and this is yours. Yeah. Hello everybody. Um, thank you so much for being here. Um, it's a great honor to be here at the rise of AI and uh, all of you. I would like to say thank you for joining this talk and uh, I don't want to disappoint you. This is not about techniques or technical stuff for the next bank benchmark we are going to uh, to break. Um what I'm what I am I'm a great enthusiast in AI and I sometimes wonder why these systems will not be used by every one of us. We all I think we all are here in the rise of AI and we believe in the AI adoption and the the transformation and disruption we are in but sometimes I believe we forget about the normal humans about the people who are going to use these systems.
We are engineers. I'm a computer scientist. And um we at at Vinmer and Hera we are machine building factory company um German militant and uh we we came to the same problem. We have good systems very very often and we have a good solution in AI and but what is needed I I believe we need a better story. We need a mascot something which gives orientation and and helps each other uh to to go further in the AI adoption. So a few words about me not so much swarm intelligence football playing robots. Uh I've been a leader for uh robotics and machine vision. So very technical everything until I got a a job at W&H in VMA and right now I'm with Vin Hersa since more than 10 years and have beside other jobs the co-lead for the AI center of excellence.
So but who are we? Um we are here um since more than 150 years. So we have we are company which started with team building machines with steam steam powered machines. So this was the first thing we have done. So and we been through several disruption since electrical revolution and so on and right now we are in industry 4.0 and we are a global leader. I think every one of you have touched today a product of us. So we make products or or machine produce products which um protect goods, protect food um and we are global global global leader for that. So no one else is selling more of these machines than us. So but I think t typical in Germany typical German middleand um we are a hidden champion and um we are always here um for a better solution.
This is something which is kind of DNA from us. We are not selling machines. We are not selling products. We are here to help our our customers for good solutions. We won't leave when the machine is delivered or commissioned. No, we just leave when the customer gets the best best out of its machine. And since we are very successful, we have grown very much. And I think it's also typical for every one of us, for every for every German milant. Um, we have doubled revenue during the last decade. Employees grown global connected hubs. Um and since we are in Germany we are let's say 2,000 more than 2,000 people and all around the world more than one and a half thousand.
And um this leads to a challenge and I believe most of you who are working in such a medium or larger company are facing the same challenge the growth in complexity we are in. So the complexity rose because of regulatories, because of technical advancements, because of global conflicts and of course because of IT systems. I sometimes believe it systems does not do not help. If I'm saying uh to to our people, hey, let's introduce a better system, a new system, they're saying, forget about that. I just have 15 six systems which each of everyone must to be um be used and I'm totally lost in these in these complexity things. So the people are overwhelmed about the the systems we are bringing them.
We are not bringing them solutions and remember VSW and H are always here for solutions and um of course we also observe that we have to grow more efficient. The the market pressure is very high even for all the specialized products. We feel that very much and we must more efficient one um employee must be must earn more money um than in the past and I think it even counts more in the future and this great growing in complexity of course we would say okay come on let's use AI to makes lives better and yet let's see how well we tried that um one step on one one thing is also counts for complexity since we have grown very much we tried as typical good Germans to document everything in every single step.
Me as as a manager, as a leader, I I I love that because I say, "Okay, everything is under control." Okay, come on. Here there's a process diagram. You must exactly know what to do. But is that true? I don't believe that because who as a human wants to uh have a have a answer question be answered by looking at a wallpaper of diagrams? That's not the way humans are. I believe humans are other than that these are good systems or good tools for computers. Um and so we believe that we must need a bridge a bridge from this complexity and such diagrams even let's say um takes care of even more complexity and not not helping people.
So we need a bridge and a good idea to have a bridge is the use of AI and we have discussed let's say very much of of topics right now and I believe that uh large language models or other foundation models are very good in transforming these data complex data to make it more understandable. This is good for let's say big data like Excel sheets or time scale databases and so on. But I also believe that we can use these systems to make it to make processes more understandable. So and this is what we have done and um this won't be so much technical but a few two technical sites if you don't mind. Of course we know about of all these kind of things and it has been worked very very well.
So lots of data engineering inside of that and and large language models and so on. And this is not simply rack, not simply retrieval augmented generation we use but um these these process diagrams are being built with JSON files and was a great um great topic for us to together with SVA um to transform these data in human understandable data and lots of other things like infrastructure system engineering and so on. But for human that's all under the wood. It must work seamlessly. I think that's the first point. You need good systems which work seamlessly so that you have a good user experience at at at the beginning. And yeah, let's have a look level deeper like multi- aent systems.
I think most of you know about what multi- aent systems are but for those who don't know about that um multi- aent system are with multi- aents and each agent does have some specific skills and some will have let's say the knowledge of pre-process data um some knows about other databases and in front of that there's always a delegation um agent so-called delegation agent and this he uh this this agent takes care of understanding the user's request address. Um it would be good let's say if the agent knows that if I'm asking hey what is my next step to be done that he needs to ask agent one because he knows about the process and knows exactly how to answer this question another question would be like give me an overview about a process so and the delegation agent all of these kind of things will be aggregated and be answered in a good user interface since we have that suddenly we are able to um to answer these simple questions.
