 Thanks. Thanks for the introduction, Gabor. Yeah, so this is a little bit about me, but I think you have seen my face around the conference enough. So this is my contacts, and I'll publish the slides. So as you see, I'm a partner of a big consultancy. And actually, what do I do when I'm in my daily work? So I work with clients. And the promise from Koenigsegg is we enable clients. We don't try to sell them PowerPoint slides. They are maybe 5% sometimes, but we try to enable them on the level they need. So my daily work is to train people, to consult on strategy, how to build data pipelines, where to go, and what the possibilities of AR are. And interesting, how did I come to this talk, like AI for managers? Because actually, last year, I basically had a blind spot because my background, a little bit more above my background. My first career, actually, was 27. I was the COO of an independent record company. So I know the management had a bit. And how employees, and we had a software problem. And we were not able to solve it. There was no software for our use case. So I taught myself programming. And this is probably one reason I'm standing in our days. So how did I come to this talk? It was actually, I talk a lot to people at conferences, to data scientists, developers, engineers. I talk to clients. I talk to managers. And actually, what I found out was quite surprising. Because mostly, the tech is not the main problem. So after all these good talks, and they were honest, and in safe environments. And before I share what I learned with you on the perspectives, let me a little bit learn about you. So who is more like a tech role, like developer, data scientist, or, yeah, OK, most? And who's more like in a managing role? So let me explain you managers a little bit better. So I give you some perspectives. So first, the data scientist slash engineer developer role. So what I learned, what people tell me when I talk to them. Many data scientists are frustrated in companies because they cannot live up to their full potential. Many tell them I'm highly trained, but I basically work as an analyst. They say the company is not well-prepared for data science in general. It's problematic to get access to data because everybody knows we should do something, but many people don't know how. Also, especially in enterprises, being along for a while, the whole tech stack is just like a big patchwork. So they were like 220 good decisions over 40 decay, 40 years, but now, if you align them, it's a very bad decision now. So it's a patchwork. It's really hard to struggle. Also, people like they, they, they miss people making decisions. Can I do this with the data? And basically, very often in a large enterprise, they have to say, I don't know is the better option. It's not like, OK, let's try. Let's there. Let's bring this forward. So this is where I hear from data scientists and companies. But also, I don't want to generalize. It's not like every data scientist. So I also know happy data scientists. And they say they have easy access, easy and then compliant access to data. They have access to domain expertise in their company. So they are the companies aware. It's not just like, OK, there's a data science problem. Let's move this to the data science department and solve them. Though it's teamwork with the actual department. So they work in teams. They feel they can drive innovation and also bring value for the company. And let me also give you some manager perspectives. So I did some research on that. What do managers actually do? So this is what managers actually spend time on. And you see, most of the stuff is more administration. There is some, and only a small part is actually a strategy and thinking about innovation. It's only like 10% than I got this from Harvard Business Review. So I think this is a very good source for that. And also, the same in the same study asked, which skills do you think you will require? And this is also very quite interesting, because they think you will read more digital technology, data analysis, interpretation. So this is actually like data science is like one of the top topics. But you see the people skills are pretty low. And I think this is a big mistake. And this is a big mistake. I, from my programmer or data scientist perspective, made the same. Because all the tech I do is great in a way that we can change the world basically, like as we start up, can change the world with this tech. And I learned, no, like working with people, like making safe environments, like talking to them, like really having real conversations, and not just like buzzword bingo. It's a skill and it's something we need to develop. And both sides needs to understand each other better. And this is basically what the talk is about. So what I usually do is I do a talk or do this very often. So let's Google it, AI for Managers. And this is what we see. Of course, it's my favorite white robot again. It's always like the same images you see at Google. But interestingly, what happens if you Google Replacing Managers? So this is the perspective. And what basically, yeah, but basically Google just crawls the web and makes this information searchable. So actually, there's a lot of stuff around embracing managers with AI, technology replacing managers. And there's a big buzz going on, like, managers can just replace by AI because AI is super smart. And once you