Complex systems vs. complex adaptive systems: why the difference matters

1 Introduction

Complexity science is now widely recognized as an interdisciplinary effort to understand systems made up of many interacting parts whose collective behavior cannot be reduced to the properties of those parts alone (Weaver, 1948; von Bertalanffy, 1968; Ashby, 1956; Simon, 1962; Anderson, 1972; Mitchell, 2009; Ladyman et al., 2013; Bar-Yam, 2002). The field draws on long historical threads in general systems theory, cybernetics, nonlinear dynamics, statistical physics, and artificial intelligence (Weaver, 1948; Prigogine and Stengers, 1984; Gell-Mann, 1995; Kauffman, 1993; Pickering, 2010; Minsky, 1986; Rosenblatt, 1958). Within this evolving landscape, the terms complex systems (CS) and complex adaptive systems (CAS) play a central role.

In broad terms, a complex system is any system composed of many interacting components that exhibit nonlinear dynamics, feedback loops, and emergent patterns. Classic examples include turbulent fluids, ecosystems, global climate systems, financial markets, and cities (Ladyman et al., 2013; Bar-Yam, 2002; Mitchell, 2009). A complex adaptive system is more specific: it is a complex system whose components adapt, learn, or update behavior as they interact (Buckley, 1968; Holland, 1992; 2006; Lansing, 2003). In CAS, agents or institutions change their rules, strategies, or internal structure in response to the system’s state and to one another. The immune system, evolving ecosystems, economies, and many social-ecological systems are paradigmatic instances (Levin, 1998; Lansing, 2003; Mitchell, 2009).

Historically, the distinction between CS and CAS developed within academic discourse. Buckley introduced the phrase “complex adaptive system” in sociology in the 1960s (Buckley, 1968). Holland and colleagues at the Santa Fe Institute later elaborated CAS as a unifying framework for systems with adaptive agents, from economies to ecosystems (Holland, 1992; 2006; Waldrop, 1992; Lansing, 2003). In parallel, physicists and mathematicians advanced general theories of complex systems, including chaos, self-organization, and critical phenomena, often without explicit adaptivity (Lorenz, 1963; Anderson, 1972; Prigogine and Stengers, 1984; Gell-Mann, 1995; Ladyman et al., 2013; Bak et al., 1987). This chronology matters. The CAS tradition did not emerge merely as a correction to theories of self-organization or criticality; rather, it developed in parallel around a distinct concern with how agents, populations, or institutions learn, remember, and revise behavior over time (Buckley, 1968; Holland, 1992; 2006).

In the past decade, however, the language of complexity has migrated far beyond specialized research communities. Policymakers, sustainability practitioners, and business strategists now routinely speak of “complexity thinking,” “complex systems approaches,” and “complex adaptive governance.” Complexity is invoked in discussions of climate change, pandemic response, urban planning, financial stability, and polycrises—interacting crises across tightly coupled social, ecological, and institutional systems (de Jager et al., 2025; Nel and Taeihagh, 2024; West, 2018; Steiner et al., 2024; Krakauer, 2024; Raman Lab, University of Chicago, 2025). This diffusion is largely positive: it reflects a growing recognition that many contemporary challenges cannot be understood or managed using only linear, reductionist models.

Yet the success of complexity as a buzzword has also brought semantic confusion. In practice, “complexity,” “complex systems,” and “complex adaptive systems” are often treated as synonyms. A city plan described as “complexity informed” might rely on network maps, causal-loop diagrams, agent-based simulations, or simply a general sense that “everything is connected.” In the resilience literature, terms like “complex systems,” “complex adaptive systems,” and “social-ecological systems” are used in overlapping ways that complicate systematic review and knowledge synthesis (Lade and Peterson, 2019).

