"Interactive complexity" is a measure of the degree to which we cannot foresee all the ways things can go wrong. This may be because there are simply too many interactions to keep track of. More likely, it is because our various theories are simply not up to the task of modeling socio-technical interactions.
"Coupling" is a measure of the degree to which we cannot stop an impending disaster once it starts. This may be because we don't have enough time, because it is physically impossible, or because we don't know how.
The greater the degree of interactive complexity, the less our capacity to prevent surprises. The greater the degree of coupling, the less our capacity to cure surprises. The greater the degree of interactive complexity and coupling, the greater the likelihood that a system is an accident waiting to happen.
In such systems, "operator errors" merely serve as triggers. Trying to find, let alone blame, the particular straw that broke the camel's back is therefore an exercise in futility--a "fundamental attribution error."
Strategies for dealing with Type III messes are therefore quite different from those appropriate for tame problems. Strategies logically follow from the ways problems are conceptualized. Thus, increasing our capacity to prevent unanticipated interactions among components entails simplifying systems (KISS); increasing our capacity to cure them entails de-coupling major components (e.g., build in longer times-to-respond).