 I just wanted to welcome you all to our Lectures in Planning series and thank you all for being patient with the room changes, but I'm very pleased to welcome Professor Michael Batty today. Michael Batty is the Bartlett Professor of Planning at University College London where he is Chair of the Center for Advanced Spatial Analysis, he has worked on computer models of cities and their visualizations since the 1970s and has published several books such as Cities and Complexity and The New Science of Cities. His most recent book Inventing the Future Cities was published by MIT Press in late 2018. He blogs at complexcity.info and then covers the science underpinning the technology of cities and has posted lectures on big data and far cities are at spatial complexity that info. So prior to this print position he was a professor of city planning and dean of the School of Environmental Design at the University of Wales at Cardiff from 1979 to 1990 and the director of the National Center for Geographic Information and Analysis at the State University of New York at Buffalo from 1990 to 1995. In 2015 he received the gold medal of the Royal Geographic Geographical Society for Geographic Information and Analysis at the University of Wales in 2016. He received the Senior Scholar Award of the Complex Systems Society and the gold medal of the Royal Town Planning Institute. In 2018 he was awarded the Walder Toppler Prize for GI Science of the Austrian Academy of Sciences and in 2019 he was elected as a Fellow of the Regional Science Association. But perhaps all this has not yet captured the depth and his founding and continued impact on the field of urban analysis. So I just want to again welcome Michael Batty to the Columbia DSAV and with that I'll turn it to you. Thank you very much. Thank you. I'm very pleased to be here like New York City. It's a great diverse complicated kind of place, something new every time. OK. When they said that my two recent books, Inventing Future Cities and the New Science Cities, both published by MIT Press, I'm very proud of the fact that both of them have been translated into Chinese. And I've only seen one of them basically so I don't know how many people in China have seen them but they're in Chinese, right? So there's a lot of Chinese in the audience here. I presume you still speak Chinese even if you love Chinese. So anyway, if you search on the web you can find it. OK. Now I'm going to talk about digital twins. In some senses this is a concept that's been around a long time but it's become very popular recently in terms of all kinds of things, all kinds of digital models being built of real systems, not just in cities but in lots of areas in organizations and so on and right throughout the sciences. So I'm really going to talk about the idea of digital twins. I mean to cut a long story short, digital twins are really digital models. They're computer models. To some extent we've been doing this almost since computers were invented. Computers were invented in fact about 80 years ago just before and during the Second World War for very obvious reasons. And almost immediately computers came on the scene in science labs, etc. They began to be applied to real problems. So for example in city planning, if you go back to the mid-1950s the Chicago Area Transportation Study Commission developed a whole series of different computer models, very different of course from the way we do things today. Watterley Leontief at Harvard basically who developed the input output model when he was still in Russia in the 1920s made it operational using analog technology at MIT, Harvard MIT and that was really during the just before the war years and of course input output modeling is pretty wide spread within urban and regional economics in that context. So again an example of an early model, a digital twin if you like although we didn't call them in those days. So I've actually put this PDF of this talk on my website. So if you go to spatial complexity, just go to spatialcomplexity.info that's the link that came up and I just shoved it on recently but spatialcomplexity.info and digital twins and you'll actually find the PDF and I'll tweet it a little bit later so anybody following my tweet and there's my handler basically at J. Michael Batty you'll be able to download it directly. Okay so let me tell you what I'm going to talk about. I'm really going to talk about and give you my idea of what the smart city is because to some extent most of what we're doing at the moment in terms of making things digital is loosely badged under this notion of the smart city. Now in some sense I'm going to make the distinction between what we call the high-frequency city and what we call the low-frequency city and I'll spend a little bit of time talking about the high-frequency city is most of what we think about in terms of smart cities today. They're basically what's happening over seconds, minutes, hours, etc. what's happening during the typical day whereas the low-frequency city is to some extent if you go back 10 or 20 years then most of what we did in cities was looking at them over much longer periods of time. So for example master planning and urban design and so on is really the low-frequency city. Time is not of the essence in some sense. You could argue that although planning is about the future in a sense previous planning is really in some sense timeless in some sense in that sense it's mainly about space. So the low-frequency city is something that we need to contrast with the high-frequency city and basically the high-frequency city is something that emerged really with a vengeance over the last 10 or 15 years and that's largely because of what we see around us here. I've got a laptop, etc. I'm linking to this thing showing you the PowerPoint somebody's on an iPad over there. There's quite a few Macs you could tell because of the Apple logo sort of shining out. I can see that in the audience. People have probably got parallel posting on their phones and so on. So in some sense this is really the high-frequency city. This is what's being embedded into the environment all around us and it's not going to get less. It's not stable. It's going to be massive. It is massive in some senses that I think the number of smartphones worldwide is something of the order of about two to three billion. Well of course there's only seven billion people seven to eight billion people actually on the planet basically. So we're talking about 20, 30, 40 percent of people everywhere that is who have smartphones and therefore they probably have access to the net in that context. Access to a great load of information. That's the high-frequency city in that sense. And the idea of big data many of the talks in this series deal with the high-frequency city but big data in particular is the kind of data that really streams out of the senses it streams out of our phones it streams out of the embedded sensors. If you go to London, one of my examples later on will be the embedded sensors on the transit system on the London Tube basically which is almost entirely digital now both in terms of when trains arrive and when passengers kind of tap in and tap out and get on trains in that sense. And one of my examples of the digital twin will be exactly that kind of data on the high-frequency city, the transit system and how that relates. I'll begin to also say where the digital twins come from to some extent I anticipated that because in a way we're thinking a lot about it at the present time in that sense and then I'm going to actually look at a slight digression of cornucopia of twins if you like maps as analogies. I want to actually develop the idea that models are close to the real thing. Digital twins in some sense are closer and closer to the real thing they can never be the real thing basically but there are lots of examples in some sense in thinking about twins in that context through the idea of maps and I'm going to actually look back historically at what people have said about maps and that's a slight digression but it illustrates this idea of digital twins. Now in some senses the digital twin idea can also be seen as hardware turning into software in that sense and what's actually happening is that much of the software that's being invented layer upon layer is actually penetrating the urban fabric in some sense and building information models are a good example building information models are now writ large in terms of designing buildings many of those models are actually entering the building fabric itself that you can't really run the building a complicated building without actually using the same software to actually maintain it and to run its functions and so on as we're actually used for its design in that sense. Okay and then I'm going to look at three very different types of twins three very different types of examples and I'll probably spend about possibly about half the lecture on this so the first half will be in the next 20 minutes or so will really be all of that what you see on the screen and then I'm going to spend about 20 minutes talking about particular examples three different types of twins the first sort of twin which is quite close to the real thing superficially to the city is virtual reality and augmented reality 3D city models and building information