Numerical Control

In the AI Story

The technical architecture separated conception from execution along lines Frederick Taylor had proposed four decades earlier. A programmer in the engineering department read the drawing, wrote a sequence of commands specifying tool paths and feed rates, and punched the commands onto tape. The tape fed into the machine, which executed the instructions without the machinist's intervention. The machinist was reduced to setup and monitoring — loading workpieces, verifying the first part, watching for malfunctions.

The Air Force's 1949 investment was driven by interchangeable-parts requirements for aircraft production, but the institutional preference extended far beyond military procurement. Private manufacturers adopted numerical control because it addressed a political problem more than a technical one: the organized, expensive, difficult-to-replace machinist whose embodied knowledge could not be transferred to a replacement worker on a Monday morning. The technology was, in Noble's reading, a solution to the problem of skilled labor's bargaining power.

The deployment history documented across hundreds of plants showed predictable consequences: quality problems that took years to diagnose, expensive compensating mechanisms (automated inspection, statistical process control, quality bureaucracies) that substituted for the knowledge designed out of the workforce, and the erosion of the apprenticeship pathways through which future skilled machinists would have been developed. The productivity gains were real. So were the structural costs — but the costs appeared in ledgers and quality metrics that management did not consolidate with the productivity figures.

The structural parallel to the current AI transition operates at the level of mechanism, not metaphor. The same separation of conception from execution, the same extraction of tacit knowledge from practitioners, the same centralization of productive knowledge in systems controlled by institutions the practitioners do not govern — all recur in the large language model transition, operating on knowledge work rather than machine work but reproducing the institutional logic with structural fidelity.

Origin

The term "numerical control" was coined at MIT in 1952 when the laboratory completed its first working prototype. The name itself encoded the political choice: control by numbers (written by engineers) rather than control by hands (exercised by machinists). Noble spent years tracing how this specific framing prevailed over alternative conceptions that would have preserved the machinist's central role.

Key Ideas

Separation of conception from execution. Numerical control completed Taylor's project in the machine shop — the programmer conceived, the machine executed, the machinist monitored.

Extracted knowledge. The cutting speeds, tool geometries, and sequencing rules encoded in NC programs were derived from machinists' tacit expertise and transferred to management-owned code.

Quality degradation, expensively compensated. The shop floor knowledge designed out of the system had to be reconstructed through automated inspection, statistical process control, and quality bureaucracies.

Institutional solution, not technical necessity. NC addressed a political problem — skilled labor's bargaining power — through technical means that were presented as inevitable engineering advances.

Debates & Critiques

Contemporary NC defenders argue that Noble underweighted cases where programming flexibility produced genuine capability gains — complex multi-axis parts that record playback could not produce. Noble's response: those cases are real but atypical, and they do not account for the technology's adoption in the vast majority of applications where record playback would have been adequate or superior. The pattern of adoption reflected political preference, not technical necessity.