machine learning

by

‘Two hands are a lot’ — we’re hiring data scientists, project managers, policy experts, assorted weirdos…

‘This is possibly the single largest design flaw contributing to the bad Nash equilibrium in which … many governments are stuck. Every individual high-functioning competent person knows they can’t make much difference by being one more face in that crowd.’ Eliezer Yudkowsky, AI expert, LessWrong etc.

‘[M]uch of our intellectual elite who think they have “the solutions” have actually cut themselves off from understanding the basis for much of the most important human progress.’ Michael Nielsen, physicist and one of the handful of most interesting people I’ve ever talked to.

‘People, ideas, machines — in that order.’ Colonel Boyd.

‘There isn’t one novel thought in all of how Berkshire [Hathaway] is run. It’s all about … exploiting unrecognized simplicities.’ Charlie Munger,Warren Buffett’s partner.

‘Two hands, it isn’t much considering how the world is infinite. Yet, all the same, two hands, they are a lot.’ Alexander Grothendieck, one of the great mathematicians.

*

There are many brilliant people in the civil service and politics. Over the past five months the No10 political team has been lucky to work with some fantastic officials. But there are also some profound problems at the core of how the British state makes decisions. This was seen by pundit-world as a very eccentric view in 2014. It is no longer seen as eccentric. Dealing with these deep problems is supported by many great officials, particularly younger ones, though of course there will naturally be many fears — some reasonable, most unreasonable.

Now there is a confluence of: a) Brexit requires many large changes in policy and in the structure of decision-making, b) some people in government are prepared to take risks to change things a lot, and c) a new government with a significant majority and little need to worry about short-term unpopularity while trying to make rapid progress with long-term problems.

There is a huge amount of low hanging fruit — trillion dollar bills lying on the street — in the intersection of:

  • the selection, education and training of people for high performance
  • the frontiers of the science of prediction
  • data science, AI and cognitive technologies (e.g Seeing Rooms, ‘authoring tools designed for arguing from evidence’, Tetlock/IARPA prediction tournaments that could easily be extended to consider ‘clusters’ of issues around themes like Brexit to improve policy and project management)
  • communication (e.g Cialdini)
  • decision-making institutions at the apex of government.

We want to hire an unusual set of people with different skills and backgrounds to work in Downing Street with the best officials, some as spads and perhaps some as officials. If you are already an official and you read this blog and think you fit one of these categories, get in touch.

The categories are roughly:

  • Data scientists and software developers
  • Economists
  • Policy experts
  • Project managers
  • Communication experts
  • Junior researchers one of whom will also be my personal assistant
  • Weirdos and misfits with odd skills

We want to improve performance and make me much less important — and within a year largely redundant. At the moment I have to make decisions well outside what Charlie Munger calls my ‘circle of competence’ and we do not have the sort of expertise supporting the PM and ministers that is needed. This must change fast so we can properly serve the public.

A. Unusual mathematicians, physicists, computer scientists, data scientists

You must have exceptional academic qualifications from one of the world’s best universities or have done something that demonstrates equivalent (or greater) talents and skills. You do not need a PhD — as Alan Kay said, we are also interested in graduate students as ‘world-class researchers who don’t have PhDs yet’.

You should have the following:

  • PhD or MSc in maths or physics.
  • Outstanding mathematical skills are essential.
  • Experience of using analytical languages: e.g. Python, SQL, R.
  • Familiarity with data tools and technologies such as Postgres, Scikit Learn, NEO4J.

A few examples of papers that you will be considering:

You should be able to explain to other mathematicians, physicists and computer scientists the ideas in such papers, discuss what could be useful for our projects, synthesise ideas for other data scientists, and apply them to practical problems. You won’t be expert on the maths used in all these papers but you should be confident that you could study it and understand it.

We will be using machine learning and associated tools so it is important you can program. You do not need software development levels of programming but it would be an advantage.

Those applying must watch Bret Victor’s talks and study Dynamic Land. If this excites you, then apply; if not, then don’t. I and others interviewing will discuss this with anybody who comes for an interview. If you want a sense of the sort of things you’d be working on, then read my previous blog on Seeing Rooms, cognitive technologies etc.

B. Unusual software developers

We are looking for great software developers who would love to work on these ideas, build tools and work with some great people. You should also look at some of Victor’s technical talks on programming languages and the history of computing.

You will be working with data scientists, designers and others.

C. Unusual economists

We are looking to hire some recent graduates in economics. You should a) have an outstanding record at a great university, b) understand conventional economic theories, c) be interested in arguments on the edge of the field — for example, work by physicists on ‘agent-based models’ or by the hedge fund Bridgewater on the failures/limitations of conventional macro theories/prediction, and d) have very strong maths and be interested in working with mathematicians, physicists, and computer scientists.

The ideal candidate might, for example, have a degree in maths and economics, worked at the LHC in one summer, worked with a quant fund another summer, and written software for a YC startup in a third summer!

We’ve found one of these but want at least one more.

The sort of conversation you might have is discussing these two papers in Science (2015): Computational rationality: A converging paradigm for intelligence in brains, minds, and machines, Gershman et al and Economic reasoning and artificial intelligence, Parkes & Wellman. 

You will see in these papers an intersection of:

  • von Neumann’s foundation of game theory and ‘expected utility’,
  • mainstream economic theories,
  • modern theories about auctions,
  • theoretical computer science (including problems like the complexity of probabilistic inference in Bayesian networks, which is in the NP–hard complexity class),
  • ideas on ‘computational rationality’ and meta-reasoning from AI, cognitive science and so on.

If these sort of things are interesting, then you will find this project interesting.

It’s a bonus if you can code but it isn’t necessary.

D. Great project managers.

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.

The Oxford Handbook on Megaprojects points out that it is possible to quantify lessons from the failures of projects like high speed rail projects because almost all fail so there is a large enough sample to make statistical comparisons, whereas there can be no statistical analysis of successes because they are so rare.

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 — it is obvious that improving government requires vast improvements in project management. The first project will be improving the people and skills already here.