models I'm just going to talk about an example in East London of how buildings are being wired in some sense to talk about the idea that these digital models of buildings are quite close to the building itself they're part and parcel of it my second example will be the automation of the London transit system and if you're familiar with many of the new transit systems for example in large Chinese cities for example Shanghai Guangzhou and Zan and Beijing then this applies as much to them as it does to the London tube it doesn't quite apply to the New York subway system one still has to get almost a physical card in some sense you can't use Apple Pay or anything like that which you can on the London subway and in some senses because this will come basically it's not going to be very long before those systems get automated once we get that kind of automation then we're beginning to think about controlling those systems and that's where the digital twin again comes into its own the last example is really the low frequency city long term urban change I'm going to tell you about a model that we're building for not the London but for Great Britain it's the main island it's basically England, Wales and Scotland not Northern Ireland because that separates it so we're building a scaling, a land use transportation model called Quant to the whole thing and that really is designed to do traditional things in terms of the low frequency city looking at long term change infrastructure planning and so on but it's another example of a digital twin and to some extent takeaway message in all of this is is this concept useful for actually thinking about our modelling of the cities and to some extent I would submit it is largely because it enables us to actually think a little bit out of the box about these things in a sense okay first of all then what is the smart city well to some extent I've really anticipated everything on this slide we make this distinction between the high frequency city if you like you can think of it as the 24 hour city or the city over a relatively short period of time that's become particularly popular the idea of finding out what's happening over 24 hour cycles in some sense and it's in contrast to the low frequency city which most of our urban theory is all about if you look at urban economics and if you look at urban geography if you look at urban sociology it's really about how segregation takes place over long periods of time how markets get established and so on it's not really about the what's happening in the next 10 seconds or the next 10 minutes or the next 10 hours in some sense in that sense it's over much longer periods of time so that's the distinction now of course digital twins apply to both and to some extent smart city technologies which are essentially digital also apply to most but most of them apply to the high frequency city rather than the low frequency and that will become a power okay so let me actually define the smart city from a simple diagram I've used this quite a lot before but it actually helps to focus ideas on this diagram the top left hand box which is a sort of well they're both orange aren't they these colours that in some sense the real built in social environment sort of yellowish box basically and that represents the city it's what's out there basically in some sense it's what we're studying it's what you're studying in some sense it's separate from you you're separate from it in a sense and we developed theories related to that built in social environment and these theories in fact are in this second box which is the orange box basically our theories of the city and to some extent there is a circularity between them we draw data about the real built environment in other words you observe it when you want to call it data or not but nevertheless we get some sense of what's going on in the real built environments and that conditions through induction normally or inference etc our theories about the city our hypotheses about how it works and then we test those against the real city in some sense that's almost like the scientific method in some sense where we draw data from the real city we produce our theories and so on and then we test them and there's nothing in what I've just said that necessarily means it's got to do with computers or or digital in this particular context that in fact over the last 50 60 70 years people have begun to use computers to actually help us build models of these theories and then to test them against the real built and social environment and to improve them in this particular context so you can think of this as really being kind of quasi scientific method in some sense we may enter the method as any particular point if we're people who are interested in data we may actually be much more interested in that in this hour really and if we're interested in models we might be interested in that but some extent to get completeness on this we have to be interested in both now what's happened in the last 20 years and maybe just the last 10 years many people say that the new reality of the smart city really dates from 2007 which is when Steve Jobs introduced the iPhone that's when we all became empowered in a sense to be to be able to access the net it's more complicated than that but generally speaking what's happened in the last 20 years say since the millennium since the year 2000 is that into this picture have come computers and sensors for the first time so the real built environment is now being sensed and people involved in these information technologies are basically building sensors everywhere and of course the biggest sensor of all in fact is the iPhone in this particular context because the iPhone is picking up information all the time just about everything in a sense and I'm not talking about Google or Apple actually tracking you in the iPhone they do that of course I'm talking about the kind of information that you're generating when you use these devices to do a multitude of things so that's new basically and out of that particular context in the last 10 or 15 years has actually come real time screen data, big data and all the fuss about big data is due to the fact that we're embedding computers and sensors into everything around us now this has been going on a long time it's been going on for probably 50 or 60 years but it's only really become possible to do this on any large scale since computers have been miniaturized down, they're scaled down to the point where we can actually embed them indeed in the environment without worrying too much about them they're getting smaller all the time and they say that the next great wave of embedding is going to be embedding these things into ourselves and that's very basically etc lots of ethical implications about that but modern medicine is very much geared to the notion of well modern society geared to the notion of taking these embeddings basically and using them to actually improve in our context to improve cities but in medicine to improve our possible functioning in that sense okay so real time screen data one of the features that this diagram shows is that real time screen data is not the reason why computers and sensors have been put into cities the reason why they're put into cities is if you look say at the London Tube my example the reason why you tap in and tap out how you do it on your phone and this sort of thing use a credit card etc on the phone the reason why you do that is for payment purposes it's not so weak and analyze the information system etc so to some extent the real time stream data that comes out of this is a byproduct it's often what people call information exhaust it may be useful to us but on the other hand it may not be useful at all I'll show you an example of real time stream data where there's a good deal of noise in the data so that it actually confounds it it's not as good as traditional sorts of data which are much more expensive to collect which have much more structure so real time stream data the second issue in this is that if anybody talked to you about big data they'll talk about the 3Vs volume veracity variety and so on in fact the modern 3Vs but the really big V in terms of big data is volume the reason why it's big is largely because it's streamed incessantly at the same sort of rate at which the sensor picks up data in some sense so only when you switch the sensor off does it no longer stream and indeed even in the context of phones your phone may be off but it still might be streaming data to some particular archive somewhere so to some extent big data is very much associated with this notion of embedding computers into the environment a couple of things on this the reason why computers and sensors are being embedded in that sense is for information and control the payment purposes and so on and what comes out of it the big data is information exhaust now let me actually throw onto this canvas a number of other things that are relevant in some sense low frequency models is mainly what we've been doing here so in other words if we look at the city and work out there are certain problems segregation and so on and we want to rearrange the avenues to in some sense this is really a notion of a low frequency model it takes time for us to actually implement a master plan because