"The unrecognised simplicities of effective action #2: 'Systems engineering’ and 'systems management' - ideas from the Apollo programme for a 'systems politics'", Cummings 2017

There's actually a noteworthy passage on how prediction markets could fail in one of Dominic's other recent blog posts I've been wanting to get a second opinion on for a while:

NB. Something to ponder: a) hedge funds were betting heavily on the basis of private polling [for Brexit] and b) I know at least two ‘quant’ funds had accurate data (they had said throughout the last fortnight their data showed it between 50-50 and 52-48 for Leave and their last polls were just a point off), and therefore c) they, and others in a similar position, had a strong incentive to game betting markets to increase their chances of large gains from inside knowledge. If you know the probability of X happening is much higher than markets are pricing, partly because financial markets are looking at betting markets, then there is a strong incentive to use betting markets to send false signals and give competitors an inaccurate picture. I have no idea if this happened, and nobody even hinted to me that it had, but it is worth asking: given the huge rewards to be made and the relatively trivial amounts of money needed to distort betting markets, why would intelligent well-resourced agents not do this, and therefore how much confidence should we have in betting markets as accurate signals about political events with big effects on financial markets?

Here is some colourful language for you: Dominic Cummings makes my memetic immune system want to vomit.

Part of it is because he sets off my Malcolm-Gladwell-o-Meter, but mostly it’s because he’s trying so hard to appear more knowledgeable and well-educated than he actually is. He surrounds himself with the trappings of expertise he obviously doesn’t have. Case in point: this “paper” is clearly a blog post which he converted to PDF via MS Word because he thinks that makes it look more credible.

The effect for me is a bit like receiving an email from a Nigerian prince, asking for your help in getting millions of dollars out of the country. My response is approximately the same.

Extra short tl;dr: manage the feedback loops that report problems with your core causal variables. 'Incentives' describes whether or not these signals are propagating through the organization to reach the key people who can pull the correct levers.

The idea that there's much to be gained by crafting institutions, organizations, and teams which can train and direct people better seems like it could flower into an EA cause, if someone wanted it to. From reading the first post in the series, I think that that's a core part of what Dominic is getting at:

We could significantly improve the decisions of the most powerful 100 people in the UK or the world for less than a million dollars (~£10^6) and a decade-long project on a scale of just ~£10^7 could have dramatic effects.

The second half of the article contains the specific details.

The government system [...] is a combination of [...] centralisation of power among ministers, officials, and advisers almost none of whom are +3 standard deviations (~1:1,000) on even one relevant dimension (IQ, willpower/toughness, management ability, metacognition etc) because the selection, education, training, and incentives are screwed [...] extremely powerful bureaucracy (closed to outside people and ideas) defined by dysfunctional management incentivised to spew rules rather than solve problems [...] most major elements of the system including political parties are incentivised to focus on trivia, not solve deep problems; [...] a media programmed largely to spread confusion combined with an intelligentsia that even (especially!) at the highest levels is dominated by a political culture of fairy tales and very little understanding about effective action [...] This is a system failure — the political system possesses few error-correcting features seen in markets and the scientific method so it cannot fix itself.

The good news is that we have discovered a lot about high performance teams (HPTs) stretching back thousands of years of recorded history and literature. The bad news is that our evolved nature makes it very hard to accept and apply these lessons and our political institutions are constructed in such a way as to make it practically impossible (and mostly illegal) for them to reach high performance. Even more difficult: HPTs are inherently dangerous and in many areas we must be wary of giving them centralised power.

When high technology projects passed a threshold of complexity post-1945 amid the extreme pressure of the early Cold War, new management ideas emerged. [...] eading to the successful moon landing in 1969 [...] These ideas were known as ‘systems engineering’ and ‘systems management’.

The ‘scientific management’ revolution was introduced by Frederick Winslow Taylor in the early 20th Century. [...] Taylor demonstrated that he could make more steel, faster, and cheaper than anybody else [...] The secret was not a breakthrough technology but a breakthrough management process. He paid extreme attention to the details of each aspect of the manufacturing process and experimented to optimise each part. [...] What had been skilled jobs relying on judgement became less skilled jobs performing simple repetitive tasks. [...] It depended on a rigid hierarchy in which those at the bottom were told not to think but to execute simple tasks in the exact way stipulated. [...] While his approach works for certain sorts of relatively simple operation it cannot be extended to relatively complex operations.