For stakeholders outside of complexity science, this ambiguity makes it harder to know what exactly is being imported when they embrace complexity thinking. Consulting and strategy outlets aimed at executives now routinely package “complexity” as a managerial imperative: major firms market “complexity management” services and publish guidance on the costs and benefits of organizational complexity (Bain and Company, 2026; Reeves et al., 2019). Again, this signals a welcome recognition of the importance of complexity science in addressing real-world problems—but it also amplifies semantic drift. In practice, “complexity” can mean anything from interconnectedness to organizational bloat, without specifying whether the system under discussion is merely complex or truly adaptive. The result is that stakeholders may adopt tools presented as complexity-based while importing hidden assumptions (e.g., fixed rules vs. adaptive learning), reinforcing the case for explicit CS/CAS distinctions.

In this article we use “complexity thinking” in a broad, lower-case sense to refer to the uptake of complexity-science concepts in applied research, policy, and management. This usage overlaps with, but is not identical to, Edgar Morin’s “Complex Thought,” which is a broader epistemological project concerned with contextual, relational, and non-reductionist modes of inquiry (Morin, 2008). Our focus here is narrower and more pragmatic: when practitioners invoke complexity thinking, one key question is whether the system under study is being treated as complex in general or as specifically adaptive.

The core claim of this Perspective is simple: for the purposes of applied complexity, it is useful to distinguish between complex systems in general and complex adaptive systems in particular. All CAS are complex systems, but not all complex systems are adaptive. Adaptivity—the ability of components or the system as a whole to change in response to experience—adds a layer of dynamics that has practical consequences for modeling, evidence, and intervention. Our goal is not to propose a new taxonomy, but to clarify this distinction and explain why it matters for those who use complexity ideas in policy, planning, and management. We therefore frame the CS/CAS distinction in terms of practical consequences for stakeholders selecting models and interventions.

We proceed in three steps. First, we outline working definitions of complex systems and complex adaptive systems and provide illustrative examples. Second, we summarize the main contrasts between CS and CAS in a concise comparison table and discuss why these contrasts matter for choosing modeling tools, especially system dynamics versus agent-based modeling. Third, we highlight implications for the way complexity is adopted in applied fields and for future research and funding in complexity science.

2 Complex systems: interaction and emergence

Complex systems are typically defined as systems composed of many interacting components whose aggregate behavior cannot be understood by analyzing the parts in isolation (Weaver, 1948; Simon, 1962; Ladyman et al., 2013; Bar-Yam, 2002). Their behavior is shaped by nonlinear interactions and feedback loops: small changes in one part of the system can have disproportionately large effects elsewhere, or vice versa (Lorenz, 1963; Anderson, 1972). Such systems often operate far from equilibrium and exhibit emergent patterns—coherent structures or behaviors that arise from local interactions but are not properties of individual components (Prigogine and Stengers, 1984; Gell-Mann, 1995; Mitchell, 2009).

This notion of complexity contrasts with systems that are merely complicated. A commercial airliner is complicated but, in principle, can be fully decomposed into subsystems whose behavior is specified by design. By contrast, a turbulent fluid, Rayleigh–Bénard convection, a power grid under fluctuating load, or the global climate cannot be fully understood by considering components independently: feedbacks, nonlinearities, and stochastic influences produce behavior that is “more than the sum of the parts” (Weaver, 1948; Ladyman et al., 2013). The same holds for many biological and social systems when considered at appropriate scales.

Importantly, complex systems do not have to be adaptive. Many paradigmatic complex systems in physics and engineering involve components that follow fixed rules and do not learn. Molecules in a fluid, stars in a galaxy, or components in an electrical grid interact in intricate ways, but the rules of interaction are specified by physical laws and engineering design. The system can show unpredictability, phase transitions, or critical phenomena, but the components themselves do not change their behavior in response to outcomes.

