Month: November 2017

Model, Philosophy

When Remoter Effects Matter

Posted on by Zach

In utilitarianism, “remoter effects” are the result of our actions influencing other people (and are hotly debated). I think that remoter effects are often overstated, especially (as Sir Williams said in Utilitarianism for and against) when they give the conventionally ethical answer. For example, a utilitarian might claim that the correct answer to the hostage dilemma [1] is to kill no one, because killing weakens the sanctity of human life and may lead to more deaths in the future.

When debating remoter effects, I think it’s worthwhile to split them into two categories: positive and negative. Positive remoter effects are when your actions cause others to refrain from some negative action they might otherwise take. Negative remoter effects are when your actions make it more likely that others will engage in a negative action [2].

Of late, I’ve been especially interested in ways that positive and negative remoter effects matter in political disagreements. To what extent will acting in an “honourable” [3] or pro-social way convince one’s opponents to do the same? Conversely, does fighting dirty bring out the same tendency in your opponents?

Some of my favourite bloggers are doubtful of the first proposition:

In “Deontologist Envy”, Ozy writes that we shouldn’t necessarily be nice to our enemies in the hopes that they’ll be nice to us:

In general people rarely have their behavior influenced by their political enemies. Trans people take pains to use the correct pronouns; people who are overly concerned about trans women in bathrooms still misgender them. Anti-racists avoid the use of slurs; a distressing number of people who believe in human biodiversity appear to be incapable of constructing a sentence without one. Social justice people are conscientious about trigger warnings; we are subjected to many tedious articles about how mentally ill people should be in therapy instead of burdening the rest of the world with our existence.

In “The Blues of Self-Regulation”, David Schraub talks about how this specifically applies to Republicans and Democrats:

The problem being that, even when Democrats didn’t change a rule protecting the minority party, Republicans haven’t even blinked before casting them aside the minute they interfered with their partisan agenda.

Both of these points are basically correct. Everything that Ozy says about asshats on the internet is true and David wrote his post in response to Republicans removing the filibuster for Supreme Court nominees.

But I still think that positive remoter effects are important in this context. When they happen (and I will concede that this is rare), it is because you are consistently working against the same political opponents and at least some of those opponents are honourable people. My favourite example here (although it is from war, not politics) is the Christmas Day Truce. This truce was so successful and widespread that high command undertook to move men more often to prevent a recurrence.

In politics, I view positive remoter effects as key to Senator John McCain repeatedly torpedoing the GOP healthcare plans. While Senators Murkowski and Collins framed their disagreements with the law around their constituents, McCain specifically mentioned the secretive, hurried and partisan approach to drafting the legislation. This stood in sharp contrast to Obamacare, which had numerous community consultations, went through committee and took special (and perhaps ridiculous) care to get sixty senators on board.

Imagine that Obamacare had been passed after secret drafting and no consultations. Imagine if Democrats had dismantled even more rules in the senate. They may have gotten a few more of their priorities passed or had a stronger version of Obamacare, but right now, they’d be seeing all that rolled back. Instead of evidence of positive remoter effects, we’d be seeing a clear case of negative ones.

When dealing with political enemies, positive remoter effects require a real sacrifice. It’s not enough not to do things that you don’t want to do anyway (like all the examples Ozy listed) and certainly not enough to refrain from doing things to third parties. For positive remoter effects to matter at all – for your opponents (even the honourable ones) not to say “well, they did it first and I don’t want to lose” – you need to give up some tools that you could use to advance your interests. Tedious journalists don’t care about you scrupulously using trigger warnings, but may appreciate not receiving death threats on Twitter.

Had right-wingers refrained from doxxing feminist activists (or even applied any social consequences at all against those who did so), all principled people on the left would be refusing to engage in doxxing against them. As it stands, that isn’t the case and those few leftists who ask their fellow travelers to refrain are met with the entirely truthful response: “but they started it!”