start to work with it, you learn that AI can do impressive things. But I've never seen a smart AI. The bearers just like to have a better comparison here. I also did the same for data scientists. So it's more like an open question, will AI replace data scientists? So it's really good. And actually, it also fits to that our data engineers are probably one of the best paid people currently. Actually, if you try the same with data engineers, nothing. Oh, wow. That's a safe spot. So to get perspective, so basically, I think this is how many people feel like when they hear about all the buzzwords. I mean, we have to see. The cloud providers are beating the drum. Oh, yeah, you can do data science with cloud providers. We have all these great tools. And you can have Colab notebooks, Azure notebooks, basically, they're all Jupyter notebooks anyway you can have. So they are really good. And many companies started to adapt it from startups to promise products which are basically almost ready. And everybody is really moving at a really high speed. Actually, it's quite impressive to see open source tools adopted by major enterprises at that pace. That's pretty impressive. And so in all that, people, especially managers, because they never make their hands dirty like programming. So they lack the experience and the feeling and the real obstacles we have to face when we try to realize a project. So they're more like this, buzzword here, all Kubernetes there, Hadoop over there. And so I would like to introduce a little bit what where I and my team stand, how we solve this, actually, or how we also propose to solve this for the future. Because for us, it's very important we want to enable clients. I don't want to sell ours or something like that. For me, it's OK. The client has a problem. Let's enable the client. And no matter what. So one big problem is actually tech data and AI literacy. So many people really lack deeper knowledge about the tech. Of course, everybody has heard the buzzwords. But can you really describe it? Can you really do you really know how important domain expertise is in the whole field? Because I think two years ago, the PyData Warsaw, one of the keynote speakers asked with all data scientists in the room, who is actually working on a project where the data scientist has domain expertise as well? And 20% of the people raised their hands. And I think this is also very often a misperception. You can just move the, like this is how many enterprises work. You have a problem and you try to move it to the desk of a colleague, just to get rid of it. And I think or to another department because then we have less work with humans. We have to see this as well. So just give the problem to a data science department which might be allocated in IT. And then people trying to solve a problem. But the one question is, are they really solving the real problem? So the first question is always, what's the real problem? How would we solve this without data science or AI? And I just want to work you through some misperceptions I have learned in the field. So one misperception is bigger is better. So let's get in the Hadoop cluster. Because we have data, Hadoop can handle any amount of data. So we're in a safe spot. We even haven't asked the question, how much data do we actually have? So do we have a lot of data to deal with? Or do we have less data and maybe have more a complex problem to solve? This is like two very different technologies required. But also in defense of people, I heard that companies born in Hadoop cluster before they had the first data scientist. But if you also, yeah, we all laugh from our perspective, ha, ha, ha, stupid. But actually, it's not stupid. It's just a different thinking. For example, think about you are a company producing things. You basically, you order the machine, and then you hire the people operating the machine. And for them, the perception, for us, Hadoop is software. For them, it's probably more a machine crunching data. And this is like one misperception. Or for me, especially, I like Hadoop. Actually, Hadoop ecosystem is facing problems currently. There's three companies, two merged. One has problems, so it's not everything's prospering Hadoop. So actually, it's quite interesting what's happening there. And nowadays, basically, for me, the new Hadoop, which is also a great technology, don't get me wrong, for me, the new Hadoop is Kubernetes. If you go to a client, everybody says, oh, yeah, we need Kubernetes. But yeah, we are just building models and trying to solve the problem. What do you think we can solve with Kubernetes? So yeah, it's Kubernetes. It's basically, it manages all the machines and the resources, and there's nothing we have to do. Sometimes it's really hard to stand up, say, no, we don't need Kubernetes. We maybe just need one server to self-models to see where we're going, to further the project. Of course, in the future, we will use Kubernetes, but we will use it when we go to production, when we have maybe 100 clients, 1,000 clients, million clients, we have to say, OK, then we need Kubernetes. But we have to focus on solving the problem and not solving the problems of Kubernetes, because there's not one Kubernetes. There's Kubernetes on Azure works a little bit different, and Kubernetes on Azure, or AWS, and Google Cloud. So it's not just like Kubernetes is the all-solving problem, and