 Thank you very much indeed, thank you very much. It's very nice to be here, there's a whole lot of police over here for just a simple group meeting, I don't know why that is. It's flattering, it's flattering, I have so much, you know, it's a secure feeling. So before I begin, I have a little disclaimer which is nowadays important in our area, in the era of data. You know, I cannot emphasize enough, artificial intelligence depends on data, the availability of good data, of clean data, reliable data, of non-fake data, I guess, is an incredibly important thing. And so if we are to build solutions, I noticed that most of the AI solutions that we're currently looking at, at least many of them, are let's say consumer oriented. So we see a lot of personal data of individuals including all of their preferences and we see code as an industry in terms of being able to digitalize, for example, and to be very predictable in terms of consumer preferences. And having said that, we all feel that there's a very thin line between being extremely intimate towards people and being predictive and being seamless about feeling, about what they're looking for and what they're moving towards versus being downward-creepy. And downward-creepy, and as you may know in the US, there's even a creepy line, so you can call that line if you feel your personal data has sort of been dated. I guess there have been a few calls in the past few weeks to do that creepy line probably. So we need to be very much aware of this and I just want to use this disclaimer. I'm not here today this morning to discuss all the ethical aspects of it, although we might imagine that this afternoon, during the over-platform treat all those sessions, we might be diving a little bit more into that as well, right? But for now, I'm just moving a little bit towards this idea of what a reference architecture might look like. And particularly the journey towards a reference architecture I think is interesting. So let's not forget this whole ethical thing, the fine balance between being very predictive and almost psychic about what the world or a person or an individual needs versus being downward-creepy is, I think, a very fine balance here, that we all need to navigate. And hopefully architecture can help us a little bit to achieve just that. The funny thing ever is that currently, I see people every now and then they tend to take themselves way too serious. Just, I can still remember, it must be already 10 years ago, one of my esteemed colleagues at that time who was a well-known enterprise architect took himself so serious that he believed that we needed an auto-hypocrates for enterprise architects. Because enterprise architects, you know, they were in full possession of immense powers to change the world. And the industries and companies, enterprise architects, were clearly even more important than doctors designing about life and death, right? So he felt that a real enterprise architect was swear and oaf about the proper use of enterprise architectures, while I'm glad to say that they really work out because I think, of course, enterprise architecture is important. Let's not exaggerate about it, right? And the same is very gaping with the eye. And again, it's an ethical discussion because maybe some people are right to say we need to have an awful hypothesis around AI as well, because artificial intelligence could change the world potentially. So anybody involved in artificial intelligence maybe should swear and oaf as well. I don't know what oafs mean nowadays anyway in this world, and, you know, what it means to be ethical is also not too clear anymore, just press the mills. But anyway, just as a disclaimer, these are the things that we need to be aware of as well. But my real topic today, by the way, John, not a timer running, John, there's not a timer running, so I'm not sure where I am in terms of time, but anyway. I'll take two hours then probably. So there's no topic more popular right now, it seems, in the boardroom than artificial intelligence. Many of you are probably IT people, but at least a few of you look like IT people and some of you probably are not on the business side, but if there's one item which is popular in the boardroom right now, that distance people, it's AI. So obviously there's a lot of reasons for it and a lot of non-reasons probably as well, because it's industry fries a little bit on the hypochrome creator cells. But I think there's very good reasons to be extremely enthusiastic about AI and also the business side. Many of the big AI engagements I've seen in the past year, or let's say a year and a half, I would say most of them were driven by business people and often there were even very few IT people involved in the strategy sessions that were set up around artificial intelligence. So if we understand that enthusiasm and let's say the transformative potential of AI, sooner or later we start looking at architecture because correct me if I'm wrong, I've always believed that architecture, particularly enterprise architecture, that there's only one reason in life not to be certified on it by the way or swear an oath about it, but to actually enable change. I think the only role of architecture in the end is to be a foundation for change, to enable change. It's not a meaning of life, it's a tool to enable change. And if we understand that AI, if we appreciate that AI will change business against the price, we realize that architecture will be a crucial tool to achieve that. Now we all know that it's very challenging to dive right into, okay, so what deep learning framework do I need then? Right, and what cognitive API tool set should I adopt? And okay, how should I manage my data so that it's a bit secure, predictable? So