The field, mapped

Organisations and leadership

Stacey et al. (2000) asked whether complexity was a management fad or a radical challenge to systems thinking, and answered that it was the second. Stacey (2001) argued that organisations are not systems that managers stand outside and design. They are complex responsive processes of relating, conversation and power, in which managers are themselves participants. Strategy emerges from local interaction. The formal plan is one input among many.

Cynefin gave practitioners the most widely used complexity heuristic (Kurtz & Snowden, 2003; Snowden & Boone, 2007). It separates clear, complicated, complex and chaotic contexts, plus a state of disorder, each with its own decision logic. In complex contexts cause and effect make sense only in retrospect, so leaders should probe, sense and respond: run parallel safe-to-fail experiments, amplify what works and dampen what does not. Uhl-Bien et al. (2007) reframed leadership as three functions rather than a role: administrative, adaptive, and enabling leadership that manages the tension between the first two.

Health care is the proving ground. Plsek and Greenhalgh (2001) argued that health systems are complex adaptive systems, and that in some settings minimum specifications should replace exhaustive protocols. It remains the most empirically developed sector for complexity-informed improvement, and the closest analogue to education.

Policy and public administration

Rittel and Webber (1973) defined wicked problems: no definitive formulation, no stopping rule, no true-or-false solutions, and no chance to learn by trial and error without consequence. Ackoff (1974) said organisations face “messes”, systems of interacting problems, rather than discrete problems. Both anticipated complexity’s core claim. The problem cannot be separated from the system that defines it.

Cairney (2012) reviewed complexity’s uptake in political science and concluded that its main contribution is to explain why policy outcomes are rarely the ones intended. Policy systems show path dependence, emergent outcomes, and long periods of stability broken by rapid change. Byrne and Callaghan (2014) give the fullest sociological account, and argue for case-based, configurational methods that suit complex causation.

Ramalingam (2013) showed how the linear logic models of international aid fail in complex settings. In domestic public services, the Human Learning Systems movement (Lowe & Plimmer, 2019) argues that outcomes are emergent properties of whole systems, so commissioners should fund learning, relationships and system health rather than buy specified outputs. The OECD (2017) and its work on anticipatory innovation governance (Tõnurist & Hanson, 2020) mark the arrival of these ideas in mainstream public-sector guidance.

Economics

Arthur’s (1989) work on increasing returns showed that small historical accidents can decide which technology dominates. Path dependence became a mechanism rather than a metaphor. Arthur (2015, 2021) generalised this into complexity economics: the economy is in perpetual disequilibrium, agents act on expectations their actions jointly create, and novelty comes from inside the system. Beinhocker (2006) and Farmer (2024) extend the programme with agent-based modelling, and argue that it does better than equilibrium models exactly where those fail, in crises.

Ostrom’s (2009) framework for social-ecological systems is complexity’s most important empirical result for governance. There are no panaceas. Durable institutions are polycentric, nested and adapted to local conditions. Her work shows that self-organisation is not the opposite of governance. Well-designed rules can make it reliable.

Social-ecological systems and resilience

Holling (1973) separated stability, a return to equilibrium, from resilience, the persistence of relationships under disturbance. He later developed the adaptive cycle and panarchy: nested cycles of growth, conservation, release and reorganisation running at different speeds (Gunderson & Holling, 2002). Walker et al. (2004) added the distinction between adaptability and transformability, and Folke (2006) traced how resilience became a perspective for analysing social-ecological systems.

Scheffer (2009) showed how complex systems can shift abruptly between alternative stable states, and how early-warning signals such as critical slowing down can come before those shifts. Lenton et al. (2008) and Armstrong McKay et al. (2022) applied this to the Earth system, and Rockström et al. (2009) turned it into planetary boundaries. This body of work supplies the most rigorous mechanisms behind the talk of polycrisis.

Education and learning

Davis and Sumara (2006) argued that learners, classrooms, schools and knowledge itself are nested complex systems. Teaching, on this view, should create conditions for emergence (redundancy and diversity, neighbour interactions, distributed control and enabling constraints) rather than transmit fixed content. Mason (2008) assembled the philosophical case, including the warning that complexity can become a fashionable vocabulary that changes nothing in practice.

Jacobson and Wilensky (2006) showed that students find emergent phenomena hard to understand because they default to centralised, single-cause explanations. That finding reaches well beyond science, into civic literacy. Jacobson et al. (2016) went further and proposed a complex-systems framework for learning itself. Long-running debates, such as cognitive against situative accounts of learning, dissolve once learning is treated as a multi-level emergent phenomenon.

In international policy, the OECD Learning Compass 2030 (OECD, 2019) names reconciling tensions and dilemmas as a transformative competency. It is the closest thing to an explicit commitment to complexity literacy, though it stops short of naming it.

Epistemology

Cilliers (1998) made the most important philosophical argument in the field: complex systems are incompressible. Any model of one must leave something out, and what it leaves out can matter in ways nobody can know in advance. So knowledge of complex systems is always provisional and perspectival. It is not relativist, but it is bounded. Cilliers drew an ethical conclusion. Because every framing excludes, choosing one is a responsibility, not a technicality.

Morin (2008) reached a parallel position from the humanities. He set “complex thought” against a paradigm of simplification that separates domains and explains wholes by their parts, and proposed thinking that distinguishes without separating and integrates without reducing. Funtowicz and Ravetz (1993) argued that when facts are uncertain, values disputed, stakes high and decisions urgent, science must become post-normal, with quality checked by an extended community of the people affected. Sardar (2010) widened this into “postnormal times”, a framing that has shaped futures practice.

Knight (1921) separated risk, where probabilities can be calculated, from uncertainty, where they cannot. Kay and King (2020) argue that most consequential decisions are made under radical uncertainty, and that treating them as risk produces false confidence. The Prairie Oyster (2025) essay makes the same point for leaders: uncertainty handled as if it were risk only looks tamed.