Agent-based modeling spurs demographic innovation. Photo: TUCHONG

Artificial Intelligence (AI) creates systems with intelligent capabilities by simulating human learning, calculation, reasoning, and thinking patterns, enabling them to perform tasks once possible only with human intelligence. Demography, the science of human populations, focuses on quantitative characteristics and emphasizes statistical research, providing numerical descriptions or explanations of population factors at the group level. According to the evolutionary stages of scientific research paradigms, demography entered the second paradigm—centered on mathematical models—as early as the 19th century. Since then, however, the paradigm has not undergone any major breakthroughs. In recent years, computational demography, which advocates the use of digital and computational approaches to study human behavior, has seen some progress. Research in geography and economics that incorporates demographic issues has also advanced population studies through data-driven methods, using data and computing power to explore frontier questions. Yet these approaches have never become mainstream within demography. Building on advances in algorithms, data, and computing power, AI enables the co-evolution of cognition, behavior, and knowledge innovation, driving the transformation of scientific knowledge under a new research paradigm.

Status quo of demographic research paradigms

The evolution of research paradigms reflects progress in scientific stages and innovation in research methods. Before the 16th century, empirical science—based on observation, recording, description, experimentation, induction, and summary—was considered the first paradigm. The second, theoretical science, relied on hypothesis testing, model building, and deductive reasoning. In the mid-20th century, with the development of computing technologies, the third paradigm—computational science—emerged, using simulation and quantitative analysis as common research tools. Traditional demographic research paradigms can be roughly divided into two main lines: demographic statistics (normative demography), which focuses on concepts, indicators, and statistical methods; and demographic research, which emphasizes the interrelationships and dynamic mechanisms between internal elements of the population system and external natural, social, and economic factors. Both have largely remained within the framework of the second paradigm.

Mainstream domestic demographic journals have published only a small number of studies using computational simulation, which remain marginal compared with quantitative empirical methods. Internationally, many researchers have applied computational simulations to demographic questions, even integrating them with AI, though these works appear primarily in related fields such as economics and medicine. Current applications of simulation methods in demography generally fall into two categories: the innovation and application of demographic statistical models under the paradigm of normative demography, and the embedding of population models into social, economic, resource, or environmental models under the demographic research paradigm. The first approach constructs predictive models based on demographic statistics, combines them with external models, and calculates how different demographic process parameters affect external systems. In essence, it merges normative demography with broader demographic research, though the external models involved are often relatively simple. The second approach embeds population-related variables into relatively large, multi-factor models—such as endogenous growth or input-output models—to analyze how demographic changes affect the overall system. Here, demographic statistical analysis tends to be relatively limited, often introduced only as exogenous variables.

Such studies cannot systematically capture the dynamic mechanisms and processes of interaction between populations and their external environment, making them difficult to integrate into complex system modeling. Contemporary population science must therefore investigate how population systems interact with various social, economic, resource, environmental, and other subsystems.

Necessity of complex system modeling and simulation in demography

First, “computationalization” aligns with the developmental logic of demographic research paradigms. In the humanities and social sciences, paradigms have accordingly evolved from qualitative to quantitative and now to computational modes. Complex system modeling allows researchers to simulate complex phenomena, trace their underlying causes, and project future trajectories.

Second, innovation in methodology is essential for any discipline to advance to a new research paradigm. Simulation and emulation research in population studies involve both the dynamic mechanisms of various internal elements of population systems and their interactions with external systems. This requires the establishment of a brand-new methodological system—one that is compatible with population statistics and analysis methods, allowing the internal dynamic mechanisms of the population system obtained through population statistics to be incorporated into complex systems. It should also be consistent with existing population research—mainly based on quantitative empirical methods—to study, analyze, and verify the connections between population factors and other external factors.

