Introduction

In July 2022 we entered the international year of basic science for sustainable development (IYBSSD). There are two ways in which basic sciences can play a crucial role in attaining a more sustainable planet. The first has to do with applying the wealth of knowledge we have accumulated in basic sciences so far, to issues relevant to sustainability. The second focuses on addressing what we still do not understand about sustainability. The first is the most direct use of basic sciences for sustainability and it is currently carried out within well defined disciplinary boundaries (e.g., physics and climate change, material science for energy efficiency, etc.).

As argued long ago [1], the sustainability crisis arises from the incompatibility of the expansion of human activity with a finite planet. Increasing the efficiency of key processes of human societies, such as e.g., more efficient ways of producing energy, can help mitigate the forthcoming crisis. This pushes further the limits under which human activity can continue expanding without impairing global stability but it may not remove the fundamental problem. It is likely that ultimately sustainability will entail a change in our patterns of behavior, based on the understanding of the limits we and our planet face. Such an understanding needs to acknowledge the interdependencies between the different systems that our planet harbors: the atmosphere and the oceans, ecosystems, economies, societies and cultures, technological infrastructures, the financial system, etc. Hence, a comprehensive approach to sustainability requires the integration of hard with soft sciences, and in particular interdisciplinary dialogue across many disciplines. In brief, sustainability issues often entail highly interdependent systems, each of which is often addressed only within strict disciplinary boundaries. For example, climate change affects soil microbiomes, water resources, marine and terrestrial ecosystems, which in turn affect our consumption patterns, our economies, our health and lead to demographic changes, such as forced migrations if not wars. Such changes in the human dimension, in turn, are going to affect ecosystems and the resource-base on which the subsistence of human societies rely, and ultimately climate itself. Any prediction, not only of what will happen, but also of the systemic stability of the biosphere, requires the integration of knowledge which is currently fragmented into many disciplines.

We believe that two aspects are key in addressing these issues: first the inclusion of the human dimension with the dynamics of physical (e.g., climate) and natural (e.g., ecosystems) systems. Second, an approach based on empirical data and rooted in the basic sciences. We believe indeed that mathematics and quantitative analysis can become the shared language that may foster interdisciplinary dialogue and promote a consensus on potentially divisive issues.

In what follows we shall explore three different modeling approaches that we believe will be important in a scientific approach to what we refer to as Quantitative Human Ecology, i.e., a complex planet system where the human dimension is key. We shall start from conceptual models, aimed at capturing the qualitative behavior of emergent phenomena, then we shall discuss more detailed data-driven computational approaches and finally machine learning approaches. We shall conclude with some final remarks.

Conceptual Models

It has been reported [2] that the production of bio-fuels, promoted with the best intentions in order to mitigate climate change, “have forced global food prices up by 75%” [2], causing the 2008 food crisis. The unfettered expansion of credit derivatives, introduced in order to provide more instruments for risk management, has likely made the world riskier [3], leading to the 2007-08 financial crisis and to the following decade of economic recession.

Unintended consequences such as this arise from the neglect of systemic effects of man-made policies. Developing models that can relate “micro-motives to macro-behavior” can provide key insights in this direction. In 1969 Robert Shelling showed by a very simple model that even mild preferences for homophily can lead to large scale segregation in residential districts [4]. This example shows how extremely simplified models can capture, though in an admittedly stylized way, the interactions and non-linearities which are responsible for the rich phenomenology observed at the collective level. This modeling strategy relies on the fact that collective behavior is ultimately governed by statistical laws, which are rather insensitive to microscopic details [5]. The theories developed in the last decades in statistical physics and complex systems science provides a toolkit of methods that can be deployed to address challenges related to sustainability issues. There are already countless examples along this line of research. An exhaustive review is beyond the scope of this contribution. Yet it might be worth mentioning few examples, from our experience, in order to highlight few aspects.