we know by the time we get to that point it will change in that sense but nevertheless the low frequency activity is largely what we've done in the past what's actually entered big time into this context is high frequency models in some sense now high frequency models have always been there and this is an excellent test bed here in New York because really for 50 years the police department transportation and so on have used high frequency models in some sense often non-digital highly manualised in a sense throwing people at the problem rather than computers high frequency models have been used to emergency services for police fire and so on in that sense there's a long history here in New York with this which relates back to the Ram's Corporation Research and National Development Corporation with their New York City project in the late 1960s the 1970s when New York was having all of those awful problems of increased crime and housing problems and so on okay so in some sense the high frequency models have always been there but of course with all of this new data we're getting much better in some sense in quotes high frequency models than we've ever had before now broadly speaking models in this particular context are referred to really in some sense as urban analytics none of these terms are cast in tablets of stone all of these words are contingent on who is using them in some senses so the semantics of it are like all the semantics in this context a little bit vague in that sense of urban analytics is planning tools planning techniques if you like range of different things but to some extent it's a term which is now being used in relation to this notion of the smart city and all of this of course is pertaining to big data so that diagram sort of shows you all the kind of bits basically that give a kind of potted summary of the idea of the smart city and what we're going to do now is to talk about how digital twins fit into this okay now let me turn to this idea of digital twins in some sense in an audience like this you'd say that you'd probably be able to say that a digital twin is a model in some sense and models are simplifications or abstractions of the real thing when we build a model we actually throw most of what were what might appear to go in the model away and we actually keep the kind of kernel of what we consider we can actually articulate in terms of theory within the model so often it's said the best models are the simplest the best is the simplest much depends on purpose and so we can have lots of different models of the same thing indeed you can say that everybody approaching the same problem would have a different model of the problem in that context and there's nothing wrong in that that's simply the variety of life in that sense so the model can never really be a twin because a twin in that sense sounds far too close to the real thing in a some sense so in other words as we're simplifying and abstracting in terms of models they're bound to actually not be the same as the real thing the whole purpose about the model is that it isn't the real thing it extracts the essence of the real thing that we're interested in so if we're interested in transportation problems in some sense we actually keep in the background a whole range of other problems that may or may not link to them a good model would be able to put them on one side and be able to inform on the the problem in focus etc without worrying too much about other problems so to some extent the model can never be the same the same as the real thing but in the move from the high from the low frequency city where models are clearly not the real thing it's a long long way away from it and it's the difference between the real thing and the map in some sense and the movement from the low frequency city to the high frequency city does appear as though our models are coming a little bit closer to the real thing now if you're a kind of a purist in that sense the digital twin can never be the real thing because if the digital twin were the real thing it would be the real thing basically in this particular context so the talk really is about how close the model gets to the original system to the real system in this particular context so we're saying here a digital twin cannot be a mirror image it can't be the same as the original and in this context we assume it's a simulation but I'm putting a lot of question marks about this because to some extent the more we think about it the more we begin to pick up ideas of where the digital model etc is entering in some sense the embedding itself into the thing that's being modelled in a sense now first of all models depend on their media and the media in this particular context is digital so if the original real thing is digital we might actually build a digital model of something which is digital now because if you look at production processes automation basically then a vast array of our production processes and manufacture processes are digital now in some sense they're never completely digital because we're producing material products in that sense but in that particular context the digital model is getting closer in terms of those artifacts to the where those artifacts are being manufactured CADCAM for example computer aided design, computer aided manufacturing is really all about that in a sense now the second issue here is that the digital twin can't really be used for predicting a future and that's because models may be I'm sorry a digital twin can be used for predicting the future whereas a future that does not exist let me go back and say that again a digital model we're asking the question can a digital model a digital twin be used for predicting a future that does not exist in some sense now models can be used for predicting that futures that don't exist but in some sense the digital model the digital twin would no longer be a twin in some sense because most of the systems to which it's applied and I'll show you the examples in a moment are really not those that predict the future the digital twin might be used to control the process in some sense but not necessarily to predict the future so models differ from digital twins in some sense largely because the digital twin can't really be used for predicting the future in terms of the definition relative to the real thing twinning the real thing then of course the real thing is not predicting the future predicting the future in the function okay third feature is the digital twin is reactive, rather than predictive if it's predictive how close is the prediction to the real thing and there is a degree of latency a degree of mismatch between a digital process operating in real time and a digital model of that particular process for the model to be useful it can't be the same as the real thing otherwise it would be the real thing it has to in some sense be different from it and if it's operating in real time the real time in the digital twin must be a little bit different a little bit behind if you like than the digital model itself now sometimes digital twins transfer information between the real process and the model and in fact most of the models that we're involved in in this particular context don't really transfer information some of the things we'll look at are getting close to that but most of our models in the past in the low frequency city and certainly the ones we have now for the high frequency city don't transfer information in that context there's still the notion of abstracting the real thing now the last point here on this slide is that the digital twin might be a kind of controller to some extent it might be a mechanism which is actually embedded in the system itself so we can think of the twin as containing elements of the system but it's embedded in the system itself and this is hardly a model in this particular context okay so where do digital twins come from now the term appears to have first been coined it's like all these origin stories you can trace it back to Michael Greaves in 2014 you can probably trace it back to John von Neumann in 1937 or Alan Turing or something like that you can trace it back to Alan Turing you can probably trace it back to Leonardo you can trace it back to Leonardo you can trace it back to the Greeks basically so all origin stories are like that but it does appear that this chap Michael Greaves wrote a paper in the early 2000s about production processes and in production engineering there's a great deal of computer aided design and computer aided manufacturing going on in that sense and he argued that the twin was very close to the real thing he didn't go as far as to say it was the real thing in some sense and the fact that these used the word twin I suppose plays on this idea of maybe it's not identical twin identical twins and non-identical twins and so on in that context and it opens up this whole notion about if we have digital twins are they identical are they not identical can we have digital triplets can we have and so on and in some sense thinking out of the box like this forces us to think about what the nature of the model is in some context now the closest in our world we are to the components in terms of the city is this idea of the building information model the idea of the model that actually not only actually designs