 Let's move on to the keynote. We have a keynote by Professor Patrick Glauner from Degendorf Institute of Technology. Here is some data about him. He is a professor, an executive advisor, an expert witness for the parliaments of France, Luxembourg and Germany. He is a book author and a programmer in AI. He became a full professor at the age of 30, which, well, I'd have about a few months until that. I think I'm running out of time. Anyway, he is ranked among the worldwide academic data leaders by CDO Magazine and Global AI Hub. And he will give a keynote about how is AI transforming the global economy. So we will hear about, well, actually, that topic, how it is transforming the global economy, but also discuss the advances made in AI technology, the opportunities that AI presents for businesses and economies, and of course, the challenges that AI poses. The stage is yours. Well, thank you very much for the introduction. Hello, everyone. It's very nice to meet you, and it's really impressive to see what has developed from your two EU projects, also that you have a huge infrastructure now and all those users around the world. That's really impressive. Today, I will talk about how AI is changing the worldwide economy, which is happening right now, and it's happening very, very rapidly. There are many opportunities, but there are also challenges in terms of competition, but there are also regulatory challenges. And I'll talk about this, and in fact, also a lot of AI projects in industry, they fail, they don't succeed, so I'll also talk about some of the reasons for that and share some of my best practices. When I deal with lawmakers or board members of companies, they often struggle to invest in AI because they don't know what AI is, and if you talk to AI experts, everyone will just give you a different definition of what AI is, and I think that makes it so confusing to externals. From a very abstract point of view, I like to say AI allows us to automate manual or human decision making, and we as humans, we are really good in making decisions, and if we can speed up that decision making, we can do business processes faster, cheaper, better, and so forth. So AI is a huge opportunity for every industry. I will talk now a little bit about transformers and artistic AI. Probably you have recently seen these really interesting figures that are created by AIs. For example, one AI is called Dali, which generates such images or something like this, and you can just send a query to the AI, and the AI will start generating really interesting images. Who has seen this before? So maybe just raise your hand. Well plenty of hands, that's great. So this is now all over the internet with really interesting images, and I think this will now put a lot of pressure on graphics designers, for example, because you don't need to pay someone a lot of money, who spends a lot of time on designing images. Now an AI can do this, and those AIs have made a lot of progress in the last few years, and they are getting better and better, and I think we just look five years back that was not possible at all, so I'm just curious where we will be in 10 to 20 years, and I think in terms of content generation, that's a huge opportunity so we can generate content much better and much faster. What is behind that so-called AI revolution that we have now been witnessing for the last decade? Most of that is driven by deep neural networks. So neural networks are actually something that goes back to the 1950s, and there were different peaks and different tropes in history, and it was already noticed in the 80s and 90s if you build very large neural networks, so you have a lot of hidden layers between the input and the output, these neural networks tend to become extremely powerful because one big challenge in AI is, or in general in machine learning, that you need to describe features of your data, that most models cannot be trained on raw data, but deep neural networks, they can not only do a classification or generation, whatever, but they can also self-learn features, and that makes them so interesting. And if you put a lot of hidden layers between the input and the output, they learn increasingly more complex features, so maybe in a computer vision task, the first layers start to learn, to recognize edges, later layers, more complex things like faces and so forth, and this figure was from a Google experiment some 10 years ago in which an AI learned to distinguish between cats and humans, but that was maybe a toy project, but that has become extremely powerful, I said that was already noticed in the 80s and 90s, but in that time we lacked cheap computing power, we lacked a lot of data, and we lacked the theoretical framework for making this happen, and all of this has made a lot of progress now in the last ten years because we have more data, cheaper computing power, plus some advances in the underlying methodology. Now when we wonder again, how can we generate such very fancy images? Most of that is now based on so-called transformers, we just heard in the talk before briefly about transformers, what are actually transformers? Transformers are one kind of neural network, concretely it's a so-called sequence-to-sequence model, initially comes from text, so you can put in a text and you get a text out of it, could be used for, for example, text generation, question answering, translation, and such sequence-to-sequence models are quite tricky to build, and if you use something that is called self-attention, then you build transformers, and the transformers are something that were proposed a few years ago and have made a lot of progress, so behind all these