Guido Caldarelli

Real systems like our cities can have their digital versions, called “digital twins,” which are now possible due to the development in the fields of sensors and artificial intelligence. However, to transform these “doubles” from mere digital replicas to reliable tools for understanding the world and predicting behaviors, they need to be combined with the science of complexity. This is the key to creating cities that are genuinely human-centered. This is what a group of scientists, including Guido Caldarelli, a physicist at Ca’ Foscari University of Venice, argue in an article in Nature Computational Science. Digital twins are highly detailed replicas of real systems, including human bodies, cities, or the entire world, created by feeding representations of the elements of the respective system of interest into “black boxes.” This allows them to learn to behave more and more similarly to the corresponding elements of the real world. Digital twins can be used to study alternative scenarios and to control the real system based on artificial intelligence. However, “doubles” do not necessarily mean that digital twins behave realistically. Neither is there any advantage in creating a perfect copy of a system without understanding either the system or its simulation. Beyond issues related to big data and machine learning, local digital twins often oversimplify aspects such as social and cultural life and anything that is not represented by data. This includes everything that is immeasurable, such as friendships, love, and quality of life – things that are terribly important to humans but not to computer models, artificial intelligence, and robots. Therefore, Guido Caldarelli and his colleagues emphasize the need for digital twins to be combined with the science of complexity. This, they write, is the key to understanding global behaviors and not just a mere repetition inside the computer. The science of complex systems studies dynamic systems made up of many elements, which typically interact with each other or with other systems in a nonlinear way. Such systems often take the form of a network and can be layered several times, forming networks of networks. Taking interactions into account is essential for understanding the nature of complex systems, which cannot be understood solely from the properties of their individual parts. Nonlinear and network interactions are often the cause of emerging system properties. One can also speak of self-organization from the bottom up. Cities are full of these phenomena. They can range, for example, from the formation of uniform lines of walking directions on sidewalks to patterns of stopping and restarting traffic flows, or patterns of segregation between people with different cultural backgrounds, as Nobel Prize winner Thomas Schelling has shown. “As a consequence,” explains Guido Caldarelli, “digital twins not only need to consider the science of complexity to become useful and reliable tools. It is not even enough to plan, optimize, and control cities from the top down. To create cities for people, it is crucial to anticipate opportunities for self-organization, participation, and co-evolution.” Real systems like our cities can have their digital versions, called “digital twins,” which are now possible due to the development in the fields of sensors and artificial intelligence. However, to transform these “doubles” from mere digital replicas to reliable tools for understanding the world and predicting behaviors, they need to be combined with the science of complexity. This is the key to creating cities that are genuinely human-centered. This is what a group of scientists, including Guido Caldarelli, a physicist at Ca’ Foscari University of Venice, argue in an article in Nature Computational Science. Digital twins are highly detailed replicas of real systems, including human bodies, cities, or the entire world, created by feeding representations of the elements of the respective system of interest into “black boxes.” This allows them to learn to behave more and more similarly to the corresponding elements of the real world. Digital twins can be used to study alternative scenarios and to control the real system based on artificial intelligence. However, “doubles” do not necessarily mean that digital twins behave realistically. Neither is there any advantage in creating a perfect copy of a system without understanding either the system or its simulation. Beyond issues related to big data and machine learning, local digital twins often oversimplify aspects such as social and cultural life and anything that is not represented by data. This includes everything that is immeasurable, such as friendships, love, and quality of life – things that are terribly important to humans but not to computer models, artificial intelligence, and robots. Therefore, Guido Caldarelli and his colleagues emphasize the need for digital twins to be combined with the science of complexity. This, they write, is the key to understanding global behaviors and not just a mere repetition inside the computer. The science of complex systems studies dynamic systems made up of many elements, which typically interact with each other or with other systems in a nonlinear way. Such systems often take the form of a network and can be layered several times, forming networks of networks. Taking interactions into account is essential for understanding the nature of complex systems, which cannot be understood solely from the properties of their individual parts. Nonlinear and network interactions are often the cause of emerging system properties. One can also speak of self-organization from the bottom up. Cities are full of these phenomena. They can range, for example, from the formation of uniform lines of walking directions on sidewalks to patterns of stopping and restarting traffic flows, or patterns of segregation between people with different cultural backgrounds, as Nobel Prize winner Thomas Schelling has shown. “As a consequence,” explains Guido Caldarelli, “digital twins not only need to consider the science of complexity to become useful and reliable tools. It is not even enough to plan, optimize, and control cities from the top down. To create cities for people, it is crucial to anticipate opportunities for self-organization, participation, and co-evolution.” Link to the original article: https://www.nature.com/articles/s43588-023-00431-4

To study what happens or could happen in extremely complex networks, such as in a pandemic or in Internet interactions, it is useful to simplify the system to make it manageable and to be able to analyze it. But how can one find the right vantage point to understand at a glance the salient features of the whole without losing sight of relevant connections? A team of physicists including Guido Caldarelli, corresponding author of the study and professor of theoretical physics at Ca’ Foscari University Venice, has found a method to efficiently and effectively ‘simplify’ the complex structure of the network. The result has been published in Nature Physics and is thus available to the international scientific community. The scholars took their cue from the technique that won U.S. physicist Kenneth G. Wilson a Nobel Prize in 1982. Wilson was able to find a theory that could explain how phase transitions, such as the freezing of the surface of a lake or the formation of a traffic column of cars, work. He invented the mathematical technique of the renormalization group, which allows one to exploit a symmetry of nature (large is similar to small) to predict the behavior of certain systems. Part of this method involves rescaling the cells in which the system is defined with larger and larger cells. At each step we merge both the cells of the system and the variables that make up the system (as in the figure where we have depicted a spin system). The knowledge of the system once the series of amalgamations is finished is able to tell us how the original system behaves at large distances and toward which fixed points the evolution of the system is headed. But how do we get the same advantage when the system is not made up of cells like a spreadsheet, but of nodes and relationships between them as is the case in our brains with neurons, in contagion between infected and susceptible individuals, or with interactions on social media? In real systems very often, if not always, interactions are characterized by the presence of a complex structure of connections that makes them very difficult to analyze. The knowledge of the system once we finish the series of unifications is able to tell us how the original system behaves at great distances and toward what fixed points the evolution of the system is headed. But how do we get the same advantage when the system is not made up of cells like a spreadsheet, but of nodes and relationships between them as is the case in our brains with neurons, in contagion between infected and susceptible individuals, or with interactions on social media? In real systems very often, if not always, interactions are characterized by the presence of a complex structure of connections that makes them very difficult to analyze. “Directly inspired by ideas from statistical physics,” Caldarelli explains, “we introduced a new renormalization group procedure that has proven essential for efficiently and elegantly discovering the organization at multiple scales of complex networks and for detecting scale invariant features when present. It also defines a universal network scaling procedure that is on the one hand very useful for analyzing large data sets and on the other hand shows us one of the fundamental symmetries of nature.” Future applications the team will work on include filtering of experimental data masses, exploration of material space, and representation of information from historical archives.