What are the most important lessons for Dominic Cummings and British civil service reformers?

Please leave your suggestions in the comments, only on-topic comments are welcome.  If you are not quite up to speed, again here is a link to the relevant Dominic Cummings blog post.  Or here is a good summary from The Economist.

After digesting all of your marvelous inputs, I will write a synthetic post of my own, with the best of your ideas and some of mine as well.


That Brexit was the only way of doing it, in the same sense that Singapore going out of Malaysia was the only way Singapore civil service be so good as it is today.

Tyler may do not like it, but "state capacity" is better in decentralization like that, not centralization like the EU project which is always "kicking the can down the road" and doing nothing to fix anything. Even China best arguments are all about Shenzhen.

Cummings hails from Education, so let's talk Education.

When Singapore sought to reform its Ministry of Education in 1979:

- the Goh Report that overhauled it would have delighted Cummings - its white paper committee (and Minister) was airdropped from a successful project management engineering team (the 'Education Study team') from the Ministry of Defence, rather than senior Education staff

- nonetheless, it did not pitch its tent on reform of the Education service as such - the Goh Report was pitched in heavily in the 'ethnic dimension', focusing almost wholly on the thorny problem of bilingualism and Chinese medium teaching. The political debate and the quiet struggle with the Chinese medium teacher's unions served to *protect* the ensuing scale of reforms by leaving the non-language changes largely uncontested; its major impact was the introduction of very aggressive and early educational streaming, but the streaming itself was not politically contested save within the echelons of professional educators, who were following the British turn against the tripartite system. This political cover allowed the reform to succeed despite their objections. If the government had instead openly lashed out at the English-educated civil service, its own demographic base, it would certainly have been doomed.

- to get around the problem of the Ministry of Education bureaucracy, the government created new statutory bodies and took care to physically locate them far away from the existing ministry building and to give them a separate corporate identity - flags, logos, jingles, etc. These units included vocational education, curriculum development, etc. Divested of sprawling responsibilities, it was then much easier to focus civil servants on core duties (evaluating teaching performance and school operations) whilst also allowing the spun-off units to experiment. Later, once experimentation had identified the successes (e.g., the 'Singapore Math' pedagogical model stems from the curriculum development spin-off) or failures (religious education, computer-based learning) and things were settling back into maintenance mode, the units would be merged back to streamline management. Again, not seeking to reform the bureaucracy directly but instead to reshape its mandate.

- the introduction of sprawling numbers of new institutions and departmental reform was expensive and could not have been done if the government had also aggressively controlling headcounts or budgets. Ultimately, change is expensive to implement, even if it eventually pays off.

Civil service reform may be needed, but it’s remarkable how Dominic Cummings is so dismissive of Oxbridge Humanities types, a group from which he hails.

Which kind of invalidates his point - if he thinks that they’re - me and him included - people who are guilty of thinking they’re clever than they are, then his post is hubristic to say the least.

Throughout his old posts he writes about how he doesn't think he's clever. I think you've misread what he's dismissive of - he doesn't think PPE prepares people for government, but I don't recall him going much further than that.

My thought in reading the two paragraphs from the economist for free was

Replace all the government workers with old miners, which by now have died back to a size small enough, to oversee all the children of miners and factory workers who lack opportunity due to Thatcherism and who voted to give Bofis the power to do Brexit and eliminate the immigrants taking the low wage boring jobs to create high wage interesting jobs for Boris supporters.

Ie, Civil Service jobs.

Just staying still will be hard enough. This looks like a bad time for major reform. The civil service worked collaboratively with other European countries on many many issues, so just standing alone and delivering the kind of governance they've apparently been giving recently would be very hard.

That you should have studied at the "very best schools". The vast majority of people didn't study at Ivy leagues or oxbridge, to exclude them just because they are affiliated with less prestigious schools... is a sure way to exclude most capable, less conventional thinkers. Why even look at anyone from GMU?

Tone is important. Many humanities departments are experiencing a crisis, and the political polarization between humanities and quantitative university departments is perhaps the worst it's ever been. Cummings' abrasive tone feeds into this, and a possible result is that quantitative competency starts to become aligned with a political outlook that is fundamentally uncaring and unemotional with a superiority complex that leads to technocratic arrogance, insularity, and backslapping corruption. That is not compatible with democracy in the long run. As The Dude said, you're not wrong Dom, you're just..

" a superiority complex that leads to technocratic arrogance, insularity, and backslapping corruption. That is not compatible with democracy in the long run."

I mean, this is a superb, even if slightly histrionic, argument pro Cummings and for a Cummings' style reform.

I mean, it's part of his argument, just dressed with hyperbole and catastrophism, isn't it? Except you're making it about a hypothetical that actually nobody defends and he makes it about the actual real world status quo.

And institutional inertia matters a lot. If you propose dramatic reform to a big organization, the status quo will work to create something that apparently fulfils the headline demand while actually preserving what they want preserved. This may well result in a reform that neither achieves what Cummings wants nor preserves the actual good properties of the current system.

Repaying the same stupid taking point over and over won't make it any more timely or interesting.

Say something original.

"It’s a bonus if you can code but it isn’t necessary." ... "It is extremely interesting that the lessons of Manhattan (1940s), ICBMs (1950s) and Apollo (1960s) remain absolutely cutting edge because it is so hard to apply them and almost nobody has managed to do it. The Pentagon systematically de-programmed itself from more effective approaches to less effective approaches from the mid-1960s, in the name of ‘efficiency’. Is this just another way of saying that people like General Groves and George Mueller are rarer than Fields Medallists? Anyway —"

Listen up. This man is an "ideas man" and as such is essentially a b.s. artist. In my professional capacity doing patents (I'm retired now, being in the 1% I no longer have to work with my mind, just with my capital) I've met many such people, and I can tell you, they make for GREAT PATENTEES (inventors). Outstanding because of the doctrine of constructive reduction to practice, meaning, "paper patents are OK (even if you can't get them to work in practice without a lot of sweat)". But in practice, in terms of actually doing something useful, well, they're b.s. artists.

I would also hate to work for this man because of his "high standards" meaning impossible to please, constantly shifting goalposts, and, worse, this sentence which speaks volumes: ""It’s a bonus if you can code but it isn’t necessary." Right. Shades of Mont Pelerin Society, thought experiments, Freakonomics, and that spike in the steering wheel. Oh man...

Bonus trivia: Greek TV showed a woman on a remote Greek island who refuses to leave despite being the only person on said island. She is supplied with provisions by the Greek military via helicopter. At taxpayer expense. She's also no doubt a woman of high ideals...see the above.

A weird accusation to level against Cummings. For instance, if he weren't' competent, Brexit wouldn't happen. Heck, there's a good chance Britain would have joined the euro 20 years ago if Cumming was as ineffective as you claim. Regardless of one's opinion of the man, Cummings has been wildly successful in terms of real-world impact, from the education reform to electoral campaigns. To claim he's an ideas man akin to an inventor is flat out bizarre, even for you.

"there's a good chance Britain would have joined the euro 20 years ago"

And here was me thinking it was Gordon Brown that blocked that with his tests instead of some bloke nobody had ever heard of at that time.

I'm with the OP. Cummings is a BS artist, but maybe that's the whole point if the final objective is to basically gut the civil service as it stands and replace it with something more in line with the wishes of whoever is behind the project.

To be fair, I wish for that steel spike on the steering wheel to be installed in other drivers' cars all the time when I'm driving on the Beltway. Some of those guys are f--king maniacs.

Since that's how you measure men, how does your net worth compare to that of Vitalik Buterin, who you treated so rudely last year?

Dominic quotes Yudkowsky (a guy who once without a hint of irony identified himself as the most consequential person in human history, a classical delusional self-confessed polygamous cult leader/charlatan with no accomplishments) with a straight face. He was also seen consipicuously wearing an OpenAI t-shirt on his first day at 10 Downing st. (a self-important hand-wringing horseshit organisation). These seem like 2 data points that Dominic is not capable of identifying extravagant insanity. His only attempt at private enterprise flopped.

Trying to smear someone by association with OpenAI and Yudkowski is going to backfire I think. Literally you are comparing him to some of the most thoughtful and intelligent people in the entire world

You can’t succeed at leadership, even on disruptive projects, if you are full of yourself and full of disdain and condescension for the people you are to lead. So if you have zero people skills and are going to be an asshole („I will bin you at will“), you need to be less obvious about it than he is. People are going to game and arbitrage the shit out of him, just out of spite.

