Robert Rosen spent decades asking a question that most biologists skip past: what is it about living systems that makes them fundamentally different from machines? Not different in degree—more complex, more intricate—but different in kind. His answer was not about chemistry or DNA or evolution. It was about the organization of causation itself.
A machine can be fully understood by decomposing it into parts. Take apart a clock: gears, springs, hands. Each part has a function; the whole is the sum. But try this with an organism and something goes wrong. Remove the liver and you don’t just lose bile production—you lose the context that makes every other organ meaningful. The organism is not the sum of its parts. It is the organization that produces the parts that maintain the organization.
Rosen formalized this circularity in his (M,R) systems—Metabolism-Repair systems—and showed that it leads somewhere startling: a system closed to efficient causation. Every functional component, every “thing that makes things happen,” is produced from within the system itself. Nothing external is needed to explain why the system keeps going. The system is its own cause.
The formal structure of an (M,R) system consists of three mappings that form a closed loop. Each one produces the next, and the last closes back to produce the system that produces it.
The key insight: β is not an external input. It is entailed by the system. This is what makes the system closed to efficient causation while remaining open to material and energetic flow. Matter and energy stream through; but every function—every efficient cause—is produced from within.
Click each mapping below to see what it produces. Use the “break” buttons to remove a mapping and see what collapses.
If you have read about autocatalytic sets, you know RAF theory: a set of reactions is RAF if every reaction is catalyzed by something in the system (reflexively autocatalytic) and every reactant can be built from the food set (F-generated). RAF captures catalytic closure—every reaction has a catalyst present.
But (M,R) closure is stronger. It asks not just whether catalysts are present, but whether they are produced. The distinction is between having processors and making processors.
The interactive below shows this. On the left, a system that achieves RAF closure—every reaction has a catalyst present—but one catalyst (the red node) is externally supplied, not produced by the network. On the right, the same topology but with that catalyst now produced internally, achieving (M,R) closure.
RAF is necessary but not sufficient for life. A chemical system can have all its catalysts present without being alive—imagine a test tube with pre-mixed enzymes. They catalyze reactions, but when they degrade, nothing replaces them. (M,R) demands more: the system must be the cause of its own causes.
Rosen’s framework is abstract—deliberately so. But Jan-Hendrik Hofmeyr asked the concrete question: how does this closure actually work in real cells? His answer splits the manufacturing process into two fundamentally different kinds of chemistry.
The three layers of Hofmeyr’s closure form concentric rings. Click each layer to expand it.
The non-localizable assembler is the deepest idea here. You cannot point to the intracellular milieu. You cannot isolate it without destroying it. It is an emergent property of all the cell’s components working together—and yet it is causally essential. Without the right pH, the right ionic strength, the right macromolecular crowding, proteins do not fold. Complexes do not form. The fabrication layer becomes meaningless.
This pattern scales beyond biology. Organizational culture is a non-localizable assembler for institutions—you cannot point to it, but without it, nothing “folds” correctly. Cognitive context is a non-localizable assembler for thought—the same neural hardware produces radically different outputs depending on the milieu of attention, expectation, and prior activation.
Rosen insisted that modern biology’s commitment to mechanism—to explaining everything in terms of material and efficient causes—was impoverishing. He retrieved Aristotle’s four causes and showed that (M,R) systems engage all four in a way machines cannot.
In a machine, efficient and formal causes are external: the watchmaker (efficient) follows a blueprint (formal) to assemble the watch. The watch does not produce its own maker or its own blueprint. In an (M,R) system, efficient causation is closed—the system makes its own makers. And formal cause (DNA, in cells) is what breaks the otherwise infinite regress of “what makes the maker of the maker of the maker...”—the same ribosome, reading different mRNA, produces different proteins. Specificity comes from information, not from an infinite tower of distinct fabricators.
Final cause, which mechanistic science banished, returns naturally: the (M,R) system exists in order to maintain itself. This is not mystical teleology. It is the mathematical consequence of closure. A system whose organization produces the components that maintain the organization is, by its structure, self-referential in purpose.
Rosen drew a controversial conclusion from all this: organisms are not computable by Turing machines. His argument runs roughly as follows. A Turing machine can simulate any mechanism—any system whose behavior is determined by its parts and their interactions, where the whole equals the sum. But an (M,R) system contains non-fractionable organization: properties that cannot be localized to any component or finite collection of components. The non-localizable assembler. The closure that is not in any single mapping but in their mutual entailment.
This claim remains debated. Some argue Rosen conflated different notions of computability. Others point out that simulation and realization are different things—you can simulate weather without making it rain. But the core insight stands independent of the computability question: the organization of causation in living systems is categorically different from the organization of causation in machines.
This connects to autopoiesis (Maturana and Varela’s framework for self-producing systems), to the question of agency (what does it mean for a system to act on its own behalf?), and to the deepest question in artificial life: can we build systems that are genuinely alive, or only systems that simulate life? If Rosen is right, the answer depends not on the substrate but on the organization—on whether the system achieves closure to efficient causation, not on whether it is made of carbon or silicon.
Autocatalytic Sets — RAF theory, the bootstrap problem, and self-sustaining chemical networks
Five Levels of Self-Production — from complexity to agency
Before Replication — what came before Darwinian evolution
Function vs Process — the distinction Rosen built on
Homeostasis — two fixed points