This highlights what might be an additional requirement for positive remoter effects in the political sphere: you need a clearly delimited coalition from which you can eject misbehaving members. Political parties are set up admirably for this. They regularly kick out members who fail to act as decorously as their office demands. Social movements have a much harder time, with predictable consequences – it’s far too easy for the most reprehensible members of any group to quickly become the representatives, at least as far as tactics are concerned.

Still, with positive remoter effects, you are not aiming at a movement or party broadly. Instead you are seeking to find those honourable few in it and inspire them on a different path. When it works (as it did with McCain), it can work wonders. But it isn’t something to lay all your hopes on. Some days, your enemies wake up and don’t screw you over. Other days, you have to fight.

Negative remoter effects seem so obvious as to require almost no explanation. While it’s hard (but possible) to inspire your opponents to civility with good behaviour, it’s depressingly easy to bring them down to your level with bad behavior. Acting honourably guarantees little, but acting dishonourably basically guarantees a similar response. Insofar as honour is a useful characteristic, it is useful precisely because it stops this slide towards mutual annihilation.

Notes:

[1] In the hostage dilemma, you are one of ten hostages, captured by rebels. The rebel leader offers you a gun with a single bullet. If you kill one of your fellow hostages, all of the survivors (including you) will be let free. If you refuse all of the hostages (including you) will be killed. You are guarded such that you cannot use the weapon against your captors. Your only option is to kill another hostage, or let all of the hostages be killed.

Here, I think remoter effects fail to salvage the conventional answer and the only proper utilitarian response is to kill one of the other hostages. ^

[2] Here I’m using “negative” in a roughly utilitarian sense: negative actions are those that tend to reduce the total utility of the world. When used towards good ends, negative actions consume some of the positive utility that the ends generate. When used towards ill ends, negative actions add even more disutility. This definition is robust against different preferred plans of actions (e.g. it works across liberals and conservatives, who might both agree that political violence tends to reduce utility, even if it doesn’t always reduce utility enough to rule it out in the face of certain ends), but isn’t necessarily robust across all terminal values (e.g. if you care only about reducing suffering and I care only for increasing happiness we may have different opinions on the tendency of reproduction towards good or ill).

Negative actions are roughly equivalent to “defecting”. “Roughly” because it is perhaps more accurate to say that the thing that makes defecting so pernicious is that it involves negative actions of a special class, those that generate extra disutility (possibly even beyond what simple addition would suggest) when both parties engage in them. ^

[3] I used “honourable” in several important places and should probably define it. When discussing actions, I think honourable actions are the opposite of “negative” actions as defined above: actions that tend towards the good, but can be net ill if used for bad ends. When describing “people” as honourable, I’m pointing to people who tend to reinforce norms around cooperation. This is more or less equivalent to being inherently reluctant to use negative actions to advance goals unless provoked.

My favourite example of honour is Salah ad-Din. He sent his own personal physician to tend to King Richard, who was his great enemy and used his own money to buy back a child kidnapped into slavery. Conveniently for me, Salah ad-Din shows both sides of what it means to be honourable. He personally executed Raynald III of Tripoli after Raynald ignored a truce, attacked Muslim caravans, and tortured many of the caravaners to death. To Guy of Lusignan, King of Jerusalem (who was captured in the same battle as Raynald and wrongly feared he was next to die), Salah ad-Din said: “[i]t is not the wont of kings, to kill kings; but that man had transgressed all bounds, and therefore did I treat him thus.” ^

History, Model

Warriors and Soldiers

Posted on by Zach

Epistemic Status: Full of sweeping generalizations because I don’t want to make it 10x longer by properly unpacking all the underlying complexity.

[9 minute read]

In 2006, Dr. Atul Gawande wrote an article in The New Yorker about maternal care entitled “How Childbirth Went Industrial“. It’s an excellent piece from an author who consistently produces excellent pieces. In it, Gawande charts the rise of the C-section, from its origin as technique so dangerous it was considered tantamount to murder (and consequently banned on living mothers), to its current place as one of the most common surgical procedures carried out in North American hospitals.