Is the software release already approved? And um usually we would have to take have a look at a wallpaper full of processes and once again I believe people are not working that way. People need simple um answers for that and of course it has to be friendly and so on and then I believe we going to have a better quality because people are working accordingly to our good German processes and we have the best out of uh the machine and quality and cost efficiency and so on. So that worked very well. But then once again uh we were thinking about okay this work work good but I don't know about you I still don't understand why people are not using chat GP or these kind of tools and I was a little bit worried about that and of course we had discussions with people in user experience faces and so on and then user stories and uh also with the works console like of course you have to be compliant with these kind of things
and they say what does GPT even mean? People don't understand that people don't understand what GPT means. We as a tech all light for AI literacy where we know what the GPT is or maybe we believe you know what it is but uh we don't want to call it W&H GPT. We need an interface which is very good and we need an interface which is as good as we would have been done it to our customers. what would would the customers think about that and we must think about our as our employees as a customers. So, and I like the the the Apple AI um thing very much for the rest of us and um we as three and a half thousand people need something which gives um orientation better orientation.
So, let me introduce to you Vinnie. Um we of course Vinnie lives in Microsoft Teams um like everyone else is here in Microsoft Teams and the the communication takes place in Microsoft Teams anyway. So please don't make another system where is also make a slack interface or whatever slack integration this is where everything happens where I have communicating with my people with my colleagues um even with Eileen I'm communicating Microsoft teams and so please use that and and make it as easy as possible not another system break not another system and uh I think Vinnie does not look like an foreign uh artificial intelligent um artificial intelligence for the normal people this like an alien out of space the first time I don't know when when you were looking using shbt I was completely blown away I could not understand it and even I'm a computer scientist I know a little bit how this works but normal people I think they're afraid of that it's it's something completely it does not create familiarity something which
is normal and harmless and so And what I also like and this is a quote from from people they love Vinnie. They say I'm loving Vinnie. Who would say I'm loving co-pilot or SAP? I I don't believe that anyone would would would say that. So a good a good emotional response and like a like a teammate you know you you are not uh get angry at a teammate because he sometimes makes mistakes. No, like a teammate like some someone who can call you, you can call who can call for a specific topic like explaining processes. Um, but not the almighty AI and once again yeah of course it's built for humans and not for our techies.
I think we techies we are fine enough in this AI thing and we made a we made her or him we are not sure about this gender. We made her young because um a younger penguin I think you sometimes forgive mistakes. She's not perfect. Our AI systems are not perfect. We are in a in a journey with AI. Uh and we are hoping to have it perfect in the future. And something which is also very important, it's okay for Vinnie not to know. So you can ask Vinnie something which you don't understand and you can keep asking and keep asking and keep asking and Vinnie will always be kind to you. It's not like asking your boss about a process.
I think the pro the boss would say okay come on he does not deserve a race or something and since we believe that Vinnie must be a young penguin at that we want everyone at WH at culture to raise her to feed her to educate her to um to give her more knowledge about uh this the things we are doing and helping each other um to to be Vinnie a better penguin a better assistant for all of us. So a few examples um a diet for day in a life with Vinnie I think is quite familiar right now you can question um these questions ask these questions um what needs to be done after software release is completed and of course Vinnie knows about that and she's saying okay after release is finished it must be tech and so on so but does that really help no because you can after then of course you can ask Vinnie hey explain it like I'm a five-year-old this is normal I think it's not not magic It's
normal large language model functionalities. But of course you can do that and right now I believe even my kids knows how to uh tag a software version as table. So just keep asking be friendly and be kind to each other and um there it is. Um yeah this is let's say one of a of key key message that I'm sending to you. Please think about the normal people. Build systems which are for every one of us and think of of of humanity and of human centered solutions. It's not only counts for AI. I think it's for every solution and think we know about that very well because we know we build solutions for people and for our customers.
And um what is also very important I believe it's very important that we need something to to integrate the system like a brand like a new product like a good marketing uh we have cups we have t-shirts we have stickers everyone like uh you can stick it all around at W&H please ask winnie ask winnie ask winnie and uh of course we let's say we at rise of AI we are here to to help to make a better AI literacy. This must be at everyone in our society and not only the young people, not only the techies, but every one of us must learn how to use these systems and we need barriers which are as low as possible and as friendly as possible.