If you want an example of the sort of people we need to find in Britain, look at this on CC Myers — the legendary builders. SPEED. We urgently need people with these sort of skills and attitude. (If you think you are such a company and you could dual carriageway the A1 north of Newcastle in record time, then get in touch!)

E. Junior researchers

In many aspects of government, as in the tech world and investing, brains and temperament smash experience and seniority out of the park.

We want to hire some VERY clever young people either straight out of university or recently out with with extreme curiosity and capacity for hard work.

One of you will be a sort of personal assistant to me for a year — this will involve a mix of very interesting work and lots of uninteresting trivia that makes my life easier which you won’t enjoy. You will not have weekday date nights, you will sacrifice many weekends — frankly it will hard having a boy/girlfriend at all. It will be exhausting but interesting and if you cut it you will be involved in things at the age of ~21 that most people never see.

I don’t want confident public school bluffers. I want people who are much brighter than me who can work in an extreme environment. If you play office politics, you will be discovered and immediately binned.

F. Communications

In SW1 communication is generally treated as almost synonymous with ‘talking to the lobby’. This is partly why so much punditry is ‘narrative from noise’.

With no election for years and huge changes in the digital world, there is a chance and a need to do things very differently.

We’re particularly interested in deep experts on TV and digital. We also are interested in people who have worked in movies or on advertising campaigns. There are some very interesting possibilities in the intersection of technology and story telling — if you’ve done something weird, this may be the place for you.

I noticed in the recent campaign that the world of digital advertising has changed very fast since I was last involved in 2016. This is partly why so many journalists wrongly looked at things like Corbyn’s Facebook stats and thought Labour was doing better than us — the ecosystem evolves rapidly while political journalists are still behind the 2016 tech, hence why so many fell for Carole’s conspiracy theories. The digital people involved in the last campaign really knew what they are doing, which is incredibly rare in this world of charlatans and clients who don’t know what they should be buying. If you are interested in being right at the very edge of this field, join.

We have some extremely able people but we also must upgrade skills across the spad network.

G. Policy experts

One of the problems with the civil service is the way in which people are shuffled such that they either do not acquire expertise or they are moved out of areas they really know to do something else. One Friday, X is in charge of special needs education, the next week X is in charge of budgets.

There are, of course, general skills. Managing a large organisation involves some general skills. Whether it is Coca Cola or Apple, some things are very similar — how to deal with people, how to build great teams and so on. Experience is often over-rated. When Warren Buffett needed someone to turn around his insurance business he did not hire someone with experience in insurance: ‘When Ajit entered Berkshire’s office on a Saturday in 1986, he did not have a day’s experience in the insurance business’ (Buffett).

Shuffling some people who are expected to be general managers is a natural thing but it is clear Whitehall does this too much while also not training general management skills properly. There are not enough people with deep expertise in specific fields.

If you want to work in the policy unit or a department and you really know your subject so that you could confidently argue about it with world-class experts, get in touch.

It’s also the case that wherever you are most of the best people are inevitably somewhere else. This means that governments must be much better at tapping distributed expertise. Of the top 20 people in the world who best understand the science of climate change and could advise us what to do with COP 2020, how many now work as a civil servant/spad or will become one in the next 5 years?

G. Super-talented weirdos

People in SW1 talk a lot about ‘diversity’ but they rarely mean ‘true cognitive diversity’. They are usually babbling about ‘gender identity diversity blah blah’. What SW1 needs is not more drivel about ‘identity’ and ‘diversity’ from Oxbridge humanities graduates but more genuine cognitive diversity.

We need some true wild cards, artists, people who never went to university and fought their way out of an appalling hell hole, weirdos from William Gibson novels like that girl hired by Bigend as a brand ‘diviner’ who feels sick at the sight of Tommy Hilfiger or that Chinese-Cuban free runner from a crime family hired by the KGB. If you want to figure out what characters around Putin might do, or how international criminal gangs might exploit holes in our border security, you don’t want more Oxbridge English graduates who chat about Lacan at dinner parties with TV producers and spread fake news about fake news.

By definition I don’t really know what I’m looking for but I want people around No10 to be on the lookout for such people.

We need to figure out how to use such people better without asking them to conform to the horrors of ‘Human Resources’ (which also obviously need a bonfire).

*

Send a max 1 page letter plus CV to ideasfornumber10@gmail.com and put in the subject line ‘job/’ and add after the / one of: data, developer, econ, comms, projects, research, policy, misfit.

I’ll have to spend time helping you so don’t apply unless you can commit to at least 2 years.

I’ll bin you within weeks if you don’t fit — don’t complain later because I made it clear now. 

I will try to answer as many as possible but last time I publicly asked for job applications in 2015 I was swamped and could not, so I can’t promise an answer. If you think I’ve insanely ignored you, persist for a while.

I will use this blog to throw out ideas. It’s important when dealing with large organisations to dart around at different levels, not be stuck with formal hierarchies. It will seem chaotic and ‘not proper No10 process’ to some. But the point of this government is to do things differently and better and this always looks messy. We do not care about trying to ‘control the narrative’ and all that New Labour junk and this government will not be run by ‘comms grid’.

As Paul Graham and Peter Thiel say, most ideas that seem bad are bad but great ideas also seem at first like bad ideas — otherwise someone would have already done them. Incentives and culture push people in normal government systems away from encouraging ‘ideas that seem bad’. Part of the point of a small, odd No10 team is to find and exploit, without worrying about media noise, what Andy Grove called ‘very high leverage ideas’ and these will almost inevitably seem bad to most.

I will post some random things over the next few weeks and see what bounces back — it is all upside, there’s no downside if you don’t mind a bit of noise and it’s a fast cheap way to find good ideas…

by

State-of-the-art in AI #1: causality, hypotheticals, and robots with free will & capacity for evil (UPDATED)

Judea Pearl is one of the most important scholars in the field of causal reasoning. His book Causality is the leading textbook in the field.

This blog has two short parts — a paper he wrote a few months ago and an interview he gave a few days ago.

*

He recently wrote a very interesting (to the very limited extent I understand it) short paper about the limits of state-of-the-art AI systems using ‘deep learning’ neural networks — such as the AlphaGo system which recently conquered the game of GO and AlphaZero which blew past centuries of human knowledge of chess in 24 hours — and how these systems could be improved.