Jerico et al. [6] addressed the relation between inequality and economic growth in a very stylised model of an exchange economy. They assumed a given wealth distribution and observed that larger levels of inequality result in slower economic activity, as observed in data from the US economy. Even though highly stylised, these result show a clear causal link between inequality and growth that arises for purely entropic reasons. This does not exclude causal links in the opposite direction, but it suggests that the direct link is sufficient to reproduce the empirically observed behavior. Furthermore, this approach also clarifies when inequality becomes untolerable, which is when the economy freezes. Bardoscia et al. [7] investigate the relation between inequality and social mobility, within a simple network model of a society. They show that when individuals intensely seek centrality in a social network, the society develops a strong hierarchical structure, characterised by dramatically reduced social mobility. The model may shed some light also on the observed relation between income inequality and status anxiety [8].

Stylised models have been particularly successful in shedding light on mechanisms at the origin of complex phenomena and global stability in finance. The Minority Game, for example, suggests that anomalous fluctuations in financial markets arise precisely when markets become informationally efficient [9]. Scaling up the insights of global games [10] to the systems level, Anand et al. [11] show that the phenomenology of the 2008 financial crisis can be captured in a simple model, shedding light on determinants and possible policy measures.

Toy models are the first step in a scientific approach to complex phenomena. They can expose non-trivial mechanisms at the origin of complex phenomena and inform further empirical research or more computationally demanding modelling approaches. Because of their simplicity and transparency on micro-foundations, they can help inform and discipline policy debate.

Another advantage of conceptual models is that they allow to explore scenarios in context where data are scarce or not existing at all.

Despite their intrinsic simplicity, toy models can reveal the existence of abrupt transitions and catastrophic shifts that are impossible to predict simply by data extrapolation. One paradigmatic examples is given by regime shifts in ecology. Relatively simple (non-linear) models describing the abundance of a biological population predict the existence of saddle-node bifurcations and alternative stable states, leading to catastrophic shifts [12]. By analogy, such behavior can potentially occur at the level of the biosphere [13], implying that a planetary scale tipping point, which a catastrophic loss of biodiversity and ecosystems services, could occur even in presence of a moderately paced deterioration of the environment.

The central goal of conceptual models is therefore not to provide accurate forecasts, but rather to define a “phase diagram”: reduce the uncountable infinity of possible worlds into a finite discrete set of qualitative scenarios. In the context of sustainability, the goal of conceptual models is not to predict accurately the loss of ecosystem services and human demography under certain global warming scenario. The main goal is to define under what conditions the planet and its human population can survive in a way analogous to the present one.

Clearly this kind of questions require the integration in the same model of multiple factors, many of which are very distant from having being described with any for of data. For instance, in [14] the authors integrate in the same conceptual model human population growth, ecosystem services and technological innovation. The relatively simple model show very different outcomes of population growth and level of technological innovation, depending on two critical parameters which define the rate of technological innovation and the feedback of technology on the environment. While this approach does not predict (or even try to) what is going to happen in the next 100 years, it identifies what could be the relevant observable we should pay attention to and the data we should aim at collecting. In this context, one relevant bottleneck for sustainability science, appears to be our understanding of the dynamics of technological innovation and its feedback on the biosphere.

Data-Driven Computational Models: Network Science

Network science is an interdisciplinary field studying complex systems through their representation as a set of distinct elements–usually a large number–and a set of connections between them [15]. In the last decades, the availability of large datasets has paved the way for data-driven computational approaches that allowed network models to gain realism and explain the patterns found in empirical networks [16]. Given its focus on interdependencies and emergent phenomena arising from them, network science is well suited to study sustainability as a complex problem entangling environmental and socioeconomic aspects.

The study of social and economic phenomena through the lenses of network analysis has a long tradition dating back to the 1930s [17]. Network approaches to study the natural and physical world are somewhat more recent, spanning from ecology [18, 19] to climate [20, 21]. The two aspects have been considered separately for a long time, but the increasing interest in the complex interplay between humans and the environment has eventually lead to the investigation of social-ecological networks [22] as well as of “networks of climate change” describing the interplay between natural and anthropogenic processes [23].

Different aspects of sustainable development have been addressed using the tools of network science. Examples include network-based interventions to optimize the diffusion of information about poverty-reducing programs and their uptake [24], the impact of globalization on the resilience of the food supply chain [25], the global spatio-temporal patterns of human migration [26, 27] and of immigrant community integration in world cities [28], and the impact of socioeconomic inequalities and environmental factors such as climate change on the emergence and spread of infectious diseases [29,30,31].

