WEBVTT 00:00:06.440 --> 00:00:10.440 A man you work a company that uses digital data to make 00:00:10.440 --> 00:00:13.880 predictions about the repair needs of water pipes. 00:00:14.680 --> 00:00:15.640 Your company. 00:00:15.640 --> 00:00:17.840 If I know correctly, no bank owns. 00:00:17.840 --> 00:00:22.840 And you never go out to check the condition of any line. 00:00:22.840 --> 00:00:26.840 How is it possible that your algorithms still know the state 00:00:26.840 --> 00:00:30.120 of the water infrastructure in a certain city? 00:00:31.440 --> 00:00:34.640 The answer, in short, is data and models. 00:00:35.360 --> 00:00:39.360 So, if I know which water pipes have had a leak in the past, 00:00:39.360 --> 00:00:43.360 for example, and I know what properties these pipes and 00:00:43.360 --> 00:00:46.920 the environment have, so things like How old is 00:00:46.920 --> 00:00:49.920 this pipe made of what material is it? 00:00:50.080 --> 00:00:53.160 How is it connected to other pipes in the network? 00:00:53.760 --> 00:00:57.760 This gives me, for example, information about what 00:00:57.760 --> 00:01:03.040 pressure it is exposed to, what typical water flow it usually experiences. 00:01:03.600 --> 00:01:05.160 Where is that? 00:01:05.160 --> 00:01:06.800 The land and the soil? 00:01:06.800 --> 00:01:10.560 And of course, the texture has an influence on things like corrosion 00:01:10.560 --> 00:01:13.560 and other deterioration processes in such objects. 00:01:14.280 --> 00:01:18.280 So if I know this kind of stuff and have collected it 00:01:18.280 --> 00:01:21.400 systematically, then I can build a model from the data. 00:01:21.400 --> 00:01:24.400 And that answers my question in the end. 00:01:24.800 --> 00:01:28.800 If a pipe has the following characteristics, how 00:01:28.800 --> 00:01:32.800 likely is it that it will break or leak in the next year 00:01:32.800 --> 00:01:36.320 or in the next five or over the next ten years? 00:01:36.320 --> 00:01:39.640 And this kind of approach can be generalized 00:01:39.640 --> 00:01:42.640 to different infrastructure objects. 00:01:43.920 --> 00:01:47.200 Okay, and where do you get the data you need for something like this, which you 00:01:47.200 --> 00:01:48.840 have all listed now? 00:01:49.400 --> 00:01:53.400 Yes, uh, a large part is of course in the hands of the municipal 00:01:53.400 --> 00:01:57.400 utilities that manage this infrastructure, or the 00:01:57.400 --> 00:02:00.720 companies that manage such infrastructure. 00:02:00.720 --> 00:02:04.720 And it's mainly about organizing them and putting 00:02:04.720 --> 00:02:08.040 them in the right format, which is necessary. 00:02:08.040 --> 00:02:12.040 This means that we usually work with our customers 00:02:12.040 --> 00:02:17.600 to put this data into the right format so that it can be used. 00:02:18.120 --> 00:02:22.120 Part of that is, I alluded to it, is simply publicly available. 00:02:22.120 --> 00:02:26.120 So something like soil conditions, for example, these 00:02:26.120 --> 00:02:31.320 are public data sets that anyone can download and combine with them. 00:02:31.840 --> 00:02:35.840 Things like weather data, which are also interesting, because of course they 00:02:35.840 --> 00:02:38.880 also have an influence on temperature differences, etc. 00:02:38.880 --> 00:02:42.880 The way this happens in different places is public 00:02:42.880 --> 00:02:46.880 data and our task is to bring it into a usable format, 00:02:46.880 --> 00:02:50.880 so to speak, where you can then build these kinds 00:02:50.880 --> 00:02:54.160 of models and other functions. 00:02:56.000 --> 00:02:59.760 I think that here for the participants in our course. 00:02:59.840 --> 00:03:03.840 We have already become acquainted with such a data platform, namely the 00:03:03.840 --> 00:03:07.840 Sustainable Municipalities Portal, where at least some of the data 00:03:07.840 --> 00:03:12.120 that is very interesting for a company like yours is also freely available. 