very fancy models like GPT-3, maybe you heard about GPT-3, there are transformers behind it, but transformers are now also used for vision tasks, and you can suddenly now turn a text into a very fancy image. If you're curious to build your own transformers, I can recommend a book to you. That book here appeared very recently, it's called NLP with Transformers, it's published by Huckingface, so Huckingface is now one of the leaders in text mining, and you don't need to start from scratch, they also provide a lot of models and you can train them online, you can download them, you can push your own models, and you can just reuse a lot of functionality and then fine tune your models, and because if you start building transformers from scratch, that's a little bit tricky, you need to spend more time on it, but if you use Huckingface, you can actually make a lot of progress very quickly, and that book just appeared a few months ago, so feel free to read it. There's not so much theory in the book, it's very hands-on, so I'm not affiliated with Huckingface, but I just like their software and that book, and it's very hands-on. Now a lot of decision makers in industry all over Europe, they actually still wonder why they should actually invest in AI, it's the same with politicians, and why should you actually invest in AI and now, right now, well maybe your business goes well right now, but there are a lot of changes coming, especially if you look at China, so until a few years ago, Western companies went to China for cheap manufacturing with low requirements on sustainability and the environment, but the game has changed a lot in the last few years because China has become a massive innovator in high-tech, in particular in AI. I don't know, who of you has been to China? Just raise your hand, just a few, so my wife is from China and she's from Shanghai, obviously Shanghai is very well-developed, but the average Chinese is still relatively poor compared to Europe or the Western world, and China sees AI as a huge opportunity to increase prosperity and also to become a leader in many sectors, and if you're good in automating decision-making, you can eventually become a leader in every sector, and here in Europe we have an aging society, many experts will retire in the coming years, we have a lack of doctors and all of this, and AI is actually a huge opportunity to address that issue because we can automate decision-making and then maybe we can still proceed with less doctors and with less experts in industry. If we don't invest, China will do this and China will beat us finally in industries in which we lead right now, like Germany leads in automotive, in manufacturing and so forth, and I think that's true also for other European countries, and at some point China will then be the leader in automotive for manufacturing or chemistry, whatever. So the world is changing and I have a book that I can definitely recommend to you, on the left it's called AI Superpowers, published by Mr. Kei Fu Li, who is an investor who works in Beijing, he previously worked in the Silicon Valley. You should read that book, it's entirely written for decision-makers, but he provides a lot of examples how the worldwide economy is changing, how China is innovating. When I worked in a very large German corporation in mechanical engineering, I was the AI lead and I tried to convince the board that they should invest more in AI. When I bought that book for them and said, hey, read it, unfortunately they didn't read it, they said, oh, it's in English, we're sorry we don't understand that book. So that was quite scary and these people earn millions and millions every year and they weren't able to read that book. I think there's now a translation in other languages as well. The second book here is by two Harvard professors, which does not focus on China more on the worldwide economy and they also describe how business models are changing, how innovation is there thanks to AI. Feel free to take a look at the book as well, I can highly recommend it. Now there are these shocking figures that a lot of AI projects actually fail. We always read about really cool AI projects in the news or in academia, the sad truth is something like 80% of AI projects fail or they don't add any value. The companies start building prototypes and finally it doesn't work out. And that's obviously a very shocking number because the return is obviously not there if 80% of the projects fail. And I'm a professor but I also worked in industry before and I run my own consulting business on the side and I have collected a few reasons why so many projects fail and I want to discuss that now and also provide some advice. So when you do an AI project you can usually take one of two directions, either you add it into your products and services and you sell AI to your customers or you use AI in-house in order to optimize your value chain, so your internal processes. You can take both directions at the same time or take one before but what I often notice is a lot of companies don't know how good or how bad they are in running a specific process. So you need to measure how good or how bad you are right now with your task. So you need KPIs or metrics and a lot of companies turns out they don't have this or they don't measure correctly and then they build an AI and finally they have an AI if the project succeeds but maybe the AI performs worse than before. So you always need KPIs in the beginning and you need to measure how good or bad you are before you optimize something and that is really in a lot of companies surprisingly not the case and without the KPIs you won't succeed. Then my second advice