Cities have been a natural focus of network-based sustainability studies, since they are an exemplary model of networked complex systems in which humans interact with the physical space [32]. In this context, the ultimate goal is to design environmentally sustainable and socially equitable human settlements [33]. Thanks to the current availability of new data such as high resolution satellite imagery, cell phones metadata and GPS traces, this goal can be achieved with the help of mathematical analysis of street and building networks and the development of service optimization algorithms, specifically to solve the problem of informal urban settlements (e.g., slums), lacking services and facing poverty, health and environmental degradation challenges [33]. Examples of network-based approaches for urban sustainability include optimization of car sharing to reduce costs and emissions [34], of bicycle [35] and sidewalk [36] networks to improve the cycling and pedestrian infrastructure, and of facility distribution over the road network to reduce travel costs [37].

Finally, game theory research studying social cooperation to manage common-pool resources and achieve environmental goals should be mentioned. Specifically, a few studies have focused on the influence of the network structure on the cooperation dynamics, studying the role of social norms and showing that the emergence of cooperation is only possible in specific settings [38,39,40].

Machine Learning

The last two decades have seen an exponential increase in the importance of data-driven technologies in every field of human endeavour, with a transformative impact in areas as diverse as business and astrophysics. Conceptually, the reasons at the roots of this phenomenon are simple: data science technologies can derive easy-to use approximations of complex systems by simply interpolating across very large instances in a data set. In this sense, algorithms can provide useful shortcuts in cases where the complexity of a system prevents the discovery of causal mechanisms, substituting prediction for understanding (which may be practically acceptable in some cases). A similar approach can in principle also be adopted for addressing questions of sustainability; however, at present the role and potential of data science as a force for good in the efforts for an equitable development is controversial, particularly concerning the subfields of machine learning and artificial intelligence (AI).

A recent study focused on the potential of AI to help the attainment of the sustainable development goals through a series of interviews with AI practitioners [41]. The overall picture described a significant potential towards a positive impact; however, as the authors themselves acknowledge, the chosen focus group (AI practitioners) might have led to an over-positive perception. Additionally, and perhaps more importantly, the focus of the study was the potential for positive impact, which may well be distinct from the actual impact of the current usage patterns of the technology.

When discussing the impact of AI and Data Science more in general on sustainable development, a few home truths cannot be avoided. First of all, as forcefully argued in [42], the environmental costs of AI systems are punishing, from the high energetic requirements of data centres and supercomputers, to the mining of the rare materials needed for processors. Secondly, current AI systems are strongly susceptible to embedding human biases in black-box decision-support tools, with the potential to seriously hamper efforts towards a more equal society, for example in the context of gender and minority rights [43]. Finally, and perhaps most importantly, the vast accumulation of data, and the related development of advanced analytics, by a handful of state and non-state actors has amplified already large economic and power imbalances, contributing to an exacerbation of inequality and the potential for significant distortion of the political discourse [44, 45]. All of these are well documented facts, pointing to the serious societal and environmental consequences of an unregulated harvesting of data and deployment of AI.

On the other hand, multiple examples exist of applications of machine learning and AI technologies that concretely point towards its potential for achieving the SDG goals. Familiar examples include the use of machine learning techniques in health: while ethical concerns need to be addressed [46, 47], concrete results in fields such as early detection of retinopathy [48] already demonstrate the usefulness of the technology, although the vision of precision medicine might be oversold and certainly remains distant [49]. Another prominent example in a completely different field is the deployment of adaptive traffic signalling in the city of Pittsburgh: based on an application of planning and multi-agents systems [50], the deployment of the Surtrac system has already led to a 21% reduction in traffic carbon emissions in the city, and a 26% reduction in journey times.

Alongside these direct applications of AI technology, a major indirect role is emerging as a tool to extract difficult to access information from indirect measurements. A prominent example is the use of night-light measurements, readily available from satellite imaging, to estimate levels of poverty in rural communities [51], a type of data that is otherwise expensive and difficult to obtain. In general, automated processing of satellite images is rapidly emerging as a key tool to obtain reliable and inexpensive estimates of parameters such as land and resource usage and air quality, for the purposes of informing policy and monitoring compliance with environmental regulations [52,53,54,55].