00:03:12.920 --> 00:03:14.560 And now? 00:03:14.560 --> 00:03:18.520 That sounds super exciting, but maybe also a bit 00:03:18.520 --> 00:03:21.520 magical if you haven't dealt with it yet. 00:03:21.520 --> 00:03:24.080 What do you do with such data in concrete terms? 00:03:24.080 --> 00:03:25.760 How does it work then? 00:03:25.760 --> 00:03:29.480 Yes, so I would first distinguish 00:03:29.480 --> 00:03:32.480 what kind of data it is. 00:03:32.480 --> 00:03:36.400 And you can perhaps talk roughly about two classes of problems, depending on how 00:03:36.400 --> 00:03:38.400 structured the data is. 00:03:38.960 --> 00:03:40.800 The one 00:03:40.800 --> 00:03:42.840 are, 00:03:42.840 --> 00:03:46.840 what you would call traditional machine learning, or machines in English 00:03:46.840 --> 00:03:50.840 the other, and that's the newer thing, are all applications that have 00:03:50.840 --> 00:03:54.200 something to do with AI and language models, so to speak. 00:03:55.400 --> 00:04:00.240 Let's start with traditional machine learning: What do you do with it? 00:04:01.200 --> 00:04:05.200 yes, so that's more the case where you, what I've described 00:04:05.200 --> 00:04:09.200 and where you have very structured data, so where you can say 00:04:09.200 --> 00:04:13.200 exactly okay, for how for each ear or water meter in my system, 00:04:13.200 --> 00:04:17.200 I can say exactly, this is the time that this was installed or 00:04:17.200 --> 00:04:20.880 that's just the diameter that this object has, etc. 00:04:21.520 --> 00:04:23.040 Um, they are. 00:04:23.040 --> 00:04:27.040 So you can really think of it as just large tables 00:04:27.040 --> 00:04:31.040 of datasets and what you do with them is basically, 00:04:31.040 --> 00:04:35.040 um, yes, so machine learning isn't magic, it's 00:04:35.040 --> 00:04:38.320 just kind of a trick, I'd say. 00:04:38.440 --> 00:04:43.120 A trick for what traditionally human civil engineers would have done. 00:04:43.600 --> 00:04:47.600 Namely, to say okay, if I know what the before and 00:04:47.600 --> 00:04:51.600 in the past, what were problems in my system in 00:04:51.600 --> 00:04:55.600 the past and what were things that were broken, 00:04:55.600 --> 00:04:59.600 I can somehow statistically find out how I can 00:04:59.600 --> 00:05:03.600 conclude from these properties that I mentioned, 00:05:03.600 --> 00:05:07.600 and machine learning is basically just a somewhat 00:05:07.600 --> 00:05:11.600 principled way, a somewhat systematic way to 00:05:11.600 --> 00:05:15.600 find exactly these answers to these questions 00:05:15.600 --> 00:05:20.480 and to build exactly this kind of statistical models. 00:05:20.480 --> 00:05:24.480 And they come with strong guarantees that the predictions 00:05:24.480 --> 00:05:28.400 are good etc and are often better in this sense than if humans would do it 00:05:28.400 --> 00:05:30.360 manually. 00:05:30.360 --> 00:05:33.360 But that's basically what we do with structured data and I wouldn't call it 00:05:33.360 --> 00:05:34.880 magic. 00:05:36.120 --> 00:05:40.120 So in principle, very traditional statistics, but with the difference, 00:05:40.120 --> 00:05:44.120 as we have already seen in the course here or in the lesson on smart cities, 00:05:44.120 --> 00:05:48.680 simply with large amounts of data and with a lot of computing capacity. 00:05:49.360 --> 00:05:53.360 And if we now come to the second class of methods that you 00:05:53.360 --> 00:05:57.360 mentioned, the newer, the more exciting, artificial 00:05:57.360 --> 00:06:00.720 intelligence, what, what does that bring us? 00:06:01.400 --> 00:06:05.320 Yes, the interesting thing about this is that traditional machine learning. 