the building in some sense which can be a computer aided design in some sense also actually controls the functioning of the buildings and in some senses buildings are being wired in some way and the software which is associated with their wiring is very close to this idea of the digital twin in some sense the twin can exist within the building etc but the building can still exist without there being a twin etc simply wouldn't be wired in this particular context okay now let me actually digress quite dramatically and this is a bit of fun basically before we then steer back to talking about examples of digital twins okay and I'm going to look at this notion about the map basically and how close is the map that we might produce which to model in some sense to the real thing itself now there's a very famous paper by Ben Wyn Mandelbrot called how long is the coast of Britain written in science in 1967 and this really is one of the origin stories about the development of fractals okay fractional objects in mathematics and in a whole range of things etc and Mandelbrot introduced the idea of the fractal by saying how long is the coast of Britain now normally what you would do to measure the coastline is to take a map so we take a map of Britain in a sense and we we take a measuring stick a ruler in some sense a scale in that sense and we measure it according to that scale and basically we get an answer basically we get the perimeter in that sense how long is the coastline of Britain and if we do it again with a more detailed scale map a finer scale map we might start to the map which is we're in America here so probably one inch to one mile kind of works here basically one inch to one mile map and then we move to six inches to the mile these are classically old-fashioned British measures of then the six inch to the mile map is much more detailed if we take our six inch to the mile ruler and we go around the coastline we get an answer which is different from the original one inch to one mile map and that value is normally greater etc as we increase the scale of the map as we get finer and finer detail in the map we have more and more things to measure around imagine yourself going onto the beach and being given a take measure and you're going to walk all around the coastline of Britain and where do you actually you've got this problem about what is the coastline at this point a definitional problem like what is the boundary of the city but nevertheless you might say well we're going to measure around every little pubble every little rock etc and you can even go down to the microscopic scale and measure all of the particles in the sand and things of that sort and each time as you increase the finest of the scale the perimeter would get longer so the answer to this question is that the length is infinite but the real answer of course is that it actually depends and it depends on the scale basically so to some extent the map is immediately a model of the real thing but we can change the model quite dramatically by changing the scale etc now in this context the best story I've ever heard and the one that's widely close to quoting is from Lewis Carroll who was the author of the the famous well it's hardly a child's fairy tale but Alisson Mudlin basically and Alisson Ludinglas and in his last book called Sylvia Grunig Concluded he tells of a conversation between himself and a German gentleman about making a map close to the real thing so this in other words is our digital twin so let me actually read out the thing that in Lewis Carroll's book the conversation goes like this between me ok so I'm the first asking the questions and mine her which is mh in a sense I'm just going to read these out to you I should say this with a German accent for mine here but I'm not pretty good at the German accent etc English accent so me ok so the conversation goes like this what a useful thing a pocket map is I remount mine her the German gentleman says that's another thing we've learned from your nation says mine her map making but we've carried it much further than you what do you consider the largest map that would be really useful me about six inches to the mile mine her only six inches exclaimed mine here we've had a student came to six yards to the mile I can't say then we tried a hundred yards to the mile and then came the grandest idea of or actually made a map of the country at the scale of one mile to a mile me have you used it much I inquired it's never been spread out yet said mine her the farmers objective they said it would cover the whole country shut out the sunlight so now we use the country itself as its own map and I assure you it does nearly as well so in other words we've moved our digital twin to the real system basically in this particular context now of course this was written long before any of what I've talked about today was thought about in that sense but lots of people have actually developed this that and Borges has a very nice thing and I'll show you that in a moment Gregory Bateson Robinson the Economist Baudrillard David Gallant and so on even D.H. Lawrence said in one of his books the map appears to us to be more real than the land that was in 1925 and I'd be remiss if I didn't actually say that this example I've just introduced from from Lewis Carroll basically was actually quoted by my late colleague Martin March who was professor of architecture at UCLA for a number of years and it says when he was at Cambridge UK and in a paper in the 1974 urban development models symposium talking about models basically then he repeated I've copied it out here the quote from Lewis Carroll he also went on to say that there's another essay by Carroll called The Hunting of the Snark and it is Carroll says the map was perfect and absolute black all models must be so active in that sense and in some senses this notion of the map being the same as the real thing or the model being the same as the real thing really has excited people and energised people in this field for a long time Lewis, George Lewis Borges in his essay on exactitude in science he tells the same story of cartographers so obsessed with their ass that they decided to produce the most detailed map of their empire that they could make of a scale to one to one now the next generation he says and that's an amade of cartography that those who made the map had little use for it but Borges concludes by saying in the deserts of the west still today there are tattered ruins of the map I don't know whether you've ever seen the movie The English Patient with Ralph Fiennes clutching Herodotus the history is basically in the western desert during the the Second World War after he's been shot down in this particular plane but the imagery of the desert and the tattered ruins of the map to some extent are highly characteristic of the Borges essay in this particular context ok to conclude my digression my interlude I couldn't come to New York without actually showing this picture of the New Yorker so this is what New Yorkers think of the world basically viewed from here Broadway I presume and this of course is what the Chinese think of the world couldn't come to New York without actually showing you China as well in that particular context so these are perceptions so to some extent if you like these are digital twins if you're a New Yorker or a digital twin some sense of a map like a Chinese in that sense ok let's actually move on to how are we doing for time that's the key thing 15 minutes or so ok we'll get onto these examples so in some senses back to this idea of the digital twin I've made these points before that there is always some latency between the model in real time in this particular context in a way but what is beginning to happen is that a good deal of hardware is turning into software that hardware turns into software and the idea of software being embedded in the building doesn't mean to say that software becomes hardware in some sense of hard becoming software in this particular context but software is merging into our environment one of my colleagues Martin Dodge with Guy from Rock Kitchen wrote a book in MIT Press called Code Space and it's all about how software is actually about 5 or 6 years old but how software is penetrating everything around us in some sense and that's really what I mean by the notion of the digital twin getting closer and closer to the real thing itself ok now there's one last quote that I'd like to make before I move on to my examples and this is another famous book where there are many different models which are talked about in that sense many different models of cities etc so there's a book by Italo Calvino called Invisible Cities and they're conversations between Kubler-Klan and Marco Polo and one of the stories there goes as follows I used it in a lecture about 12 years ago in Liverpool which was reproduced in my model cities paper in the town planning review back in 2007 I think so this is basically what he said so all Calvino stories are like this so Kubler-Klan says and yet in my mind I've constructed a model city from which all others can be deduced it contains everything corresponding to the norm since the cities that exist diverge in varying degrees I need only to foresee the exceptions to the norm and calculate the most probable combinations Marco Polo