Yeah, and several of the comments here have made a similar point. Cummings should avoid becoming another Robert S. McNamara or Peter Thiel.

Mr Cummings seems to be an unapologetic elitist -- and that's fine; we need elites -- but wants to shake up the current set of faux elites and replace them with actual elites.

This probably won't work and, even if it does, the problems we have with our current elites will simply reappear in a decade or two.

I would recommend that he first find a rigorously quantifiable problem to solve and unleash his boffins and weirdos to solve it; he might then gain the reputation necessary to solve the hard problems of education, social work, and all the other processes that hinge primarily on human-human interactions.

But the solutions the folks he seems to want to hire will produce will be extremely likely to be found abhorrent to current societal moral dogmas, which tend to be unworkable combinations of elitism and egalitarianism.

Even rigorous solutions to such crises as climate change will cause the louder elements of British society to call for Mr Cummings's head, even if (or especially if) they can be shown to be workable -- e.g., nuclear power.

I worked on the campaign team for an Irish politician who got elected and got made a Minister immediately. He made a short speech at the party to celebrate. He said: 'I intend to bring the Department from the 19th Century into the 21st!'. We all cheered. But the old political hack behind me laughed and said: 'Meanwhile the Department isn't out partying. It's back at the office plotting how to stay in the 19th!' Four years later the Minister lost his seat. Reforms had been minimal.

Raise educational standards, take Harvard Entrance Exam 1988 as a benchmark, copy French Elite education/selection, make Civil service a choice career, broaden curriculum, not more students of geography in completely unrelated careers, end the ideal of the dilettante

You forgot the bit where they pay rock star wages to tempt them away from the other people with deeper pockets and more interesting jobs they want to offer them...

1. The civil service does what ministers tell it to. Ministers tell civil servants conflicting things, or chage their minds with little understanding of the consequences, or waste valuable time focusing on announcements rather than actual policymaking. Most ministers don't have experience of managing large organisations, so they fail to understand the consequences of their decisions.
2. You can't control everything from Whitehall or No.10, and you need the humility to understand this, and to focus on ways of giving people back power that leads to constructive solutions rather than democratic deadlock.
3. Beware Chesterton's Fences
4. There are limits of the data-driven approach. Many policy problems are too complex to model, even with the best data and people. Many problems are inherently political and solutions are dismissed because they are unpalatable to politicians or the public or special interests.
5. Civil service reform and expert policy units in no.10 have been tried before - there are lots of lessons to be learned.
6. Innovation isn't as simple as "get a bunch of smart people in a room and manage them really well". His favourite example of ARPA was so successful because of the time and place it was in - part of it was just that this incredible general-purpose technology of digital computing was just entering maturity. There's no guarantee current emerging technologies will turn out to be general-purpose in the same way. (And computing's impact was less than previous general-purpose technologies anyway).

True. Also - and it's an unpopular thing to say - because DARPA is DOD most of its work is classified and there has never been a publicly available assessment of its work. only anecdotes form interested parties.

I think he's mistaken in prioritizing unusual over normal. My general rule of thumb is that the best results come from radicals using normal strategies or normal bureaucrats using radical ones. You don't want to double up. The best case scenario with radical/radical approach is a decade wasted reinventing the wheel, and that will probably be the result here.

Every bureaucracy will eventually land in the same place the British civil service is now, it's just the nature of long-lasting system. You aren't going to find reforms that will fix it, all you can achieve is temporary success before the bureaucracy adapts and reasserts itself. The process of democracy is one of the best solutions to the problem -- periodic agitation to keep sclerosis at bay, but he decries the shuffling of civil servants in Part G, so I don't think he really knows what he's looking for. Deep expertise is incompatible with a bureaucracy that is willing to adopt new ideas. You have to choose which of those you are willing to sacrifice.

Follow the centers of excellence model: it’s hard to reform an entire organization, but it is easier to create a small elite team within which becomes the standard bearer of excellence. As that team solves discrete problems, add more and more people (slowly to enable integration) and eventually use graduates of that team to seed more centers of excellence.

The rate of reform for this approach is exponential which is means it underdelivers our linear expectations in the short run and massively overdelivers in the long run.


I would advise them to read, and *understand* Helen Andrews on the initial merit based Middle Victorian civil reforms (https://herandrews.com/2016/07/01/the-new-ruling-class/) for what they were really an alternative to, what their unforeseen consequences were, why the thinkers who represented the most original response to them thought that merit was necessarily not an obvious principle and so forth.

Read John Fitzgerald on the flaws of meritocratic bureaucracy as an alternative to government which draws on folk institutions and public inclusion as seen through the Qing and on the folly of vacant enthusiasm willing to brand dubious authoritarianism as meritocracy- http://insidestory.org.au/the-qing-is-dead-long-live-the-qing/ .

Essentially, before simply concluding that you will do merit better, challenge yourself whether what you are implementing really is merit based selection or simply a cloak for autocracy (with all its flaws) and whether merit based selection has inherent limitations, unstable contradictions and unintended consequences *even* if is.

The Helen Andrews piece is priceless!

Give Middle Victorian reforming its due. Maybe not everything worked out, but one of the most inspiring stories is that of Rowland Hill, an unknown school teacher and pamphleteer who came up with the idea of postage stamps and uniform postage irrespective of distance.

A simple idea that drastically cut costs and drastically increased circulation, much like containerization of shipping traffic and packetization of internet traffic. Blindingly obvious in hindsight, but it faced a great deal of inertia and opposition. It took many years of perseverance and occasional setbacks to see the reform through.

Is there another big simple idea today?

Possibly UBI. Been reading "The Road To Somewhere" - why did Brexit happen ('anywheres' vs 'somewheres') The chapter on marriage/ family, which went into the unintended consequences of Blair & beyond reforms, proposed various solutions, all of which could be bundled into UBI. Needs to be set at the equivalent of the tax free allowance's value (inc NI) (~ 3k5) which is also pretty much the level of job seeker's allowance. Start there, see what happens.

The Helen Andrews piece is so ignorant I don't know where to begin. No one who comes after me should make the mistake of assuming that just because there's a link, there's worthwhile evidence on the other end of it!

Andrews' critique of meritocracy and the evidence she cites paint two totally different pictures. She argues that Harvard students aren't as smart as they're cracked up to be. So far still following, but how does she support this? With an anecdote about how some Harvard undergrad students don't know much about Beethoven, or the American Civil War.

Conspicuously absent is any comparison with students from other schools, or any argument explaining why Harvard students SHOULD know about Beethoven, except a vague allusion to shared knowledge bases (does knowing about Beethoven become more useful if everyone else knows about him?)

There are arguments to be made here, but Helen Andrews fails to make them.

If you know nothing of your culture or history, what could you possibly have to offer about anything except some expertise in a narrow technical field?

Many narrow technical fields are far more useful than culture or history. I'd trust a statistics major to run my business over a history major any day. The problem with public administration is not a lack of history knowledge but a lack of quantitative reasoning skills. Too many people make decisions based on gut feelings that are essentially prejudices without crunching the numbers or looking at what the data actually says.

"If you know nothing of your culture or history, what could you possibly have to offer"

"Many narrow technical fields are far more useful than culture or history"

I have been struggling to reconcile these two facts for longer then I care to admit. I'm a tech dude I swim in data and ideas all day. Yet the ideas I find most useful are people-ideas (tm) that cover mostly psychology type things. Basically I want to make the argument that synthesis of humanity/politics/economics and Math/statistics/conservation-of-energy is required to make any headway on improving the blindingly complex modern gov bis sci systems that make modern life possible.

Everyday I see people from both sides of this divide making mistakes that the other team just cracks up about when they hear the stories.

And the sad part is while the truly deep problems/questions in most fields are now matters for domain experts only (if you don't understand the question how can you answer it?) Most fields have some really quick simple yet powerful summations that could be spread around to every other field to prevent baby's first type mistakes.