there's many updates, and many things you have to learn. So it's not just like, click here, Kubernetes, everything's working. You need people to operate and have experience how to operate this, as well. This is very often forgotten. Another pre-perfection is data lakes are where there was. So we have a big company, we have a lot of data, so let's just put them all in the data lake and draw the data out of it. So we have this clean lake with all the data. We just need a hub. Can just pull the data in, or as we basically describe. So basically, we tell our clients, this is just like an abstract. It's just a concept. So a data lake can involve many, many systems, file storage, databases, ERP systems, basically everything. It's more like a concept to think about. But it doesn't really solve your real data problems, because it only solves probably a storage problems, and it maybe helps accessing the data. But still, you have to think about the company culture where we come back to the humans, because you need a company culture. People appreciate data. Also, save data in the right spot. Add metadata, the right metadata to the data they save. But I think many just throw them on some drive, and it's forgotten. And I know companies that built crawlers to basically understand what data they actually have, because there's many people involved. So this is just lateral. Somebody is more or less careful, leaves the company in a good tone, in a bad tone, and more or less maybe is cleaning up or not. So actually, company culture can help. The governance of the data, and also curation of the data is important. It's not just like, it's not a tech solution. It's, again, a human. We have to solve this as humans. We have to evolve, build a healthy culture. We also see very often, well, we've seen this before. This is just hype, sorry for the typo. So actually, let me put some evidence for this. Here you see some covers of MIT's technology review. MIT technology review goes back to 1879, and MIT has, they put them online. So we see 1985, we see technology automation, fix them up automation. We see 1986, will artificial intelligence ever deliver on its promise? And in the middle, we see like in 1984 as experienced nowadays with Stranger Things. Another fun fact though, this continues actually. There's something from the end of the 90s. Can computer create literature? And how to keep major, major industries innovative in 1987. So also like, we think we're having all these new AI and data problems and perspectives and dreams what computers can do and help us build. And actually, this goes like really way back. And if you had been in Europe, Python and Rimini, also there's a really interesting keynote by Catherine German. She actually told what people thought that innovative AI, because many people are not aware AI is like super old. AI predates databases. This goes back to the 40s. So there was always like obstacles. And so there were multiple stops and dark ages or AI winters where somebody important wrote a book and said, okay, the XOR problem is not, it's not solvable actually. And this put AI to sleep. But which were the challenges? Not enough data, not enough computation of power. But we see as the electronics evolve, software evolves, we see more and more of these problems are being solved and we know after basically the beginning of the 2000s deep learning researchers were like exods on these conferences. Now they are the superstars. So the times, the tides have changed. Basically it's upside down now. So you see, this is like a history, but actually it's like waves and then we hit certain limits until data, more data or the technology and computational power solvers. And also where do we actually stand with deep learning and artificial intelligence and machine learning? Because nowadays everything's artificial intelligence and yeah, okay, it's basically just this. There's artificial intelligence, basically the whole field with machine learning as a part and deep learning as another part here. So basically it's not a new technology. Many people don't know, oh, we talked about data science, machine learning like two years ago, it was a big hype and now it's deep learning. Is this something different? No, it's basically the whole thing, just the family. Also, it's not actually new and actually all the questions and challenges because actually I visited a applied machine learning conference in Lausanne earlier than the year. And actually, Gary Kasparov was giving, was talking, who remembers or knows who Gary Kasparov is? Okay, okay, I have a few. So let me give you a quick talk. Gary Kasparov was the superstar of chess. He was basically the, what's it called? Magnus Carlsen of its time. And you see here, and he was basically playing chess against an IBM computer. Basically this is the very similar moments we experienced with DeepMind and playing StarCraft or solving Go nowadays. So at that time, many of the innovators of, or like theoretical innovators of the technology we use here like Turing or Robert Weiner or Ferdinand, they all thought if the machine is able to, like chess is the game of kings, so once a machine is able to be the human in chess, the machines have won it. So obviously this was like in the 90s and Gary Kasparov was the brain's last stand. He was our last hope and actually nowadays he's giving talks. Actually, he's also like saying, hey, this was