says in the same conversation I've also thought of a model city from which I can deduce all others it's a city made only of exceptions exclusions incongruities and contradictions if such a city is the most improbable by reducing the number of abnormal elements we increase the probability that the city really exists and this is the idea that out there there are multiple cities in some sense which we can think of as models of cities and the ultimate models which we would actually apply to cities so to some extent they're all kind of they're all conversations and speculations on how we might think of many different models of the same thing and indeed many different things which elaborated from the same model okay now for my examples okay so now we're changing tack yet again and I'm going to look at three very different conceptions of digital twins the first example I'm going to look at is a physical representation of the elements of the defined buildings in the city so it's buildings in digital representations virtual reality augmented reality and then spin which is building information models in this context that's a whole series of ideas or sets of ideas which are writ large in architecture and building construction at the present time and due to some extent it relates once we scale them up to city scale to ideas about virtual realities of entire cities in a sense so this is really the high frequency city what can we say about that and I'll show you the example from this one second is my tube example where for example we have digital data pertaining to passengers and trains basically and the modelling problem or the management problem is linking passenger movements to train movements for a whole range of things that I'll tell you about and this is really the high frequency city but it has long term implications long term implications meaning that this data that's being collected is actually showing how the system is responding over time and it's showing once we examine this big data which is very short term but over very long time periods and thereby hangs a tail that's not really been done yet for a whole range of political reasons then you can begin to identify secular changes in the transportation system how some stations are losing patronage others are gaining and so on and my third model is of the low frequency city simulating long term change in urban growth now to introduce the idea of the 3D city this was a paper that we wrote back in 2005 in the General Architectural Design and we had a 3D block model of the city basically and these are all pictures taken from it and I'll show you some examples of that in a moment so these models really these models existed oops sorry these models existed really the first models in fact existed in the early 1980s but really by the early 2000s they were being developed in bulk and if you go to London or New York here for example there are countless of these 3D models basically so this was our model back in 2005 it was built in ArcGIS Arc info software was in those days in 3D 3D scene I'm going to start seeing it built and you can actually see that most of this is a block model objective of blocks and bits of it are rendered in a bit more detail now in the south bank basically just to fix our ideas you can see a lot of iconic buildings at the moment that's walking through the station which used to be the north side and we're crossing now to Parliament bits of Parliament are rendered in more detail and then there's the what's called the Lending Wheel basically which will put up to separate the year 2000 and so on and on and on so you see most of this model is a block model and the reason why we built it is not so we can render every building it's a kind of 3D GIS basically in that sense in fact this actually shows you the same thing now if I click on this let's see if this works because these were the first block models ever this was done by Skidmore Owings & Merrill back in the late 1970s early 1980s now that was almost prior to the PC revolution they were done on Vax computers so if I click on this let's see if this is going to work yeah it is the triumph of eduro basically to your eduro etc and I've got I've got okay I've got this model so I'm going to turn the sound off basically you don't need the sound there's some music in the background so this is what this is what when people made movies back in 30-35 years the screen back in those days was amazing anyway there's a difference the first example if you go to youtube and Skidmore Owings & Merrill and wireframe or something like this then you get that there are several US cities in that context so again this has been really developed very extensively in the last 20 years let me just move on to the next one here is the kind of thing that's going on now one of the actual features and this is very relevant to digital twins is the notice that once we have the model basically so this is part of East London then the model can be translated into many other virtual environments so for example that's a sandbox that has got the model so this is the model of London it's highly impressionistic because what they're doing in this model is they're adding information into it you can see lots of different types of visualisation but I think a little bit there if we run this a little bit longer it scans up and you can see that embedded in this are things like tube lines and so on so in other words all of these data visualisations have some sort of purpose in some sense and the purpose of course is as the captain suggests to page the other plans so lots of augmented reality in that sense now the same group are actually involved in thinking about wiring buildings so this is where the digital twin comes into its own but this sensor for example has been put into a building which we have a new campus in UCL which is at the Olympic Games site in East London in the Queen Elizabeth Park basically and the building is being wired not in terms of it's constructed it's being wired after the fact basically and therefore all of these sensors actually enable one to actually create the building in some sense so from the sensors themselves the building is created now there's nothing very special about this except of course it's the process of actually controlling and designing the building etc so there's the real building and this is the model that's actually constructed by sensing all of this sort of stuff using these boxes basically and a variety of other media in this particular context so to some extent this is a digital twin in the sense that it really is very close to the real thing but it isn't the real thing and to some extent the whole notion it's still a digital model the whole notion of it being a twin is problematic so you see from this we can actually produce a whole range of different measurements etc which really pertain to the performance of the building in this particular context okay let me move on to my second example which again shows how close we can actually get to the real thing this is the subway system in London all public transport is controlled by Transport for London that's the public agency and they use a smart payment card called the Oyster card I think basically they copied the Hong Kong MTR Optimus card basically but this has been around for 20 years now of course you can pay with you can use a well, not exactly what basically but you can use what your credit card on a smart phone Apple Payments on to actually tap in about 85% of people traveling on public transport are using Oyster card which is a charge card where you charge it on basically you put cash on it etc and it records your period and you tap out now the tube system is particularly convenient to actually study it because it's a closed system you have to tap in and tap out the bus system you only tap in and the reason for that is to do with the notion of one-man buses and the driver actually acting also surveillance of passengers to some extent and it's not possible for the driver to actually look at people getting on at the same time so hence the reason for that and the data is much less useful because we don't have options of destination but from this tube data we do. Now that's the passenger demand data this thing over here is the supply of trains data once you get into the system you want to know when the train is coming and of course down here this is Traconet so this is on the central line as it's by Newbury Park and this is wrapping this is the central line two trains going off that we've got here in that sense telling you how long it's going to take so this is the data that really comes from Traconet so tube is Traconet here and this is an API and we can collect that data there's a latency of about three minutes basically in that sense obviously the tube train drivers have this material in the cab basically. Now the key thing this is basically demand data passengers and this here is supply data what we'd really like to do is to build a model of the process of linking demand to supply now you'll appreciate this directly but in the London tube like here in New York there are some very complicated underground passages that if you if I go into Tottencourt Road tube station and go down to the central line because I do it every day I know exactly which way to go I know how to take