there's a lot of interest around to the actual tooling around this, but I believe in this proper enterprise architecture we should build it up so, so I'll take you through a little journey that we've been through so far in terms of understanding actually what it means to work towards a reference architecture. So that's really what I intend to do with my little speech, which I'm not sure about timing, but let's see what time did I start. Okay, well, I'll sort of try to guess, where, oh, you're working on this, oh yeah, technology, yeah, and you like books and stuff, yeah? Yeah. I don't know, no, it's not, it's a very nice slide, but it's my size. Yeah. Nevermind. So, before of course we dive into AI and what it does to enterprise is we need to of course understand what it is. Now, if there's one thing that we can really spend our life on, or even, you know, bombard it with minimal fire, it's discussing what AI is and what it means. And so I decided to certainly not pass on that opportunity to add my own definition as well. So here would be my definition of what it is. I think also the definition of AI is it's human intelligence while artificial, and I think it doesn't do justice to the actual potential of AI because I think it goes way beyond that. I think artificial intelligence is not only mimicking human intelligence as we know it, but also other forms of intelligence that we as humans still would perceive as intelligence, but might go way beyond what we've seen ourselves as humans so far. So I like that idea. It's our perception of humans of something that is intelligent, it's a perception, right? And then of course if you're talking about perceptions, I found until now that the most effective way to explain AI is by examples. So rather than to really get nitty gritty about the details of the definition, I'm sure by the way that this esteemed community would be pretty good at that. Discussing these different definitions, right? Until the end of the time. But it's probably more interesting at this point to simply to discuss, you know, just mention a few examples as you can see them all here as well. And we all realize we already saw some examples this morning. The whole ability in terms of natural language, understanding and processing in generation is a clear example. Image recognition, autonomous systems, audio, video, indexing, but also I would say very complex algorithms and predictive algorithms with a level of accuracy as we've not seen before. I would, all of these, I would, you know, sort of see as intelligence perceived by humans. And even at least within our community there's a discussion of pretty straightforward rule-based stuff like robotic process automation, I'm sure most of you are aware, RPA. It is by many of our clients at least perceived as intelligence. And hence they consider it part of the whole AI portfolio as well. And who are we to, you know, who are we to discuss that or to talk about that? If we feel as humans that that is a level of intelligence as well, look at the way automatic things with SAP screens, like humans used to do before. Now the system does it magically on itself. If we perceive that as intelligence-fines, it's just a set of rule-based scripts usually, but if that's the case, we would perceive that as probably as AI as well. So that's really what I would like to begin with. Yeah, we're not powered at all, yeah, that's definitely. No, it's still not in our lab at the moment. So I don't want to go too technical, as I said, because as many different technologies, and actually as a matter of fact, some of the natural language technologies, natural language solutions that have been discussed earlier, I would say that they might not necessarily be based on deep neural networks or deep learning, if you like, on neural networks. But on the other hand, it's fair to say that the increased enthusiasm in the past two, three years to a large extent, except maybe the natural language area, are due to deep learning or deep neural networks. And this is a very simple technology activity, just talking to another person in this thing. Many of the veterans in IT will tell you, hey, I did this in the 80s, and in the 90s, and this is all true. Some people wouldn't even claim, hey, I did this in the 50s, but that's... The thing is, many of these technologies have been very well known to us for quite a few years. But the breakthroughs we're currently seeing, both in our ability to collect and store data in amounts and structures and unstructured that we've never seen before, absolutely not able to do that in the 80s or the 90s. And also, of course, the highly-optimized processor architectures like GPU-style, FPGA-plus type of processor architectures that are particularly suitable for neural networks all have brought a breakthrough in terms of, for example, autonomous driving cars and the ability to really understand speech with an accuracy that we haven't seen only just a few years before, right? So there's a breakthrough over there. Partly ventured to big data, not a part of the illustrious, open-platform, free-to-load digital in-neighbors is, of course, big data. Big data brought us the capability to store huge amounts of data, structure to a non-structure, make it available very quickly. I think that's an important thing that drove the current popularity of AI. Combined, indeed, with optimized processor architectures, massive, massive parallel, but particularly geared towards, you know, building processes. Like your iPhone X, for example, some of you know about the iPhone X. It has a complete neural network on board, right, that recognizes your face and it's just on the chip itself and it's there and that's the reason why it's so fast right now. Things we