The human ability to interrogate stored representations of their environment with counter-factual questions is fundamental and, for now, absent in machines. (All bold added my me.)

‘If we examine the information that drives machine learning today, we find that it is almost entirely statistical. In other words, learning machines improve their performance by optimizing parameters over a stream of sensory inputs received from the environment. It is a slow process, analogous in many respects to the evolutionary survival-of-the-fittest process that explains how species like eagles and snakes have developed superb vision systems over millions of years. It cannot explain however the super-evolutionary process that enabled humans to build eyeglasses and telescopes over barely one thousand years. What humans possessed that other species lacked was a mental representation, a blue-print of their environment which they could manipulate at will to imagine alternative hypothetical environments for planning and learning…

‘[T]he decisive ingredient that gave our homo sapiens ancestors the ability to achieve global dominion, about 40,000 years ago, was their ability to sketch and store a representation of their environment, interrogate that representation, distort it by mental acts of imagination and finally answer “What if?” kind of questions. Examples are interventional questions: “What if I act?” and retrospective or explanatory questions: “What if I had acted differently?” No learning machine in operation today can answer such questions about actions not taken before. Moreover, most learning machines today do not utilize a representation from which such questions can be answered.

‘We postulate that the major impediment to achieving accelerated learning speeds as well as human level performance can be overcome by removing these barriers and equipping learning machines with causal reasoning tools. This postulate would have been speculative twenty years ago, prior to the mathematization of counterfactuals. Not so today. Advances in graphical and structural models have made counterfactuals computationally manageable and thus rendered meta-statistical learning worthy of serious exploration…

Figure: the ladder of causation

Screenshot 2018-03-12 11.22.54

‘An extremely useful insight unveiled by the logic of causal reasoning is the existence of a sharp classification of causal information, in terms of the kind of questions that each class is capable of answering. The classification forms a 3-level hierarchy in the sense that questions at level i (i = 1, 2, 3) can only be answered if information from level j (j ≥ i) is available. [See figure]… Counterfactuals are placed at the top of the hierarchy because they subsume interventional and associational questions. If we have a model that can answer counterfactual queries, we can also answer questions about interventions and observations… The translation does not work in the opposite direction… No counterfactual question involving retrospection can be answered from purely interventional information, such as that acquired from controlled experiments; we cannot re-run an experiment on subjects who were treated with a drug and see how they behave had then not given the drug. The hierarchy is therefore directional, with the top level being the most powerful one. Counterfactuals are the building blocks of scientific thinking as well as legal and moral reasoning…

‘This hierarchy, and the formal restrictions it entails, explains why statistics-based machine learning systems are prevented from reasoning about actions, experiments and explanations. It also suggests what external information need to be provided to, or assumed by, a learning system, and in what format, in order to circumvent those restrictions…

[He describes his approach to giving machines the ability to reason in more advanced ways (‘intent-specific optimization’) than standard approaches and the success of some experiments on real problems.]

‘[T]he value of intent-base optimization … contains … the key by which counterfactual information can be extracted out of experiments. The key is to have agents who pause, deliberate, and then act, possibly contrary to their original intent. The ability to record the discrepancy between outcomes resulting from enacting one’s intent and those resulting from acting after a deliberative pause, provides the information that renders counterfactuals estimable. It is this information that enables us to cross the barrier between layer 2 and layer 3 of the causal hierarchy… Every child undergoes experiences where he/she pauses and thinks: Can I do better? If mental records are kept of those experiences, we have experimental semantic to counterfactual thinking in the form of regret sentences “I could have done better.” The practical implications of this new semantics is worth exploring.’

The paper is here: http://web.cs.ucla.edu/~kaoru/theoretical-impediments.pdf.

*

By chance this evening I came across this interview with Pearl in which he discuses some of the ideas above less formally, HERE.

‘The problems that emerged in the early 1980s were of a predictive or diagnostic nature. A doctor looks at a bunch of symptoms from a patient and wants to come up with the probability that the patient has malaria or some other disease. We wanted automatic systems, expert systems, to be able to replace the professional — whether a doctor, or an explorer for minerals, or some other kind of paid expert. So at that point I came up with the idea of doing it probabilistically.

‘Unfortunately, standard probability calculations required exponential space and exponential time. I came up with a scheme called Bayesian networks that required polynomial time and was also quite transparent.

‘[A]s soon as we developed tools that enabled machines to reason with uncertainty, I left the arena to pursue a more challenging task: reasoning with cause and effect.

‘All the machine-learning work that we see today is conducted in diagnostic mode — say, labeling objects as “cat” or “tiger.” They don’t care about intervention; they just want to recognize an object and to predict how it’s going to evolve in time.

‘I felt an apostate when I developed powerful tools for prediction and diagnosis knowing already that this is merely the tip of human intelligence. If we want machines to reason about interventions (“What if we ban cigarettes?”) and introspection (“What if I had finished high school?”), we must invoke causal models. Associations are not enough — and this is a mathematical fact, not opinion.

‘As much as I look into what’s being done with deep learning, I see they’re all stuck there on the level of associations. Curve fitting. That sounds like sacrilege, to say that all the impressive achievements of deep learning amount to just fitting a curve to data. From the point of view of the mathematical hierarchy, no matter how skillfully you manipulate the data and what you read into the data when you manipulate it, it’s still a curve-fitting exercise, albeit complex and nontrivial.

‘I’m very impressed, because we did not expect that so many problems could be solved by pure curve fitting. It turns out they can. But I’m asking about the future — what next? Can you have a robot scientist that would plan an experiment and find new answers to pending scientific questions? That’s the next step. We also want to conduct some communication with a machine that is meaningful, and meaningful means matching our intuition.

‘If a machine does not have a model of reality, you cannot expect the machine to behave intelligently in that reality. The first step, one that will take place in maybe 10 years, is that conceptual models of reality will be programmed by humans. The next step will be that machines will postulate such models on their own and will verify and refine them based on empirical evidence. That is what happened to science; we started with a geocentric model, with circles and epicycles, and ended up with a heliocentric model with its ellipses.