00:06:05.320 --> 00:06:08.680 As soon as I, as soon as something about my dataset fits me 00:06:08.680 --> 00:06:11.680 into this structure, it just doesn't work at all. 00:06:12.280 --> 00:06:16.280 And the interesting thing here is that 00:06:16.280 --> 00:06:20.280 I can do much more with AI in the sense that 00:06:20.280 --> 00:06:25.320 I can also process structured inputs and that. 00:06:25.800 --> 00:06:29.760 So the point here is this, the applications are a little bit 00:06:29.760 --> 00:06:32.760 different, but are much closer to reality. 00:06:32.760 --> 00:06:36.760 Because only if I now assume that I have such a model, which 00:06:36.760 --> 00:06:40.760 tells me that for all my pipeline in this city here is exactly 00:06:40.760 --> 00:06:46.080 the risk profile that every single one of them has, then that's a great start. 00:06:46.640 --> 00:06:48.440 But that doesn't help me yet. 00:06:48.440 --> 00:06:51.800 Um, or that in the end it is not yet the decision What 00:06:51.800 --> 00:06:54.800 do I do now with it and for these decisions. 00:06:55.680 --> 00:06:59.680 These problems are often much more complex, because it is 00:06:59.680 --> 00:07:04.320 usually not only about water pipes, but there are also gas pipes. 00:07:04.320 --> 00:07:06.840 The water pipes are located under a road. 00:07:06.840 --> 00:07:12.000 I can't just go there and replace them without touching the street. 00:07:12.000 --> 00:07:16.000 Things, that requires a lot of planning 00:07:16.000 --> 00:07:20.000 and coordination, so to speak, and, um, 00:07:20.000 --> 00:07:23.200 um, yes, and a kind of work that 00:07:23.200 --> 00:07:26.200 also works a lot with estimates. 00:07:26.200 --> 00:07:30.200 So something like, for example, if I now open this one construction site 00:07:30.200 --> 00:07:33.520 because I have to replace this and that infrastructure object, what does it 00:07:33.520 --> 00:07:35.200 actually cost me? 00:07:35.320 --> 00:07:39.320 These are such estimation tasks that such a highly structured 00:07:39.320 --> 00:07:43.320 model cannot answer, cannot answer, because you don't have 00:07:43.320 --> 00:07:48.920 such gigantic data sets that you can simply build the one prediction model. 00:07:48.920 --> 00:07:52.920 So, that's now that this new AI is incredibly 00:07:52.920 --> 00:07:56.880 helpful, because it significantly accelerates 00:07:56.880 --> 00:07:59.880 the processing of such data. 00:08:00.200 --> 00:08:04.200 So to speak, what we would do as humans, namely 00:08:04.200 --> 00:08:08.200 combine the outputs from such machine 00:08:08.200 --> 00:08:12.200 learning models with, for example, um, 00:08:12.200 --> 00:08:16.200 actual cost calculations from previous 00:08:16.200 --> 00:08:20.200 construction projects, etc., because AI 00:08:20.200 --> 00:08:25.600 can also do that and are clear to us, simplify the job, so to speak. 00:08:25.600 --> 00:08:29.320 That's actually something that traditional civil engineers 00:08:29.320 --> 00:08:32.320 or civil engineers who are employed by a city do. 00:08:32.320 --> 00:08:36.320 To put it bluntly, they will soon no longer be needed and they 00:08:36.320 --> 00:08:40.280 will be unemployed, because it is software from a startup. 00:08:41.040 --> 00:08:42.600 Yes, I don't think so. 00:08:42.600 --> 00:08:45.400 So at least we don't build our software. 00:08:45.400 --> 00:08:48.920 So there, there will have to be a few decades before 00:08:48.920 --> 00:08:51.920 we would somehow have reached such a point. 00:08:52.840 --> 00:08:58.280 Our software works more like we support decision-making. 00:08:58.320 --> 00:09:02.320 But you just have to keep in mind that in the end it's about 00:09:02.320 --> 00:09:05.880 millions of budgets that are really distributed. 00:09:05.880 --> 00:09:09.880 And if you just imagine tearing up in the street and 00:09:09.880 --> 00:09:14.000 replacing pipes there, that's really expensive. 00:09:14.000 --> 00:09:15.560 And that has enormous. 00:09:15.560 --> 00:09:19.760 This influences life in the city. 00:09:19.760 --> 00:09:21.920 Yes, it also causes a lot of trouble. 