absolutely couldn't achieve in the 80s or in the 90s. Neural network is not sure some of you probably have heard of it and probably many haven't. It's the dumbest thing ever in neural network. No, really, it's the dumbest thing ever. It has nothing to do with data science as we typically know it, so there are no algorithms, there are no statistics, there's no math, there's nothing. It's just a very silly dumb machine that measures input with output. It has potentially thousands, 10 thousands even if necessarily 100 thousands of tiny little buttons notes that they're all connected and it just tries to, for any real input record, try to adjust itself in the dumbest way ever to match the output that you get and it sort of gets better and better. It doesn't know what it's doing. It's just, you know, pushing all the little buttons and turning it a little bit to further adjust itself to ensure that you have a proper match between input and output. That's all it is, it is as dumb as you can possibly imagine. It's too dumb to even, you know, to explain. The thing is, it works. And if you don't get what a normal network is, think about us as children. We all learned how to catch the ball. Right? We all learned how to catch the ball. We don't remember because we were too young probably because it was, right? But we all learned how to catch the ball and the way we did it, we didn't have a clue. So they throw a ball at us and we don't know what to do with the thing, right? So, you know, we failed, all of us failed because we have no clue how to catch the ball. And then they throw another ball and have it again and again and it's frustrating. Somehow you get better, you adopt yourself. As sooner or later you have the thing. And if you would ask us, little children, how did you do that? I don't know. I just adjusted myself. I, you know, all these notes over there are sort of changed a little bit so that I could catch the ball. But if you later on would ask, how did you do that? You're like, I don't know. You know, I just adapt to it. And that is the fascinating but also the scaring thing about the deep neural networks. They are completely opaque. There are black boxes. They just adjust themselves by almost little tiny buttons. And we now have tiny little buttons that they actually work. But on what basis it exactly matches input to output in the end? Based maybe on hundreds of thousands of records, input records and output. We don't know, right? So it's opaque. It's like asking 70 feet with a function of feet. How it works. You know, the moment you ask the 70 feet, it's like, you know, it's free, it's a special fight. It doesn't know how it works. And this is the same with the deep neural network. That's all really I want to say about it. But I think it's crucial to understand that this is one of the great fruits that we've seen. So if we're looking currently at some technology options for artificial intelligence, although I would say that a lot of it is ramped among it, right? So it could be a robotic cross-search automation. It could be the typical language type of facilities that are not necessarily all based on neural networks. Regional IBM Watson was also not based, you know, the Japanese thing was not necessarily based on neural networks. So language is another aspect of it. Much of the advanced algorithms and advanced analytics, as we know it, still feature very fancy data sites. As we know it, they're still part of the AI wave as well. But I think it's fair to say that particularly the neural networks are crucial in this. So we understand a little bit what it is and how we see some great fruits currently happening. Now let's move a little bit further into our architectural journey. Because in the end, if we create an architecture, we first of all appreciate that AI in this case, where we want to design architectural foundations for, actually has the change potential, right? Actually has the enabling qualities for an organization to transform. So you could say if we create something and then we create an architecture, we want to ensure that any project we will unleash on top of that architecture actually will enable us to elevate our corporate IQ a little bit. Steve just mentioned TechnoVision over here. Thank you very much by the way for plugging that into that square. But still, all go there, please. Captain, I'd love to call and slash TechnoVision, but it's each our yearly trend series. And one of the big, let's say design principles, as architecture, we'd appreciate design principles. One of the design principles we put over there is the notion of IQ up. Whatever project you put in your portfolio, whatever initiative you would launch, it would sort of add to the corporate IQ. It would add something to your profit and loss balance of corporate IQ. And we all realize it in the era of artificial intelligence. The more we can raise our intelligence artificially or not as a corporation, the better it is. So we need to define an architecture and a change approach that actually leverages and, you know, grows that corporate IQ if you like. So the first thing to look at that, and this is another design principle we've used in TechnoVision for quite a few years is to look at companies that already successfully done so. One company, I would say, that does that successfully is bigger in distance than Amazon probably. I'm sure that some of you might already be almost a cliche, but rest assured that this company is not only, I think, one of the most innovative companies in the world. They truly are. They have an innovation strategy and innovation culture and a process which is second to none. It's absolutely impressive the way they do it. For example, Werner Krogels, the CTO of Amazon Web Services once