‘We’re going to have robots with free will, absolutely. We have to understand how to program them and what we gain out of it. For some reason, evolution has found this sensation of free will to be computationally desirable… Evidently, it serves some computational function.

‘I think the first evidence will be if robots start communicating with each other counterfactually, like “You should have done better.” If a team of robots playing soccer starts to communicate in this language, then we’ll know that they have a sensation of free will. “You should have passed me the ball — I was waiting for you and you didn’t!” “You should have” means you could have controlled whatever urges made you do what you did, and you didn’t.

[When will robots be evil?] When it appears that the robot follows the advice of some software components and not others, when the robot ignores the advice of other components that are maintaining norms of behavior that have been programmed into them or are expected to be there on the basis of past learning. And the robot stops following them.’

Please leave links to significant critiques of this paper or work that has developed the ideas in it.

If interested in the pre-history of the computer age and internet, this paper explores it.

by

Unrecognised simplicities of effective action #1: expertise and a quadrillion dollar business

‘The combination of physics and politics could render the surface of the earth uninhabitable.’ John von Neumann.

Introduction

This series of blogs considers:

  • the difference between fields with genuine expertise, such as fighting and physics, and fields dominated by bogus expertise, such as politics and economic forecasting;
  • the big big problem we face – the world is ‘undersized and underorganised’ because of a collision between four forces: 1) our technological civilisation is inherently fragile and vulnerable to shocks, 2) the knowledge it generates is inherently dangerous, 3) our evolved instincts predispose us to aggression and misunderstanding, and 4) there is a profound mismatch between the scale and speed of destruction our knowledge can cause and the quality of individual and institutional decision-making in ‘mission critical’ institutions – our institutions are similar to those that failed so spectacularly in summer 1914 yet they face crises moving at least ~103 times faster and involving ~106 times more destructive power able to kill ~1010 people;
  • what classic texts and case studies suggest about the unrecognised simplicities of effective action to improve the selection, education, training, and management of vital decision-makers to improve dramatically, reliably, and quantifiably the quality of individual and institutional decisions (particularly 1) the ability to make accurate predictions and b) the quality of feedback);
  • how we can change incentives to aim a much bigger fraction of the most able people at the most important problems;
  • what tools and technologies can help decision-makers cope with complexity.

[I’ve tweaked a couple of things in response to this blog by physicist Steve Hsu.]

*

Summary of the big big problem

The investor Peter Thiel (founder of PayPal and Palantir, early investor in Facebook) asks people in job interviews: what billion (109) dollar business is nobody building? The most successful investor in world history, Warren Buffett, illustrated what a quadrillion (1015) dollar business might look like in his 50th anniversary letter to Berkshire Hathaway investors.

‘There is, however, one clear, present and enduring danger to Berkshire against which Charlie and I are powerless. That threat to Berkshire is also the major threat our citizenry faces: a “successful” … cyber, biological, nuclear or chemical attack on the United States… The probability of such mass destruction in any given year is likely very small… Nevertheless, what’s a small probability in a short period approaches certainty in the longer run. (If there is only one chance in thirty of an event occurring in a given year, the likelihood of it occurring at least once in a century is 96.6%.) The added bad news is that there will forever be people and organizations and perhaps even nations that would like to inflict maximum damage on our country. Their means of doing so have increased exponentially during my lifetime. “Innovation” has its dark side.

‘There is no way for American corporations or their investors to shed this risk. If an event occurs in the U.S. that leads to mass devastation, the value of all equity investments will almost certainly be decimated.

‘No one knows what “the day after” will look like. I think, however, that Einstein’s 1949 appraisal remains apt: “I know not with what weapons World War III will be fought, but World War IV will be fought with sticks and stones.”’

Politics is profoundly nonlinear. (I have written a series of blogs about complexity and prediction HERE which are useful background for those interested.) Changing the course of European history via the referendum only involved about 10 crucial people controlling ~£107  while its effects over ten years could be on the scale of ~108 – 109 people and ~£1012: like many episodes in history the resources put into it are extremely nonlinear in relation to the potential branching histories it creates. Errors dealing with Germany in 1914 and 1939 were costly on the scale of ~100,000,000 (108) lives. If we carry on with normal human history – that is, international relations defined as out-groups competing violently – and combine this with modern technology then it is extremely likely that we will have a disaster on the scale of billions (109) or even all humans (~1010). The ultimate disaster would kill about 100 times more people than our failure with Germany. Our destructive power is already much more than 100 times greater than it was then: nuclear weapons increased destructiveness by roughly a factor of a million.

Even if we dodge this particular bullet there are many others lurking. New genetic engineering techniques such as CRISPR allow radical possibilities for re-engineering organisms including humans in ways thought of as science fiction only a decade ago. We will soon be able to remake human nature itself. CRISPR-enabled ‘gene drives’ enable us to make changes to the germ-line of organisms permanent such that changes spread through the entire wild population, including making species extinct on demand. Unlike nuclear weapons such technologies are not complex, expensive, and able to be kept secret for a long time. The world’s leading experts predict that people will be making them cheaply at home soon – perhaps they already are. These developments have been driven by exponential progress much faster than Moore’s Law reducing the cost of DNA sequencing per genome from ~$108 to ~$103 in roughly 15 years.

screenshot-2017-01-16-12-24-13

It is already practically possible to deploy a cheap, autonomous, and anonymous drone with facial-recognition software and a one gram shaped-charge to identify a relevant face and blow it up. Military logic is driving autonomy. For example, 1) the explosion in the volume of drone surveillance video (from 71 hours in 2004 to 300,000 hours in 2011 to millions of hours now) requires automated analysis, and 2) jamming and spoofing of drones strongly incentivise a push for autonomy. It is unlikely that promises to ‘keep humans in the loop’ will be kept. It is likely that state and non-state actors will deploy low-cost drone swarms using machine learning to automate the ‘find-fix-finish’ cycle now controlled by humans. (See HERE for a video just released for one such program and imagine the capability when they carry their own communication and logistics network with them.)