Good. So, and what is the future with Vinnie? Um, this is just a saying we're having. Isn't that a task for Vinnie? So, everyone must ask him or herself what am I doing the whole day? So Vinnie gives orientation so and everyone can ask himself hey isn't that a task which Vinnie cannot be used and we have lots and lots of use cases and we made it that easy that you can um talk to Vinnie and the use case implementation is quite quite easy for us. Good. So and of course in the future um who are let's say the people who who deserve the best knowledge for all of us I think our service technicians we have more than 400 of them they're traveling all around the world because we are global player um they don't see their families for days they are traveling let's say more than 250 days a week a year sorry here and of course this is the next step we are going to do so this will help our service
technicans uh when they get the best of the knowledge and on the other hand we also believe that our systems for our customers as I said I believe in the transformal power of these systems and sometimes our machines which are let's say five 15 mters uh in length and 7 mters in height and so on are large complex industrial machines and we are going to simplify the interaction with these machines with interaction of all the data like all the industrial uh IoT data so you can simply ask how was the last shift, what are ways to improve and we have showcased that on the last trade fair and I believe in the future we're going to have that even more so and once again this is a symbol for us a symbol for the AI transformation and um Vinnie is getting more and more important for us and we need a good good way to operate these kind of things and Eileen is going to show us how this will work.
Sorry. Thank you. So what also lies within the close future is the next big chapter of the UI act coming into force in August and with it we need to decide which of our models counts as high risks and Vinnie isn't really in that category. No one's life or livelihood depends on its output and there is no strict regulatory mandate beyond being transparent that this is a model and not a human being. So you might want to ask why then AI governance in this talk at all. So Jung as you already said how much do you depend on Vinnie being reliable in your day-to-day operations? Yeah, right now I think it's okay if if she's sometimes wrong, but in the future they must be more reliable these systems.
So and this is where AI governance comes into place by covering a lot of those topics like fairness, accountability and so on. you will pro you probably also cover most of the things that the UI act is now requiring for more high-risk models. So this is not only required by the UI act but it also is really just best practices in my view for very complex productive AI systems and that is actually not really something new. Machine learning teams have done this in the past by applying machine learning operations but there is a little bit of a change now. There's a lot more non-technical stuff that needs to be aware of what's going on and be deeper involved in risk management.
And because we haven't already talked about enough buzzwords today, there's another one I want to introduce to you called LLM Oops, which is more focused on large language models and their specific nature. And there is a lot of there is a lot of old stuff that we know from MLOps like automation and version control. for example, you still need to track your models, your data, your code, and you want to do it as automated as possible. Especially if you want to compare runs or roll back runs, but there's also a lot of new tasks on the blade, like for example, prompt and context management. You really want to track your prompt and context windows as closely as you do model code because slide changes will drastically alter responses.
And I don't know if you have experience already, but if you tested out different models from different suppliers that are comparable, then you probably figured out they do have their own personality. They are not interchangeable, let alone new versions of the same model. And for response quality evolution, the problem is that the output is inherently non-deterministic. So the same input will lead to different output. That's the whole idea of large language model as well. And you still want to track the standard quantitative metrics like rouge, blue, coherence and so on. But you will have additional challenges um for example a human in the loop and especially an informed human in the loop and also something like hallucination detection and so on.
And I could have a completely separate talk just talking 30 minutes about the safety requirements for Adams. So if you want to incorporate all of those things uh into an newly designed system already existing one um then you need to consider all those points and with Vin Miller and HA for example you already have a very comprehensive technical setup so that addresses most of those points however there is still lack of visibility so people outside of development teams do not really have a chance to see what's going on and be involved And this lack of visibility, transparency, and control might become a problem in the future. So this is the point where you might want to consider a governance tool, especially if it's getting more complex and newer AI solutions are introduced and that's why I brought to you today to introduce to you what's an XG governance.
So what's an XG governance is a tool that works as a governance layer. You can put it into basically every infrastructure cloud, hybrid, onrem. All of this is possible and it allows you to define your own workflows something like approval processes or documentation requirements while tracking models matrix and risk along the way. So it basically is allows you to build a unified platform your for your whole AI's life cycle. everything from model experimentation, versioning, development, deployment and all along the risk cycle according to the risk catalog of the UAI act but also your internal policies. So in summary, this is a tool that does documentation, reporting, risk analysis. You can define your own workflows and it basically is integratable into every infrastructure by using REST RPS or Java RPs.
And with it, Jurg and his colleagues will be able to track Vinnie's next step and get her from puberty into adolescence and also all future brothers and sisters, I hope. Okay. Thank you. Thank you. Adolescence. Next one. Any questions from the audience? Okay. Uh I have one. I mean um Alen, you have been uh instrumented over 30 major enterprise AI projects in uh Germany. Um assuming it's always about human centered technology. Um how do you measure the um success rate with these tools with with one tool or well um the let's say it differently um the change management initiatives how do you really measure the success that this actually really helped in the change process that people not just said oh yeah it's fun but also integrated that also there was a real change.
That is a really hard question. It's really qualitative. I think it's a little bit like you you already said that people are starting talking about it. The more people start talking about what is implemented and actually do it by themsel instead of just being pushed to use new tools, the more you are sure they actually want to use it. And that also helps you then with the next project to basically sell the next project saying, "Oh, Winnie was successful. Everybody wants to use it." Yeah. And thank you. Any questions otherwise? Nobody wants that's all right because we are really good in time and well then I would say round of applause for Yung Brinkman and Eileen.
Thank you.