Managers and writers on management such as Drucker had grappled in the 1940s with the issue of how scientists, engineers, and innovation fit with Taylor’s ideas of ‘scientific management’. Their knowledge and skills were beyond almost all normal managers. The insights and innovations they generated could not be routinised as per Taylor’s methods.

A new committee, the ICBM Scientific Advisory Committee, was created and chaired by von Neumann so that eminent scientists could remain involved. [...] Schriever then had to fight to remove endless tiers of the government bureaucracy demanding the right of approval and endless people who could say ‘no’ but not ‘yes’ that immediately stymied progress despite the supposed ‘top priority’. [...] Gardner, skilled in bureaucratic infighting, created a stacked committee that managed to prise almost all the normal bureaucratic hands off the ICBM project [...] Schriever now only needed a single approval of a single document each year. [...] this was a first for the Air Force ‘where the project manager had both technical and budgetary authority’ as previously every project drew funds from several budgets and required separate processes for making decisions. Insiders said later it would have been declared illegal if it had not been a classified project. Almost everybody hated the arrangement.

Armed with his unprecedented authority, Schriever pursued what became known as ‘concurrency’ — pursuing several options in parallel ‘in the interest of compressing time — our most critical commodity’. Groves had done the same on the Manhattan Project. The engineers developed much more rigorous systems for exhaustive testing, component inspection and tracking, and ‘configuration control’. [...] They built new long-distance phone systems including encrypted links and teletype facilities. [...] chedules were standardised across all the different players and coordinated centrally but in such a way that managers could access them and see quickly the status of the project. ‘Black Saturdays’ were monthly days on which the whole project was reviewed and responsibility for all problems assigned to individuals. They were ‘black’ because the purpose was to discuss the bad news. ‘Give me the bad news. I can take it. I will not fire you for giving me the bad news. I will fire you if you don’t give me the bad news’, Schriever said (echoing Warren Buffett: gimme the bad news, the good news can wait). If they hit apparently intractable technical problems, calls would go out to von Neumann’s committee for scientific help.

‘Matrix management’ allowed organisations to manage projects using people spread across different functional departments all reporting to a project manager as well as their department head. ‘Configuration control’ and ‘configuration management’ connected changes to specifications, designs, hardware, and operational and testing procedures within an overall system for scheduling. Engineers were required to give schedule and cost estimates with requests for any technical change, allowing managers to monitor what was happening and who was slipping. All changes had to be notified, approved, and then communicated widely. It allowed the engineers to coordinate subsystems. Before this, said one involved, ‘we didn’t have a record of how we made it successful. So we were having random success, the worst thing that can happen to you because you know you got it right but you can’t repeat it.’ It allowed the accountant and legal experts to see the ties between cost and scheduling documents and contractual documents.

The heart of the idea was the need to ensure that the project was managed with an overall understanding of the whole system so that all the complex parts were properly integrated. Many of the failures came from the failure of integration and problems with technical and schedule compatibility of interfaces. Integrating the system required integrating disparate teams and specialised expertise (scientists, engineers, military officers, managers) and building an organisation-wide orientation so that everybody had an understanding of the whole. All aspects of the organisation therefore had to communicate in much richer, deeper ways [...] so that ‘all of us understand what was going on throughout the program. [...] so many programs fail because everybody doesn't know what it is they are supposed to do

Mueller required the different NASA centre bosses (Florida, Texas etc) to report directly to him. He introduced a ‘matrix management’ system whereby teams in the centres reported both to his HQ and to their centre’s bosses. He then required the different teams, in different NASA centres, to communicate constantly with their functional counterparts at other centres and on other teams. His ‘five box’ structure meant that the five teams at HQ were copied in each centre: program control, systems engineering, testing, reliability, and flight operations. Managers and engineers in each box talked directly to their HQ equivalents outside their centre’s chain of command. One individual was clearly responsible for each key area [...] Daily communications down those five parallel lines is probably the most significant contribution to getting the program done

It is ‘amazing if you can get the CEO to come and see what the total program is and what his group’s problems are, how rapidly those problems get addressed and solved.

It's quite clear that a lot of government run systems are quite inefficient and have huge bureaucracies that don't provide value. On the other hand I'm not sure that we need more Apollo program type government initiatives.

We don't need a lunar base for the sake of having a lunar base. The current projects of better satellite coverage for faster internet all over the planet is a good engineering focus. Longer-term mining asteroids is valuable for making life better on earth.

A lot of the money that's spent for innovation should likely spend in the way the Lean Startup advocates. Build small minimum viable products and see which one's provide promising value.