00:09:21.920 --> 00:09:23.080 Fundamental. Exactly. 00:09:23.080 --> 00:09:26.640 So it's also simply about such human dimensions. 00:09:27.000 --> 00:09:30.600 That means I really want to know very well that what I'm doing is the right 00:09:30.600 --> 00:09:32.400 thing to do. 00:09:32.400 --> 00:09:37.440 And since that will happen, we won't entrust it to an algorithm anytime soon. 00:09:38.120 --> 00:09:42.120 And even the one that, um, uh, who decides the 00:09:42.120 --> 00:09:46.120 legal liability for procedural proceedings, 00:09:46.120 --> 00:09:50.120 how this money is spent, etc., it's also not 00:09:50.120 --> 00:09:53.840 clear how to transfer that in a machine. 00:09:54.200 --> 00:09:56.840 So I wouldn't worry about that. 00:09:56.840 --> 00:10:00.840 What it means is simply that you will do better. 00:10:01.080 --> 00:10:05.080 That means in these jobs, they will shift and the kind 00:10:05.080 --> 00:10:09.080 of decision-making you make about the way you work 00:10:09.080 --> 00:10:13.720 will look different than it did five or ten years ago, of course. 00:10:13.720 --> 00:10:17.000 But even five or ten years ago, things looked different than before the 00:10:17.000 --> 00:10:18.640 calculator existed. 00:10:19.320 --> 00:10:24.920 At that time, machine learning was not yet so traditional and boring. 00:10:25.360 --> 00:10:29.360 So if a little all-clear maybe for our students, 00:10:29.360 --> 00:10:34.200 if that's maybe quite good, if I come back to the problem now. 00:10:34.200 --> 00:10:37.600 We also dealt with Orban urbanization in this course. 00:10:37.680 --> 00:10:41.680 Now you are talking about the first wave of urbanization in North 00:10:41.680 --> 00:10:45.680 America and in Europe, and then you also talked about the construction 00:10:45.680 --> 00:10:50.640 and construction of this infrastructure, which we are talking about right now. 00:10:50.640 --> 00:10:53.440 But that also means that a lot of it is really old. 00:10:53.440 --> 00:10:58.480 Now 100, 120 years old and coming to the end of its life. 00:10:58.480 --> 00:11:02.480 We see this everywhere in the Western world, that many things are in need 00:11:02.480 --> 00:11:05.680 of repair and at the same time public funds are scarce. 00:11:05.680 --> 00:11:11.120 So I can't repair everything overnight now that should actually be repaired. 00:11:11.880 --> 00:11:15.760 Can such a technology as yours such a smart city technology help with such a 00:11:15.760 --> 00:11:17.720 problem? 00:11:18.480 --> 00:11:22.480 Yes, that's exactly where the business case comes 00:11:22.480 --> 00:11:26.480 from and why it makes such a company possible in the 00:11:26.480 --> 00:11:30.480 first place, so to speak, gives us the option to say 00:11:30.480 --> 00:11:34.480 Hey, there's a problem here that needs to be solved 00:11:34.480 --> 00:11:39.320 and money can be made with it, because you've hit the nail on the head. 00:11:39.320 --> 00:11:40.000 Mettmann. 00:11:40.000 --> 00:11:42.960 Because there is simply not enough money to solve all these problems. 00:11:42.960 --> 00:11:45.840 And that means just as important as answering the question: Where should I 00:11:45.840 --> 00:11:47.320 put my money? 00:11:47.320 --> 00:11:50.000 It too, where don't I do that? 00:11:50.000 --> 00:11:53.600 Because just because an infrastructure object is old, for example, doesn't mean 00:11:53.600 --> 00:11:55.440 that I have to replace it. 00:11:55.440 --> 00:11:56.400 It may well be. 00:11:56.400 --> 00:11:59.400 Okay, it can live on for another five years. 00:11:59.800 --> 00:12:02.680 Um, and I don't have to do anything about it, because I. 00:12:02.680 --> 00:12:05.200 Because I have more important problems to solve. 00:12:05.200 --> 00:12:08.440 There are a number of concepts and there is not everyone, not every city 00:12:08.440 --> 00:12:10.080 administration. 