La Porte's 1996 High Reliability Theory accepts Perrow's tight-coupling and complexity preconditions but denies they make catastrophic failure inevitable

high-reliability-theoryla-portenormal-accident-theorycharles-perrowtight-couplingai-infrastructure

Todd R. La Porte's 1996 paper "High Reliability Organizations: Unlikely, Demanding and At Risk," the summary statement of the Berkeley High Reliability Organizations (HRO) research program, does not dispute the structural precondition Charles Perrow's Normal Accident Theory (NAT) names. La Porte describes the organizations his group studied as exhibiting "pervasive patterns of complexly related, tightly-coupled technical and social relationships that shape their social, structural and decisional character (Perrow, 1984; Weick, 1987; ...)," and states Perrow's own argument directly: "Perrow (1984) argues succinctly that knowledge available to regulators, planners, managers and operators declines compared to knowledge requirements as technical complexity and tight coupling increase, increasing errors ... Indeed, Perrow claims that safety measures themselves can become sources of error if they serve to further increase complexity and coupling."

What High Reliability Theory (HRT) contests is not the structural description but the inference from it to inevitable catastrophic failure. La Porte's own closing assessment of the HRO project's findings: "Whilst these conditions may be necessary, they are not sufficient. Indeed, one of our work's most striking lessons is that these conditions are so demanding that they may not be attainable in other areas without great hazard, travail and social costs along the way." The paper's own title states the position plainly: reliability under tight coupling and complexity is possible, but "unlikely, demanding, and at risk" — achieved through costly, continuously maintained organizational culture and structure, not guaranteed or foreclosed by system architecture alone.

HRT is the standard theoretical counterweight cited against NAT-based AI-risk framings such as Williams & Yampolskiy's, and bears directly on Bit2Watt's claim that tight coupling between GPU compute load and grid stability is dangerous: HRT's answer is that tight coupling alone does not settle the outcome, sustained organizational practice does.