Finally, complex system modeling and simulation are critical for cross-disciplinary integration. As one of the most prominent interdisciplinary approaches in computational research, complex system modeling draws on multiple fields. Interdisciplinarity itself is a defining feature of scientific innovation in the AI era. Since the rapid rise of computational social science, disciplines such as economics, management, and geography have all entered the “computational” era. Integrating theories, methods, and tools across disciplines to build comprehensive complex simulation and emulation systems has become a hotspot in the field of interdisciplinary integration.

AI-empowered ‘population-nature-society’ system

The “population–nature–society” complex system involves multiple disciplines such as demography, economics, management, and computational science. It is characterized by the integration of multi-disciplinary theories and methods, large-scale models, complex structures, and the need to process vast amounts of data. AI empowers such systems by meeting demands for knowledge, algorithms, data, and computing power, while also reducing the barriers for demographers to engage in cross-disciplinary modeling, thereby increasing the feasibility of a transformation in demography towards simulation-driven demographic research.

First, AI enhances data acquisition and processing. In the “population-nature-society” complex system, constructing the population system and coupling it with other subsystems requires extensive demographic as well as resource, environmental, and socioeconomic data. By combining AI with rapidly advancing digital technologies—big data, machine learning, the Internet of Things, and cloud computing—researchers can collect, organize, and refine data more efficiently. AI can further generate predictive outputs based on past data and knowledge, accelerating model validation and improving research efficiency. With AI assistance, demographers can integrate big data, machine learning, and other methods into their research, generating real-time, multi-scale, high-precision, low-cost, and information-rich datasets covering populations, resources, the environment, and social economy.

Second, AI facilitates knowledge discovery regarding the dynamic mechanisms linking population and external factors through three pathways. First, it assists with collecting, processing, and integrating existing population research data and results, enabling the “rediscovery” of knowledge about the influence mechanisms and dynamic relationships between population elements and external elements. Second, it assists with simulation and data analysis to generate “new discoveries” of such mechanisms. Third, it empowers computational demography theory and methods, introducing new methodologies into population research and establishing a new set of demography research methodologies, advancing the “computational” transformation of the demography research paradigm.

Finally, AI drives system innovation toward “human–machine symbiosis.” Beyond data processing and knowledge discovery, in the field of complex systems and simulation, AI contributes deeper autonomy and innovation across the entire life cycle of complex systems—from modeling to operation and maintenance. By integrating AI into complex systems, researchers can create systems capable of proposing hypotheses, formulating research questions, actively conducting analysis and validation, and autonomously optimizing and making decisions, fostering a collaborative environment for human–machine decision-making.

AI + ABM: a new direction for complex system modeling

Agent-Based Modeling (ABM) is a new interdisciplinary method that integrates AI, computer science, and knowledge systems, and is one of the most important tools in computational social science. ABM takes a micro-level perspective, employing a bottom-up modeling approach. By defining agents to represent real-world micro-entities, it constructs virtual environments in which computer simulations of numerous micro-level interactions give rise to emergent macro-level dynamics and complexity. This approach helps explain intricate social, economic, and demographic phenomena.

The integration of AI with ABM (“AI + ABM”) opens new possibilities for building “population–nature–society” complex systems in demography. The key is training heterogeneous, human-like AI agents that represent real people—what may be termed “anthropomorphic” training of AI. This process requires drawing on the theories and methods of experimental economics.

The procedure could proceed along the following stages. First, construct a “population–nature–economy” complex system framework and embed AI agents without prior anthropomorphic training. Second, establish a human–machine interaction platform for experimental demography outside the system and link it to the model. Third, guided by the theories and methods of experimental economics, design human–machine experiments in which heterogeneous, real human participants engage in the simulation together with AI, allowing AI systems to learn human behavioral characteristics. Fourth, once anthropomorphic training is complete, disconnect the human-machine interaction platform and replace human participants with large numbers of trained, heterogeneous AI agents in simulations. Fifth, test the robustness of simulation results with AI participation, complete system training, and initiate operation of the “population–nature–society” complex system.

Gu Gaoxiang is an associate professor from the Institute of Population Research at East China Normal University.