I already did mention the importance of VVUQ when considering operational digital twins.
More elaborated is the work from teh ETH Zurich team (Dirk Helbing & Javier Argota Sánchez-Vaquerizo), Digital Twins: Potentials, Limitations, and Ethical Issues.
Rather than aiming for perfect digital twins, a predictable future, and total control, one should use computer simulation technology to create better opportunities, e.g. for digital assistance of creativity, innovation, self-organization, coordination, cooperation, and co-evolution.
The Web 3.0 now allows us to develop entirely new solutions that are based on a distributed, flexible adaptation to local needs, on digitally assisted selforganization, and on co-evolution. While this may come with less predictability and control, it is expected to improve sustainability and carrying capacity, quality of life, prosperity, and peace.
They summarize the chapter with 12 statements on digital twins
- On Data: It has become an attractive idea to create digital twins of everything, including the Earth, climate, and the human body. While the benefits of this approach may be huge, there are limits. All in all, one must realize that a data science rather than a merely data-driven approach is needed, which require sharing a lot more data with a lot more people.
- On Complexity: Creating an accurate digital twin for infrastructures, which change little over time, is easy. However, it will probably never be possible to produce an exact digital twin of life on Earth, even if nanotechnology is being used on a large scale. One is faced with fundamental challenges and
measurement limits when models of complex dynamical systems are built, for example, of weather, climate, life, behavior, or health, so one needs to be prepared for uncertainty.
- On Machine Learning: The biggest publicly known modern machine learning models try to learn a trillion parameters or so. Unpublished corporate, governmental, or military models may contain even more parameters. While this is impressive, more predictive power is often achieved by simpler models (think of “over-fitting”). Surprisingly, noisy or little data can sometimes generate better models. But no matter how many variables are being considered, there are many orders of magnitudes of interaction effects that are not captured, hence neglected. This can produce a wrong picture and bad forecasts, which can be dangerous.
- On Artificial Intelligence: So far, Big Data has not been able to replace science, nor do we have a universal AI. Even if we had one, this could still be dangerous, and particularly if not retaining meaningful human control. Suppose, for example, one would task an intelligent system to solve the sustainability problems of the Earth or to maximize planetary health. This might result in depopulation and trigger an apocalyptic scenario, even though a better future for everyone might exist. Moreover, as many of today’s AI systems operate like “black boxes”, one might not even realize some of the harmful effects AI systems are causing.
- On Optimization: The concept of “optimizing the world” is highly problematic because there is no science that could tell us what is the right goal function to choose: should it be GDP per capita or sustainability, life expectancy, health, or quality of life? The problem is that optimization tries to map the complexity of the world to a one-dimensional function. This leads to gross oversimplifications and to the neglect of secondary goals, which is likely to cause other problems in the future. Using (co-)evolutionary approaches would probably be better than optimizing for one goal function. Coordination approaches may be more successful than control approaches.
- On Qualities: A largely data-driven society is expected to perform poorly with regard to hardly measurable qualities that humans care about. This includes freedom, love, and creativity, meaning, dignity, and culture, in short: quality of life…
- On Innovation: Something like a “digital crystal ball” is unlikely to see disruptive innovations, which are not included in the data of the past. Hence, predictions could be too pessimistic and misleading. For example, consider the forecast of the world population. According to some future projections, about one-third of the world’s population is sometimes claimed to be “overpopulation”. Consequently, these people may get in trouble when managed by the system. However, such projections do not sufficiently consider alternative forms of running our economy. Probably, “overpopulation” is not the main problem, but lack of economic (re-)organization.
- Humans vs. Things: In a highly networked, complex world, where almost everything has feedback, side, or cascading effects, ethical challenges abound. For example, people should not be managed like things. In times where many argue with “trolley problems” and “lesser evils”, if there is just a big enough disaster, problem, or threat, any ethical principle or law may be overruled, including human rights and even the right to life. Such developments can end with crimes against humanity.
- On Dual Use: A powerful tool, particularly when applied on a global scale, may cause serious, large-scale damage. It is, therefore, necessary to map out undesired side effects of technologies and their use. Effective measures must be taken to prevent large-scale accidents and dual use. Among others, this calls for decentralized data storage and distributed control. Moreover, transparency of and accountability for the use of data and algorithms must be dramatically improved.
- On Alternatives: We should carefully consider alternative uses of technology. Here, we would just like to mention the idea of creating a socio-ecological finance system: a finance system, which would use the Internet of Things to measure externalities that decisions of people and companies cause. The measurement of externalities would define multiple new currencies, which could locally incentivize positive behavioral change. This novel real-time feedback and coordination system is inspired by nature. Nature has already managed to develop a circular economy based on self-organization and distributed control. Hence, introducing real-time feedback into our socio-economic system could create forces promoting a sustainable re-organization. A sustainable circular and sharing economy would result through a co-evolutionary process. It would be a system consistent with freedom, self-determination, creativity, and innovation, with human rights and democracy. This is probably the best path to sustainability currently known.
- On Governance: As people are increasingly an integral part of socio-technical systems, a technology-driven approach and technological innovation are not enough. We first and foremost need social innovation to unlock the benefits of the digital age for everyone. A platform supporting true informational selfdetermination is urgently needed. Moreover, the classical war room approach needs to be replaced by a peace room approach, which requires, among others, an interdisciplinary, ethical, multi-perspective approach, in other words, a new multi-stakeholder approach to achieve better insights and participatory resilience.
- In Conclusion: Societies are not machines, and optimization is a too narrow approach to manage them. It is, therefore, important to recognize that complexity is an opportunity for new kinds of solutions, not “the enemy”.
Planning should be increasingly replaced by flexible adaptation, optimization by co-evolution, and control by coordination. Obviously, all of this can be supported by digital assistance, if used wisely and well.
One response to “Digital Twins: Societies are not machines, & optimization is a too narrow approach”
[…] concepts attempt to build these consideration, but the path to the engineering and realisation of real complex systems is still […]