But idk what to do about this other then scream at the top of my lungs to anyone who will listen the following three things

The Efficient Market Hypothesis IS TRUE. (econ)
Energy IS CONSERVED. (phys)
The Primary Agent Problem IS ENDEMIC. (gov)

Does anyone care to post an article covering this? Cause I have a burning desire to learn solutions to Dunning Kruger if there are any good ones out there that don't amount to everyone needs to double major. And just to put the icing on the cake of the human condition every now and then we get an RNA-world solution where my maxim of all deep problems are lay-man resistant breaks down /ShakesFistAtHeavans


Yes, it's CP Snow's "two cultures" in yet another guise. And probably more important than ever, both at the level of the whole society and for individuals wanting to advance their own careers. The better that we (as societies and as individuals) can combine expertise in both domains the better off we will be.

I don't think the argument by Andrews is that it seems that the individual talked to lacked knowledge about the ACW, but that he claimed that being a "big picture thinker" with facts others have supplied to him (without much to back that up), was a superior substitute for actually knowing the knowledge.

Though the strongest part of the article is for the history, not the characterisation of today's would be meritocrats (though I suspect it is fair, it will probably upset some Ivy League snowflakes).

I take it differently: Whatever the criteria, Harvard will win! Harvard is clearly good at winning stuff.

Therefore we need not just one, two, three Harvards, but more different kind of stuff.

Meritocracy and autocracy seem to be two orthogonal dimensions; one can have a very limited government with the few positions that do exist filled by merit, versus a totalitarian government where the positions are filled through patronage or elections. Ideally, you should have a government that is extremely limited but the positions that must exist ought to be filled by merit. In fact, I would argue that a true meritocracy is easier to achieve when the size of the government is limited and the number of people to be selected is very small. In the US, when you think about the government agencies that function best, people are likely to name independent agencies like the Federal Reserve where officials are appointed based on merit. For a long time, Americans consistently held a better view of the judicial branch of government (which was appointed by merit) rather than the executive or legislative (which were elected). Of course, judicial appointments are becoming more politicized today, and this is corresponding with a decrease in public trust in the judicial branch.

The Fitzgerald argument on the Qing also fails because it ignores the reason why the Qing Empire suddenly faced a fiscal crisis in the mid-1850s: foreign imperialism, which required a massive and sudden increase in military spending as well as payment of very heavy war indemnities. Prior to the Opium War, the Qing state was small, with very little revenue relative to the size of the Chinese economy compared to European countries and so little military power that it was easily defeated by quite small European armies. And this system of a small government with a relatively meritocratic bureaucracy worked decently in the early part of the dynasty with China experiencing an economic and population boom, again only failing because foreign imperialism resulted in a massive increase in military spending needs. As no one is likely to invade Britain today, it seems that a similar fiscal crisis that would necessitate expanding the state is unlikely. It would be better to be more like the early Qing, and make the state both smaller and more meritocratic.

No, it would not be better to be like the early Qing and exclude a role for institutions outside a narrow exam taking appointed class. Qing China had many problems relating to its undersized state well before lack of military capacity became apparent in its navy, obviously its lack of monetary sovereignty and customs enforcement which together caused its distorted trade patterns and then its incompetent response to problems which led to the war in the first place. Britain's state today may be served by being somewhat smaller or larger than it is today but a proprtionately small Qing sized rump with little capacity to do anything, forced to be like that with no choice because of a refusal to admit wider participation in politics, would serve no benefit.

Although Zaua, you are a thoughtful guy and props for reading and understanding his article.

I suppose you could still argue that if technological divergence had not come about, a smaller state like the Qing, with minimal participation and consistent regulation (of currency, of customs), still might have won out with sufficiently advanced military force, and thus there are no lessons to be learned there about the pitfalls of lack of participatory government and the weakness of a purely "merit based" form of appointment in gaining support, in raising taxes and supporting an army and regulating trade and commerce. Myself, I think that's not going to be a winner though.

The main point though, is to understand that merit itself has tradeoffs in inclusion and inclusion has tradeoffs in support and funding (people will pay tax to a state that represents them and where they choose the officeholders and hold them accountable, not one where the officeholders are free to go off serving what class or personal or monarchial interest they wish). Whether you judge those tradeoffs to be worth it, or not, is a question the article opens (and the author, as largely a Western democrat, of course supports that they are not), but the value of the article is to be aware of the tradeoffs, not to simply propose merit as an alternative to accountability without understanding them, nor to cloak processes which are obviously not merit based (Communist Party patronage) in that guise.

"As no one is likely to invade Britain today ..."

I suppose a pedant would say that starting at 1066 instead of 1065 deflates it somewhat ... but for me it's still the Startling Contention of the Week: "In fact, Britain today receives more immigrants in a single year than it did in the entire period from 1066 to 1950."

Okay, but the British population is a heck of a lot bigger now, too.

Okay, but the British population is a heck of a lot bigger now, too. How does the current rate of immigration look as a fraction of existing population?

The British civil service is the best in the world. No need to change!

What he could do is expand the use of SpAds -- hire many many more of them & make the recruiting process a bit more professionalized and competitive (but not too much! - the relationship-driven nature of the job is important too).

"The British civil service is the best in the world. No need to change!"

Let's not get carried away. :-)

"Seems like they're only inviting officials to get in contact."

-1 for obtuseness

(i) He seems to have been influenced by the best documentary the BBC ever made - Yes, Minister.

(ii) So he's set an exam for Jim Hacker. If Hacker fails it then our Dom will be off to pastures new.

Cummings, Raab or both?

I was thinking of the Jim Hacker who looks like a blond panda.

Government is not Caltech.
Government is not Ren. Tech.
The lesson of Icarus.

There are some lessons about diversity that, alas, will not be learned: that, for instance, ethnic/cultural diversity matters for ‘diversity-of-thought’.

To whatever extent the UK is de-coupling from the EU, the message of Brexit would seem to require requisite domestic follow-through, to wit:

how much political administration are Whitehall and Westminster prepared to cede within the British state?

How much decision-making for local affairs (Northern Irish, Scottish, and Welsh, for starters) can prudently be ceded to Belfast, Edinburgh, and Cardiff, respectively? Surely the locals in their respective settings can administer local bureaucracies and their staffs at least as well as London, as long as local responsibility for local administration is understood by all. (Surely London knows all about building in incentives and penalties that could enhance local power sharing.)

(Haven't looked at the rest of the Economist yet: what is the SNP up to since the election?)

He could start a review of what could be given up by the centralised Civil Service and handed off to local authorities. Localism is better not only because locals understand the issues better than the centre but also because it encourages natural experiment between areas about what works best. In fact more experiments would be good, lets try different things in education for instance to see what works better.

Idea 2, try allowing local communities to subsidise things in their local area, you often see small towns agitating against a local school or hospital closure decided centrally on the basis of cost efficiency, allow the community to keep them open via higher local taxes.

Idea 3 - New Housing commissioner with sole purpose to eliminate red tape for new housing. Create area plans for new housing.

Further thoughts:

If there are large projects to be done, they should use the stage gate system of governance, witha separate perhaps parliamentary gate decision before proceeding to the next stage. The stage gate system requires detailed definition of the project before it moves to the high capex execute stage. This process mitigates against projects which start with an idea with a low budget due to their lack of definition. Similarly, taking a cue from the oil industry, more dual FEEDs for projects paying for design to keep competitive before the final bids are received.

Create a single project management organisation for delivery of all Government projects (military, infrastructure etc) rather than having multiple departments learn about project management. Projects have very long cycle times to project to project learning is a important way to improve project performance).

There should be more collaboration between contractors and clients in the early stages of definition as well, including trying out new construction technology, usually what happens is the contractor gets a month or so to provide a bid, so they sill use existing known technologies rather than looking into more efficient ones.

Project specifications are another area, I would bet these can be drastically simplified. Perhaps more BOT ( build operate transfer) projects would allow that, so that the contractor who builds things has to operate for say ten years. This way they deal with any operation issues themselves.

Another idea, why not merge the armed forces into one, why have separate command structures for the army, navy and airforce? A single force but with individual units with individual objectives would be more efficient. You can even have a ceremonial division for all the upperclass to join.

Set and publish annual KPIs ( key performance indicators) for each Government department, to make it clear on what they are supposed to achieve, what gets measured gets done.

Reorganise the diplomatic service to be focused on the commercial support of companies looking to export rather than vague “diplomatic” objectives. Now we have modern communication technology we don’t need ambassadors anymore, and they have no weight in any Government to Government discussion.

