Floor 2 — Turing: The Tape That Forgot It Was Continuous

The floor

In 1936, Alan Turing wrote On Computable Numbers, with an Application to the Entscheidungsproblem. To solve a problem about which numbers could and couldn't be computed in principle, he invented a fictional device: an infinite tape of discrete cells, a head that reads and writes one symbol at a time, a finite table of rules.

The Turing machine was never meant to be a real computer. It was a piece of mathematical minimalism, stripped down to the barest parts a "computation" could be said to have. The power of the argument came from the fact that it was so minimal. Anything a richer machine could compute, this little thing could compute too, given enough tape.

And then the world built the rest of itself inside this metaphor.

What was picked

State as a discrete cell. The machine is in state q₃ or it is not. The tape cell contains a 0 or a 1. The head is at position 47. There is never a moment where the head is "mostly at 47 but drifting toward 48", or the state is "between q₃ and q₄ while the input settles".

The metaphor became concrete. Registers. Program counters. Flags. Finite state machines. Every computer built since is, at the logical level, one of these tapes.

What could have been picked

The continuous version was on the table before the ink was dry.

In the same decade, cyberneticians like Norbert Wiener and Ross Ashby were writing about systems whose state was a point in a continuous phase space, drifting under the pull of attractors, settling into basins. The differential equation was the native language of biology, of physics, of feedback. The tape was the odd cousin.

Analog computers existed and, for some problems, outperformed their digital descendants for decades. The Norden bombsight, the mechanical integrator, the differential analyser at MIT — these were not toys. They were solving real problems by letting state flow rather than step.

The tape won because it was cheap to reproduce, cheap to reason about, and cheap to debug. The flow lost because analog noise was hard to manufacture out of, and a generation of engineers found digital error correction easier than analog precision.

What we missed

A machine whose state is a position, and whose computation is a drift, does not need an if/else to respond to almost. It is already almost. The whole pathology of "handle the edge cases" is a symptom of state-as-cell: a cell is either one thing or another, and the boundary between them has to be legislated.

In the alternate timeline, computation is settling. The question "what does the program do in this case?" becomes meaningless, because there are no cases. There is only where the system is, and where the inputs are pulling it.

The Turing tape was a modelling choice. It became a prison.

What the next floor will ask

If the state is a cell, how do we model a neuron?

That's Floor 3, and it's where the prison got built in biology.