told me that the only way we can stop an innovation proposal by anybody is to prove that it won't work. So this is fairly different, right? Usually if you want to launch the innovation project, we have to move heaven and earth in order to get it done, right? To convince people, here what you need to do is to have somebody convince you that it won't work. You know, just as an idea. Not so long ago, a month and a half ago, I was over here in London at the MRS conference, which is one of the biggest marketing research conferences in the world, and that they did some research themselves instinctively enough to see what's the most trustworthy brand. And it also turned out to be Amazon right now. Probably not Facebook at this point. Most trustworthy brand in the world, for some reason, right? Yes, you know, it's like the image, some slight mis-problems, probably. But Amazon is also a random owner. It's actually the most trusted company in the world. I also would say, if you want to understand what an AI first company is, I think they're doing pretty well themselves as well. And I believe as an architect, we always, first of all, need to ask the question, you know, what type of enterprise do we want to enable with our architecture? Because we don't know that. No, we just want to do fancy things with AI and the neural networks. That's not interesting, but we do want to understand what does a company look like that actually uses AI and everything that it's doing? And look at Amazon. Everybody knows, of course, that they started as a online bookstore, right? And grab their famous recommendations. Their recommendations got better and better. Nowadays, no longer based on, let's say, statistics, like, you know, adaptive filtering. But instead, nowadays, it's based on neural networks that are simply, literally like that, black box in terms of, there's so much preference over here, there's so much next choice over here. You know, we're trying to match that. We don't even exactly know what we're doing. But they're so good nowadays with recommendations that they are at the level of the so-called cycling pizza, right? They can deliver cycling pizzas. You don't know what those are like. So, cycling pizza, you know, imagine, it's Friday, Friday evening, 7 p.m. You start to get hungry, suddenly realize, start to get hungry, doorbell rings. Pizza delivery. They just knew, they knew. They knew you wanted to pizza before you knew it. But they're so psychic, they're so aware of your preferences and how that will evolve over time that they can deliver the cycling pizza. The pizza knows it. You want it before you know it yourself, right? Same reason that they've already been contemplating for Amazon Prime members, certain selected members, to deliver an Amazon box every two weeks. It's a box and you open it and you're like, oh, that's exactly what I wanted. I didn't know. I didn't know. But now I didn't even know myself, I'm one of them. But this is so small long, right? And they're actually working on this stuff, believe me. Better believe it or not, is there a fine balance between being very intimate and downright creepy? Yeah, sure. Sure, but hey, they can do this, this is what they do with AI. Obviously, everybody knows Amazon's Alexa. I'm quite sure many of you have it, or it's Kova, or whatever. The thing is, there is deep learning underneath here, and you're right for it to recognize your voice. There's also, by the way, language processing, but it's certainly also speech recognition, which is very much driven, obviously, by AI. Then we have their unmanned warehouses, okay? There are 43,000 different automated robots driving around over there in these warehouses. They're virtually unmanned in these warehouses. There's a lot of AI behind these things. There are, of course, they're the three drones that are still working on. Very much depend on AI in order to not create an accident and actually get where they should be, despite all sorts of unpredictable events that might occur during the flight. And by the way, you don't realize what's one of the very first things they really start to deliver with these drones is, of course, pizza, because that's one of the favorite topics for delivery drones, is pizza, so that's funny. So, psycho pizza delivered through the drone. This sort of AI first, I guess, by them. And then, of course, I was in Seattle myself just last week, Amazon Go store is, of course, that sort of pinnacle of what you can achieve with AI. I'm sure all of you know it by now, Amazon Go. It's a checkout-less store, right? You just walk in with your Amazon ID and it automatically starts to follow you with tons and tons of cameras and other centered technology. They know exactly what you're picking up from the shelves. Also, if you're very sneaky, try to pick it up. You know, do like this one. You know, you just try to do it like that or you pretend that you're putting it back and so on, it doesn't work. So, it's completely aware. Again, intimate versus creepy, you get the point. A lot of people love it anyway because you just get into the store and you walk out with the stuff and there's nothing, there's no check-outs, there is no gates to go through or whatever. It just follows you and knows what you've taken with you. And trust me, you have the invoice before you actually left the store and they already invoice you because you're pretty good at that as well. So, if you are in Seattle, it's next to the Amazon Sphere which is their little indoor jungle that they created to work in, talking about work-life balance. They have sort of thousands of different plans from across the world, like a little indoor zoo and you're supposed to work there. Any Amazon employee can work there as well. So, yeah. Sort