Related: Safety management in NASA
We have the “Apollo Project for AGI” and the “rocket alignment problem”. Can we learn anything about how humans organize themselves to create rockets?
Below are some notes from The Secret of Apollo: Systems Management in American and European Space Programs
The book presents a nice contrast: A European project (ELDO) was set up to compete with NASA. Apparently the starting technical capabilities were roughly the same. What made the difference were superior systems management practices of the Americans.
How did this happen?
- The hero of NASA’s systems management in the 60’s, George Mueller, worked in Bell Labs prior to his NASA gig.
- Mueller made internal information flows significantly more efficient by introducing so-called “GEM boxes“
In November 1963, Mueller reorganized the Gemini and Apollo Program Oﬃces, creating a ‘‘ﬁve-box’’ structure at headquarters and the ﬁeld centers. The new structure (see ﬁgure) ensured that the ﬁeld centers replicated Mueller’s concept of systems management and provided Mueller with better program surveillance. Inside these ‘‘GEM boxes,’’66 managers and engineers communicated directly with their functional counterparts at headquarters and other ﬁeld centers, bypassing the ﬁeld centers’ normal chain of command. As one NASA manager put it,‘‘Anywhere you wanted to go within the organization there was a counterpart whether you knew him or not. Whether you had ever met the man, you knew that if you called that box, he had the same kind of responsibility and you could talk to him and get communication going.’’
- As often happens, the path to a higher local optimum was not a pleasant one. Engineers didn’t always have an easy transition to more managerial roles.
Mueller’s new organization initially wreaked havoc at NASA headquarters, because the change converted NASA engineers who monitored speciﬁc hardware projects into executive managers responsible for policy, administration,
and ﬁnance. For several months after the change, headquarters was in turmoil as the staﬀ learned to become executives.6
NASA’s organizational structure changed as a result of Mueller’s initiatives, but he could not always ﬁnd personnel with the management skills he desired. Shortly after assuming oﬃce, Mueller wrote to Webb, stating that NASA could use military personnel trained in program control.
- One thing I’m confused about: when the Japanese tried to copy NASA’s systems (in the 80’s, I think), they measured a noticeable decrease in performance. A likely explanation is that NASA became dysfunctional in the meantime as people like Mueller left
- Systems Management and operations research might end up being important in AI development. It seems to me that
- (a) Management and coordination of people and AI systems is likely to be a major bottleneck in development and deployment,
- (b) transformative behavior is likely to emerge from multi-agent systems and heuristics built up in these fields might provide important insights,
- (c) learning more about how humans (can) decompose tasks is convergently important (e.g. Paul’s agenda)