00:12:10.560 --> 00:12:12.680 He manages it right away. 00:12:12.680 --> 00:12:16.680 But basically, it's always about return on investment, 00:12:16.680 --> 00:12:20.480 so how much do I get back from what I invest? 00:12:21.160 --> 00:12:25.160 A core concept here in this area is the useful life and you can 00:12:25.160 --> 00:12:28.840 calculate it, so to speak, the economic remaining useful life of an 00:12:28.840 --> 00:12:30.720 infrastructure object. 00:12:31.160 --> 00:12:35.160 And roughly speaking, it works in such a way that you look at okay, if I just 00:12:35.160 --> 00:12:38.680 receive it now, then I might have to fix it every now and then. 00:12:38.680 --> 00:12:43.640 This has certain costs, but it can still be cheaper than replacing it completely. 00:12:44.120 --> 00:12:47.960 At a certain point, however, I'm at the point where it just doesn't make sense to 00:12:47.960 --> 00:12:49.880 repair it anymore. 00:12:50.040 --> 00:12:53.240 Everyone knows this from their washing machine or something at home and it just 00:12:53.240 --> 00:12:54.880 has to be replaced. 00:12:54.880 --> 00:12:58.880 And the decisive thing here is rather than simply knowing where 00:12:58.880 --> 00:13:02.360 this is the case in each case, so that I can make these 00:13:02.360 --> 00:13:05.360 decisions in a targeted manner, where do I use money? 00:13:05.760 --> 00:13:08.400 And I can then do that in different ways, for example, by saying, okay, I have a 00:13:08.400 --> 00:13:09.760 fixed 00:13:09.760 --> 00:13:13.520 budget, now I want to get the maximum risk reduction out of this budget, get 00:13:13.520 --> 00:13:15.440 the maximum risk reduction out of it. 00:13:16.200 --> 00:13:20.200 I can also do it the other way around and say okay, if I know, say, 00:13:20.200 --> 00:13:24.200 what is the risk profile that I'm aiming for, how much money do 00:13:24.200 --> 00:13:28.440 I have to spend on it in order to argue for this budget, so to speak? 00:13:28.440 --> 00:13:32.000 But that's about the way we are approached. 00:13:32.320 --> 00:13:37.440 And there you see again, to tie it back to the, to the last question. 00:13:37.440 --> 00:13:41.440 These are questions that no simple model will answer for me, which 00:13:41.440 --> 00:13:45.440 predicts whether a line will break or not, but these are just network 00:13:45.440 --> 00:13:49.400 organizational questions, rather like How do we want to do that? 00:13:49.400 --> 00:13:52.360 Perhaps even political questions in some cases. 00:13:52.360 --> 00:13:54.720 I would like to drill a little further. 00:13:54.720 --> 00:13:59.400 You've already talked about the business case and the economic logic and so on. 00:14:00.040 --> 00:14:04.040 What I find quite interesting in this smart city area 00:14:04.040 --> 00:14:07.800 is that you are a startup and a private company, which 00:14:07.800 --> 00:14:10.800 means that there is no state funding. 00:14:10.800 --> 00:14:14.800 It is also not financed by donations, although one could perhaps 00:14:14.800 --> 00:14:18.800 argue that you do something charitable, but you have to find 00:14:18.800 --> 00:14:22.200 customers who specifically pay you for these services. 00:14:22.200 --> 00:14:24.840 Otherwise the whole thing won't work. 00:14:24.840 --> 00:14:26.040 Who are these customers? 00:14:26.040 --> 00:14:29.040 Exactly: And what is their interest in doing so? 00:14:29.520 --> 00:14:33.520 Yes, yes, these are indeed municipal utilities 00:14:33.520 --> 00:14:38.000 in the USA, but mainly in other countries as well. 00:14:38.640 --> 00:14:42.640 And the logic behind it is, first of all, we sell these 00:14:42.640 --> 00:14:46.640 solutions as a service and the idea behind it is, instead of 00:14:46.640 --> 00:14:50.480 every individual, every single municipal utility, having to develop this 00:14:50.480 --> 00:14:52.400 expertise. 