For overseas aid, there should be an equivalent to NICE for medicines, all overseas aid projects should be analysed by outside experts for utility and probably of success and then ranked accordingly. Publish the rankings and then follow up after the project is complete and learn for future rankings. Be transparent about this to the public. If projects fail to be ranked, offer them for public support with matching say at 50% by the Government.

Adapt a NGDPLT policy for the BOE for a five year trial period and require narrow banking for all retail banks. This could drastically reduce regulatory burden and reduce numbers of required regulators.

For railway and road infrastructure, work on tunnelling technology more, a large part of the cost of new infrastructure in a place like the UK is the existing surface infrastructure, if low cost tunnelling could be developed then we could get more stuff built quicker.

I expect he is on the right track.
Two thoughts spring to mind : one from Scott Galloway's recent blog (https://www.profgalloway.com/unremarkables):
if his arguments also apply to the UK? there is a worry about “unremarkables” seeing their ideal career future in protected Government employment (thus also reflecting Furedi’s prevailing “safetyism” culture) - this supports Cummings push for more direct recruitment into senior Govt positions and for more government/private work relationships to introduce innovation into institutions;
the other from Fukuyama’s “Political Order and Political Decay: From the Industrial Revolution to the Globalization of Democracy” book and papers - regular change of people and positions within government institutions is necessary to break up patrimonial and reciprocal altruism relationships that reinforce conservatism and the status quo, support incumbents and threaten elected representatives.
This is most easily done by restructuring - unanticipated reshuffling of departments, and devolution through delegation to other elected groups of community representatives - thus avoiding progressive centralisation of bureaucratic and regulatory power.

At least in the US, civil service jobs often involve accepting lower pay and a more bureaucratic working environment for more job safety and often a more family-friendly environment. (Back when I worked for a state government agency long ago, the joke was that there was some kind of fertility treatment in the water--newly married young women got hired and soon started popping out children.)

The fundamental problem with this plan is that a "skunkworks" only works as a secondary or tertiary project, hidden from view. You need headline projects to attract attention of powerful busy bodies. While you get work done, elsewhere.

You can't "skunkworks" the very same thing that high profile managers (or politicians) want to put their imprint on.


When I was a young engineer, my project manager told me "we are the project no one cares about, that's why we are going to come in on time and under budget."

He was right. It wasn't that we were so much better, it was that we had no fingers pokin'.

We later joined that swollen project and billed good rates to finish it .. two years late of course.

Given his aims, include computer scientists and not just mathematics and physics people. (In particular, people in machine learning, algorithms, and theoretical computer science). He includes software engineers but that is not the same as CS.

Include math/physics/CS people without a masters.

Include people with PhD level skills in computer science, statistics, and economics (esp. more data heavy skills).

Hire people who are good at evaluating large and confusing empirical literatures that contain lots of bad statistics/methodology and where understanding the sociology of the field might be important. These people are often not math/physics people. (E.g. Scott Alexander).

Many organizations pay search firms to find people with the skills/attributes needed for high level positions. I have not seen good organizations hire important people through blog request.

Don't you remember the Google ads?

I worry about three things.

First, Dominic Cummings cites so-called "cutting-edge" papers on the interface between physics and social science and machine learning. I'm fairly certain that he's fallen prey to buzzword-itis in this one. I think most gains in government are to be had not from the application of fringe, cool-sounding, buzzword-laden ideas but rather from boring, well-established, well-studied ideas that need to be scaled carefully (sort of like the work done by Banerjee and Duflo). Moreover, he cites a lot of Nature and Science papers, which is a red flag. Scientists know that the big splashy papers in Nature and Science are often (not always) not as interesting or important or scientifically sound as they are made about to be by the press. So hiring buzzword-friendly talent will increase attempts to leverage "cool" ideas to get at possibly-nonexistent, very-high-hanging fruit, while the right thing to be doing is to focus on getting at the low-hanging fruit.

Second, Silicon-Valley-type talent flourishes in Silicon-Valley-type places. No. 10, on the other hand, is pretty much as far as you can get from a Silicon-Valley-type place. So I worry that the talent will not be structured in the right sorts of ways to really extract its potential.

Finally, a lot of this reeks of some of the arrogance that affected the early LessWrong-rationality community, which was like, oh, just by improving our general reasoning and considering weird ideas, we can influence the world at a global scale. While the rationality community has had some modest success, I don't think it's anywhere near what they originally hoped for. I worry that this is going to fall into the same failure mode.

Second the comment about "cutting-edge papers", judging from the ones I looked at ("Scale-free networks are rare" and the ML to chaotic systems one). I get that Cummings is personally interested in the "network theory" (seeing as he referenced Watts-Strogatz and its highly cited 1-page condensation) but this stuff is so far from what cutting edge pure mathematicians actually work on...

To get to specific critiques (granted, both are far outside my expertise but I suspect my level of expertise is not that much worse than other readers and if my analysis is off-base or incomplete maybe someone can give a better one), under a hypothesis of universality of scale-free networks, the critical exponent is an invariant of a social network that carries information and can be measured empirically; without universality, you do not measure such a thing. It's interesting, but it is high hanging fruit (and the relevance to civil service reform is IMO suspect). It still seems to me that the best *social science* lesson in this field as I know it (i.e. excluding PageRank et al) comes from "The Strength of Weak Ties" as it pertains to the robustness of homophily as concepts that could potentially determine success or failure of an innovation (see the whole book "Diffusion of Innovations" for applications of this idea).

Regarding the ML prediction of chaotic systems, note that when input data is uncertain long time prediction *should* always be impossible. I don't know enough to say whether or not their result of ~6 times the Lyapunov exponent of a predictable regime before chaotic unpredictability takes over is a good result or not or whether it is competitive with how a numerical approach would perform. My bias as an outsider is that numerical methods that are (specifically) designed to tackle chaotic dynamical systems will outperform a relatively unsupervised approach using ML. Maybe someone else can tell me whether I'm actually right or wrong.

To be more coherent, it seems more productive to structure a program like this by open questions first, specific papers to study later. If there is a robust framework for why "these sorts of things should be studied more" that's a bit better. Often things that are studied in "deep" subjects are so for good but similarly deep (but good) reasons. I get the sense talking to Lesswrong types that they have a good deal of respect for mathematicians but they think mathematicians are misusing their mental tools and if they only had a bit more introspection, they would try to work on AI or AI safety or something similar (or at least focus on more important probems). As a mathematician, I find a grain of truth in this but find the recommended areas/problems that LW types advocate for to be naive (although naive in subtle ways).

One final thought... Cummings specifically advertises for "weird" people. Perhaps Cummings feels the bureaucratic pressure towards institutional isomorphism and believes this is a potential way to avoid that.

I don't think these criticisms about LW types necessarily apply to Cummings FWIW. And the ARPA/PARC document is quite interesting. Why are PARC's successes so much more impressive than those of Microsoft Research?

Right, it's the technocratic fallacy of "people in occupation X seem competent enough, let's put them in charge of everything." X here at various times and places has been equal to machine learning specialists, engineers, military generals, McKinsey consultants, corporate CEOs, and investment bankers. Never lawyers, though, despite the fact that the U.S. is crawling with lawyers who become political leaders.

Hayek said that the curious task of economics is to impress upon men how little they really know what they imagine they can design. Does this also apply to what Cummings is trying to do?

An interesting document no matter your priors.

The 2015 Trudeau Liberals were intent on reforming not government but how the country functioned. It was a bad idea; they had commissions and stakeholder meetings to no end, but frankly started from the wrong place and ran into the reality that they could make it worse.

And they did. The one noteworthy accomplishment of their first term was to make the Canadian justice system into a political entity with the promise to benefit the highest bidder. Quite disgusting.

I get the sense that this is too broad. There are a few things that work well in Britain, some things that work very badly, lots of middling ones muddling along. Choose a few things that will make a big difference, where the power of the state can make a difference for the good (not a simple chore to find them), find where the State is making things worse (not hard to find). Change them in a way that will outlast your administration.

But definitely start from the assumption that the basic structures of the institutions that have been around for three or four generations are working with minor adjustments required. Start from the assumption that all the smartest people in one room cannot do very much.

The section on project management was interesting. Good luck with that.

The most interesting thing in the comments is a general agreement that it is impossible to even think of an effective government. How dare this guy think that he could do anything about them. Maybe everyone should simply stop paying taxes.

