Several complexity theorists draw a sharp and ontologically robust distinction between (merely) complicated systems and (truly) complex systems (e.g. Cilliers, 1998; Poli, 2013; Kauffman, 2019; Smith & Landgrebe, 2022):
Complex systems: Roughly, complex systems are systems whose behaviour is irreducible to any comprehensible algorithm, set of rules, or simpler constituent parts. Such systems are recalcitrant to exact modelling, prediction, or manipulation, and they cannot be understood completely. Our epistemic grasp of complex systems is necessarily partial and limited. This means that we cannot know whether the parts of a complex system we isolate during modelling constitute the essential characteristics of that system.
Complicated systems: Complicated systems (or what are sometimes called closed, simple, or logic systems) may appear complex, but are, in fact, simple. Their behaviour is reducible to some comprehensible algorithm, set of rules, or simpler constituent parts. Complicated systems can, in principle, be modelled, predicted, manipulated, and understood precisely.
Crudely put, complexity is, in principle, irreducible to something simpler, while complicatedness is, in principle, reducible to something simpler (even if we do not currently possess the tools to do so).
Importantly, proponents of this metaphysical distinction generally consider it to represent a joint-carving – i.e., qualitative rather than quantitative – ontological difference in the world (Van der Merwe, 2023). Examples of complicated systems putatively include motor cars, jumbo jets, computers, and snowflakes. Examples of complex systems putatively include living organisms, language, society, and the brain (Cilliers, 1998, 41–42).
This kind of metaphysical dualism is surfacing in the currently ‘hot’ debate around artificial intelligence (AI), specifically the possibility of artificial general intelligence (AGI). Barry Smith and Jobst Landgrebe, for instance, write as follows:
The solar system, your toaster, your car radio, are logic systems [i.e., complicated systems] – their behavior can be predicted using logic and laws of physics. But for complex systems… – including human beings – we are unable to create mathematical models that can emulate anything more than consistently repeating patterns of their behavior (such as the sleep-wake cycle). (2024, 158; see also 2022)
According to Smith and Landgrebe, AIs (or algorithms) can model and predict the behaviour of complicated systems, but they will never be able to do so with truly complex systems. They conclude that AGI is impossible.
I tend to agree with Smith and Landgrebe. But, from an analytical point of view, the problem is this: when we look closely, a toaster, for instance, is continuously interacting with its environment. Like all things, a toaster’s composition and form change; it will rust and degrade over time. And it will do so in a way that is unpredictable, non-linear, and ostensibly ungoverned by deterministic laws. There is ongoing micro-physical activity at the interface of any object and its environment. Ontologically speaking, this renders that object de facto an open or complex system at a certain level of analysis. At the micro-level, chemicals interact, atoms bond, and various quantum events occur. These include particle annihilation and creation, not to mention entanglement, decoherence, and tunnelling.
Mostly, the toaster slowly disintegrates. Yet, there might also even be moments of construction (or ‘creativity’) caused by chemical reactions and/or quantum effects. This can occur even while the system (like all systems), on average, obeys the second law of thermodynamics. Although the toaster appears to be a simple or complicated system at the level of medium-sized dry goods, events and processes usually associated with complexity can occur both within the toaster and between the toaster and other (complex) systems. This suggests that a toaster is, in fact, a complex system (even if only minimally complex). As a system, it is evolving and adapting to its environment at the physical and chemical level (see also Van der Merwe, 2023). What appear to be complicated systems at the macro-level can then be complex systems at the micro-level. When we consider a system from multiple perspectives, it becomes apparent that it exhibits features of complexity and cannot be easily ontologically categorised as merely simple.
Next, consider the solar system (mentioned in the above-quoted passage). Smith and Landgrebe state that, when it comes to “simple physical systems such as the solar system”, the “types of relations among the elements of such systems do not change over time” (2022, 126). Yet, things look otherwise when we adopt a different perspective. Although astronomers model the solar system using the kind of mathematical equations that Smith and Landgrebe (and Cilliers, Poli, and Kauffman) associate with complicated systems, the solar system is not a perfectly mechanistic and predictable (i.e., merely complicated) system. The planets move in ever-so-slight ellipses, and the system is, strictly speaking, unstable (at least on million-year time scales). There is a very little bit of unpredictability, and therefore chaos, in the planets’ orbital phases (see Laskar, 2013 for detail), and astronomers have to approximate or simplify their models accordingly.
Smith and Landgrebe will be aware of this. Their mistake is, however, to frame it as an ontological, rather than epistemological, issue. The fact that astronomers produce useful models by ignoring or glossing over the complexity involved does not mean that the complexity goes away. All things considered, the solar system is a complex system, even if it is depicted as a merely complicated system in astronomical models. We should surely consider all aspects of some system before making claims about its ontological status. And we should be aware of mistaking the model for the world.
This is not to say that systems do not tend to clump into two general, ‘fuzzy’ groups: complicated and complex. And it is not to say that AI tends to master the former and struggle with the latter. Indeed, my thesis is unlikely to have any impact on future technological developments or discourse in Silicon Valley. Still, from a metaphysical point of view, it is, strictly speaking, incorrect to delineate systems into two qualitatively distinct types. Instead, the universe might be inherently complex; it’s complexity ‘all the way down’.
Works Cited:
Cilliers, P. 1998. Complexity and Postmodernism: Understanding Complex Systems. London: Routledge.
Kauffman, S. A. 2019. A World Beyond Physics: The Emergence and Evolution of Life. New York: Oxford University Press.
Laskar, J. 2013. Is the Solar System Stable?. In: Duplantier, B., Nonnenmacher, S., Rivasseau, V. (eds.) Chaos. Basel: Birkhäuser, pp. 239–270.
Poli, R. 2013. A Note on the Difference Between Complicated and Complex Social Systems. Cadmus 2(1), 142–147.
Smith, B. and Landgrebe, J. 2022. Why Machines Will Never Rule the World: Artificial Intelligence Without Fear. New York: Routledge.
Smith, B. and Landgrebe, J. 2024. Intelligence. And What Computers Still Can’t Do. Cosmos+Taxis 12 (5+6): 104–114.
Van der Merwe, R. 2023. Collapsing the Complicated/Complex Distinction: It’s Complexity all the Way Down. Interdisciplinary Description of Complex Systems 21 (1): 1-17.
By Ragnar van der Merwe
Ragnar van der Merwe is a post-doctoral research fellow at the University of Johannesburg. His research interests are in the philosophy of science, philosophy of truth, pragmatism, and complexity science. He is specifically interested in how we can have knowledge of the world given its complex nature and, most importantly, what this can tell us about scientific progress and scientific truth.