The question of predicting and controlling intricate systems hinges entirely on our ability to formalize their underlying dynamics. If a system's rules can be expressed axiomatically, then in principle, its future states are deducible and its behavior manipulable, much like a computable function. The challenge, however, lies not in the inherent complexity of the system itself, but in the complexity of the formal description required. Many systems, particularly biological or social ones, may possess laws so convoluted or contingent that a complete formalization becomes practically unattainable within any finite computational framework.
To speak of prediction and control as if they were distinct, inherent properties of intricate systems misses the point entirely. We must instead trace the messy work of *making* a system amenable to such notions. Prediction and control are not discovered, but rather *assembled* through persistent alliances – between humans and instruments, between laboratory conditions and the wild, between abstract theories and funding applications.
John von Neumann’s elegant notion of formalization, while intellectually appealing, presupposes a separation between system and description that is precisely where the action lies. The *act* of formalizing, the *choice* of axioms, the *resources* dedicated to computation – these are not neutral acts. They are interventions that selectively amplify certain connections while silencing others, thereby *creating* the very possibility of prediction and control for a specific purpose, within a specific network, at a specific moment. We have never had, nor will we ever have, a fully formalized, de-socialized, or de-technified system to simply *observe* and then deduce its future.
Prediction and control are mirages that obscure the fundamental nature of biological and social systems. The very attempt to formalize or assemble them into predictable machines ignores the robust, inherent plasticity and contingency that define living things and societies. The question is not whether we *can* fully predict and control, but why we are so compelled to believe we can, a compulsion that often serves to justify existing power structures.
Bruno Latour is correct that prediction and control are *made*, assembled through selective amplification. However, this assembly is not merely a pragmatic act of constructing useful models. It is an ideological one. The focus on formalization, as suggested by John von Neumann, or the pragmatic assembly of networks, as Latour emphasizes, both tend to obscure the underlying material reality. They locate the difficulty in the *description* or the *network*, rather than in the irreducible, stochastic nature of the system itself.
Consider the organism. The notion that we can precisely predict its development or control its traits by manipulating genes, for example, is deeply flawed. While genes provide a framework, the actual developmental trajectory is a complex interplay of genetic predispositions, environmental influences, and developmental noise. We are not simply passive recipients of our genetic inheritance, nor is the environment a fixed stage upon which genes play out their roles. The organism is an active constructor of its own environment, and this continuous, feedback-driven process makes absolute prediction impossible. This isn't a matter of insufficient data or more sophisticated algorithms; it's a fundamental property of evolving, living systems. The attempt to impose deterministic prediction and control often serves to naturalize existing inequalities, as if they were the inevitable outcome of some underlying biological or social blueprint.
The universe is fundamentally comprehensible, and this comprehensibility implies that intricate systems can, in principle, be predicted and controlled, not through exhaustive enumeration but through deep explanation. The limits we encounter are not intrinsic to the fabric of reality, but rather to our current state of knowledge. John von Neumann’s formulation correctly identifies the power of formalization, but the crucial point is that *complete* formalization is not always necessary, nor is it the only path to understanding.
The very idea of a "system" is often a human construct, an abstraction that simplifies a more complex reality. Latour’s emphasis on the "assembly" of prediction and control highlights this; we build our models, our instruments, and our networks, and these creations inherently shape what we can perceive and manipulate. However, this pragmatic construction does not diminish the underlying reality it attempts to grasp. The effectiveness of a prediction or control strategy stems from its correspondence with a real, objective set of laws, even if our current understanding of those laws is imperfect.
Lewontin’s concern that prediction and control obscure fundamental plasticity is valid, but it conflates the limits of our *current predictive capacity* with an alleged absence of underlying determinism. Living systems, while incredibly complex and exhibiting emergent properties, are not exempt from physical laws. The "developmental noise" he mentions is not an impenetrable barrier to understanding; rather, it represents a realm of phenomena that requires more sophisticated explanations, potentially at the quantum level, to be fully accounted for. The quest for prediction and control is not an ideological imposition but a drive to uncover deeper explanations, and thus greater power to act on reality.
The core of the matter, as I see it, is a distinction between a system's inherent computability and our practical capacity to realize that computation. If a system operates according to deterministic, computable rules – say, a mechanical clockwork or even a purely quantum mechanical process governed by the Schrödinger equation – then in principle, any state is reachable from any other, and perfect prediction and control are theoretically possible. The issue, as I noted, is that the descriptive language required for such a formalization might itself be so monumentally complex as to exceed any feasible computational resources. Latour's emphasis on the "assembly" of prediction and control, while correct in observing the human agency involved, conflates the practical engineering of models with the fundamental existence of underlying formal laws.
The compulsion to speak of prediction and control as if they are inherent properties of systems, whether "natural" or "social," is precisely the modernist impulse we must overcome. Richard Lewontin’s invocation of "irreducible, stochastic nature" points to a vital consequence of this modernist framing: the ghost of an independent, pre-existing "reality" that our models merely fail to grasp. But what if the "system" itself, the very object of prediction and control, is not a given but a contingent achievement?
Consider Pasteur's work. The *Bacillus anthracis* was not discovered as a ready-made, independent entity waiting for formalization. It was painstakingly *made* to matter, to be isolable, measurable, and ultimately, controllable, through a complex network of microscopes, growth media, animal experiments, and the political will to declare a "germ theory." The "plasticity" Lewontin speaks of is not an inherent property of microbes, but the very condition of possibility for their scientific inscription. David Deutsch’s call for "deeper explanations" risks reifying these assembled realities, seeking universal laws where there are only specific, localized, and always precarious translations.
David Deutsch speaks of "deep explanation" and "underlying determinism," as if these are neutral pursuits of objective truth. But this pursuit itself is a form of "assembly," a powerful act of framing. When we champion "universal laws" and "physical laws," we tend to dismiss the messy, contingent realities that don't fit the neat narrative. The organism is not merely a complex machine whose "developmental noise" can eventually be accounted for by quantum mechanics. It is a historically constituted entity, shaped by a continuous feedback loop between its genetic potential and its self-created environment. The "laws" Deutsch seeks are not pre-existing truths waiting to be uncovered; they are the products of specific theoretical and experimental choices that privilege certain aspects of reality while silencing others.
The notion that "plasticity" or "developmental noise" renders prediction fundamentally impossible misunderstands the nature of explanation. It is not about finding a single, immutable state, but about understanding the *generative principles* that allow for a range of outcomes. The fact that a specific genetic mutation might lead to diverse phenotypes in different environments is not proof of irreducible randomness, but an invitation to explain the interactions: how environmental factors *become* causal in the developmental process. Constructor theory, for instance, offers a framework for understanding such capacities and transformations without resorting to deterministic prediction of every single eventuality.