of an inspiring company at this point, I guess. You never know, but it looks like today or tomorrow. But for now, I'd say they're a very nice example. And then, of course, they have developed all of that fancy AI technology stuff. And they actually realized they're a retailer. Hey, why don't we sell it? So, the funny, the most funny thing which I always consider a real landmark of company that they're reinventing themselves, they're like, hey, we develop all of these tech technology for our purposes, the purposes you see over here. And many other examples, by the way. But why not sell that technology itself as well? Do anybody else who wants it? Even other retailers. Trust me, there's a lot of retailers that currently say we wouldn't use any cloud except, of course, for AWS. Because retailers, you know, they're a competition. But trust me, there's a lot of retailers that won't admit it, but are actually using the cloud as well. The Amazon cloud for their purposes. Simply because it's very cost effective and the highest quality, right? And if you could possibly imagine. So that's, I think that's the landmark of a retailer, an open AI first company that actually sells all the technologies that develop themselves as well to whoever wants to have it. Just as an example of what does an AI first company look like? There's even an AI first country, by the way. Did you know? I'll think about who might be it. So it's certainly not the US. I can tell you, it's also not the UK. Now it's not even commonwealth. I believe, oh, yeah, no, that wouldn't be commonwealth. Anyway, I'll leave that a little bit on even AI first countries. Because what these AI first companies do is also very interesting example to us as an architect. What it would mean to actually enable that. What you need to set up in order to achieve that. Nice thing about Amazon is also that they, you know, introduce quite some time back the notion of the mechanical term. You may have heard of it already more than 10 years old. At that time, we have already had this idea of if we could launch a web service that would trigger some artificial intelligence, that would be nice, right? And the only thing was that behind that web service often we still need a human intelligence to perform that intelligence task. Mechanical term is mentioned after the famous chess computer in the 18th century, developed by Hungarian scientists. It was a, you know, it had a doll, it was a marionette, it had a turban, so they called it a turb, what did they know? It had a turban, so hey, it's a turb probably. And so they called it the mechanical turb, it was a chess computer, and it beats all the grandmasters and also the royalties and anybody else who wanted to play that machine, the machine beat it. But actually inside there was a human, tiny little chess player, grandmaster, that had an ingenious set of mirrors, so you could see what's happening on the board. And I think it lasted something like 17 years before somebody actually found out how they did it. So this was a big thing. It's, you know, artificial intelligence is quite old as you can see. So this is the end of the 18th century, mechanical turb, and Amazon realized well, you know, we would have all web serves interface for any human intelligence task, and then you launch it. As a web service crew in the API, as we all know it, but behind it are still humans. It's a crowd-sourced community that very quickly picks up a human intelligence task, you know, subscribe to interesting tasks that they want to do and then carry it out. And the nice thing is if you nowadays do, they're still active, but if you look at the type of tasks that were typically done through the Amazon Mechanical Turk system, you start to realize that there are different variants of intelligence, and that helps us also to put together an architecture. There is, I would say, an level of intelligence that would be taking the role out of the human. So this is simply to automate things that we all realize that humans do themselves as well in a very automatic, repeatable, predictable way. So you could, for example, say, well, go to the third line of any website, you know, you copy and paste the first seven words, put it in a database, and make it available. A lot of things that knowledge workers nowadays do themselves as well behind screens, right? And it's just that 10 years ago, it's still something that was worth to dispatch to humans that would do it behind a web server, so then, often in seconds, you would have the result, right? And it's being done for you, or maybe it's done a few thousand times by all sorts of different crowd-sourced, you know, members of this particular community. So that's one level, taking the role out of the human. I would say the second is more interesting, and often we nowadays don't consider that being the highlight of what we can achieve with AI. This is about cognitive capabilities. And cognitive capabilities are human-like capabilities in terms of communication and reasoning and, you know, making others understand and solve an issue. Human capabilities, if you like. And we see a lot of these things on Amazon as well. Why don't you write a little summary about the political essay? Or why don't you describe the few-line scenario? And, you know, or why don't you sort out the purple unicorns on these pictures? You know, select the ones that feature purple unicorns. These are human capabilities, cognitive capabilities, if you like, that at that time, 10 years ago, systems were usually not able to do at all. See all of the text analysis, for example, that we saw this morning, that that's pretty advanced stuff, not able to do it 10 years ago. So it was a human intelligence task to combine that service. And I would