A canonical example is the weather. Atmospheric dynamics are governed by fluid mechanics and thermodynamics. Interaction among air masses, heat flows, moisture, and topography leads to emergent features such as storms and jet streams. The system is open, nonlinear, and sensitive to initial conditions; forecasting remains challenging despite sophisticated models and data (Lorenz, 1963). Yet neither air parcels nor pressure systems “adapt” in the sense of learning: they do not change their governing rules based on past forecasts or policy decisions.

Typical methods for studying non-adaptive complex systems include differential equations, nonlinear dynamical systems, statistical mechanics, and network theory. System dynamics models—aggregate models of stocks, flows, and feedbacks popularized in the systems thinking community—are especially well suited to systems where we can reasonably treat the interaction rules as fixed over the timescale of interest, even if the resulting behavior is nonlinear and counterintuitive. In these settings, the main challenge is to understand how feedback and structure generate emergent patterns, not how agents change their behavior.

3 Complex adaptive systems: adding learning and evolution

Complex adaptive systems share all the hallmarks of complex systems—many interacting components, nonlinearity, emergence—but add a crucial feature: adaptivity. Holland defines a CAS as a system “that involves many components that adapt or learn as they interact” (Holland, 1992; 2006). In CAS, components (often called agents) change their state, strategies, or rules in response to experience. Adaptivity typically implies some form of memory or internal structure that can be updated, and often leads to coevolution between agents and their environment (Levin, 1998; Lansing, 2003; Mitchell, 2009).

Examples of CAS abound in the life and social sciences. In an ecosystem, species evolve traits that affect and are affected by the traits of other species and by the abiotic environment. Predator–prey dynamics, mutualisms, and competition can all change over evolutionary time. In an immune system, cells learn to recognize and respond to novel pathogens, and the repertoire of responses is shaped by previous encounters. In economies and financial markets, firms and investors adjust strategies in response to prices, regulations, and each other’s behavior. Cities, too, are complex adaptive systems: households, firms, and institutions adapt to infrastructure, policies, and one another, producing changing patterns of land use, mobility, and inequality (West, 2018; Mitchell, 2009).

Adaptation can occur at multiple, nested levels. An agent at one level (e.g., a person) may itself be a CAS composed of interacting subsystems while simultaneously functioning as a component of a higher-level CAS (e.g., a city). In many biological and social CAS, this adaptivity is hierarchically organized: higher-level structures constrain and enable local behavior even as local interactions generate emergent system-level patterns. The scientific traditions informing complexity science have produced different accounts of how adaptivity initially emerges in such nested CAS hierarchies. Maturana and Varela’s notion of autopoiesis—systems that self-maintain their identity through internal organization and continuous regeneration—captures a biological view of adaptivity (Maturana and Varela, 1980). Cybernetics and control theory likewise studied how negative feedback allows systems to maintain goals and self-regulate, from thermostats to guided missiles (Rosenblueth et al., 1943; Pickering, 2010).

Artificial intelligence and computational models provide synthetic CAS. Minsky framed a “society of mind” as many simple agents combining to produce intelligent behavior (Minsky, 1986). Early neural networks and perceptrons demonstrated how systems of simple units could learn from data (Rosenblatt, 1958). Agent-based models (ABMs) now allow researchers to specify heterogeneous agents with adaptive rules and to explore how macro-level patterns emerge from micro-level interactions (Miller and Page, 2007; Axelrod and Cohen, 2000; Lansing, 2003).

From this Perspective, the key point is that CAS are not just complicated; they respond to intervention. When we intervene in a CAS—for example by introducing a new policy, regulation, or technology—agents respond and may change their behavior in ways that undermine or amplify the intervention. Pathogens evolve resistance, financial actors find loopholes, and communities reorganize in response to planning decisions. That adaptive response is what distinguishes CAS from non-adaptive complex systems and makes them both fascinating and difficult to govern.