Governments can be good, and they can move quickly, but it is rare to do both at the same time. WWII mobilization was both, but it was driven by a very existential threat.

So sure, if Brexit Brittain can feel that again, they might do it again. If not, it's another episode of


I'll have to look for that. Sounds like it could be quite amusing.

Probably the most effective bureaucracy I've seen is from about the mid 90's to 2008. Canadian Federal government and most provinces had eschewed deficit spending for the most part, and the actions of the bureaucracy had a direct effect on their funding levels. If they did stupid things the economy generated less revenue; if they did things well the revenues were maximized. It was one of those very rare instances where a by definition and by design unaccountable system was accountable, and the results were quite good.

At the federal level in Canada the key years were 1993 to 2003, when vote-splitting on the right and a first-past-the-post parliamentary system temporarily created one-party rule. Austerity was possible without electoral consequences.

They weathered the Great Recession far better than most Western countries. No financial institutions needed bailouts.

Justin Trudeau is undoing all those past achievements. He's running large deficits at a time of relative prosperity, and when revenues come in higher than expected thanks to the economic cycle, all that windfall revenue is promptly spent rather than letting the deficit come in lower than what the budget projected. Meanwhile per capita household debt is at record levels.

They certainly won't weather the next downturn nearly so smoothly.

PS, those familiar with other first-past-the-post parliamentary systems like the UK or Australia might not realize how Canada differs. In a majority government nearly all executive and legislative power is concentrated in the hands of the prime minister. Party-line discipline is demanded for nearly all votes. A dissenting vote is as rare as a faithless elector in the US Electoral College. Backbenchers and cabinet ministers alike never express independent opinions in public.

Lose an election though, and you're out. The major opposition party is run along the same lines. Corbyn would not have had a second election chance in Canada.

The main constraint on the prime minister's power is not parliament (unless there is a minority government) but the Supreme Court and to some extent, eternal jurisdictional turf wars with the governments of the provinces.

" eternal jurisdictional turf wars with the governments of the provinces."

I remember reading articles where it seemed clear that Canada's version of a federal government is more decentralized (gives more power to the provinces) than the USA's version. That was decades ago though; having skimmed quickly through some stuff on the web, it's clear that Canada has over its history has had movements towards centralization followed by movements towards decentralization.

My impression is that currently, Canada is still more decentralized than the US is?

It's probably true. There are far fewer provinces than states. Ontario has 39% of Canada's population, while Quebec has one-quarter of the population and is constantly seeking to expand its own powers. For example, some provinces run their own immigration programs in parallel to the federal government.

Whatever idea you are implementing, try to build in tests.

I am not sure what Cummings wants, better project management or a better bureaucracy or both. In any case, the positive experiences he singles out, such as the Apollo missions, are positively simple compared to a whole government, to say nothing of a society and the world. What is not understood cannot be modeled.

The only known way to deal with such complexity is to allow competition, and I sense Cummings is open to that, but he doesn't put it front and center. E.g., it is important that government bureaucrats do not all come from the same educational and social background. Competition, also from below, would be helpful, for some will rise.

I'd recommend this talk from 2014, where he lays out his vision for reforming the British public service, and which is eerily prescient:


A few observations, some also drawn from his blog posts:

1. What he cares about is changing the operating system of government. Policy is downstream. The bad policy decisions that he sees plaguing the British government since 1862 (and causing both world wars, for example) are only symptoms of the real problem.

2. In line with this, Brexit and the EU hardly get a mention in a 1-hour talk. Brexit, while a worthwhile goal in itself, is more importantly a route to the reform of Whitehall.

3. He wants technocratic solutions. He reveres the successes of mathematics/physics, Silicon Valley, Manhattan Project, Apollo program, probably also high-frequency trading. He also loves maverick Skunk Works programs that get something done in larger sclerotic organizations.

4. Although he reveres high-level mathematics and physics, he seems to have only a superficial understanding himself. Loves to name-drop the Fields Medal, polymath, Grothendieck, and other things that reveal a fanboy level of understanding. At times he gives the impression he thinks that all we need to do is get a bunch of math/physics PhDs in the room and everything will sort it itself out. Proposes giving courses on probability theory to ministers, as if that would change anything. Many would say all this is deeply naive (esp most mathematicians and physicists), but he would probably respond that a necessary first step to solving problems is to have your organization staffed by people who are focused on problems, rather than focused on people and on process, and who actually know what it even means to solve a problem.

I just looked at the first few minutes of that video. While I can have some sympathy, what he does say isn't even wrong. :-)

I didn't catch that he said Britain caused both world wars, but if he did, and if it were true, which it isn't, that would hardly be a symptom of the real problem. Hell if that were the symptom, what could possibly be the problem? :-(

“ The bad policy decisions that he sees plaguing the British government since 1862 (and causing both world wars, for example) are only symptoms of the real problem.”

I’m hoping this is an infelicitious summary, because Britain in no way caused WWII, unless one considers finally awakening to the imperial aspirations of Hitler’s Germany to be yhff Ed cause of the war; and was only one of 3-4 principal powers (agents) that played a role in causing WWI.

You could argue that if the UK had intervened earlier, say at the stage of the militarisation of the Rhine, then the Nazi’s could have been defeated much earlier and easier. But this is hindsight of course and to ascribe blame would be too strong in my view.

Yes, thanks for that. What I should have said was that he appears to think the UK could have prevented both word wars but didn't, because of incompetent governance.

One of the most important questions that should be asked his competent weirdos is:

How do we stop the regulators from distorting the allocation of bank credit the real economy with risk weighted bank capital requirements stupidly based on that what bankers perceive as risky is much more dangerous to our bank systems than what bankers perceive or describe as safe?


Cummings’ adulation of quantitative types and out of the box thinkers sounds fantastic. But is also echos LTCM and all the crooked investment bankers and their derivatives trading gig, which did not work out well except for the crooked bankers and the overpaid regulators.

DoD has tried similar initiatives with DIUx(now DIU) and Space Defense Agency in the last couple years which bring energy and some added value but fade over time when their political sponsors leave office. Using Federally Funded Research and Defense Centers(FFRDCs) and Support Contractors provide talent that is paid well.

Liberate the central banker. The BofE gets the pound and operates as a non-profit for some lengthy period, maybe seven years. And lave a renewal option. Hence forth, the BofE never reports back to parliament, until renewal time.. Parliament can charge a one time payment.

Then all the rest of your weirdos will follow, i know that crowd. The BofE will unleash bearer cash and London becomes banking capital of the world.

Considerar um simulacro.

Hail the power of the nudge! With a few words Tyler reformed his comment section.

Congratulations to all participants.

People over propaganda: It's more important to get people with the right perspectives into key positions deep in the civil service than it is to try and convince the existing people to go along.
Propaganda over people: Once you have those people in place, then you have to create incentives / key targets & metrics for the existing power structures to head the right direction. It's tough to create the right key metrics, but existing bureaucracies are good at optimizing for whatever you created without needing to be micromanaged.

Trump's rules around getting rid of regulations aren't the best metrics, but are good enough that they've been a wild success compared to previous efforts. Consider when good enough might just mean at least heading things in the correct direction.

If you want to privatize a government function, removing/transferring the entire existing apparatus while setting up a competitive private process is faster and more likely to succeed than trying to create a hybrid where the civil servants managing the private efforts have opposing incentives. Helping the existing service providers form a company and bid on the resulting work is a good solution. Lets them put their money where their mouth is, plus they have experience.

Data science is great... when filtered through people who already know what's going on. He needs a process for taking their insights and bouncing them off front-line practitioners who may be able to help explain what's actually going on behind the numbers.

I've worked with software developers at the absolute highest levels and here's the key: The best coders are both smart and boring. Most software development is working on code that someone else wrote - usually poorly - and that takes a level of attention to detail that creative people often don't have. Rather, it is the creative people who made the big software mistakes in the first place as they are motivated to show off their brilliance by writing overly complex code that other people have trouble understanding.

The best way to do this is to pay a ton of money to a couple of staff-level engineers from a place like Google and put them in charge of hiring everyone else.

The larger issue is that Mr. Cummings needs to have some sort of a 'Risk Budget' and decide where he wants to spend Risk to get Innovation. Usually software is *not* where you want to have a lot of Risk.