In the medium-term, many billions are being spent on finding the secrets of general intelligence. We know this secret is encoded somewhere in the roughly 125 million ‘bits’ of information that is the rough difference between the genome that produces the human brain and the genome that produces the chimp brain. This search space is remarkably small – the equivalent of just 25 million English words or 30 copies of the King James Bible. There is no fundamental barrier to decoding this information and it is possible that the ultimate secret could be described relatively simply (cf. this great essay by physicist Michael Nielsen). One of the world’s leading experts has told me they think a large proportion of this problem could be solved in about a decade with a few tens of billions and something like an Apollo programme level of determination.

Not only is our destructive and disruptive power still getting bigger quickly – it is also getting cheaper and faster every year. The change in speed adds another dimension to the problem. In the period between the Archduke’s murder and the outbreak of World War I a month later it is striking how general failures of individuals and institutions were compounded by the way in which events moved much faster than the ‘mission critical’ institutions could cope with such that soon everyone was behind the pace, telegrams were read in the wrong order and so on. The crisis leading to World War I was about 30 days from the assassination to the start of general war – about 700 hours. The timescale for deciding what to do between receiving a warning of nuclear missile launch and deciding to launch yourself is less than half an hour and the President’s decision time is less than this, maybe just minutes. This is a speedup factor of at least 103.

Economic crises already occur far faster than human brains can cope with. The financial system has made a transition from people shouting at each other to a a system dominated by high frequency ‘algorithmic trading’ (HFT), i.e. machine intelligence applied to robot trading with vast volumes traded on a global spatial scale and a microsecond (10-6) temporal scale far beyond the monitoring, understanding, or control of regulators and politicians. There is even competition for computer trading bases in specific locations based on calculations of Special Relativity as the speed of light becomes a factor in minimising trade delays (cf. Relativistic statistical arbitrage, Wissner-Gross). ‘The Flash Crash’ of 9 May 2010 saw the Dow lose hundreds of points in minutes. Mini ‘flash crashes’ now blow up and die out faster than humans can notice. Given our institutions cannot cope with economic decisions made at ‘human speed’, a fortiori they cannot cope with decisions made at ‘robot speed’. There is scope for worse disasters than 2008 which would further damage the moral credibility of decentralised markets and provide huge chances for extremist political entrepreneurs to exploit. (* See endnote.)

What about the individuals and institutions that are supposed to cope with all this?

Our brains have not evolved much in thousands of years and are subject to all sorts of constraints including evolved heuristics that lead to misunderstanding, delusion, and violence particularly under pressure. There is a terrible mismatch between the sort of people that routinely dominate mission critical political institutions and the sort of people we need: high-ish IQ (we need more people >145 (+3SD) while almost everybody important is between 115-130 (+1 or 2SD)), a robust toolkit for not fooling yourself including quantitative problem-solving (almost totally absent at the apex of relevant institutions), determination, management skills, relevant experience, and ethics. While our ancestor chiefs at least had some intuitive feel for important variables like agriculture and cavalry our contemporary chiefs (and those in the media responsible for scrutiny of decisions) generally do not understand their equivalents, and are often less experienced in managing complex organisations than their predecessors.

The national institutions we have to deal with such crises are pretty similar to those that failed so spectacularly in summer 1914 yet they face crises moving at least ~103 times faster and involving ~106 times more destructive power able to kill ~1010 people. The international institutions developed post-1945 (UN, EU etc) contribute little to solving the biggest problems and in many ways make them worse. These institutions fail constantly and do not  – cannot – learn much.

If we keep having crises like we have experienced over the past century then this combination of problems pushes the probability of catastrophe towards ‘overwhelmingly likely’.

*

What Is To be Done? There’s plenty of room at the top

‘In a knowledge-rich world, progress does not lie in the direction of reading information faster, writing it faster, and storing more of it. Progress lies in the direction of extracting and exploiting the patterns of the world… And that progress will depend on … our ability to devise better and more powerful thinking programs for man and machine.’ Herbert Simon, Designing Organizations for an Information-rich World, 1969.

‘Fascinating that the same problems recur time after time, in almost every program, and that the management of the program, whether it happened to be government or industry, continues to avoid reality.’ George Mueller, pioneer of ‘systems engineering’ and ‘systems management’ and the man most responsible for the success of the 1969 moon landing.

Somehow the world has to make a series of extremely traumatic and dangerous transitions over the next 20 years. The main transition needed is:

Embed reliably the unrecognised simplicities of high performance teams (HPTs), including personnel selection and training, in ‘mission critical’ institutions while simultaneously developing a focused project that radically improves the prospects for international cooperation and new forms of political organisation beyond competing nation states.

Big progress on this problem would automatically and for free bring big progress on other big problems. It could improve (even save) billions of lives and save a quadrillion dollars (~$1015). If we avoid disasters then the error-correcting institutions of markets and science will, patchily, spread peace, prosperity, and learning. We will make big improvements with public services and other aspects of ‘normal’ government. We will have a healthier political culture in which representative institutions, markets serving the public (not looters), and international cooperation are stronger.

Can a big jump in performance – ‘better and more powerful thinking programs for man and machine’ – somehow be systematised?

Feynman once gave a talk titled ‘There’s plenty of room at the bottom’ about the huge performance improvements possible if we could learn to do engineering at the atomic scale – what is now called nanotechnology. There is also ‘plenty of room at the top’ of political structures for huge improvements in performance. As I explained recently, the victory of the Leave campaign owed more to the fundamental dysfunction of the British Establishment than it did to any brilliance from Vote Leave. Despite having the support of practically every force with power and money in the world (including the main broadcasters) and controlling the timing and legal regulation of the referendum, they blew it. This was good if you support Leave but just how easily the whole system could be taken down should be frightening for everybody .

Creating high performance teams is obviously hard but in what ways is it really hard? It is not hard in the same sense that some things are hard like discovering profound new mathematical knowledge. HPTs do not require profound new knowledge. We have been able to read the basic lessons in classics for over two thousand years. We can see relevant examples all around us of individuals and teams showing huge gains in effectiveness.

The real obstacle is not financial. The financial resources needed are remarkably low and the return on small investments could be incalculably vast. We could significantly improve the decisions of the most powerful 100 people in the UK or the world for less than a million dollars (~£106) and a decade-long project on a scale of just ~£107 could have dramatic effects.