4 Why the distinction matters

For complexity researchers, distinguishing CS from CAS may seem obvious. For many practitioners adopting “complexity thinking,” however, the difference often remains implicit or unstated. Yet it has concrete consequences for how we model, how we search the literature, and how we design interventions. Table 1 summarizes the main contrasts, drawing on standard treatments of complex systems and complex adaptive systems (Bar-Yam, 2002; Ladyman et al., 2013; Mitchell, 2009; Holland, 1992; 2006; Lansing, 2003).

TABLE 1. Comparison of complex systems (CS) and complex adaptive systems (CAS). In practice, many real systems combine features of both; the distinction highlights where adaptivity is central to the questions at hand.

4.1 Model choice and misuse

A first reason the distinction matters is that it guides model choice. System dynamics models, for example, are powerful tools for representing nonlinear feedbacks and accumulations at an aggregate level. They are well suited to complex systems whose behavior is dominated by structural feedbacks rather than by explicit agent-level adaptation. Infrastructure networks, resource stock-and-flow systems, and some aspects of climate economics can often be usefully approximated this way.

However, system dynamics models generally do not represent individual agents adapting their rules or strategies over time. When applied to systems where agent heterogeneity and learning are central—such as urban mobility, financial markets, or the spread of misinformation—purely aggregate models can miss key dynamics. An urban planner, for instance, might construct a system dynamics model of vehicle flows and road capacity and conclude that adding a new road will reduce congestion. In reality, drivers adapt their routes, developers change land use, and the new capacity induces additional demand, potentially leaving congestion unchanged or even worse. An agent-based model that explicitly represents adaptive travelers and developers would be better aligned with the system’s adaptive character (Miller and Page, 2007; Axelrod and Cohen, 2000; West, 2018).

The same logic applies in epidemiology and public health. Simple compartmental models of disease spread treat individuals as homogeneous and non-adaptive: they do not change contact patterns or behavior in response to risk or policy (Bansal et al., 2007; Funk et al., 2010). In actual epidemics, people adapt; they reduce contacts, alter compliance, and respond to information. Treating an epidemic as a non-adaptive complex system may be adequate for early, short-term forecasting, but for long-term policy evaluation a CAS perspective—and corresponding models that incorporate behavioral responses—are critical.

4.2 Literature and evidence

A second consequence concerns how we search for and interpret evidence. In their work on resilience in complex systems, Lade and Peterson (2019) found that differences in terminology—including whether papers used “complex systems,” “complex adaptive systems,” or “social-ecological systems”—affected which papers were retrieved in systematic reviews. Important studies of adaptive ecosystems or social-ecological regimes could be overlooked if search terms were restricted to “complex systems” in a narrow sense. Conversely, treating all complex systems as adaptive risks importing results from physical systems into domains where agent behavior fundamentally changes the dynamics.

Explicitly tagging work as CS or CAS, where appropriate, helps build cumulative knowledge. It allows researchers to identify when results from one domain (say, network cascades in power grids) might transfer to another (financial contagion) and when adaptivity makes such transfers more fraught. For practitioners browsing the literature, seeing which models assume fixed rules and which include adaptive agents can help them gauge which findings are relevant to their own context.

4.3 Policy, governance, and funding

A third reason the distinction matters is that complex adaptive systems behave differently under intervention than non-adaptive complex systems. In a non-adaptive complex system such as a power grid, interventions change flows and feedbacks, but the components do not change their rules in response. One useful heuristic is that non-adaptive complex systems react to perturbations according to fixed interaction rules, whereas adaptive systems can respond by altering behavior, strategies, or internal state in light of feedback and experience. We might still see unintended consequences—blackouts or cascading failures—but we do not have to worry about deliberate strategic responses by the system’s elements.