The C-section – and epidurals and induced labour – have become so common because obstetrics has become ruthlessly focused on maximizing the Apgar score of newborns. Along the way, the field ditched forceps (possibly better for the mother yet tricky to use or teach), a range of maneuvers for manually freeing trapped babies (likewise difficult), and general anesthetic (genuinely bad for infants, or at least for the Apgar scores of infants).

The C-section has taken the place of much of the specialized knowledge of obstetrics of old, not the least because it is easy to teach and easy for even relatively less skilled doctors to get right. When Gawande wrote the article, there was debate about offering women in their 39th week of pregnancy C-sections as an alternative to waiting for labour. Based on the stats, this hasn’t quite come to pass, but C-sections have become slightly more prevalent since the article was written.

I noticed two laments in the piece. First, Gawande wonders at the consequences of such an essential aspect of the human experience being increasingly (and based off of the studies that show forceps are just as good as C-sections, arguably unnecessarily) medicalized. Second, there’s a sense throughout the article that difficult and hard-won knowledge is being lost.

The question facing obstetrics was this: Is medicine a craft or an industry? If medicine is a craft, then you focus on teaching obstetricians to acquire a set of artisanal skills—the Woods corkscrew maneuver for the baby with a shoulder stuck, the Lovset maneuver for the breech baby, the feel of a forceps for a baby whose head is too big. You do research to find new techniques. You accept that things will not always work out in everyone’s hands.

But if medicine is an industry, responsible for the safest possible delivery of millions of babies each year, then the focus shifts. You seek reliability. You begin to wonder whether forty-two thousand obstetricians in the U.S. could really master all these techniques. You notice the steady reports of terrible forceps injuries to babies and mothers, despite the training that clinicians have received. After Apgar, obstetricians decided that they needed a simpler, more predictable way to intervene when a laboring mother ran into trouble. They found it in the Cesarean section.

Medicine would not be the first industry to industrialize. The quasi-mythical King Ludd that gave us the phrase “Luddite” was said to be a weaver, put out of business by the improved mechanical knitting machines. English programs turn out thousands of writers every year, all with an excellent technical command of the English language, but most with none of the emotive power of Gawande. Following the rules is good enough when you’re writing for a corporation that fears to offend, or for technical clarity. But the best writers don’t just know how to follow the rules. They know how and when to break them.

If Gawande was a student of military history, he’d have another metaphor for what is happening to medicine: warriors are being replaced by soldiers.

If you ever find yourself in possession of a spare hour and feel like being lectured breathlessly by a wide-eyed enthusiast, find your local military history buff (you can identify them by their collection of swords or antique guns) and ask them whether there’s any difference between soldiers and warriors.

You can go do this now, or I can fill in, having given this lecture many times myself.

Imagine your favourite (or least favourite) empire from history. You don’t get yourself an empire by collecting bottle caps. To create one, you need some kind of army. To staff your army, you have two options. Warriors, or soldiers.

(Of course, this choice isn’t made just by empires. Their neighbours must necessarily face the same conundrum.)

Warriors are the heroes of movies. They were almost always the product of training that starts at a young age and more often than not were members a special caste. Think medieval European Knights, Japanese Samurai, or the Hashashin fida’i. Warriors were notable for their eponymous mastery of war. A knight was expected to understand strategy and tactics, riding, shooting, fighting (both on foot and mounted), and wrestling. Warriors wanted to live up to their warrior ethos, which normally emphasized certain virtues, like courage and mercy (to other warriors, not to any common peasant drafted to fight them).

Soldiers were whichever conscripts or volunteers someone could get into a reasonable standard of military order. They knew only what they needed to complete their duties: perhaps one or two simple weapons, how to march in formation, how to cook, and how to repair some of their equipment [1]. Soldiers just wanted to make it through the next battle alive. In service to this, they were often brutally efficient in everything they did. Fighting wasn’t an art to them – it was simple butchery and the simpler and quicker the better. Classic examples of soldiers are the Roman Legionaries, Greek Hoplites, and Napoleon’s Grande Armée.

