Dominic Cummings - An 'Odyssean' Education [review]

post by Louis_Dixon (bdixon) · 2020-05-29T16:52:28.504Z · score: 66 (43 votes) · EA · GW · 10 comments

Thoughts are entirely my own and not those of my employers past or present.

In 2013, Dominic Cummings posted An Odyssean Education: Some thoughts on education and political priorities, an essay covering a range of topics relevant to government and decision-makers. Cummings is often portrayed in the media, as a curmudgeonly genius, and the term 'smartest man in Britain' is bandied about, not entirely ironically. Is this a man who's really figured it all out?

I know several EAs who have at least some admiration for a man they see as not playing by the rules, but getting things done. Cummings is a polarising character, and I figured I’d give his essay a read to try to make up my own mind. Is he really the genius he’s reputed to be?


This paper is Cummings' attempt to sketch a curriculum of what modern education should be. The first thing to say is that the essay is long - 237 A4 pages, or 135,000 words (the first Harry Potter book is ~77,000), and it’s packed full of footnotes and references. He references a very wide range of topics in technology, science, maths, and philosophy. It is impressive that an individual in government has read so widely: it references the energy transition, synthetic biology, deep learning, and quotes Kant, Nietzsche, and Cicero.

Parts of it are fantastically fun. He opens a discussion on political systems with an excerpt from the James Bond film From Russia with Love, and he brings up interesting ideas from synthetic biology, computing power, and has a desire to guard against major risks (he does raise the risk of pandemics in this essay). It aims to have the same sort of broad sweep as Yuval Noah Harari’s books. Longtermist EAs will be excited to hear Cummings raise concern about catastrophic risk, and he even argues that governments should defend against asteroid risk by installing lasers in space.

I can see why some people might find Cummings so appealing. Many of us have encountered dysfunctional bureaucracies, and many EAs share with Cummings an interest in Tetlock and Yudkowsky, among other topics. I share his scepticism of classical macroeconomic models, built both on wobbly conceptual ground and with limited explanatory power. But there is more to say about this article, and the thoughts behind it.

The first is to realise that this tract is not an education. It's one man’s parochial and partial view on a couple of topics, and a series of disconnected thoughts in many others. In an education, you'd hope to develop critical thinking skills, and then go off into diverse fields and work out your own answers. But Cummings takes a number of set positions on complex and contended issues, and is more of a stubborn hedgehog than a truth-seeking fox.

He argues that European democracies outperformed China from 13C to 19C because of decentralised markets,which could be a bit of an oversimplification but seems roughly fair. He throws in a sentence about the Ming dynasty, another reference to the Ottoman empire, and concludes that these three examples prove that laissez-faire systems are always superior, everywhere. He then moves on to three charts showing improving global trends from the 1870s to the early 2000s, which many people, including The World Bank, argue were largely driven by China's mixed state-led markets approach. Further discussions of China’s phenomenal growth, led from the top down are conspicuously absent from the rest of the piece.

Fabians who caricature Cummings as a free market evangelist after Nozick and Thatcher I think are oversimplifying his view. It’s true, he does quote Hayek at length over and over, but he also spent years at the Department of Education, and his fascination with government projects like DARPA and the Apollo mission is evident throughout. He agrees that markets can fail, need to be regulated, and that the government plays an important role in fostering innovation.

The word ‘institutions’ appears 57 times in the piece, and in every instance it’s a critique. They don’t adapt fast enough, they don’t incentivise the ‘best’ performance, they get in the way of something better. I studied Economics and Politics as an undergraduate, and don’t get me wrong, I think markets are great. But even advocates of market approaches, like Harvard academics Iversen and Soskice (2019) argue that successful capitalism has to always be embedded in the institutional features of democratic states.

Lant Pritchett, one of the top-cited developmental economists at both Harvard and Oxford describes development as four interlocking transformations across four dimensions:

‘an economic transformation from lower productivity to higher productivity; a political transformation to governments more responsive to the broad wishes of the population, an administrative transformation to organizations (including those of the state) with higher levels of functional capability for implementation, and a social transformation to more equal treatment of the citizens of the country (usually with a sense of common identity and, to some extent, shared purpose).’

Pritchett’s thesis was at the heart of one of the most popular forum posts to date, ‘Growth and the case against randomista development [EA · GW]’.