a good experience. He's not at all against the tax. He said, here's the straight tools and this is really great. So this is actually like machine AlphaGo zero now with compared to human chess. They also like very concepts as we would think. Why not combine deep learning AI with humans? Actually they had the same ideas then like having like work with a computer to learn how to play better chess or have strategies. But he also put it like, he put a like a really nice citation of Pablo Picasso also there. So computers are useless. They can only give you the answer which is also something really nice to think about. So the problem very often is the problem seems simple as well if you look at yourself being really critical and so, or other people always keep in mind very often the problem seems only simple to solve because you actually have no clue and experience in the problem. For example, you watch the video about some new technology and after an hour to the video and you see, oh, I totally got what the people told me. And so until you go to your computer, somebody asked you, can you please reproduce what you have just learned? And then you will probably have this moment. Oh, again, what like, because the brain actually is a very good in making us believe, feel good. Oh, if you've understood, just like these cloudy ideas and very often. So this is nice, this is a nice experience but we have to also be critical, say, okay, maybe it's also like our brains just giving us a good feeling and tricking us as a bit. So very often we see over and biggest imagineering, like, okay, we can solve this and this and this and that. We can like forecast things without actually having experimented on it at all. And also very often because there's a deadline and there's some deliverable premature shipping of technology which has not really been understood, just produces some outcome put which looks nice but I think it's very important to check that to explain what actually models deliver and why and whether there's a bias in the data set and whether it has an ethical standard as well. So it's, yeah. So there's also personalities. So because I have been bragging a lot about like managers now let's brag a little bit about data scientists. So actually just like last week, I saw a talk by Peter Baumgartner and you see already and he had basically built some really good personas for projects in his work. He's working at a research institute. So you see there's like, Anna-Dune Andy doesn't know the problem but wants to see what things can do. You have EasyEd, you have show of Sarah who just wants to do AI. So maybe you have a look later at the slides I don't wanna go for all of them. But you see it's always like really important also to understand what are the personalities of the peers and the team you're working and also how to deal with them and to handle them but also in a respectful way because show of Sarah wants to do AI but basically maybe this is not a bad thing. She's maybe just like a curious person like she wants to do in a way. So this is not necessarily a bad thing but we have to basically agree upon step by step use a minimal version first and then improve and see how far we can get. There's also a blog post that put the link there. And also like for big enterprises, IE, IT is the wrong track very often. So I Googled IT and I was very surprised. So maybe it gets the tense how IT department feel but actually, yes, I didn't expect that but very often data science and AI projects are allocated to the IT department which is I think very wrong because data science and IE is research and development. So we need an open culture, a curious culture to learn things. We have to do experiments. We have also budget wise prepare something might fail but we might have another finding. And it's not just like, yeah, it's not just like, okay, buy this from Microsoft or any other cloud provider and they solve it anyway. It's very different. So we also have to see what do we have the data? Which domain expertise do we have? Also, many companies actually also struggle that senior people leave the company and we lose the knowledge and there's not new people coming. So it's not just like about replacing people. Actually, we're losing a lot of knowledge nowadays because many people like I think, well, I call the baby boomers are getting, ah, yeah, they're becoming pensioners now. So we have to deal with this as well. It's not about replacing. It's like basically also keeping the knowledge and keep things going. And actually quite unsurprisingly, clients reached out to me and said, okay, we've seen your talk. So I have given different talks on deep learning for fun and profit last year. And actually clients reached out and said, can you do this, give this talk at our in-house conference as a keynote? And I was surprised why because I always feel like a little bit like, yeah, this is funny and experimental and I learn a lot. Where do you actually see the value for your people there? And they tell me, no, I think how you present it is very good to make them understand the way. So I wanna share a little bit of what I did. So if you have probably heard about style transfer, learn on a French comic and take a modern DC comic and apply it. Apply it. So basically I drew a modern DC comic into a French comic style. And so, and you can also do like this, take a Piazza de Farnese in