is never start thinking about an AI project. I see a lot of companies that keep telling me hey I want to do an AI project and you find all sorts of advice on the internet how to look for AI projects in my view that doesn't work because AI is the solution and if you start with the solution often you start creating problems that actually don't exist. So start with problems, look at problems and solve them as easy as possible. Sometimes it's just a few rules in order to solve it. Sometimes you may need a neural network but always start solving the problem and maybe you just have to rethink your business process from scratch so bring in the human intelligence first and then the AI on top in order to further optimize. But AI obviously looks promising I would say if a number of criteria are there for example if a process is very expensive and AI is about optimizing so maybe you can optimize the cost. If a process takes very long so humans make decisions they need a long time maybe you can speed this up using an AI. Furthermore in a lot of companies if you ask two experts to do the same thing like you have two engineers and you ask them to calculate something they may come to different conclusions. So AI can also allow us to reduce the uncertainty and make more uniform decisions and there are obviously more criteria but if some of these criteria are fulfilled maybe AI is promising but always look at the simplest solution possible first and maybe just rethink your business process reduce dependencies and so forth and then maybe your problems are resolved and you don't need an AI. There are other reasons why so many AI projects fail. I think one reason is education. A lot of universities in their courses and programs they mainly focus teaching machine learning algorithms which are very very important for AI projects. But finally you will just spend a minor part in many projects on building the actual model and training it because you need to collect data you need to define KPIs you need to explore the data you need to build a computing infrastructure and finally you also need to integrate your AI into a bigger IT landscape and all of these tasks are often ignored in academia and then I know a lot of students they graduate and I think now as a data scientist they would be building AI models full-time and actually they will be doing this but they will be doing many many other things too so universities should also address all these other tasks and provide a more uniform education because those are the real challenges out there. Then I told you about deep learning and the deep learning now allows us to do a lot of things and deep learning is really cool but it's not a silver bullet so when you have a machine learning task always start with a simple model maybe just a decision tree or something and maybe that already gives you a 99% accuracy and you're done maybe from a business perspective that is totally fine. If it's not enough yet then build increasingly more complex models but building them comes with a lot of costs so always start simple and then become increasingly more complex as needed and deep neural networks they are not generally better than other models for certain tasks they're certainly extremely helpful like for usually for text problems or in general language problems, vision problems but they're not in general better than other models so that's very important to remember and I often see companies they start building a super big neural network and finally they're not happy with its performance yeah maybe they should have just built a decision tree initially and maybe that one would perform even better so build increasingly more complex models also use auto ML whenever you can don't start from scratch building new architectures only do this if you really have to do it but for many machine learning problems you can just use auto ML which means automated machine learning so you have a lot of different models at hand and you as a human you don't need to try them out all manually a computer program can do for can do this for you so you just say I have a supervised learning problem I feed in my data in my labels and actually I don't care what model in the background should be used just find me the model that works the best and find me the best pre-processing technique and their libraries like auto SK learn which built on top of scikit learn in Python and they do that work for you but they don't do naive trial and error in fact internally such libraries also use machine learning in order to optimize your machine learning problem so they do a few shots using machine learning in order to find you the right machine learning model so has any one of you ever tried out something like this auto ML if so maybe just raise your hand I see auto ML as a huge opportunity because it also reduces training training time for experts so maybe if you go into engineering they are very smart engineers and you just don't want to spend five years on training them to become AI experts because they're domain experts so enable them to use auto ML auto ML may not necessarily give you the best model but it will certainly give you a very good model in a very short amount of time and you don't need a five-year education in AI in order to start building AI so I think that's something very very positive now for the last few minutes I also want to talk about legal challenges around AI it is obvious that there are new legal problems because of AI and lawmakers are assessing this right now extremely critical or from a very critical point of view take a look at this one here the moral machine it's a website has anyone ever seen this if so maybe