By contrast, most policy-relevant systems are complex and adaptive: economies, cities, governance networks, and social-ecological systems all contain agents that respond to incentives and information (Levin, 1998; Colander and Kupers, 2014; Nel and Taeihagh, 2024; Steiner et al., 2024). As Colander and Kupers argue, complexity-aware public policy should focus less on top-down control and more on creating conditions under which adaptive agents self-organize toward desirable outcomes (Colander and Kupers, 2014). Nel and Taeihagh highlight the “soft underbelly” of complexity science adoption in policymaking: non-technical challenges such as institutional incentives, capacity, and trust can make or break attempts to use complexity-informed tools (Nel and Taeihagh, 2024). Recognizing when we are dealing with CAS rather than non-adaptive CS helps policymakers understand why static, one-off interventions often fail and why iterative, adaptive governance is needed.

Finally, the distinction has implications for research funding. Calls for “complexity science” can attract very different projects: some focused on foundational mathematics of nonlinear systems, others on agent-based modeling of social processes, still others on hybrid approaches. Clarity about whether adaptivity is central to a funding program can help align projects and evaluators. As Raman and colleagues argue, investing in complexity science is essential for addressing contemporary global challenges, but doing so effectively requires understanding the diversity within the field (Raman Lab, University of Chicago, 2025; Krakauer, 2024). Being explicit about CS and CAS is one small but important part of that.

5 Discussion

Complexity science now sits at a critical juncture. On one hand, its concepts and tools are increasingly institutionalized, with dedicated schools, research centers, and journals (San Miguel, 2023; Krakauer, 2023; Lichtenstein, 2025; Krakauer, 2024). On the other hand, the rapid uptake of “complexity thinking” in applied domains risks diluting its meaning. In this Perspective we have argued that distinguishing complex systems from complex adaptive systems is a simple, practical step that can improve how complexity ideas are used.

Our central claims are: (i) complex systems and complex adaptive systems are related but distinct; CAS are the subset of complex systems in which agents or institutions adapt or learn (Holland, 1992; 2006; Ladyman et al., 2013; Lansing, 2003); (ii) this distinction matters for model choice, literature synthesis, policy design, and funding; and (iii) greater conceptual clarity can help practitioners avoid misusing tools and over- or under-estimating what “complexity” can deliver. We have highlighted weather and cities as intuitive examples of non-adaptive and adaptive complex systems, respectively, and frame key contrasts in Table 1.

We emphasize that the CS/CAS distinction should not be read as forcing a rigid either/or ontology or a single preferred method. Many research problems require complementary descriptions: CS models may be appropriate for structural, physical, or network constraints, while CAS models are needed to represent learning, strategic adjustment, or evolution. The distinction is therefore methodological as much as semantic; it helps researchers decide when such complementary or hybrid approaches are necessary.

Looking ahead, we see several directions for deepening this perspective. First, there is a need for more hybrid modeling approaches that combine strengths of system dynamics, network science, and agent-based modeling, especially for socio-technical systems that mix physical infrastructure with adaptive agents (Nguyen et al., 2024; Howick et al., 2024). Second, as large-scale data on human behavior, ecosystems, and infrastructure proliferate, there is an opportunity to test when non-adaptive approximations suffice and when explicitly modeling adaptation is essential. Third, education and training in complexity science—for students, policy analysts, and decision-makers—should explicitly address the CS/CAS distinction and its practical implications.

Most importantly, when we export complexity into applied settings, we should be honest about what is and is not being imported. For some problems, adopting a complex-systems lens means paying attention to feedback, nonlinearity, and emergent patterns in systems whose components do not adapt. For others, especially in life and social domains, complexity thinking must include adaptivity, learning, evolution, and, in some cases, goal-directed, purposive, and self-regulating forms of behavior. Collapsing these cases under a single, vague notion of “complexity” risks disappointment and misdirected effort. By contrast, using the language of complex systems and complex adaptive systems with care can make complexity thinking a more precise, powerful, and trustworthy guide for action.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. Generative AI was used to check the final draft for grammatical and spelling errors, and to identify any narrative redundancies.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Georgios Balasis, National Observatory of Athens, Greece

Reviewed by: Byomkesh Talukder, Florida International University, United States

Srdjan Kesić, University of Belgrade, Serbia