00:14:52.400 --> 00:14:56.400 How to build such models, how to map such 00:14:56.400 --> 00:15:00.400 development decision-making processes, 00:15:00.400 --> 00:15:04.400 I should say, in software. Um, no, let's build 00:15:04.400 --> 00:15:08.400 up this expertise, so to speak, and we then 00:15:08.400 --> 00:15:11.960 rent it out, so to speak, via a, um, one. 00:15:11.960 --> 00:15:15.960 We have it as a small group, similar to how you can take out 00:15:15.960 --> 00:15:19.960 a subscription with Spotify or Netflix or something like 00:15:19.960 --> 00:15:24.120 that, and that, so to speak, how the business model works. 00:15:24.120 --> 00:15:28.120 And the efficiency comes from the fact that, as I 00:15:28.120 --> 00:15:32.120 said, not everyone has to do it individually, but we 00:15:32.120 --> 00:15:35.960 simply offer it in exchange for money, so to speak. 00:15:35.960 --> 00:15:39.960 Um, of course, that only works if our thing and 00:15:39.960 --> 00:15:43.440 this software that we produce produces 00:15:43.440 --> 00:15:46.440 enough value for the end customer. 00:15:46.440 --> 00:15:50.440 And as you rightly say, in our case it is often municipal utilities, 00:15:50.440 --> 00:15:54.440 sometimes it is private companies that have been outsourced to the 00:15:54.440 --> 00:15:57.440 management of this infrastructure, so to speak. 00:15:58.320 --> 00:16:01.680 But in most cases, they are actually public institutions. 00:16:01.680 --> 00:16:05.680 And how they work, at least in the USA, is 00:16:05.680 --> 00:16:09.160 that they manage from the money 00:16:09.160 --> 00:16:12.160 they earn from the water bill. 00:16:12.160 --> 00:16:16.160 This means that every household pays for others, for drinking water and 00:16:16.160 --> 00:16:20.160 wastewater and that is regulated on a first-come, first-come, first-come, 00:16:20.160 --> 00:16:23.680 first-come, first-come, anti-Semitic basis, how much each has to 00:16:23.680 --> 00:16:26.680 pay for it somehow and everyone has to pay for it somehow. 00:16:27.480 --> 00:16:31.920 And that, so to speak, makes up the budget of such a municipal utility. 00:16:31.920 --> 00:16:36.400 These can be very large amounts, especially for large municipal utilities. 00:16:37.000 --> 00:16:39.360 And they manage it. 00:16:39.360 --> 00:16:43.360 And that means that in order for it to make any sense for them to 00:16:43.360 --> 00:16:47.880 buy such software, they have to recognize the economic value in it. 00:16:47.880 --> 00:16:51.880 And that's usually because we can argue that if you 00:16:51.880 --> 00:16:55.880 have this functionality, then you can make decisions 00:16:55.880 --> 00:16:59.880 so and so much better, which either ensures that you 00:16:59.880 --> 00:17:03.880 can get by longer on your budget, which reduces your 00:17:03.880 --> 00:17:08.160 risk, which means things like burst water pipes. 00:17:08.160 --> 00:17:11.360 When it's in the middle of the city, no one wants 00:17:11.360 --> 00:17:14.360 to have things like that don't happen. 00:17:14.560 --> 00:17:17.440 And that's the logic behind it, so to speak. 00:17:17.440 --> 00:17:21.560 But in fact, that it is a very free-market logic. 00:17:21.560 --> 00:17:25.560 So we don't do this with donations or through tax financing, 00:17:25.560 --> 00:17:28.960 and the municipal utilities themselves are not 00:17:28.960 --> 00:17:31.960 tax-financed, but by selling the water. 00:17:33.640 --> 00:17:34.120 Okay. 00:17:34.120 --> 00:17:35.920 Yes, thank you very much. 00:17:35.920 --> 00:17:39.920 Um, data models, startups, software as a service, 00:17:39.920 --> 00:17:43.000 as this business model is called. 00:17:43.200 --> 00:17:46.240 These are important topics in the field of smart cities, in the field of 00:17:46.240 --> 00:17:47.800 sustainable cities. 00:17:48.200 --> 00:17:50.240 And we will also deepen this in this course. 00:17:50.240 --> 00:17:52.080 And that was a great start. 00:17:52.080 --> 00:17:54.280 Thank you very much. Thank you!