But what is missing from Cumming's piece is exploring exactly how states and markets should be best interlinked to drive improvements for their citizens. There’s a tension between his desire for decentralised and responsive institutions, and his failure to engage with the key topics that limit their ability to do so: electoral dysfunction, regulatory capture, corporate monopolies, corruption, and inequalities of opportunity.

Cummings admits that inequality is increasing, 'Corporate profits as a share of GDP are at a 50-year high while compensation to labor is at a fifty year low', but quickly moves back to talking about technology. There's no discussion of austerity, which even Boris Johnson recently declared to have been a failed experiment. In America, home of the free market, real incomes and life expectancy have been declining, while ‘deaths of despair’ from suicide, drug overdose, and alcoholism have risen dramatically, as analysed by Anne Case and Nobel prize winner Angus Deaton in the recent Deaths of Despair (2020).

Cummings doesn’t give much thought to this, concluding that the ‘winner-takes-all’ approach ‘can be economically efficient and socially damaging, argue some.’ This thread, arguably one of the most important for Western democracies, astonishingly, is taken no further. He briefly discusses automation and the labour market, then goes back to a fresh rant on quarks, evolution, chaos theory, economics and plasma physics. I start to get the feeling like I’m being cornered by a man in a pub.


In Brexit: The Uncivil War, Cummings was played by the smooth Benedict Cumberbatch, whose previous roles include Sherlock and Alan Turing. I wonder if Cummings compares himself with Turing. He’s put two links on his site to Guardian articles which describe him as a genius. He hurls footnotes and references at the reader, with some pages having more footnotes than text itself, and it's easy to find the piece bewildering.

But at points like his discussion of China, you can see the threads are wearing thin, and holes start to appear. After the tenth reference to Thucydides or the twentieth reference to Feynman, I started to wonder how broad his reading really was. Often the text moves on to another topic without linking to the previous one, and without having developed the ideas much further. Several sections, particularly that on energy, simply repeat the conclusions of one book, and do a poor job of grasping current debates in the fields. Climate modelling is an entire academic field, the clean energy transition is another entire field and a set of political questions, and Cummings reduces both to a set of bullet points from one book.

One of the great failings of the piece is his lack of engagement on climate change (while space exploration gets three full pages). In his only paragraph on the topic, he first describes global warming as a 'debate', but concedes the 'probability of very bad case scenarios are impossible to answer with precision', and that we should take them seriously.

Yet carbon emissions get very little airtime in his discussion of energy systems, and the impacts of climate change (heatwaves, floods, food shortages, and others) are not discussed at any point in the 235-page essay. It seems that people who adore Hayek struggle to engage positively with climate change, a massive externality - a market failure, where most roads show that the government needs to step in with a carbon tax.

Much of the piece covers potential future developments in technology - he’s fascinated by Moore's Law, algorithms, and how advances in machine learning could revolutionise existing fields. But it soon becomes hyperbolic - he draws false equivalences between the count of neurons in a brain and the processing power of computers, when this field still has many deep uncertainties.

There's no discussion of the societal impact of advancing technologies. In a section on artificial intelligence, he has no discussion on labour market displacement, the ethics of drones and autonomous weapons, or even the potential challenge of AI alignment. There's nothing on how algorithms are interacting with our social structures, locking in biases, and systematically discriminating against women. Cumming's essay has no real discussion of people, or reflection on lived experiences.


Notably vacant from Cummings' essay is any discussion of political representation, democracy, or even voting systems, which are in desperate need of reform. Our institutions mediate our private desires and provide us with the infrastructure in which markets and creativity can flourish. This argument was powerfully made in Why Nations Fail by Daron Acemoglu and James A. Robinson, reviewed favourably on the forum in a popular post here [EA · GW].

To conquer humanity's greatest challenges now and in the future, we need political structures. Arise, Leviathan. Only when our political structures are built on strong foundations in the lives of individuals can they deal effectively with our most pressing challenges.

Cummings would do well to listen to his own advice: be humble, listen to the data, give caveated conclusions, and explore complex fields with expert guidance carefully. But when it’s his turn, he’s up on the soapbox, arguing with weak data that we should push hard on technical economic growth, and drastically reform everything else.