Challenges

The challenge of integrating the feedback between human actions, biosphere, and climate in a unique predictive framework is daunting and will likely keep us busy for decades. At present, we lack an established quantitative, data-controlled framework for many factors. For instance, we cannot quantitatively predict how ecosystem services function (e.g., how much nitrous oxide is emitted by microbes performing nitrogen fixation) depends on abiotic factors (e.g., temperature or precipitation) since we lack an understanding of how community composition impact ecosystem services (e.g., how the change in soil microbial communities due to temperature affects the emissions of nitrous oxide). Similarly, it is extremely hard to predict human collective decisions (e.g., migration patterns or demographic changes) over decades, which will have a strong impact on human societies and the biosphere.

While this level of knowledge is still lacking, much progress has been made. Further investment in these problems can produce a significant increase in our predictive power and conceptual understanding, because they may allow us to test models that attempt to capture the inter-dependencies between different factors thanks to the availability of large data-sets. Basic sciences will play a key role in this.

Such efforts should proceed across a wide range of quantitative approaches, ranging from the highly stylized conceptual models to fully data-driven forecasting tools. This range of approaches can be characterized along many axes: simple versus complex, parameter-poor versus parameter rich, low- versus high- dimensional, conceptual versus predictive, etc. Both of the approaches (and all the ones in between) are obviously characterized by their own challenges. Identifying synergistic approaches, to combine data-driven tools with low-dimensional stylized descriptions, is a challenge in its own, which could present great opportunities for addressing the issues related to sustainability.

The expertise in connecting the dynamics at the micro-scale to that of the meso- and macro-scale, which has been developed in theoretical physics and complex systems science, is an indispensable asset in this endeavor. This can also shed light on the scale (e.g., communities, cities, regions, states or the world) at which policies can be more effective in particular system. Identifying robust stylised facts and developing models that can reproduce them in the simplest possible way, requires however to have the right “intuition” about which factors and processes can be neglected in the description. Such a-priori choices are hard to formalize and can only proceed by trial and error.

On the other hand, data-driven methods can aid the development of such models in at least two complementary ways. First of all, data driven methods can be used as a “shortcut” to approximate complex functions directly from data: at a very low level, this is what is already happening when satellite images are transformed from a collection of pixels to annotations about land use through a classifier such as a neural network. Conceivably, such methods could be used in many other situations where data needs to be summarised efficiently and no a priori model is available. A second major potential use of data science techniques is in the development of rigorous statistical methods to test the validity of models, for example through the use of Bayesian model selection or model criticism techniques [56]. Such approaches can provide firm statistical foundations to the “intuition” used in stylised models.

An alternative, potentially fruitful synergistic way of combining the two approaches would be to use stylized facts and models as constraints for data-driven tools. For instance, simple and regular laws emerge in the patterns of human mobility and migration. While these patterns do not capture the heterogeneity of how individuals take decisions, they encapsulate statistical constraints which appears at the collective level. Such collective patterns could be put as statistical constraints of ML algorithms.

In brief, the Sustainable Development Goals demand not only the application of our current know-how to address specific challenges, but also the expansion of our scientific base and its integration. We strongly believe that the growth and consolidation of “Sustainability Science” will be one of the main trends in the next decades. Besides initiating research in-house along these lines, we also believe that it will be important to promote interdisciplinary dialogue by international workshops on key aspects of sustainability, that can be addressed by different angles.

There is a growing trend of interdisciplinary research along these lines. Yet the community is rather fragmented, and we believe more efforts has to be put in reaching a critical mass. For example, departments that can host researchers that venture in these domains are rare, as well as cross-disciplinary funding opportunities. This renders a career path in these areas much less well defined than the one of a researcher that invests within traditional disciplinary boundaries.

We are at the beginning of a long journey. Ultimately, we believe, the real challenge will be to develop programmes to train a new generation of scientists, endowing them with a background in basic sciences that fosters interdisciplinary dialogue and a curiosity driven approach to sustainability science.