shortcuts basically you know these directions that say you don't go down there there's most people going down there I know how to do all of that basically if you were to come into Tottencourt Road station the first time it might take you five minutes longer to get down to the platform where the train is not very complicated there's some really complicated ones right here and in China some massively complicated ones but what we don't have is a way of tracking the passenger to get on which of the train now that's very important because if you're interested in improving the quality of the journey you really need to know what passenger gets on what train because a train that gets disadvantaged with stalls and it's like a full load of passengers who've been disrupted and sitting in tunnels for sort of 15-20 minutes and there's something to train them on the platform people get on who have not been disrupted etc so at any point in time disruption is quite complicated so we need to link tap in tap out to the trains themselves and of course we can't link them because of privacy you can't track people in the tube the bylaw says we can't do this London Transport says it's a judicial thing the cameras you can't take photographs in the tube people do but you're not supposed to the cameras are all there for security purposes out by London Transport so in other words this notion of connecting a demand supply which would be really useful in terms of managing the system is almost impossible in that sense let me show you some examples in the next few slides that's not the tube system that's the overground railway and network rail and the tube system the typical tube map is sort of embedded in the middle here that's the abstract one that's a complicated system like New York like Tokyo and Sam just as complicated don't think that the Tokyo system is any more complicated in London or New York in that sense it's just that the maps are drawn differently so we're just doing this on the closed system here which is the tube this is the kind of data we've got we've got lots of data from Transport for London what John Reid who worked on this problem he's a professor of kings in geophonetation group now he basically worked on this when he worked in castles and IRA and he puts together the use of a flow map we have one people tap in and tap out what you see is the tube that little counter at the top there this is Wednesday midday we go to the peak let's go to Thursday Thursday morning, 7 o'clock the peak after the middle of the day then the evening peak and the late evening peak you can just see at that point it's sort of entertainment peak so in this kind of data we've got all the kind of things that we're interested in terms of actually sensing what they're doing now of course we don't know what person gets onto which train they tap in and we know where they tap out so we know that origin and destination but we don't know their route because we can't track it in a sense and there are lots of different routes that you can take to go from A to B and so what we use is the shortest route algorithm the standard Dijkstra algorithm or some variant in that sense and we know that that is not not perfect in some sense so even this data is slightly problematic now there's some pictures of John but there is a YouTube movie actually of this basically so what I'm going to do here is I'm just clicked on the you can see what you see you can see this is the weekly cycle with the morning and evening peak can't really make it out to the top here this is the trains data I'm just going to click on this trains data here these are the trains now this is particularly interesting in terms of big data because this is simply a day days worth of trains data so it's a different frequency this is actually a week's worth of passenger data so I've not been able to I've not needed to coordinate them I'm just showing you let me just go to the next slide in that sense this is showing you in a bit more detail to show you some of the problems now these trains are supposed to be on these lines now look at the purple line and some of those trains are falling off the lines so I said to my angry Richard would you put the trains back on the lines and he said to me he said you write and talk about big data how am I possibly going to put these trains back on the lines there's literally millions and millions of observations once the train gets on off the lines through noise the track and edge is resetting itself because I'm very old technology analogue technology probably so it's resetting itself and I could he said put the trains back on the lines I'd have to pick everyone every one of these every one of these things that's off the line and physically move it back to the line and he said he can't do that with big data so in other words here's a classic example looks like good data it's got noise in it we know what the noise is due to but it's very difficult to correct now when we look at passenger data passenger data is very interesting because they gave me a days worth of data in November 2010 and I found that in the data they give me just from the tube station they didn't give me the routing data I found that in total 5.2 people tapped out and 6.4 6.4 million people tapped in so there's 5.4 6.2 5.4 tapped in and 6.2 tapped out there's a difference of about 800,000 passengers and where did that difference come from well dead easy basically the barriers are left open at night by the porters some of the suburban stations you have to self validate on the thing you're supposed to self validate your card even if a barrier is open you go down the platform and you'll see a thing that's validated there if you're a season digging holder or if you are a freedom card holder I'm a freedom card holder that means I'm over 60 over 60 and that means you get what's called a free bus pass in Britain every local authority issues free travel pass basically now we only have one subway system in Britain well two because one in Newcastle but it's only small so basically living in London you get this you go freedom pass anywhere on the system so that's no difference if the barriers are open I don't need to tap of course data is being missed so these are some of the pitfalls of big data now we do like to these are the kind of things we get from it and I couldn't resist showing you a picture of big data in 1939 this is actually taken from the Getty I like Getty's got it London transport workers ladies very sexist I think but it is 1939 this is six months before the war in Europe broke out when the Britain and France were on the globe war on Germany basically so they're analysing and counting these tickets to look at origin and destination so in some senses there's nothing new under the sun these problems are still the same as they've always been it's just that we now are working with them digitally now my last example my last example I'm going to actually go through this very quickly because I appreciate I've been going on for a bit we're building a land use transportation model you don't need to worry about these things except we drive it by predicting employment and trips which is the job to work and we have a link back to the German the demand for commerce and retailing so this is the kind of employment model it's a very standard land use transportation model built for the cross section and one of the things we've been doing is actually making these models very visual because we want to communicate them to people who are interested in looking at planning alternatives so this is London and the outer metropolitan area and big ideas it looks like but basically Greater London is in the middle here that's about this was 2001 it's about 15 million people and about 8 million they've been what's called the Greater London Authority area which is the inner area if you go to London and you go to Heathrow then you are on the very edge of the Greater London Authority I think the next the next example I'm sorry it doesn't show it but I'll went through these quickly but basically essentially the Greater London Authority where the tube meter was that I was talking about was in this bit here but this is basically Greater London I should say this is Reading over here all the London airports are contained in this but Oxford and Cambridge which arguably are really part of the London are not in there but they're just on the edge of this region what we're interested in this is a major problem in building any models is boundary etc. so what you're actually seeing here predictions from the model it's very visual in that sense but we're very interested in drawing the boundary basically because we draw the boundary in the wrong place there are a lot of the movement in and out it's not captured so this boundary is drawn in such a way that it minimizes the outflow you can see that even here for example of Reading you've got a big town 300-400,000 people on the edge of the region although if you go west of Reading it drops up very quickly basically the population density in that context so visualization is all important we wanted to make this model visual in this context and what we wanted to do was to scale it up to England, Wales and Scotland now that's a picture of England and Wales at the same sort of scale so this is the region that you're seeing here