For Cummings, institutions (at least as they currently stand) get in the way, and what we need to do is drastically reform them to be as responsive as possible. The drive for the new is nothing new. And tearing down the system is often not a good move. In 1910s Italy, Futurism championed speed and technological development, and shared a dislike of fussy institutions. In 1919 the Partito Politica Futurista merged with Mussolini's newly formed fascists, and Italy slid into a fascist dictatorship for over 20 years.

But Cummings hasn’t learnt this lesson from history: in 2019 the current government, with Cummings as a senior advisor, tried to shut down parliament to quel debate on EU Exit, before their actions were ruled to be unlawful by the Supreme Court of the United Kingdom. Chilling.


All told, what do I think of Cummings? I think the paper raises some interesting topics. Individuals often do think irrationally, and politics could do a lot better. Technology has contributed to huge gains in welfare for humanity over the past century, and I think a raft of new technologies offer cause for both hope and concern. I would love for governments to become more effective and evidence-based, but also to act with an understanding of the complexities of our societies.

What I do like about Cummings is his ability to think on the macro-scale of human history, and look ahead at what developments might come in the future. But when you overreach and try to create a theory of everything, I think you end up with a theory of nothing. We need sensible, modest and well-informed people in government to weigh the evidence carefully and act with the public in mind.

On the most crucial topics, and in capturing the nuance and complexity of the real world, this piece fails again and again: epistemic overconfidence plus uncharitable disdain for the work of others, spread thinly over as many topics as possible. The real content of politics, and the lived experiences of individuals now and in the future, will be shaped by our cultures, our institutions. It's a messy struggle, and it really requires engaging humbly with experts in a range of fields, and thinking compassionately about the complex and intricate nature of our personal and political lives.

10 comments

Comments sorted by top scores.

comment by Commentator · 2020-05-30T10:52:34.858Z · score: 36 (26 votes) · EA(p) · GW(p)

I didn't find this review very helpful. Hopefully I'll be able to explain why and hopefully people will read Cummings's essay for themselves.

As you note, this is not (straightforwardly) an essay about education. It's a wide-ranging discussion of his views on a handful of core themes, and a series of disconnected thoughts on many topics. You might think these are vices in an essay, but I think it's only fair to evaluate the essay in terms of what it's actually trying to do.

As far as I can tell, it's not trying to be a comprehensive political manifesto, but most of your criticisms seem to simply be objecting that he's not talking about things you'd like him to talk about (or just not talking about them as much as you'd like):


- "One of the great failings of the piece is his failure to engage with climate change (while space exploration gets three full pages)."

- "Yet carbon emissions get very little airtime in his discussion of energy systems, and the impacts of climate change (heatwaves, floods, food shortages, and others) are not discussed at any point in the 235-page essay."

- "There's no discussion of the societal impact of advancing technologies... There's nothing on how algorithms are interacting with our social structures, locking in biases, and systematically discriminating against women. Cumming's essay has no real discussion of people, or reflection on lived experiences."

- "Notably vacant from Cummings' essay is any discussion of political representation, democracy, or even voting systems, which are in desperate need of reform."

You also complain that "Often the text moves on to another topic without linking to the previous one, and without having developed the ideas much further." But this is inevitable given that the essay ranges across an enormous variety of different topics of discussion. It might be frustrating to you as a reader if you didn't want to read a loose collection of Dominic Cummings's personal thoughts about multiple topics, but that's what you signed up for in reading the essay. There's nothing wrong with writing a personal essay which alludes to a wide range of topics without going into depth on most of them.

Notably, practically the only specific concrete criticism of anything in Cummings's essay that I found in your review was concerning one of these many digressions on technical issues of interest to him that he goes into: "he draws false equivalences between the count of neurons in a brain and the processing power of computers, when this field still has many deep uncertainties" which, for one, is a fairly mild objection but also seems only tangential to the main themes (to the extent there are any) of the essay.