Florence, Google some Avengers, just copy and paste them. No Photoshop involved. Just copy, paste and preview here on the Mac and you, it's like a five minutes project actually and apply style transfer. So basically I trained an eye model to learn the style of the comic. Apply this and said, okay, this is a nice picture. If we used to have the resolution of the input picture. So now the picture size is now half. It will look different and also the style transfer will look different. If you, for example, you can also go and say, we take a higher contrast, which looks like this them. And if you put it there, also like the style looks a little bit different. But fun fact, for example, we can also use a black and white version. And if we apply the style, unexpected is not changing so much. So here you have a really and very good combination on input output levels to see, okay, data quality matters, which we will see in the next demo. And also to have a very good connection input data, output data model and also maybe mistakes or artifacts being produced. So if you wanna learn more about this, talked about this extensively last year, zero Python, PyData Berlin and also especially on the audio stuff I'm going to show you now in, just like in PyData Amsterdam a few weeks ago. So yeah, you can just Google me. So the next example I wanna show you is Tachotron 2. Tachotron 2 is a neural net which is a little bit more complicated. The key takeaway is you can just like take audio snippets, take the text and Tachotron 2 will solve the rest. So it basically learns on spectrograms. And this is what I did. Once upon a time there was a little mermaid named Siren who lived with her stepmother under the sea. She didn't get to go out of the sea like any other. So this took like nine days of training. It cannot read any English text without any cloud. And this is the equipment used. This is very MacBook and an eGPU. Nine days of training with a very good data set. So this looks impressive. And I think, okay, this is my call for we have to experiment more and also I'm not trying to build like a speech synthesizer. But for example, we are discussing with the client we can apply this or like have like a transfer of this technology for predictive maintenance project where sound is involved to see how the machines work. So actually this experimentation, this was very helpful. But also I have to point out this is now of course my cherry pick impressive example. Let me play you how this sounded like after 10 hours of training. You see like this is like we could have really judged this is leading anywhere. If you want to know more, check the video. So also it's really nice to explain AI makes mistakes as well. And there's sometimes easier to trick than a human. That's like this nice video I was researched. So actually this is about object detection and the researcher on the right. Hey, basically he's not recognized as a person for us. We see a person just like having this crazy picture there. And we see basically the object detection which is basically very good system. YOLO doesn't really see him, which is okay. We have to also rethink how this technology works. Always I call it like you have like expert idiots as an eye system. Yeah, something like that. So watch this on YouTube. Wow, speed up. So I think to sum it up, it's a lot more about people in the technology and the thing because actually we need an open space to discuss new technologies. And everybody from where he comes from, from management, junior, senior, domain experts and we have to get everybody on the same page. So actually it's a big human challenge. It's actually a big, all of these projects are R&D and also like change projects. We have to get everybody like to open up, ask questions, admit I didn't get that. So this is like a really big challenge and this is we need to put more power and energy into this. If we really want to further this technology, we have to basically involve people on any level and in many roles. It's also like another nice saying as well like another experiences. If you do, for example, an AI data science project, you, in a production industry, of course you have all the data. You never have to talk to anyone on the shop floor as we call it, operating machines, but you can get like the best insights from people working with these machines for 20 years because they have like an amazing experience and they can tell you way more than probably the data or make you, helps you to understand the problem, the solution and whether you're actually on the right way and this applies to any time. So this talk, my talks are never finished. So of course I'm going to finish in a bit but I like to learn more. So if you have any experience you want to share with me to add to further iterations of this talk, please talk to me or also my business partner Ingo is over there. He's also, he's the care bearer in the company. So I was like, give us input, share the experience because I think it's very important. We understand each other way better. So thank you very much. Thank you so much. I think we have still time for one question. So any questions? All questions answered. Okay. Somebody has an experience you want to share? Well, we need to work on the safe zones actually. Okay, yeah, but we can also talk later I'm around at the conference, just like free fee to me. Yeah, thank you very much. Thank you.