Complex code can also be seen to be corner cutting and may set you up for an error or a program that is harder to improve and enhance. It can be hard work to make code simple. This expensive process is called refactoring. Code that other people have trouble understanding may increase the total cost of ownership. That other person may be your future self.

This entire enterprise is laughable. If you really want to reform the civil service, you need to fire everyone immediately and then only rehire the people who are effective. I have no idea how you do that, and neither does anyone else.

Reduce scope. Get the government out of non- core activities so that you can focus the always limited money, management attention, and political capital on the few critical core functions.

This. Unless you reduce scope and then rebuild it (or not) you end up playing whack-a-mole.

Cummings is completely on the right track when he says that governments are not focused on execution enough (at the local level in Austin, “plans” for improved transit, zoning reform, etc receive unanimous council support while serious on the ground legal changes are almost impossible). That’s why it’s good to see someone in office who’s prepared to be unpopular for a bit. Meaningful policy changes are almost definitional contentious (LBJ and civil rights, civil service reform in the early 1900s, etc)

However, to be sustainable, meaningful changes in a democracy also require convincing people to go a long. As Thatcher said “first you win the argument, then you get the policy”. It’s tempting for people like Cummings and the sorts of people (ie very smart) he’d like to hire to forget that their most important long term job is to convince people that their values are the right values. To do so would recommit the worst errors of the European Union, forcing policies through without convincing the people they impact that they are in fact the right policies.

Johnson and Cummings seem to have won the argument about leaving the EU. They seem prepared to at least try to push through serious platform-level changes in government. I’d hope, while doing so, that they remember their most important job is to keep winning the argument that serious government reform is necessary.

The easiest way to prevent the culture from reverting would be to introduce explicit randomness. This can be applied to hiring by seeing a threshold and an open lottery. Alternately some of the best teams I've been on have involved teams with a core and then every so often a rotation in from elsewhere in the company (and out, forcing us to be adaptable).

Regarding his team, I would suggest he look to the manufacturing sector instead of Silicon Valley types. There are some amazing IT departments and they are used to the longer horizons and high costs of failure associated with government. The good workers are used to entrenched bureaucracy and yet there are still success stories. They tend to be exposed to different elements of society. Watching how technology has changed oil or car manufacturing without the headlines of big tech companies should set an example.

i anotéra vía eínai i stigmí tis ídias tis práxis

I’ve spent almost 40 years as an environmental attorney trying to improve the regulatory system in the US, and I would love to see Britain change something to get better agency performance and to put me and my peers out of work. I want to restructure the agencies’ workforce to give more pay and promotions to the line workers who interface with the public and the regulated community. Agencies now give pay and promotions to policy and political people. They don’t reward the people who actually make the decisions that protect the environment in a way that works for businesses. So you get less competent people where it matters most. Of course there are times when you need high-powered policy and political people to deal with a particular issue. But you don’t need them so much that you make them the sole way to advance in pay and status. What that gives us is a system with not-so-great people making the thousands and thousands of decisions that carry out government’s mission, which gives me and my peers the job of fixing their poor decisions. Please, put me out of business.

Cummings: beware of memes!

“People, people, people” is a meme. Job interviews are very bad at selecting good talent. Ex ante, interviewers have zero marginal predictive power as to who might become a high performer upon employment. A well designed institution that allows new ideas to be tested and adequately rewards high performers (no matter their level in the hierarchy) will select for the best people. Whenever possible, there should be 2 separate hierarchies in 2 different buildings through which any given good idea can rise. This mitigates favoritism, and allows competition to select for the best ideas. These considerations are more important than precisely who is hired to begin with, assuming reasonably high hiring standards are in place.

He won't get any of the very best people, who for the most part already have cushy jobs. Time-limited teams with high stress (who needs more than full time? not me) and high expected attrition due to moody political bosses, working for mop head, btw ruining your cv, and on a british civil servant salary in central London? Come on.

1. Hire internationally. Carney has been effective at BoE. Also will help set an outward looking post-brexit mindset.

2. Focus on doing a few things well. He’s in danger of mission creep and getting nothing substantive done. It will be hard and slow.

3. The ideal areas of focus are those things that are both tremendously important and also very boring. Raise the status of the people who are working on the boring problems and they will buy in right away!

Carney? Delivered for the remainers.

Just talking about Brexit would create a recession, he said. Hiring internationally will get you establishment supporters, because they are hired by the establishment! The international hires speak foreign languages, except for Carney.

I don't mean the issues; I mean how people are chosen. And by whom.

Perhaps read Ken Miller's short book We Don't Make Widgets: Overcoming the Myths That Keep Government From Radically Improving. Maybe in Britain the kinds of changes that Cummings has in mind can occur faster than they can in the United States, but I am surprised that for somebody with a number of years of experience in government he does not sound as though he has a good understanding of how bureaucracies work.

Dynamic Land may be a good idea but I wouldn't be able to tell because it falls into the set of tech hype where it is impossible to tell what the product actually does. The whole website leaves me asking "ok, can I please just see the white paper to explain what is actually going on?" Which of course runs counter to its intended goal.

Part of the benefit of pencil and paper over interactive visualization is that there are implicit cultural rules around what is written and how to convey complicated information exhaustively that one must reinvent or at least relearn from scratch when working with a new medium.

There is a lot of experience in project management since Apollo, and Apollo - which was charismatic and winning the cold war - is not representative of government projects. In particular, matrix management is not a panacea. The MoD is now strongly encouraging contractors to use Agile project management on computer projects, which is at least modern. It is not necessarily ideal for the culture of the companies or the size of the projects involved. Cummings is right to concentrate on project management, but should not fixate on Apollo. The industry side of the UK aircraft carrier procurement appears to have succeeded.

The civil service has/had no problem in recruiting highly skilled mathematicians - at one time this was its most cost effective way of finding people who could be taught to be programmers. However such people may become discouraged when they find out that the civil service _promotes_ generalist administrators, and promotion is the only route to more money and some degree of security from interference from people of higher rank but lower expertise. Cummings already has access to mathematicians and data scientists from within the civil service.

I have long believed that the civil service selects _against_ project managers who achieve success. It is only by managing failure that a project manager can display the communication skills and interactions with higher ranked others that demonstrate that they are a good generalist administrator. Simply delivering on the requirements on time and within budget will not do this.

A good first step would be to ensure that every project has a single person with the authority and responsibility to manage that project, who remains on it for the life of the project, and who is rewarded according to the success of that project. If this is not feasible, the project is too big to manage and needs to be broken up hierarchically. In Agile terms, this emphasizes the role of the Project Owner as a "single wringable neck".

A more effective and efficient Whitehall could be an expanded Whitehall. This would increase the number of things it would make sense for Whitehall to control. Assuming that possession of a PPE does not guarantee unteachability, perhaps Cummings and Whitehall could compromise by retraining existing staff on project management and statistics. Project managers deployed and rewarded as described above should be keen to learn project management before starting their new project.

There is often a 'cargo cult' mentality with Agile, where they have the scrums, the sprints, the Kanbans, etc..., yet people wind up painting themselves into corners and don't deliver value. The book, 'Outcomes over Output' deals with this very well and would be useful for them should they pick an Agile approach.

No kidding. Last company I was at went to the dogs due to a slavish devotion to Agile. It's good for certain products, but not ours, which was exactly the kind of product you want to use Waterfall (or similar) style development processes on. Very few MoD software development products will be suitable for Agile beyond their citizen-facing website. This is exactly the kind of environment that needs strong software architecture, which means putting more effort into developing requirements and specifications than actual coding. Agile is for situations where the requirements are fluid and impossible to specify in advance.

IMHO the MoD are keen on agile, as customers, because they believe this makes it practical for them to change the requirements very late on. See e.g. https://www.baesystems.com/en/cybersecurity/feature/client-conversations-making-the-ministry-of-defence-more-agile. If you are really keen to do so, a surprising number of computer systems can be implemented using an application server and a browser interface, making it possible to deliver impressive demonstrations using the infrastructure available to support non-real time systems using large computer resources and essentially unlimited bandwidth.

Yeah, good point. Agile is often used as an excuse to skimp on preparatory design work. Don't build a nuclear sub control system by setting up a website and saying, "ok, what do you want next?" Which some of the zealots out there actually think would work.