The real obstacle is not a huge task of public persuasion – quite the opposite. A government that tried in a disciplined way to do this would attract huge public support. (I’ve polled some ideas and am confident about this.) Political parties are locked in a game that in trying to win in conventional ways leads to the public despising them. Ironically if a party (established or new) forgets this game and makes the public the target of extreme intelligent focus then it would not only make the world better but would trounce their opponents.

The real obstacle is not a need for breakthrough technologies though technology could help. As Colonel Boyd used to shout, ‘People, ideas, machines – in that order!’

The real obstacle is that although we can all learn and study HPTs it is extremely hard to put this learning to practical use and sustain it against all the forces of entropy that constantly operate to degrade high performance once the original people have gone. HPTs are episodic. They seem to come out of nowhere, shock people, then vanish with the rare individuals. People write about them and many talk about learning from them but in fact almost nobody ever learns from them – apart, perhaps, from those very rare people who did not need to learn – and nobody has found a method to embed this learning reliably and systematically in institutions that can maintain it. The Prussian General Staff remained operationally brilliant but in other ways went badly wrong after the death of the elder Moltke. When George Mueller left NASA it reverted to what it had been before he arrived – management chaos. All the best companies quickly go downhill after the departure of people like Bill Gates – even when such very able people have tried very very hard to avoid exactly this problem.

Charlie Munger, half of the most successful investment team in world history, has a great phrase he uses to explain their success that gets to the heart of this problem:

‘There isn’t one novel thought in all of how Berkshire [Hathaway] is run. It’s all about … exploiting unrecognized simplicities… It’s a community of like-minded people, and that makes most decisions into no-brainers. Warren [Buffett] and I aren’t prodigies. We can’t play chess blindfolded or be concert pianists. But the results are prodigious, because we have a temperamental advantage that more than compensates for a lack of IQ points.’

The simplicities that bring high performance in general, not just in investing, are largely unrecognised because they conflict with many evolved instincts and are therefore psychologically very hard to implement. The principles of the Buffett-Munger success are clear – they have even gone to great pains to explain them and what the rest of us should do – and the results are clear yet still almost nobody really listens to them and above average intelligence people instead constantly put their money into active fund management that is proved to destroy wealth every year!

Most people think they are already implementing these lessons and usually strongly reject the idea that they are not. This means that just explaining things is very unlikely to work:

‘I’d say the history that Charlie [Munger] and I have had of persuading decent, intelligent people who we thought were doing unintelligent things to change their course of action has been poor.’ Buffett.

Even more worrying, it is extremely hard to take over organisations that are not run right and make them excellent.

‘We really don’t believe in buying into organisations to change them.’ Buffett.

If people won’t listen to the world’s most successful investor in history on his own subject, and even he finds it too hard to take over failing businesses and turn them around, how likely is it that politicians and officials incentivised to keep things as they are will listen to ideas about how to do things better? How likely is it that a team can take over broken government institutions and make them dramatically better in a way that outlasts the people who do it? Bureaucracies are extraordinarily resistant to learning. Even after the debacles of 9/11 and the Iraq War, costing many lives and trillions of dollars, and even after the 2008 Crash, the security and financial bureaucracies in America and Europe are essentially the same and operate on the same principles.

Buffett’s success is partly due to his discipline in sticking within what he and Munger call their ‘circle of competence’. Within this circle they have proved the wisdom of avoiding trying to persuade people to change their minds and avoiding trying to fix broken institutions.

This option is not available in politics. The Enlightenment and the scientific revolution give us no choice but to try to persuade people and try to fix or replace broken institutions. In general ‘it is better to undertake revolution than undergo it’. How might we go about it? What can people who do not have any significant power inside the system do? What international projects are most likely to spark the sort of big changes in attitude we urgently need?

This is the first of a series. I will keep it separate from the series on the EU referendum though it is connected in the sense that I spent a year on the referendum in the belief that winning it was a necessary though not sufficient condition for Britain to play a part in improving the quality of government dramatically and improving the probability of avoiding the disasters that will happen if politics follows a normal path. I intended to implement some of these ideas in Downing Street if the Boris-Gove team had not blown up. The more I study this issue the more confident I am that dramatic improvements are possible and the more pessimistic I am that they will happen soon enough.

Please leave comments and corrections…

* A new transatlantic cable recently opened for financial trading. Its cost? £300 million. Its advantage? It shaves 2.6 milliseconds off the latency of financial trades. Innovative groups are discussing the application of military laser technology, unmanned drones circling the earth acting as routers, and even the use of neutrino communication (because neutrinos can go straight through the earth just as zillions pass through your body every second without colliding with its atoms) – cf. this recent survey in Nature.

by

On the referendum #20: the campaign, physics and data science – Vote Leave’s ‘Voter Intention Collection System’ (VICS) now available for all

‘If you don’t get this elementary, but mildly unnatural, mathematics of elementary probability into your repertoire, then you go through a long life like a one-legged man in an ass-kicking contest. You’re giving a huge advantage to everybody else. One of the advantages of a fellow like Buffett … is that he automatically thinks in terms of decision trees and the elementary math of permutations and combinations… It’s not that hard to learn. What is hard is to get so you use it routinely almost everyday of your life. The Fermat/Pascal system is dramatically consonant with the way that the world works. And it’s fundamental truth. So you simply have to have the technique…

‘One of the things that influenced me greatly was studying physics… If I were running the world, people who are qualified to do physics would not be allowed to elect out of taking it. I think that even people who aren’t [expecting to] go near physics and engineering learn a thinking system in physics that is not learned so well anywhere else… The tradition of always looking for the answer in the most fundamental way available – that is a great tradition.’ Charlie Munger, Warren Buffet’s partner.

During the ten week official campaign the implied probability from Betfair odds of IN winning ranged between 60-83% (rarely below 66%) and the probability of OUT winning ranged between 17-40% (rarely above 33%). One of the reasons why so few in London saw the result coming was that the use by campaigns of data is hard to track even if you know what to look for and few in politics or the media know what to look for yet. Almost all of Vote Leave’s digital communication and data science was invisible even if you read every single news story or column ever produced in the campaign or any of the books so far published (written pre-Shipman’s book).