I think it would have been more useful if you engaged substantively with some of the arguments Cummings makes and explained specifically where you think he goes wrong. For example, you grant that "[Cummings] agrees that markets can fail, need to be regulated, and that the government plays an important role in fostering innovation." But then you write "even advocates of market approaches, like Harvard academics Iversen and Soskice (2019) argue that successful capitalism has to always be embedded in the institutional features of democratic states" as though this you are pointing out a flaw in Cummings's position. As it stands, if I hadn't read Cummings's essay, I would have virtually no idea from reading your review what his distinctive views are and I barely know what you actually disagree with him about.




comment by Louis_Dixon (bdixon) · 2020-05-30T11:28:15.009Z · score: 18 (13 votes) · EA(p) · GW(p)

Thanks for your feedback! I agree that people should read the essay and make up their minds for themselves.

To address the points you raised:

  1. As a set of notes, it claims to address the most important economic and political priorities, and I think this is the criteria on which it should be judged. My view is that it fails to do so.

  2. My main beef with Cummings is that he overreaches in areas he's not familiar with, and he has uncharitable disdain for the work of others, and I think this is consistent throughout the piece.

  3. In my view there is a tension between his admiration of big government projects, but his failure to talk about democracy and link government activity to public needs. If I rewrite the piece I will make this more explicit.

comment by Commentator · 2020-05-30T13:33:44.872Z · score: 11 (6 votes) · EA(p) · GW(p)

Thanks for your reply!

1. I don't think this is the right standard (or a fair standard) for a few reasons.

1a. I don't recall Cummings claiming that this was a _comprehensive_ description of the "most important economic and political priorities." But it's been a while since I read the essay, perhaps you can correct me.

1b. Even if this essay was presented as a description of the top economic and political priorities, it's always trivially easy to think of _some_ potentially highly important issue that wasn't discussed. I think it's a bad idea to dismiss an article on that basis. One could easily think of multiple potentially huge global issues that aren't discussed at any length in _Doing Good Better_ for example. As such this seems like an unrealistic standard.

1c. This strikes me as an asymmetric demand for rigor. If someone wrote an article or blog post discussing their personal thoughts on how climate change, poverty, various forms of discrimination are all inter-linked as top priorities, I think it would be churlish and inappropriate to respond that this didn't discuss (at sufficient length for your liking) technological risks or democratic reform or whatever.

2. Maybe he does, perhaps especially in one of the various digressions he goes into where he just starts talking about Quantum Computing or Set Theory or whatever. But I didn't see anything particular in the review to suggest that he does. Per my original comment, it seems to me that you mostly just observe that he hasn't discussed certain topics.

3. I think you presenting an argument for this would be worthwhile. Notably, he _does_ discuss democratic and democratic institutions at numerous points throughout the essay, so I think this would require engaging in more detail with the substance of his arguments.

comment by Louis_Dixon (bdixon) · 2020-06-01T08:22:53.733Z · score: 2 (2 votes) · EA(p) · GW(p)

1a,1b. My interpretation is that the essay aims to provide a sketch of what should be covered in an education, e.g. p7: 'In order to provide some structure to such an enterprise, a schema of seven big areas and some material is sketched'. I think it is a poor sketch.

1c. I think that if DGB claimed to be a sketch of the most important areas there are, and it missed out a huge area, e.g. global poverty, then it could be rightly challenged as a poor guide.

2. In the review I mention that his digressions into neurons and processing power draws false conclusions. He is also particularly scornful of the social sciences, and arts and humanities.

3. It is true that he briefly discusses voting in a section on decision-making and in an endnote, but in both places it is incidental to his point rather than a main topic. A discussion of voting and democracy is absent from the section I was expecting it, ' 7. Political economy, philosophy, and avoiding catastrophes.', and I cannot see it covered elsewhere in the piece.

comment by Commentator · 2020-06-01T10:56:28.054Z · score: 6 (6 votes) · EA(p) · GW(p)
My interpretation is that the essay aims to provide a sketch of what should be covered in an education, e.g. p7: 'In order to provide some structure to such an enterprise, a schema of seven big areas and some material is sketched'. I think it is a poor sketch.

I note that you've switched from suggesting that the essay is supposed to give a comprehensive account of "most important economic and political priorities" to suggesting this is supposed to give "a sketch of what should be covered in an education", but I think this is a better characterization of the point of the essay so that's fine.

However, I think it's only plausible to think that the essay is aiming to sketch some necessary prerequisites of an education, not as claiming that the contents are sufficient for or fully exhaustive of a good education. Complaining that he hasn't discussed certain topics (in enough depth) seems an unreasonable complaint: of course, no single essay is going to describe every topic that should be in an education.