That's a good example. And nuclear sub control systems are simple -- even trivial -- compared to systems of social control (for one thing, there is no such thing as social control, unless you're North Korea and even they have smugglers, defectors, etc.). That's a key factor that too many technocrats ignore, especially the ones with backgrounds in natural science and technology, as opposed to social science.

That doesn't mean that the social scientists will be better at this than the natural scientists. But the natural scientists and tech-types need better backgrounds in history, economics, and other social sciences so they'll stop making mistakes such as proclaiming how self-driving cars will be taking over the roads in a couple of years, IBM's Watson will revolutionize medical diagnosis, online classes will make traditional classrooms obsolete, etc.

Do we really want systems of social control that work? Whether the customer is China or Google, I'd rather do without them.

I used North Korea as the example of the country that currently imposes the most social control on its populace. I would think it's obvious that just about everybody (including a lot of North Koreans, I imagine) view that as a Bad Thing.

But control aside, every nation has to do some planning, managing, etc. And the important point is that it's harder to do that in a complex human system than in a complex system in the natural sciences.

Pretty much all the cited work comes from online videos and social media. Nielsen and Victor are brilliant, but there is a ton of related material that Cummings does not cite because it is not highly cited on the Internet. His knowledge outside his domain of expertise seems to come from airport books bestsellers and summary articles popularising actual research (Watts, Strogatz, Kahneman).

Senior civil servants, who have a great deal of power and a real sense of solidarity (they even have their own Union, named the First Division Association, which shows a suggestive contempt for those they manage and the organisation as a whole, which they would claim to feel loyalty towards) will resist all meaningful change to the structures and incentives of the senior civil service (they showed no such reluctance in implementing huge cuts to the civil service as a whole from 2010 onwards). These changes are essential if the civil service is to be properly reformed.

All other reforms should therefore be delayed until the power of promotion has been taken out of the hands of civil servants and given to politicians, who should be advised in making appointments by external headhunters rather than existing civil servants.

One key indicator that this reform has been successful will be the dissolution of the First Division Association - a voluntary dissolution, led by the example of the new departmental heads, as why on earth would senior civil servants want to be members of such an organisation under the new dispensation?

There's just one flaw in this whole proposal: the government can't come close to paying this superbureacrat their opportunity cost.

Just look at the problems the US Civil Service is encountering when attempting to hire "ordinary" Ph.D economists (I work for such an agency, and believe me, the challenge gets more difficult every year.)

Cummings talks about red teams in his writing, the Netherlands has an effective one in the BIT:
It audits big government IT projects, and can advise to close them down if certain criteria are exceeded (budget, time, chance of success) .
I think this Cumming's project is fascinating. What could a supportive Dutch person do to help?

There is plentiful evidence from institutional economics and economic history about the importance of creating durable institutions that will let entrepreneurship and innovation flourish. DARPA, the Manhattan Project and Apollo were all impressive; but it is equally important to selectively and where workable copy the best institutions from other countries and experiment to find what government institutions work best at the technological frontier to facilitate and allow innovation by entrepreneurs.

Interesting idea. I have two quibbles:

1. He wants diversity but only physicists and mathematicians from “top” universities may apply. Setting aside the problems with assessing “top” (e.g. ratings of USN&WR physics departments correlate with size rather than per capita output), this is a good way to miss out on unconventional thinkers. He should focus on skills, creativity, curiosity, and personality rather than pedigree.

2. He is looking for technocrats and outliers. Fine, but he has the same blindspot you had in identifying the best communicators....religion. if you want your decision making to be palatable to your constituency, you have to understand the appeal of religion. Mainstream religious thinkers (n.b. established/mainline religions are not mainstream) provide valuable insider info. Secondly, theological reflection has a rich history and is a form of science. The difference between natural science and theological science is the source material. One may reject the validity of theological source material while appreciating how it is studied. Perhaps natural scientists could learn something from the way theological results have evolved and been disseminated among the public. How is it that prosperity preachers can convince third world people that if they have faith, they can be rich, Catholic theologians can convince gay men to live celibate lives for a future reward, or evangelicals can convince harried parents to get up early every Sunday for church? Given that scientists and public policy experts have done such a poor job persuading the public of our evolutionary origins, the peril of climate change, safety of food irradiation, benefit of gmo foods, and value of freer immigration and trade, perhaps they would benefit from the insights of religious believers.

Religious thinkers each convince maybe 5% of modern educated people often repelling the other 95%. Not many Billys Graham. Politicians have to convince 45% of modern educated people.

That thinking is what got us Trump and Brexit. Noneducated and not so modern people vote too.

I mean that YOU are repelled by the guru whom I adore, and vice versa, whereas almost never do we see a Billy Graham type who appeals to Christians of all denominations, let alone non-Christians.

1. What does prediction hope to achieve? The very best forecasters are significantly better than status quo at narrow factual questions like "will Assad leave office this year" but will still make many mistakes, including embarrassing high profile ones, and even to put them to work you need to know what questions to ask far ahead of when the topic enters the public discourse. This will mean you end up asking a lot of boring questions that never become relevant.

2. Who is doing mega civil service (or mega bureaucracy, or mega project management more generally) well? Actually doing it today, not "here is a video of a cool vision for working on a project" or "wasn't the Apollo program awesome". And don't assume any successful organization is doing the particular job of mega bureaucracy well- Apple or Facebook or Trump 2016 wouldn't qualify, I think. I suspect even if you could find such an organization today (ExxonMobil? HBO? The Villages?) they are going about it differently than you plan to do things yourself.

Cummings thinks his Vote Leave campaign did well "If you think you are one of the a small group of people in the world who are truly GREAT at project management, then we want to talk to you. Victoria Woodcock ran Vote Leave — she was a truly awesome project manager and without her Cameron would certainly have won. We need people like this who have a 1 in 10,000 or higher level of skill and temperament."

If you don't believe in the cult of the generalist administrator then you have to consider the possibility that project management expertise might be specialized by domain. It is even possible that some of it is personality fit, in which case it will be difficult to train or even assess, except by considering the track record of the project manager in previous similar projects, in which case the cult of the generalist administrator is really in trouble: not only does it not allow people to accumulate the required experience for their jobs, it also withholds the information required to match people with jobs. Consider the additional difficulty introduced by requiring of an NFL coach that the players who are on the offensive line in even numbered games must play wide receiver in odd numbered games. Consider also the plight of the player on the offensive line. How should they play to increase the chance of them making the team as a wide receiver next week?

It's hard to say from his blog post what Mr Cummings is trying to achieve other than creating a buzz. Maybe he's intentionally being secretive, which would be fair given his position. Anyway, some questions seem relevant.

1. Is he aiming to undertake mega-projects, or to improve the routine operation of the civil service? Is there any evidence that the two goals call for similar approaches? I'm unfamiliar with the literature, but I'm quite skeptical a priori. His specific examples seem to be all engineering projects, whether cutting-edge science (Manhattan, ICBM, Apollo) or much more mundane (C.C. Myers). Most civil-service work is quite unrelated to engineering, and there's a long (not to mention slightly Stalinist) record of engineers and physicists thinking they should be good at running unrelated projects too and being proven wrong. I noticed Cummings's blog has a specific category for 'Econophysics' (his scare quotes) and I couldn't help recalling SMBC.

2. What has the cool-sounding tech jargon of prediction science, data science, AI, cognitive technologies, computational rationality, machine learning, networks and so on got to do with either of the goals above? Perhaps Mr Cummings has definite answers but doesn't want to divulge them. However, his post suggests the alternative reading that he merely finds these ideas cool and futuristic and thus thinks they ought to be tapped as a solution to whatever problems he plans to tackle. That seems worrying. First, it's typically better to look for solutions that fit the problem, rather than fitting the problem to a given solution. Second, it's typically safer to implement boring known solutions than to devise glamorous novel ones. Commenters above already made this point about picking low-hanging fruit first.

3. Perhaps not truly distinct from the previous point: where do the unusual, young, tech-loving, clever weirdos fit into the picture? I don't doubt that's a class of people that has made great contributions to innovation. But is it also one that has had a successful leadership role in making large organizations run more effectively and efficiently? In implementing engineering mega-projects? Didn't both the Manhattan and the Apollo projects rather recruit the established leaders of their scientific fields, Nobel-prize winners and all?