Today we have made a software product available for download – Vote Leave’s ‘Voter Intention Collection System’ (VICS) – click HERE. It was named after Victoria Woodcock, Operations Director, known as Vics, who was the most indispensable person in the campaign. If she’d gone under a bus, Remain would have won. When comparing many things in life the difference between average and best is say 30% but some people are 50 times more effective than others. She is one of them. She had ‘meetings in her head’ as people said of Steve Wozniak. If she had been Cameron’s chief of staff instead of Llewellyn and Paul Stephenson had been director of communications instead of Oliver and he’d listened to them, then other things being equal Cameron would still be on the No10 sofa with a glass of red and a James Bond flick. They were the operational/management and communications foundation of the campaign. Over and over again, those two – along with others, often very junior – saved us from the consequences of my mistakes and ignorance.

Among the many brilliant things Vics did was manage the creation of VICS. When we started the campaign I had many meetings on the subject of canvassing software. Amazingly there was essentially no web-based canvassing software system for the UK that allowed live use and live monitoring. There have been many attempts by political parties and others to build such systems. All failed, expensively and often disastrously.

Unfortunately, early on (summer 2015) Richard Murphy was hired to manage the ground campaign. He wanted to use an old rubbish system that assumed the internet did not exist. This was one of the factors behind his departure and he decided to throw in his lot with Farage et al. He then inflicted this rubbish system on Grassroots Out which is one of the reasons why it was an organisational/management disaster and let down its volunteers. After Vote Leave won the official designation, many GO activists defected, against official instructions from Farage, and plugged into VICS. Once Murphy was replaced by Stephen Parkinson (now in No10) and Nick Varley, the ground campaign took off.

We created new software. This was a gamble but the whole campaign was a huge gamble and we had to take many calculated risks. One of our central ideas was that the campaign had to do things in the field of data that have never been done before. This included a) integrating data from social media, online advertising, websites, apps, canvassing, direct mail, polls, online fundraising, activist feedback, and some new things we tried such as a new way to do polling (about which I will write another time) and b) having experts in physics and machine learning do proper data science in the way only they can – i.e. far beyond the normal skills applied in political campaigns. We were the first campaign in the UK to put almost all our money into digital communication then have it partly controlled by people whose normal work was subjects like quantum information (combined with political input from Paul Stephenson and Henry de Zoete, and digital specialists AIQ). We could only do this properly if we had proper canvassing software. We built it partly in-house and partly using an external engineer who we sat in our office for months.

Many bigshot traditional advertising characters told us we were making a huge error. They were wrong. It is one of the reasons we won. We outperformed the IN campaign on data despite them starting with vast mounts of data while we started with almost zero, they had support from political parties while we did not, they had early access to the electoral roll while we did not, and they had the Crosby/Messina data and models from the 2015 election while we had to build everything from scratch without even the money to buy standard commercial databases (we found ways to scrape equivalents off the web saving hundreds of thousands of pounds).

If you want to make big improvements in communication, my advice is – hire physicists, not communications people from normal companies and never believe what advertising companies tell you about ‘data’ unless you can independently verify it. Physics, mathematics, and computer science are domains in which there are real experts, unlike macro-economic forecasting which satisfies neither of the necessary conditions – 1) enough structure in the information to enable good predictions, 2) conditions for good fast feedback and learning. Physicists and mathematicians regularly invade other fields but other fields do not invade theirs so we can see which fields are hardest for very talented people. It is no surprise that they can successfully invade politics and devise things that rout those who wrongly think they know what they are doing. Vote Leave paid very close attention to real experts. (The theoretical physicist Steve Hsu has a great blog HERE which often has stuff on this theme, e.g. HERE.)

More important than technology is the mindset – the hard discipline of obeying Richard Feynman’s advice: ‘The most important thing is not to fool yourself and you are the easiest person to fool.’ They were a hard floor on ‘fooling yourself’ and I empowered them to challenge everybody including me. They saved me from many bad decisions even though they had zero experience in politics and they forced me to change how I made important decisions like what got what money. We either operated scientifically or knew we were not, which is itself very useful knowledge. (One of the things they did was review the entire literature to see what reliable studies have been done on ‘what works’ in politics and what numbers are reliable.) Charlie Munger is one half of the most successful investment partnership in world history. He advises people – hire physicists. It works and the real prize is not the technology but a culture of making decisions in a rational way and systematically avoiding normal ways of fooling yourself as much as possible. This is very far from normal politics.

(One of the many ways in which Whitehall and Downing Street should be revolutionised is to integrate physicist-dominated data science in decision-making. There are really vast improvements possible in Government that could save hundreds of billions and avoid many disasters. Leaving the EU also requires the destruction of the normal Whitehall/Downing Street system and the development of new methods. A dysfunctional broken system is hardly likely to achieve the most complex UK government project since beating Nazi Germany, and this realisation is spreading – a subject I will return to.)

In 2015 they said to me: ‘If the polls average 50-50 at the end you will win because of differential turnout and even if the average is slightly behind you could easily win because all the pollsters live in London and hang out with people who will vote IN and can’t imagine you winning so they might easily tweak their polls in a way they think is making them more accurate but is actually fooling themselves and everybody else.’ This is what happened. Almost all the pollsters tweaked their polls and according to Curtice all the tweaks made them less accurate. Good physicists are trained to look for such errors. (I do not mean to imply that on 23 June I was sure we would win. I was not. Nor was I as pessimistic as most on our side. I will write about this later.)

VICS allows data to be input centrally (the electoral roll, which in the UK is a nightmare to gather from all the LAs) and then managed at a local level, whether that be at street level, constituency or wider areas. Security levels can be set centrally to ensure that no-one can access the whole database. During the campaign we used VICS to upload data models which predicted where we thought Leave voters were likely to be so that we could focus our canvassing efforts, which was important given limited time and resources on the ground. The model produced star ratings so that local teams could target the streets more likely to contain Leave voters.