say that this is often the area we're typically talking about nowadays. To augment ourselves with cognitive capabilities that we recognize sort of. We understand how to find a purple unicorn in a picture or how to write a summary. Some of us might be better than others. We get the point, we know our stuff. So we augment ourselves, if you like, with these cognitive capabilities. It's not an area that we need to deal with from an architectural perspective. And then the third area I would say is really, it goes beyond what we're able to do as humans ourselves. Everybody knows, I think, the famous AlphaGo story, of course, code in the Asian board game that was considered the final frontier. Computers couldn't beat human players at Go because it was a poetic, metaphysical, you know, spiritual game, if you like. But then deep learning came and this was a very simple approach. They look there, there's white and black moves. Hey, black is fun. Nothing else, no rules, nothing, no algorithms, nothing. Just they look, white moves, black moves. Oh, now, white is fun. And you feel it in a few thousand games. It starts to get better because through deep learning, it sort of tries to catch the ball, right? It gets better and better. And then you do something very smart, this is called reinforcement learning. You have to, for various instances of this system play against itself. Not a few thousand times like you were able to capture the board games as they were, but a few million times. It plays brute force against itself and it gets better and better and better and then it beats the world champion in sound and nowadays everybody realizes it's, you know, it's cannot be beaten anymore. And by the way, Google that brought this system said it's no fun anymore. To say that wasn't at that time with Jeopardy, you know, that there's no challenge anymore. So we'll do something else, something interesting, maybe in a medical aspect. And it's the same with this AlphaGo team that said, well, maybe we should use Google's deep mind for something more useful than beating somebody at all, but we wanted to make a point, right? And now it's no fun anymore. So, the interview, at least at all, was the world champion beaten by this system, said, okay, I can no longer beat the system because it's obviously a barrier right now, but it made a few moves that I considered out of this world. They were exotic alien moves that a human never would have considered to make, but they were valid. And he said, I heard as a goal player from it, maybe as a human goal player, you know, much more diverse as well. I actually learned from the system, although I never understood why it wouldn't make such a move, which I think is interesting. Same reason, by the way, that Miss Pac-Man, an optimal score in Miss Pac-Man, could never be achieved. This is not Pac-Man, this is Miss Pac-Man's female version, much more complex. So, no, it's true. Miss Pac-Man is much more complex than Pac-Man. No, really. And the optimal score, it would never be achieved, of course, and then, again, with similar technology that I just mentioned, the young Microsoft actually created a set of AI solutions that actually created the optimal score here. But I want to say, as sometimes we see solutions that go way beyond what we as humans possibly imagine and think of. So, one of the things that we need to realize from an architectural perspective is what type of impact do we want to achieve with AI? Is it automatic, so taking the robot out of the human? Is it a cognitive addition? Or are we actually thinking about fully re-imagining solutions and systems as well? We did find, by the way, that another way of looking at this in terms of shaping a portfolio based on architecture is to look at the benefits that you achieve versus the complexity. Very simple thing, but when we did our research last year in terms of where our companies taxi currently, focusing their efforts on in terms of AI solutions, we found that there's a lot of people on the upper right-hand corner. A lot of companies are in the upper right-hand corner with their current AI solutions, which means that they are looking, yes, for the high-benefit flexibility, but they're also looking for trouble. So, we're looking for the more complex projects that might take several years and are difficult to achieve on the short term and achieve tangible benefits on the short term. So, our recommendation currently often is when you start to shape your portfolio on top of your vision of an AI-first company, you may as well focus on a few areas that are high-benefits and low-achievements. Easy to do, not that difficult to achieve, but have a high benefit. And, by the way, you can download that report if you have a second download where it's this evening. You know, when you go to the party or whatever, you may as well also download it's all available. It features hundreds of different use cases. And what I found with AI, the thing that we need to do is architects more than anything else, is simply lead by example and show as many examples of the game of AI also as an inspiration for imagine that we would be the Amazon of Delta or the Amazon of healthcare, you know, or the Amazon of insurance. And then you start to sort of lead by example and I think us as architects need to do that as well. If you look at architectures, I think we need to understand, again, given what we are looking for, that there are different levels of, let's say, complexity and how deep you want to be in the infrastructure. That's the infrastructure level itself. We're really talking about the GPU and parallel processing architectures. We're talking GIT level and virtual image system and the systems that actually run it for us. On top of that, you