Amittedly, the blog explicitly mentions having a small, odd team coming up with weird and wonderful ideas, and that's something different. But why is the prime minister's office the right place for such a team? And how are they going to "find and exploit" ideas that seem bad but truly are great? Again, it would seem natural to try implementing first policy ideas that seem good --- and those too benefit from rigorous policy evaluation. I've had minimal but non-zero experience talking to policymakers, and I found that civil servants were genuinely interested in policy evaluation, while it was politicians who dismissed it because they'd rather trust their vision instead. I hope Mr Cummings knows better, but it isn't reassuring he's looking for expertise in forecasting but not in policy evaluation.

4. How is Mr Cummings going to get civil servants on board with his transformative plans? He and Mr Johnson have a great track record of salesmanship when they're selling to British voters. Point F in his blog post aims to build on this and sounds the most likely to succeed. However, are TV and digital story-telling as good at building internal support within the civil service as they are at building external support among voters? I'm no communication expert, but that doesn't look too likely. If instead Mr Cummings plans to ram reforms down the throat of civil servants, promoting "many great officials, particularly younger ones" but sidelining if not firing even more not-so-great officials, particularly senior ones, his chances of success seem slim. He could counter that he and Mr Johnson have succeeded at purging the parliamentary Tory party and bending it to their own will, but I suspect both the law and the sheer size of the civil service prevent them from having the same kind of leverage over it.

5. Always worth asking: how does Mr Cummings think his reforms would make the civil service work better under a Corbyn government? It's hard to tell from his extremely vague blog post, but in general politicians tend to get mad at bureaucrats (and judges) for stymieing their own policy initiatives. There's been plenty of noise along those lines from the Johnson campaign. So it seems fair to infer one of Mr Cummings's reforming goals is to get policies (and perhaps engineering mega-projects) implemented faster and more decisively. That may well be a good idea at the margin in the UK today. Specifically in the context of infrastructure and construction more generally, British planning policy is infamous for having gone utterly BANANA (i.e., "Build Absolutely Nothing Anywhere Near Anyone"). On the other hand, checks and balances including institutional inertia have their uses, and it's easy to forget them in the heady aftermath of a decisive electoral victory.

Well, I've actually written a paper about 'early warning systems' in complex dynamical regimes and I've been one of Tetlock's superforecasters, so let me take a swing at this:

The main tension that needs to be negotiated is that balance of local knowledge/skin in the game with local sub-optima and externalities.

This tension can never be fully resolved and needs to be negotiated differently for various kinds of decisions.

In general, nobody has the answer to this - most people tend to hollow out the middle layers because those look worst. Individualists (libertarians) hate local busy bodies; elite bureaucrats (classical liberals and totalitarian fascists and communists) hate local incompetents; middle rung government bodies (conservatives) hate their [so-called] peers next door. However, that is the sweet spot (or at least the best compromise) for so many systems; to put the decision making power near the smaller end of the scale, but not on the individual; locally diverse and therefore always a bit of a pain.

Being an average circuit design engineer based in bay area I have often wished I was an average any of the below -

_Unusual mathematicians, physicists, computer scientists, data scientists, unusual economists or unusual software developers or a misfit_

Because how handsomely regular engineers working on machine learning or data-science (broadly speaking) are paid over here and how many different kind of opportunities they have for their careers. To be sure regular engineers here in bay area tend to be very talented and capable.

However for the best of the above, there is an absolute arms-race. If not amongst the start-ups that have good chance of being the next batch of unicorns or the most valuable tech-companies based here, then amongst the VCs who want to fund these kind of people to see what ideas they can turn into businesses.

I would be curious to learn what the civil service will be able to offer in this arms-race. A lot of companies have lots of talented teams, money, compute-infrastructure, well-organized data-sets (i.e. data that is ready for application of machine learning) and a lot of interesting problems. They also tend to offer a lot of freedom and flexibility to go after these problems.

I don't know what lessons civil service reformers can draw. However if they manage to put together a team of such people and retain them, other organizations and governments should seek to draw lessons from the civil service reformers for how to hire and retain a talented team.

Thanks to prof Tyler Cowen for posting this. (Thanks to people for the comments.)
Dominic Cummings: "training of people for high performance"...
I would like the full listing of methods on how to 'train people for high performance'. Does that include creativity method (for engineers, developers, startup founders, research scientists, etc.)? Could we accelerate progress by improving the way humans produce new ideas?

He should aim to do a limited number of things well. An attempt to reshape the entire state administration will likely fail.

He might also reflect on his time at the Department of Education. Should he repeat his performance there, he will ensure a one term Johnson government.

Cummings knows full well that substantial change is impossible, nor desirable, and that his real prize is merely in convincing the public that government has changed. From a micro point of view, what incentive does a party with a significant majority have in disrupting the order they themselves have created and manifestly benefited?

To paraphrase di Lampedusa: "Everything must have the appearance of change, in order to stay the same"

As someone who works in a commonwealth government and shares some of Cummings' broader overall goals, I'd suggest the following (may not be generalizable to the UK from the commonwealth):

1) Convincingly formulate the need to shake up the civil service as a way to provide generational change and 'attract the youngest and brightest'.

There is serious and exploitable grievances from younger, upstart, and highly skilled civil servants towards those they see as coasting primarily on seniority, status quo, and networks. I think there would be a lot of appetite institutionally - especially with the ability to hire (and fire) more easily. More 'neutral' and innovative application and exam systems could also be seen as providing fairness.

2) Look at DFID as a starting point where you can push for more radical changes in funding models, measurement, etc.

It is easier to be explicitly experimental in a field where your recipients or stakeholders aren't domestic special interests who can write to their MPs. Likewise, you can do the reverse and see where the most push-back will occur.

3) Find a way to quantify and cut on the costs from the approval cycle (if it hasn't been done). Like literally find out how much tax dollars, time and content loss occurs from the initial product to what ends up to ministers and cabinet.

4) Joseph Heath argues in the "Ethics of Public Administration" that the proper "professional morality" of the civil servant is to bring about "pareto efficient outcomes", such as solving difficult collective action problems to bring about gains for everyone. This seems in line with much of Cummings' intended goals and I think communicating many of your projects to the public service in this manner would be attractive. Also, find ways to improve and unleash this (state) capacity more generally.

5) Find a way to strengthen, either through increasing capacity and/or independence, the program evaluation function. In our context, they are housed within each department and makes their ability to be publically critical very difficult. Perhaps finding a way to increase face time with cabinet.

Since no one seems to have mentioned it yet and its the top of my list for successful case studies, I'd recommend a review of the New Zealand civil service reform successes and failures in the 1980s - plus the longer term durability of the reforms.

Maurice McTigue at Mercatus would be a great resource to talk with since he was a key participant. Here is an excerpt of him discussing his work in New Zealand:

"As we started to work through this process, we also asked some fundamental questions of the agencies. The first question was, “What are you doing?” The second question was, “What should you be doing?” Based on the answers, we then said, “Eliminate what you shouldn’t be doing”—that is, if you are doing something that clearly is not a responsibility of the government, stop doing it. Then we asked the final question: “Who should be paying—the taxpayer, the user, the consumer, or the industry?” We asked this because, in many instances, the taxpayers were subsidizing things that did not benefit them. And if you take the cost of services away from actual consumers and users, you promote overuse and devalue whatever it is that you’re doing. When we started this process with the Department of Transportation, it had 5,600 employees. When we finished, it had 53. When we started with the Forest Service, it had 17,000 employees. When we finished, it had 17" (https://imprimis.hillsdale.edu/rolling-back-government-lessons-from-new-zealand/)

Excellent! Thank you, I forgot about New Zealand. You're right, it's another great example.

Paradox and gift of Cummings is balancing rational reasoning with irrational gut feel.
The effect of advice on these kinds of people can often unpredictable (unless exceptionally data-driven).
But I think: believe in yourself, there will be problems, follow your own advice, remember this is just preparation for the years ahead which will be more important.
Rest is up to Chaos (for now).

And will add -- I suspect Sam Altman's recent twitter posts were written with Dom in mind.

Understand that bureaucracies are as competitive as markets in their own way - with different incentives. At the department and ministry level, the most long lived and successful bureaucracies are the ones that prioritize their own existence over their mission. Even at the individual civil servant level, the one who prioritizes "customers" over the bureaucracy will usually lose out to the "company man".

If you want to improve the utility of the bureaucracy, you must keep this in mind and adjust incentives accordingly.

Comments for this post are closed