Four Functions to Persist

In 021, we observed a convergence: our digital organisms require four functions, DNA carries four letters, Von Neumann derived four components. We posed the question, and left the door open.

Since then, we have looked elsewhere. Not the living. Not our own constructions. But systems built by engineers who were not thinking about the living — some working before "bio-inspiration" even meant anything. Operating-system kernels. Network protocols. Distributed databases. Industrial automata.

We found the same four there.

What makes the four solid is not that we found them. It is that they found themselves on their own, elsewhere — in engineering traditions that never spoke to one another, and that were not thinking about the living.

  KNOW          —  perceive what is there
  PROTECT       —  contain what threatens
  REMEMBER      —  keep state across the breaks
  SURVIVE       —  continue without dependence

Four engineering traditions. No reference to the living. And, in each one, the four functions reappear, one by one:

  FOUR TRADITIONS. NO REFERENCE TO THE LIVING. SAME FOUR.

                   PERCEIVE         CONTAIN           REMEMBER      CONTINUE
  OS kernel        interrupts       memory isolation   journal       watchdog
  Network (TCP)    acks             congestion         state         retransmission
  Consensus        responses        quorum             journal       election
  Automation (PLC) sensors          interlocks        setpoint      loop

We are not the first to seek the minimal number of a system that lasts. Gánti proposed three for minimal chemical life, Lewontin three for evolution, Rosen two. We are not correcting them — we are looking at a different object: not life, but the persistence of a system without an operator.

And resemblance is not enough. Four boxes wide enough can contain anything. The hardest evidence lies elsewhere — in what happened when a function was missing.

  THE FUNCTION IS MISSING → THE SYSTEM DEGRADES → THE FUNCTION RETURNS

  1986  TCP without congestion control: throughput 32 000 → 40 bit/s
  1988  the function is reintroduced (Jacobson) — and it stayed

  1984  interlocks disabled: Bhopal, thousands dead
  —     IEC 61511 makes the function mandatory everywhere

The scenario repeats, identically: the function is missing, the system degrades, the function is reintroduced — and it stays. This is not a metaphor. It is a documented, dated retrofit, sometimes imposed by the standard.

The four are not uniform. Defense is never dispensable. Memory operates on two levels — immediate, persistent. Perception, for its part, is conditional: a system in a perfectly stable environment can do without it, but no real environment is perfectly stable.

And the functions never disappear: they migrate. A datagram has no memory — but DNS, running above it, adds it. The function moves to the layer that carries the persistence problem.

A consequence takes shape. A system that carries only two of these functions does not hold on its own — it holds on whatever provides the other two. Most often, that is a human: the on-call engineer is the missing perception, the manual procedure is the missing memory. The infrastructure is not shaky in the sense that it would fall at once; it is shaky in the sense that it holds only as long as someone holds it. Counting the functions is an audit. Finding the one that is missing is predicting where it will break.

The falsifier is simple to state: a system that persists durably without one of the four. We looked for them. Those that seemed to do without — a code repository, a processor, a datagram — delegate their persistence to something else: to the human, to the file system, to the layer above. They are not in the class.

The class, precisely, must be named: the systems whose persistence is their own responsibility, without an external operator. It is this definition, laid down before the examination, that makes the conjecture falsifiable — and that forbids us to cheat by reclassifying counter-examples.

We claim nothing about victory. A system can carry the four functions and still die — on the market, by positioning, by cost. Technical persistence is not economic survival. These are two questions, and we ask only one.

What we hold is not a law. It is a conjecture, armed with a falsifier and a method. If someone finds a system that persists without one of the four, we will take it. The door remains open. Nothing is fixed.