Before going into a detailed discussion about the Earth4All-Global model, we would like to make some general remarks that hold true for any (type of) model. In general computer models are tools that make explicit the modelers’ mental models and allow them to run experiments within this. While the famous saying by George Box reads “All models are wrong but some are useful”, it has been widely acknowledged that actually all models might be useful for their specific purpose. All models are idealizations of reality, and this often provides a lot of room for criticism. Criticisms are welcome and often a great source of insight in itself as it opens up the debate about different mental models of the systems in question and the underlying assumptions. In fact, some of the most famous and yet most criticized models have not only provided valuable insights into how our world might work but also shown us why some simplifying assumptions about it might lead us astray (e.g. the DICE model by Nordhaus). 

 Donella Meadows (1982, 2008) wrote several critiques/reflections about models and their role in policy. One of the main conclusions was that models should be designed for a particular purpose/to answer a particular question. This is also what defines the boundaries (i.e. what variables to include, what relationships to include etc) for that specific model. 

Some questions about the quality of the Earth4All-Global-model do not take into account what this global simulation model was made to do. The model was not made to do what many of its critics apparently expect it to do. No model can do what some critics seem to expect Earth4All-global to do – namely to forecast the future of the real world with “scientific” precision.  

What can and does the Earth4All-global model say about humanity’s long-term future? The quick answer is “very little” but also “much more than nothing”. The model illustrates the story of the Earth4All book by simulating two scenarios resulting from the model’s structure and input parameters. Together with the assessments of the 21st-century Transformational Economics Commission (TEC) and the storylines that were outlined in the book, the model paints two consistent but contrasting pictures of potential futures with broad pen strokes. 

1. This model is a global integrated systems model. It simulates the development over time of human wellbeing towards 2100 given the model’s structure – focused on trends in the wellbeing of the global majority. 
2. The Earth4All-global model is a very rough representation of the real-world system, formulated at a very high level of aggregation. Hence it is incapable of saying anything about local detail. Except in the form of informal deduction from global developments in the model system. This is why an additional regional model has been developed (Earth4All-regional model to be released in 2024). 
3. In principle any model can only say something rigorously about what happens in the model system. The relevance of such statements to the real world depends on whether the model system is similar (in the eyes of its users) to the real system in the aspects that are assessed. 
4. The Earth4All-global model can help keep foresight thinking straight. That is by ensuring that scenarios are internally consistent, both in space and time 
5. The Earth4All-global model can help clarify future trends – especially in the business as-usual (BAU) scenario. That is, it can try to describe what is likely to happen if decision making continues to follow the pattern from the last forty years as represented in the model. For example, what will happen if decision making in the energy field leads to continued decline in energy intensity by the historical 1 % per year. 
6. The Earth4All-global model can provide rough quantitative estimates of central variables in BAU based on the model’s structure. That is, it can give back-of-the envelope sketches of the order of magnitude of central variables, and their rate of change. Such estimates depend (of course) on both the specification of structure and choice of parameter values. Hence it is important to be open about the most important assumption. For example, in Earth4All we connect the estimated values of future population, GDP, energy use, emissions, etc to the underlying assumptions in the form of births, deaths, GDP growth rate, energy use, and emissions as functions of GDP per person (derived from extensive calibration with historical data (1980-2020) achieving statistically significant goodness of fit measures). 
7. The Earth4All-global model can demonstrate the sensitivity of the BAU scenario to variation in parameters and structure. That is, it can help clarify the effect of change in policy, structure and parameters. For example, the effect of experimentally introducing additional structural cross links can help to explore the magnitude of a (new) causal relationship or driver. Earth4All-global can also be used to evaluate different ‘what-if’ scenarios by changing parameters and assessing how trajectories change given the model structure.

No, the Earth4All-global model is made to “help increase broad understanding of the system of interest” (structuralist school). It is not the type of models that are made to “forecast the long-term future with high precision” (Clancy et al 2023) 

There are important differences between building “models to help policy making/help change the world” versus building “models that represent specific scientific findings”. As a result, some model builders (like us) deliberately choose to downplay quantitative precision and detail, to focus on providing an overarching understanding of the system’s main characteristics, namely its structure and behaviour. See Donella Meadow’s Thinking in Systems (2008) for an extensive discussion on the choices of both granularity and system boundary of a model. 

No, quite the opposite. All models are idealized representations that are built to answer very specific questions. The perfect all-round model of the whole wide world does not exist.  

Since people see the world differently, model builders (who also are people) differ fiercely in what they view as the best model for a given purpose. The root cause of such disagreements is the fact that all social large systems are so complex that it requires (sweeping, heroic) simplifications to avoid ending up with an impenetrable “black box” – streams of code producing voluminous outputs that may be comprehensive or exhaustive but cannot be understood, by people. As the map can never be the territory, a model can never be reality.  

Consequently, models should not be used for purposes other than the one they were made to elucidate. And they should certainly not be run further in time than their designed time horizon. 

The main message of the Earth for All book is that implementing five turnarounds (eliminate poverty, reduce inequality, increase opportunity, switch to regenerative energy and food production) will improve the wellbeing of the global majority – when wellbeing is defined as a combination of a) disposable income, b) public services, c) inequality, d) environmental quality, and e) social tension. You don’t need a model to present and defend this conclusion. And clearly the priorities and conclusions will depend on how you weigh the components. One should not trust black-box models, whether system dynamics, equilibrium models, large language models or big-data models, so much that one starts to believe in conclusions that cannot be explained and defended in plain arguments, words and reasoning. 

 Models may help us to spot counter-intuitive connections, see systemic patterns more clearly, and broaden our understanding of the world. But models can never replace values or conversations for guiding policies and common actions to achieve agreed solutions. In sum, the Earth4All-global model illustrates the two scenarios by giving them quantitative backing but does not provide details on how the world would behave. 

Simulation models can only provide if-then answers: “if we assume this structure and these parameters, then this time development will result”. So strictly speaking, a simulation model can only say something about the development of the model system. Sadly, it is not possible to “prove” that the time development of the model system says something about the time development of the real world. All one can do is to highlight the similarities between model and reality. It helps if the model is capable of reproducing history, if its cause-and-effect links are intuitively plausible and have empirical and theoretical backing, and if the set of assumptions form a logically consistent totality. But none of these prove that “the model is right”. “All models are wrong, but some are useful”, (George Box).