As an aside, I think it's somewhat perverse given that the majority of educations almost completely fail to address topics like probability, forecasting, overcoming bias, decision-making about risk (the topics which Cummings highlights and EAs are rightly concerned about) and yet the response is to single out this essay to complain that it doesn't egage with topics like climate change (which it actually does).

In the review I mention that his digressions into neurons and processing power draws false conclusions.

Yes, in my original comments I highlighted your claim that "he draws false equivalences between the count of neurons in a brain and the processing power of computers, when this field still has many deep uncertainties" because that was the only point where I could find you actually disagreeing with something specific in the essay. Having reviewed every mention of "neuron" in the text, I'm not sure that this does actually lead to any "false conclusion." But that aside, as I noted in my original comment, this putative "false equivalence" in a field with "deep uncertainties" concerning a side point that he goes into doesn't seem like much of an objection to the essay as a whole.

It is true that he briefly discusses voting in a section on decision-making and in an endnote, but a discussion of voting and democracy is absent from the section I was expecting it, ' 7. Political economy, philosophy, and avoiding catastrophes.', and I cannot see it covered elsewhere in the piece.

I think this falls under my general response that you can't reasonably criticize the (already hugely long, multi-topic spanning) essay for not discussing certain specific topics (that it discusses) more. Much of the piece is arguably about improving democratic institutions (through discussions of better thinking about risk, probability, forecasting and complex systems), so just saying that it could have discussed voting more in a specific section seems an unreasonable demand.

comment by Louis_Dixon (bdixon) · 2020-06-01T15:46:07.044Z · score: 5 (4 votes) · EA(p) · GW(p)

Thank you for your thoughtful feedback. I've thought about the points you raise and I think they are all good challenges. I agree that Cummings raises interesting and relevant points in a range of areas.

I think you and I have different views on several points. The most important one seems to be that I think the piece at times aspires to represent the whole, even as a sketch ('An Odyssean education'). And my view is that two of the most important areas (if I were writing such an essay) would be voting and political systems, and climate change, neither of which I feel get sufficient attention in the piece. It seems that on each of these topics you take a different viewpoint, which I respect. Again, thanks for your feedback. Unless there's much else you think we can do to resolve this difference, I'd probably leave it there.

comment by MathiasKirkBonde · 2020-05-30T12:15:31.774Z · score: 20 (13 votes) · EA(p) · GW(p)

I have read the entirety of "An 'Odyssean' Education" as well as another ~300 pages of blogposts from him, fascinated by how someone with someone who seemingly has formed his beliefs from much of the same literature I have, comes to such vastly different conclusions than me.

I mostly agree with your book review, your summary hit the nail on its head.

An 'Odyssean' Education, reads like a rough first draft at its best and like a stream of thoughts at its worst. But that's alright. Its purpose is to be a rough sketch of Dominic Cummings' worldview. It gives us insight to why he holds his beliefs and push for certain policies but not others.

I wish I could read a similar document for other prominent political figures, and get a similar understanding of their worldview as well.

comment by Philip R · 2020-05-31T22:36:12.009Z · score: 4 (9 votes) · EA(p) · GW(p)

I had read this following a link in the UK Guardian newspaper "Mad, Bad, or Brilliant", how could I resist?

I agree with most of the above, what I found also was that climate change was the issue that doomed the essay. Throughout the essay he discusses moving forward economically, what this means and how best to move forwards societally. However whilst he gives climate change lip service I found that all his other propositions are undermined by that he gives lip-service only. He has stated the problem but then not made any effort to figure it into any other solution. This is my mind leaves him as a climate denier in standing.

comment by Elityre · 2020-06-05T22:38:25.268Z · score: 1 (1 votes) · EA(p) · GW(p)
On the most crucial topics, and in capturing the nuance and complexity of the real world, this piece fails again and again: epistemic overconfidence plus uncharitable disdain for the work of others, spread thinly over as many topics as possible.

Interestingly, this reminds me of Nassim Nicholas Taleb.

comment by henrycooksley · 2020-06-05T22:36:02.266Z · score: 1 (1 votes) · EA(p) · GW(p)

I really enjoyed this, thank you!