By Dirk Helbing
Progress in the digital world has been phenomenal. With modern sensor networks, often framed as “Internet of Things” or “Internet of Everything”, entirely new data-driven applications have become available, among them “Digital Twins”. These are highly detailed “look-alikes” of real systems, be it human bodies, cities, or the entire world. In some cases, such Digital Twins are created by feeding learning “black boxes” with surveillance data, representing the elements of the respective system of interest. This allows them to behave in an increasingly similar way to the corresponding elements of the real-world system. The resulting Digital Twin can then be used to study alternative scenarios and to control the real system, based on Artificial Intelligence.
However, “look-alike” does not necessarily mean that the Digital Twin behaves in a fully realistic way. As we know from deep fakes, it could also be a good illusion. In case of ChatGPT, a machine learning model using hundreds of billions of parameters, many have concluded that the system often deceives users with texts that read plausible at first, but turn out to be wrong at closer inspection.
This is not just a problem of large language models such as ChatGPT or GPT-4. It is also potentially a problem of Digital Twins of cities, as a recent publication in Nature Computational Science by Guido Caldarelli and others points out.
Besides a number of issues of Big Data and Machine Learning, “Local” Digital Twins are often oversimplifying aspects such as social or cultural life, and whatever is not represented by data well. This includes everything that is not well measurable, such as friendship, love, and quality of life – things that are important for humans, while computer models, Artificial Intelligence, and robots do not really care about them.
Therefore, Guido Caldarelli et al. underline that digital twins need to be combined with complexity science. This is the science of complex dynamical system that are made up of many components (such as people or cars), which typically interact non-linearly with each other or in networks. Such networks may be multi-layered, i.e. networks of networks.
Considering interactions is essential to understand the nature of complex systems, which cannot be understood just from the properties of their parts. Non-linear and network interactions will often cause emergent properties of the system, which are novel properties not understandable from the component properties. One also speaks of bottom-up self-organization and emergent properties.
Cities are full of such phenomena. This may range from the formation of lanes of uniform walking direction in pedestrian counter-flows, to stop-and-go patterns in traffic flows, or segregation patterns among people with different cultural backgrounds, as Nobel Prize winner Thomas Schelling has shown. One could, of course, add a lot more examples.
As a consequence, Digital Twins of cities do not only need to consider complexity science to become reliable and useful tools. It is also not enough to plan, optimize and control cities in a top-down way. To create cities for its people, it is crucial to foresee opportunities for self-organization, participation, and co-evolution.
In other words, it is not enough to build seemingly optimal cities, in which humans would live like in a zoo with 24/7 surveillance delivering the data required to run smart cities. It is important that Digital Twins empower us to co-create the future, thereby, turning the cities we live in into “our” cities. This will be, in fact, the subject of an upcoming workshop at the Complexity Science Hub Vienna starting September 11, 2023.
Link to Nature publication: The
Role of Complexity for Digital Twins of Cities
Link to Vienna workshop:
