable and gives us uncertainty and uncertainty is something really important because you know uh that system say i don't know right is is is not by default designed in supervised supervise into supervision right classification scenario it's very important to
you know cope with uh certainty and uncertainty of addiction right and you know counteranalysis question right um these kinds of aspects and the third is graphs we do that for a graphical representation graphical representation and graph also deep learning applied to graphs and knowledge graphs like entities relation connected to each other and these provide also very nice um explainable um from arts what was happening right we did that for healthcare drug repurposing right can we use certain drugs and with certain paradigms model it is a graph structure and then can we use certain um you know can we repurpose uh and fourth is we combine both learnings so we you have you train you train and maybe say a blackbox algorithm and you train a system that is more explainable trades rule based right but you you you train these systems on the output and approximate towards your black box algorithm um and so it's it's not directly opening up the black box and say show me one of the 100 million neurons but it's
rather what kind of bias do we have in in um in the current algorithm okay to give some kind of okay and first is uh sorry fifth is observing the entire process because it is not about a one-time deployment if you're not observing that you if you update something change if the application change something changed so you need to have a consistent view of the entire deployment process and continuously monitoring don't stop here's my model deploy i'm out right um but consistently you know the glowing antenna okay okay so um i would like to sum here for a second with the pulse i had asked yeah what are the main reasons we are lacking behind in b2b ai integration yes in contrast to end consumers and number one answer was well we have not enough awareness in the company for ai so this training education of your team your management is key number two is lacking frameworks and mythology to implement ai i think so training them and giving them tools at the hand and
best practices having this competent centers i think this is the second stage and then it's uh it was a mixture of fear of making mistakes and top management does not consider ai critical enough so you explained really well today two factors um just to the end you next step would be opening the ecosystem opening siemens to build at the end a platform to adapt innovation from the outside and giving it back and we will have a dedicated talk about this in a couple of weeks but auto yawn about how to build an open innovation system but tell me a little bit how siemens and your role doing open innovation in the terms of ai okay so we we um i i said like you know we should talk about not only the the virtual function but also that's what prosperity so not only the depth of value but also the breadth of value okay now talking about the breadth of value breadth of value is a strategy starts obviously if you think about what we
want to do but a strategy actually also is as of importance is what you don't do defining what you don't do okay is is of importance because then you can leverage the power of the ecosystem and if you if you know this is not our core this is what we don't do and develop ourselves then you look for partners doing that okay and and this this is i think that the moment where uh we on the one side see a a continuous changing and intertwined ecosystem of partners if you know what you don't know you can leverage speed and scale with the partners outside because optimizing tower as a process is something done within an organization but optimizing towards um towers outside it means you need to have the breasts and this is what you embrace them as what is called a network effect you can optimize only to a limited amount you know within your own product but if you launch new products a new value change and you want to value the transaction
whether between your partners which you enable is only doable outside and that's why it's it's of importance to actually identify you know what value do we give to our ecosystem partners so let's take and the natural let's take this natural reflects natural reflex is you know i sell you a service right and i sell you this service right so i don't create any value between you and your other customer right because i i i would potentially even separate you because i was like i want to sell you once and i sell you the same product you know a thousand times but this doesn't scale a network fact this scales on on optimizing the products i can make it cheaper i have a higher quantity i can optimize my supply chain but this doesn't scale in after effects the network effects kicks in if i'm enabler to connect you with your other customers and the number of customers and ecosystem partners are much broader than i can sell individual partners and that means scaling over networks
means giving up value and enabling the ecosystems together by connecting system partners together and this this is i think the challenge in europe and the challenge we need to embrace how do we give a value mindset again you know leadership skills how do we change that because not optimizing you know until 100 but giving away values you're great first you're the first person answering me how i get from strategy to implementation in real case i mean a lot of people saying we need this and you gave us really really good hands-on advice today i love it and i will replay this often for everyone out there so i thank you and second i agree and we will have different talk about this how siemens adds to the value ecosystem how we can integrate but we don't have the time um therefore the two last questions or statements from you first summarize a little bit because you were into ai for 10 years summarized what achieved ai in the last 10 years where are we standing
today and what is your vision what ai could be in 10 years you know what is the next decade bringing us in terms of ai i know it's hard yeah um yeah so what brought us um it was um a a screwed um perception of attention economy which leverages attention and interaction on purely consumer space this is what it tells us um and um it it went in these kinds of attention space that i think it's uh close to an unhealthy state uh yesterday was a paper in in new york times and they did performed an experiment when people people were asked to stay 50 minutes alone with their thoughts or choose the option to to get an electric energy right to bring an electric shock for themselves right um two-thirds of man rather uh have an electric shock accumulate to themselves than being alone with her you know so it means these attention mechanisms and this this incentive it's you know predicting attention and and is is is quite i don't know quite wrong
and the challenges you see and i hope this is what an action is we embrace the right challenges and what are the right challenges i think we have a great leader board and a great blueprint board which is called the sdgs and um that means you know we we would we would need to rethink um what is the value system we want to represent and what kind of these values do we want to tackle and embrace with technology in a responsible and meaningful way and i hope that this is not only in a consumer space but also it serves more than one single partner in the world right that's that's a big one i fully agree i collect this from every of my guests now because i like it um holy i know you have the next meeting coming up yeah i assume it will be a little bit less interactive but productive for your corporation and for your career and maybe you're the the the chief ai officer one day at siemens i would
welcome this i i don't i don't care about you know about role i i kind of i care about roles and responsibilities and accountability is not not necessarily you know the naming of the things right okay and by the way um you do an amazing job um i think the ai narratives for europe right knowing about the strengths knowing about the weaknesses but identifying opportunities and for also um you know motivating startup ecosystem and the ecosystems partners needs leaders and needs you know thought leaders would say like you know what i'll bring that together and hey this is what you ask you know on an excellent way doing like who do you think thank you thank you very much very kind words that's why we have to show together you brought us a lot of knowledge olly thank you for this everyone stay please i i do the the rest of the moderation for a second but goodbye to oli um you can add him on linkedin he has his own private webpage so he's
very public about how to reach him it's worth having a chat with him i wish you a good day thanks guys thanks guys for joining in much appreciated um so for everyone else left here um announcement in two weeks we have the next um rise of ai virtual chat so one week break in two weeks with professor christian goodman and we will speak about the ai roadmap for your business so on purpose we will go deeper what ollie and i touched today i think that was really really nice and we will scale this now for companies out of the mains a little bit more neutral and let's see how everyone else can do this also would like to announce that we still still we have the summit out there it's 17th 18th of november siemens will be partner this is not official but um only put in a good worth and i thank him for this um siemens will be a partner too i think this is important because emits shows itself in the ecosystem
that's why we do events that's why we cooperate with other companies there too it's hybrid so 10 people of you can be there in person because you know we only have 10 people left everyone else can join virtual and it's free we decided to do everything for free content streaming networking it's free because we want to open the ecosystem we want that everyone can learn understand and adapt artificial intelligence so go online go and rise if i get your ticket if you haven't done it yet join us and else i wish you a great rest of the week goodbye you