see, particularly in the AI, I would say the deep learning networks and again, there was more than that of the deep learning neural networks. You see a lot of particularly open source frameworks. At two levels, by the way, there's the deep level, Linux style almost in an open source, paramental style of neural networks and then you have more friendly, high productivity frameworks on top of that. So for example, Blue Room and Keras are more friendly ways of creating neural networks, mobileing them and training them rather than than the paramental stuff which is underneath, right? So that's a decision that needs to be made. I would predict that many more of us will be architecting for the use of APIs on top of that. So imagine I need image recognition, I need voice recognition, I need natural language generation, I need natural language understanding, I need conversational technologies. I'll use APIs, it's a web service. Almost like the Amazon artificial artificial that I just showed to you. And we don't need to understand that there is somebody trying to catch a ball underneath there. That there is a deep learning neural network underneath or just supply it with data, train it, get it better and apply it through an API and use it, hey, it recognizes image, right? So I think many of us will actually be more interested in that level in the end. And then on top of that you will expect complete solutions that will really contain predefined models or you just need to link it up with your own training data and you get let's say a predefined solution that has AI in it and it would be able for example in the case of Raven to go through a contract that has been created 20 years ago and it will able to tell you what obligations you have according to that contract. And you just need it with the contracts and that's it, find the obligations, even get ultimately put it in the obligation management system or watching the health for example which is a set of solutions that use AI to facilitate us in doing all sorts of things in healthcare and we don't care anymore there's even several levels underneath what type of AI technologies actually being used over there. Having said that, for anybody with sort of a background in software engineering, obviously it's good for the sports just that it would exposure to the actual neural networks and actually having created one yourself and populated and trained and usually it's sort of, I would say exciting but that's probably me, it's a bad answer. Because you don't really need it. It helps of course as architects that we understand what is underneath over there. So, I really don't want to dive into what it actually that in the end looks like because you can see the same area as I just mentioned you see what about in more details logical architectures if you like as well. You come through the conceptual part looking a little bit over here what type of services you actually need to go from infrastructure all the way out to solutions and the API level and obviously as I began my presentation I managed the data very carefully so there's a whole series of stuff around it that we put together as well in order to feed the system with the right data and ensure that it actually learns from the proper materials but also that's the way that we deploy it of course which is being used in the proper way. It's quite a crucial one. We'll see a lot of AI pop up in the forthcoming years which means that we have a pre-test, we have a model we design it, we train it, we test it it's all right and then we can download it for example even the smallest central system you still would have on the chip the possibility to run and actually train a model somewhere over there so there will be some very interesting work to be done in the IoT space as well in terms of AI reference architectures. And then of course here you will instantiate it if the actual tooling, I did this one for Amazon we have one for Microsoft we have one for passive read open source tools you can imagine several others you could do it for the IBM ecosystem as well and then you simply pick up all the various layers sorry about it, it's not an argument but this is actually the design language that Amazon are using so they create it if you happen to know a complete library of icons that they use to create architectural diagrams Microsoft has a similar one by the way and you see all the actual products and you instantiate it with the actual products and the way they work together which I think at this stage it's not the main message I wanted to bring across I'm sure at BLEV from Frigo Low we'll have a lot of interesting work to do in the forthcoming years to work on all these different levels but as architects particularly as enterprise architects it's for us absolutely crucial to understand what transformative impact is we want to enable with our enterprise architecture and I felt I needed to emphasize that show you a little bit of the journey that we've been coming through in the past few months and even years to approach a reference like this so my time is over yeah so the country that already has declared itself AI first is United Arab Atlas not so strange, Dubai of course is the host in the World Expo 2020 which is very nearby now and rest assured I happen to know they are quite ambitious in making you know a good name for themselves around them so United Arab Atlas has declared itself and it's funny by the way city Dubai AI got it to work so that's sort of a chemical instance I guess but they declared themselves an AI first country and I think as architects it's actually very inspiring to look at what they've done because they created all the use cases on the left so they particularly focus on use cases linked by an example