Birmingham, Manchester, football distances are very short it's very small in Britain it's about 180 miles from Manchester and Liverpool to London about 90 miles to Birmingham this is the Colfield with Leeds and Jaffer all these places here and that's South Wales, Cardiff so in this box basically I mean this is essentially the most basic New York region anyway we're really talking about about 40 million people in this area here so it's urban Britain basically now we're building a model which is not only visual it's web based and anybody can run it anywhere it's for England, Wales and Scotland and it basically has a set of web services you can log on to this model and test any scenario the reason why we're interested in building it for Great Britain is because many of the things are happening in our country at the present time such as our flirtation with high speed train Britain's railway building era ended at the end of the 19th century and now we're beginning to have to renew it high speed trains and so on, new tracks high speed one is the channel tunnel high speed two is the proposed London, Manchester railway and high speed three is the one that will go up towards Newcastle and Scotland has all impacts that are much bigger than any particular town so we really need to model the whole nation to look at some of these impacts globalization too is such that we really need to do that it's web based, it's on the web so we're building an online tool basically so that any planner or policy analyst anywhere in Great Britain basically can actually test their own scenarios the scenarios are very simple put in so many jobs put in so much population and so on and that's the actual block diagram we use a a lot of well there's quite a lot of web based services being used our programmer Richard he's sort of a good officer basically he basically thinks of this as such web based services sort of GIS inside and of course he thinks of it too as being any model, not just a model like this but also a 3D model he thinks of the same kind of structure as being for any model in that sense so to some extent it's a kind of abstract architecture for the model this is what you get if you log on and this is the version for England and Wales we've added Scotland the reason why we added Scotland late was because Scotland has a different registrar general for the census and the data wasn't quite available in that sense, you have a series of pull down menus that enable you to click on so we've clicked on there and that's observed population it'll freshen up, let me actually show you working in that sense okay it's coming straight away it's not bad so explore quant basically so if I click on population etc so these are all I mean when you know all about these it's all Greek to me but anyway, this is population density so we're just exploring the data here I'm sorry, this is employment density and you can actually see it's not employment density it's employment basically and basically this is the centre of London this is the square mark these are called MSUAs middle-class super-active areas and you can see the great concentration of jobs here 2 million jobs in this area here 4 million in the whole Greater London Authority area that's the smaller area basically so highly monocentric city like New York like, no not quite like Tokyo but certainly like New York in this particular context you can query it so if I go to here and click on that then it says that's the city of London song 1 and back in the 2011 basically it's 10 years out of date now we're 356,000 workers recorded in the census at that particular point okay so anyway this is giving you an idea of what quant does it comes into its own basically when we look at scenarios so here's a case where what we've done is we've added a bunch of jobs this is Merseyside Liverpool here Manchester over there and we've added some jobs we've put in 40,000 jobs and what the model does is as you'd expect with any of these models it enables you to run it in real time you know since the model works straight away it has to work straight away so compute intensive the song is computational thinking behind how the thing works and what we're doing is looking at the impact of these things this is where the population locates basically so put these jobs in there you would expect this kind of package to take place I've got more complicated scenario where we change transport and so on so that's the extension for Scotland et cetera employment density population counts we have a version working for Cambridge at a slightly different level at a finer scale and that's the one that is actually being used with the Cambridge Peterborough combined authority and Ying Jin and Li Wan who are the academics of Cambridge who developed their model are using quant to do that now I've been going on far too long and 1426 my goodness okay the concept is the concept of digital twin useful well to some extent it has to be in some sense I'll leave you to judge basically it's like many things both good and bad it forces us to think about how cities are changing how models are perverating how do we deal with many models of the same problem that's the classic thing many models of the same problem how do we deal with that basically how do we somehow put them together that's an enormous question mark and that's what we're beginning to get forces to think about what smart cities are about and that's part of the debate about how cities are being transformed by new technologies and it forces us to think about what a model is all about and how we can link many models together and how we can have many models of different conceptions of the same thing at the same time so these are the sort of long term kind of reflections if you like that really deal with that okay so thank you very much and if you want to read a little bit about digital twins and my paper the map is not the territory these are not papers these are editorials in the journal environment and planning B urban analytics and city science environment planning B is the journal that my friend Lionel March who I mentioned earlier on many years ago and I became editor in the mid 1980s etc so have a look at those editorials they're online Sage, EPB we'll get it in Google and you can actually download the editorials because they're open access thank you very much I want to think around I think it's very interesting the notion of digital twins specifically says ask a question and I think if we look at a lot of the big data that's being produced today I think because it's not intentionally created there's a theoretical kind of identically between the data and the reality but because of the unintentionality it we generally find in your examples that are different the reality so I don't think this is a different I don't think it's just a matter of time I think there's something that I can hear about that unintentionally okay that's a very interesting point because what you're really saying is that the data that's generated by these sensors is to some extent does not reflect what we might be interested in with respect to the system so it doesn't match necessarily what we consider to be the most important features of the system whereas traditionally if we were to mount for example a household interview survey so transport is a good example because in the past people have got transportation data through quite expensive samples of households 10% samples of households were done quite routinely in the 1950s and 60s and that data was very expensive it was quite good there's obviously lots of issues to do with ambiguity of questions but the data was quite pertinent what you could get from that data was where people started the trip and where they ended now with the big data from the oyster card data set we can't do that we only know where they get onto the tube and where they get off we don't know what else they do so transport for London they're the agency public agency responsible for everything going on there they have many different data sets they do their own longitudinal data sets and so on they've not been able to use this data yet because they find it very difficult to stitch it into other data in other words to actually coordinate it with other data to merge it with other data in other words this famous quote of adding value to a data set by putting two things together they're not able to do in that particular case and this is to some extent the same with I've got a student doing looking at well, you know, mining tweets basically from the Twitter API and plotting them looking at them, basically lots of people have done this and he's particularly interested in trying to infer characteristics of the person who tweets with respect to what is in and around that particular person so typically if somebody is tweeting about you know, sort of Tottenham Hotspur just lost at home against Manchester United or something and he's near the ground, you know, the Tottenham Hotspur ground then basically he is trying to associate what's in the message what's local to the particular message to the person sending the message to try and derive if you like intensity data land use data would be a bit strong but trying to derive that now that's very problematic because there are lots of data sets that could in principle be linked at least giving you the same