Data flowed in on the ground and was then analysed by the data science team and integrated with all the other data streaming in. Data models helped us target the ground campaign resources and in turn data from the ground campaign helped test and refine the models in a learning cycle – i.e. VICS was not only useful to the ground campaign but also helped improve the models used for other things. (This was the point of our £50 million prize for predicting the results of the European football championships, which gathered data from people who usually ignore politics – I’m still frustrated we couldn’t persuade someone to insure a £350 million prize which is what I wanted to do.) In the official 10 week campaign we served about one billion targeted digital adverts, mostly via Facebook and strongly weighted to the period around postal voting and the last 10 days of the campaign. We ran many different versions of ads, tested them, dropped the less effective and reinforced the most effective in a constant iterative process. We combined this feedback with polls (conventional and unconventional) and focus groups to get an overall sense of what was getting through. The models honed by VICS also were used to produce dozens of different versions of the referendum address (46 million leaflets) and we tweaked the language and look according to the most reliable experiments done in the world (e.g. hence our very plain unbranded ‘The Facts’ leaflet which the other side tested, found very effective, and tried to copy). I will blog more about this.

These canvassing events represented 80-90% of our ground effort in the last few months, hence some of the reports by political scientists derived from Events pages on the campaign websites, which did not include canvassing sessions, are completely misleading about what actually happened (this includes M Goodwin who is badly confused and confusing, and kept telling the media duff information after he was told it was duff). There was also a big disinformation campaign by Farage’s gang, including Bone and Pursglove, who told the media ‘Vote Leave has no interest in the ground campaign’. This was the opposite of the truth. By the last 10 weeks we had over 12,000 people doing things every week (we had many more volunteers than this but the 12,000 were regularly active). When Farage came to see me for the last time (as always fixated only on his role in the debates and not the actual campaign which he was sure was lost) he said that he had 7,000 activists who actually did anything. He was stunned when I said that we had over 12,000. I think Farage et al believe their own spin on this subject and were deluded not lying. (Obviously there was a lot of overlap between these two figures.) These volunteers delivered about 70 million leaflets out of a total ~125 million that were delivered one way or another.

While there were some fantastic MPs who made huge efforts on the ground – e.g. Anne Marie Trevelyan – it was interesting how many MPs, nominally very committed to Leave, did nothing useful in their areas nor had any interest in ground campaigning and data. Many were far more interested in trying to get on TV and yapping to hacks than in gathering useful data, including prominent MPs on our Board and Campaign Committee, some of whom contributed ZERO useful data in the entire campaign. Some spent much of the campaign having boozy lunches with Farage gossiping about what would happen after we lost. Because so many of them proved untrustworthy and leaked everything I kept the data science team far from prying eyes – when in the office, if asked what they did they replied ‘oh I’m just a junior web guy’. It would have been better if we could have shared more but this was impossible given some of the characters.

VICS is the first of its kind in the UK and provided new opportunities. It is, of course, far from ideal. It was developed very quickly, we had to cut many corners, and it could be improved on. But it worked. Many on the ground, victims of previous such attempts, assumed it would blow up under the pressure of GOTV. It did not. It worked smoothly right through peak demand. This was also because we solved the hardware problem by giving it to Rackspace which did a great job – they have a system that allows automatic scaling depending on the demand so you don’t have to worry about big surges overwhelming the system.

There were many things we could have done much better. Our biggest obstacle was not the IN campaign and its vast resources but the appalling infighting on our own side driven by all the normal human motivations described in Thucydides – fear, interest, the pursuit of glory and so on. Without this obstacle we would have done far more on digital/data. Having seen what is offered by London’s best communications companies, vast improvements in performance are clearly possible if you hire the right people. A basic problem for people in politics is that approximately none have the hard skills necessary to distinguish great people from charlatans. It was therefore great good fortune that I was friends with our team before the campaign started.

During the campaign many thousands of people donated to Vote Leave. They paid for VICS. Given we spent a lot of money developing it and there is nothing equivalent available on the market and Vote Leave is no more (barring a very improbable event), we thought that we would make VICS available for anybody to use and improve though strictly on the basis that nobody can claim any intellectual property rights over it. It is being made available in the spirit of the open source movement and use of it should be openly acknowledged. Thanks again to the thousands of people who made millions of sacrifices – because of you we won everywhere except London, Scotland and Northern Ireland against the whole Government machine supported by almost every organisation with power and money.

I will write more about the campaign once the first wave of books is published.

PS. Do not believe the rubbish peddled by Farage and the leave.EU team about social media. E.g. a) They boasted publicly that they paid hundreds of thousands of pounds for over half a million Facebook ‘Likes’ without realising that b) Facebook’s algorithms no longer optimised news feeds for Likes (it is optimised for paid advertising). Leave.EU wasted hundreds of thousands just as many big companies spent millions building armies of Likes that were rendered largely irrelevant by Facebook’s algorithmic changes. This is just one of their blunders. Vote Leave put our money into targeted paid adverts, not buying Likes to spin stories to gullible hacks, MPs, and donors. Media organisations should have someone on the political staff who is a specialist in data or have a route to talk to their organisation’s own data science teams to help spot snake oil merchants.

PPS. If you are young, smart, and interested in politics, think very hard before studying politics / ‘political science’ / PPE at university. You will be far better off if you study maths or physics. It will be easy to move into politics later if you want to and you will have more general skills with much wider application and greater market value. PPE does not give such useful skills – indeed, it actually causes huge problems as it encourages people like Cameron and Ed Balls to ‘fool themselves’ and spread bad ideas with lots of confidence and bluffing. You can always read history books later but you won’t always be able to learn maths. If you have these general skills, then you will be much more effective than the PPE-ers you will compete against. In a few years, this will be more obvious as data science will be much more visible. A new interdisciplinary degree is urgently needed to replace PPE for those who want to go into politics. It should include the basics of modelling and involve practical exposure to people who are brilliant at managing large complex organisations.

PPPS. One of the projects that the Gove team did in the DfE was funding the development of a ‘Maths for Presidents’ course, in the same spirit as the great Berkeley course ‘Physics for Presidents’, based on ideas of Fields Medallist Tim Gowers. The statistics of polling would be a good subject for this course. This course could have a big cultural effect over 20 years if it is supported wisely.