don't make the difficult diagrams you know just tell what you want to enable and then we'll create your architecture to enable it and they also educated their own executives which is I think a hilarious sort of they have a minister of AI but they also educated all their other VPs in the government to actually be very much aware of this created a whole training program for them as well and actually literally it's a design principle for them whatever solution, whatever budget proposal we see within the government we'll actually ask ourselves the question is this an AI first proposal and have we actually thought about the impact of AI and whatever we want to do as a government, as a country which I think is quite an interesting have a look at it, you'll find much more about it because you know, unless they like to sort of communicate and show to the world what they're doing over there they're pretty much proud of it I guess that's a good piece of work as well so thank you very much for bearing with me so early in the morning I hope you agree with me it's a very exciting time to come and it's not only about the reference architecture I hope you'll appreciate it also about the journey of understanding what actually should be in that architecture that I try to share with you this morning thank you very much for listening and almost within time almost within the day of the dialogue yeah, technology yeah, that's good so I have to say I've got to start with one of my own Raven being able to give you contractual obligations yeah just to give you a contract to reduce the number of lawyers in your practice that is such an interesting... that would improve the world, right? I am not going to comment on that then he would... definitely, we all realise that that would be the case and it just... there's so much work involved nowadays in going to contracts that might be created 20 years in the early morning and there's no structure to it and very difficult also outsourcing companies can take off from contracts and then you realise we really need help to understand what our obligations are and the systems can do it for us powerful capability so we are a little behind the scene now, but we'll... sorry about that, no, not at all by the way given AI advances how long before machine generated architecture? yeah that's actually a topic I... I propose for a little architect summit that we will have that our little Chateau in France will take some time with our yearly architect summits and one of them will do... Alex, I'll do my architecture it's the name of it, so... I think that's a very interesting idea I like the idea of extremely detective architectures or adapters, so it's almost like a neural network to the circumstances, right? so it has still architecture by the way it's questionable, but I find it's a... stimulating idea I first would say let's look at our architecture work and see how we can implement it particularly with cognitive capabilities and that's what it looks like to re-imagine architecture which means we don't only need... no need for lawyers anymore maybe also no need for... architects anymore perishable from... yeah perishable, okay, sorry about that lawyers in the room, yeah, architects, you know so back to the oath oath, yes you believe that EA is all about enabling change isn't that a foundation principle a foundation for an oath? yeah, maybe, but then I think so many of us are in a business opportunity to change some on a daily basis and certainly around enterprise architecture I felt it's a bit... you know, we take ourselves maybe just starting a bit more seriously, sir you know, on the other hand if you look at AI for example in healthcare or as we see how AI nowadays can generate meals and can actually, you know literally influence society as a whole and you start to realize that there's definitely an ethical aspect there it's just very clear to the states and I feel like why don't you create a few who's the losers first before talking about you know, saving lives and being ethical about it but for some projects maybe you'll see that both, I don't know so next question how do you go about creating security trust models with AI? have you go about creating the security with trust models with AI or do you? well, you need to so clearly I believe first of all that there's a data foundation there that I think goes way beyond that that's what's my disclaimer at the beginning of my pitch as well we just realized that we're working with very exposed materials and you know, and we realized the public opinion is very sensitive about this as well every few seconds somewhere in the world a human driver behind the steering wheel kills someone every few seconds, right? literally and if it happens once every six months everybody's like stop all engines stop the projects there's something very dangerous going on here so I do really I think on one hand of course this asks for very established good practices in terms of data privacy and security I think the whole GDPR thing over here for example in Europe will help us a lot to ensure the proper use of private data particularly personal data but also very sensitive let's say for HD live friendly data I think these are very strong practices and the interesting thing is if we want to really enter the digital era we need to realize it's a foundation not the meaning of life to most of us people not to stop it but how to enable that's the right question to security should we do this how can you enable me to do this I think is the proper way of looking at it we're only going to leave it there sorry about that we do have a panel this afternoon at 2.45 the other room we have a panel debate hoping for many more questions in the meantime thank you very much