sort of but in practice it's simply impossible basically from social media data seems to me to be extremely difficult I mean, when you tweet you know, re-tweeting, that's hardly a network, it's a kind of network thing but so there's a whole range of issues pertaining to the nature of the data which is coming out as big data, great volumes on this, that are problematic in terms of what we want to do so I would agree with you entirely that's a very I'm not particularly aware of many people have written about it but it needs to be thought about a lot you know well, that's right now of course don't go away from this talk thinking that Mike Batty has advocated you know, the whole world will be full of digital twins which will be the same as the real thing so I would concur I agree more with you that in some senses the reason why one is thinking about it is that the nature of models are changing the data that we have for models and the models themselves and the fact that our cities are now full of different things, different messaging all of those things are changing the nature of the problem and they're changing the nature of the model in other words, I mean there's plenty of evidence of that in traditional areas so for example the notion that the city works according to the urban economics of William Alonzo and Von Thunen and all these people that's no longer the case clearly I mean, when you look at mortgage markets and rental markets and globalization and so on clearly you can't go any distance at all to explain the housing market without looking at those factors like that so in other words the models are changing anyway in that sense so the models are bound to change they're changing because of people doing different things and so on so the digital twin idea is to put the focus back on models now I agree with you it's a contradiction in terms to assume that the digital twin will ever be the same as the real system the example I gave of the building software entering the building is not the same right etc okay you can talk about embedding a digital twin in the system which is which is trying to simulate in some sense but it's still different in and of itself because at the end of the day there are models models are abstractions and in a way we can't predict the future is unknown to us in some sense so we can never have a model which is predictable in any sense for a real system really because the system is too many degrees of freedom we just don't know what's going to happen basically so there's all issues like that so they're all tied up in the same sorts of things and I think we need I'd like to really read something by people about these issues basically there's not that much out there but I think people it's growing a bit I think think about models one of the functions of models is obviously predicting right so in your discussion of low frequency which is shorter time span longer and shorter time interval which is the model so in that case the high frequency does it fundamentally change the reliability or if you observe that it changes the reliability of one of the functions of models yeah in other words you're asking the question can we we have plenty of predictions of models in the low frequency city transportation models particularly which almost universally have been demonstrated to be wrong in some sense now that does not mean that they're bad wrong does not is not bad because they you know we're not magicians we can't predict the future in that sense so it depends how they use but in terms of the low frequency city a lot of the predictions we've made over modest periods of time years and so on have been wrong and incorrect in some senses I think one question that spins off from what you said is in the high frequency city are the models that we have likely to produce better results now I know the results will be different because the people involved in using those models are doing different things from in the low frequency city but at the end of the day can we build a model of movement on the London tube system for example which might predict what happens in certain disruptions a model is very close to the real thing so the data is being generated we can't predict a disruption but we can once we know there's a disruption predict the consequences now I don't think anybody as far as I'm aware actually people will have looked at this but nobody has quite phrased it in those terms I think it's part of this notion about if we predict over very short intervals are we going to get better predictions over long intervals it's part of that general question and that's quite problematic because the conventional wisdom was I think until maybe 20, 30 years ago short term predictions are going to be better than long term predictions because changes, bigger changes happen in the long term and smaller changes in the short term but in actual fact a lot of things have proved otherwise Hurricane for example Hurricane Sandy and all this sort of business and a whole range of things in that sense the sort of black swan effect the idea that everything all swans are white until you discover a black one that's a pretty radical thing only when Australia was discovered did people ever see there were black swans before that all swans are white it's the inductive fallacy basically there's this great book about it the black swan so in other words it's a tricky one I don't know the answer to the question it's well worth thinking about though I think there's some big questions that come out of all of this to do with predictability and prediction which we've not been because we sort of have to predict in planning we have to try and second guess the future in some sense we've not been very good at thinking about prediction nobody else has either in some sense so I think that's a very important focus relative to the concerns we've had this afternoon which is talking about big data different types of models and so on I think the time scales traditionally in our planning are changing as a conception of what influence we might have in terms of future plans is changing in that context that they're getting shorter certainly in the professional institutional context of planning in Britain for example the time scales are getting shorter manifestly 50 years ago a whole variety of plans were produced for the longer term and that's no longer the case so to some extent you could say strategic planning has sort of caught up the agenda in some sense and that's actually that's actually exacerbated to some extent by the development of this smart cities movement in the sense that that a lot of people have been attracted to the smart cities movement who don't really have a kind of reflective view of cities particularly so then city planning is all about smart lamp posts and all that sort of thing and the word Cisco or IBM can make a lot of money this sort of thing past the transit systems and so in some sense that's actually changed the agenda and pushed it a little bit towards the short term it so happens I think that the people responsible for these short term things are very different from the people responsible for long term things so it's not all bad news it's not all shifting so planners still have traditionally the same concerns I think although our ability to actually demonstrate the longer term is more problematic I think in that sense so you're right that these time scales are shifting and when you add to that other time scales about big problems I mean one is aging for example and the impact particularly in probably in China but also in the west aging societies now that has an impact on cities and we are talking there about low frequency we're talking about longer time scales although some of these things are quite urgent in terms of changing the amount of facilities and so on and then you're talking about climate change all of these things have actually taken on a sense of urgency in terms of their temple frame which they didn't have 50 years ago that I was just reading on my phone the BBC website saying that Boris Johnson he's not exactly a climate denier or anything but he said he didn't get it he didn't understand why the world was getting hotter he accepted it because people have told him right there's a big fuss over one of his MPs he kicked off some climate for some reason anyway what I'm saying here is the time scale thing is quite climate change is quite problematic time scale thing in the sense that this project the modelling project I'm looking at the impact of sea level rise in London and the south-east and it was big flood plain in the London region and if the North Sea were to rise by one meter a lot of stuff would be flooded so that area where you see how the East is at the London game site would all be flooded as long as nothing was put in place it's not the flooding but since then 15 years ago we did this project people are talking about North Sea rising by about one half meter now so it's quite problematic these issues have become more urgent so the time scales are changing because of that and again I think to summarize all of these there's whole business about the time scale thing very important again there's something I think that would be nice to see some people write about that in a reflective way in terms of climate with that thank you so much again