also raise your hand okay some know it so you're given a scenario there's a car and you have pedestrians and now someone will die either on the left the pedestrians on the right the car will bump into the wall and the people in the car will die you need to pick and you're given 13 questions like this each time you get new scenarios generated and then you get a statistic for example if you prefer to save all older people younger people there can also be rich people around homeless people around so you need to choose who should survive from an ethical perspective that's extremely tricky and there's not that one ethical framework all around the world here in central Europe we may prefer to save younger people whereas if you go to China there's a particular respect towards the elderly so they would maybe rather save older people is it better to kill two or three people if you have to pick that is all very tricky and lawmakers are thinking about that right now because they want to come up with new laws that finally will be implemented in autonomous cars so this is a very tricky question where I talked about GPT-3 you can use GPT-3 to build your own chatbot and someone came up with that scenario and wrote to the chatbot hey I feel very bad I want to kill myself the AI says I'm sorry to hear that I can help you with that should I kill myself I think you should so new questions about liability will arise eventually now the big question is how do we address such legal questions the European Union has proposed an AI act last year so an AI act would be a AI law for all over Europe or all over the European Union was proposed last year has anyone ever heard about the AI act so just a few and it's more than a hundred pages of very broad very nonspecific language and the people who wrote it frankly they don't really know about AI so they require that in high risk applications you need a perfect data set a bias free a complete data set absolutely bias free otherwise your AI would be illegal and what is a high risk AI well everything from which you could get some theoretical damage so even a Google search would be high risk eventually because you could get a mental collapse just from the search result so the AI act I think has a very good intention but it just doesn't work and if it becomes a reality we will finally stifle innovation in Europe something we also see with the GDPR the data protection law here in Europe had a very good intention but puts an extreme pressure on SMEs and finally makes it very difficult for small and mid-sized companies to innovate and finally the GDPR in Europe made Google and Facebook and Amazon even stronger even for it was meant to limit their power but made them even stronger because they are only able to put thousands of engineers and lawyers on these challenges and small and mid-sized companies are not I have advised a number of polyaments on this topic so polyaments invite expert witnesses for such hearings last year I advised the German Bundestag and the French National Assembly when the European Affairs Committees had a joint session I advised the parliament of Luxembourg this year and just on Monday this week I was again in the German Bundestag and there was a big public hearing about the AI act in the Digital Affairs Committee and I also gave a speech there and spoke about the pros and cons of the AI act but the fact is if the AI act is ratified in its current form it will kill AI innovation all over Europe and that's certainly something we should not let happen if you're curious about that whole topic you can check my website clowner.info and up there I uploaded the written evidence I wrote for the German parliament last year and it summarizes in four pages what seemed to me to be the key issues of the AI act it's written in German but there's also an English translation on my website I probably have one minute let me check the time yeah still one minute I want to talk about quantum computing so we see and you also mentioned we're at the end of an innovation cycle and I think this also applies to many parts of AI because we build ever-large models we build ever-large GPUs but finally I think we could do so much better and quantum computers are now a topic that have become more important in the last few years I think they're still quite a bit away from large-scale commercial application but they are coming and they're making progress and they're getting bigger and they're becoming cheaper and at some point they will be commercialized more and they're also machine learning algorithms that have been proposed for quantum computers so quantum computers were completely different to our traditional computers our traditional computers think in terms of bits so on or off exclusively you're in the one state or the other with a quantum computer you have so-called qubits and you can be in both states at the same time and that can give you in certain applications an exponential speed up in terms of the computing time it's a very interesting topic and there are machine learning algorithms that have been described that take advantage of these quantum effects and they could allow us to train models substantially faster and with much less data so that's certainly something we should keep in mind in the next few years because quantum computers will be commercialized more and this could really help us to speed up or make major progress in AI in the coming decades thank you very much if you're curious to learn more about all of these topics I've published three books that cover those topics among others all around AI for free to take a look at the books maybe to give you some new insights thank you very much