The techniques that soldiers learned were simple because they needed to be easy to teach to ignorant peasants on a mass scale in a short time. Warriors had their whole childhood for elaborate training.

(Or at least, that’s the standard line. In practice, things were never quite as clear cut as that – veteran soldiers might have been as skilled as any warrior, for example. The general point remains though; one on one, you would always have bet on a warrior over a soldier.)

But when you talk about armies, a funny thing happens. Soldiers dominated [2]. Individually, they might have been kind of crap at what they did. Taken as a whole though, they were well-coordinated. They looked out for each other. They fought as a team. They didn’t foolishly break ranks, or charge headlong into the enemy. When Germanic warriors came up against Roman soldiers, they were efficiently butchered. The Germans went into battle looking for honour and perhaps a glorious death. The Romans happily gave them the latter and so lived (mostly) to collect their pensions. Whichever empire you thought about above almost certainly employed soldiers, not warriors.

It turns out that discipline and common purpose have counted for rather a lot more in military history than simple strength of arms. Of this particular point, I can think of no better example than the rebellion that followed the Meiji restoration. The few rebel samurai, wonderfully trained and unholy terrors in single combat were easily slaughtered by the Imperial conscripts, who knew little more than which side of a musket to point at the enemy.

The very fact that the samurai didn’t embrace the firing line is a point against them. Their warrior code, which esteemed individual skill, left them no room to adopt this devastating new technology. And no one could command them to take it up, because they were mostly prima donnas where their honour was concerned.

I don’t want to be too hard on warriors. They were actually an efficient solution to the problem of national defence if a population was small and largely agrarian, lacked political cohesion or logistical ability, or was otherwise incapable of supporting a large army. Under these circumstances, polities could not afford to keep a large population under arms at all times. This gave them several choices: they could rely on temporary levies, who would be largely untrained. They could have a large professional army that paid for itself largely through raiding, or they could have a small, elite cadre of professional warriors.

All of these strategies had disadvantages. Levies tended to have very brittle morale, and calling up a large proportion of a population makes even a successfully prosecuted war economically devastating. Raiding tends to make your neighbours really hate you, leading to more conflicts. It can also be very bad for discipline and can backfire on your own population in lean times. Professional warriors will always be dwarfed in numbers by opponents using any other strategy.

Historically, it was never as simple as solely using just one strategy (e.g. European knights were augmented with and eventually supplanted by temporary levies), but there was a clear lean towards one strategy or another in most resource-limited historical polities. It took complex cultural technology and a well-differentiated economy to support a large force of full time soldiers and wherever these pre-conditions were lacking, you just had to make do with what you could get [3].

When conditions suddenly call for a struggle – whether that struggle is against a foreign adversary, to boost profits, or to cure disease, it is useful to look at how many societal resources are thrown at the fight. When resources are scarce, we should expect to see a few brilliant generalists, or many poorly trained conscripts. When resources are thick on the ground, the amount that can be spent on brilliant people is quickly saturated and the benefits of training your conscripts quickly accrue. From one direction or another, you’ll approach the concept of soldiers.

Doctors as soldiers, not as warriors is the concept Gawande is brushing up against in his essay. These new doctors will be more standardized, with less room for individual brilliance, but more affordances for working well in teams. The prima donnas will be banished (as they aren’t good team players, even when they’re brilliant). Dr. Gregory House may have been the model doctor in the Victorian Age, or maybe even in the fifties. But I doubt any hospital would want him now. It may be that this standardization is just the thing we need to overcome persistent medical errors, improve outcomes across the board, and make populations healthier. But I can sympathize with the position that it might be causing us to lose something beautiful.

In software development, where I work, a similar trend can be observed. Start-ups aggressively court ambitious generalists, for whom freedom to build things their way is more important than market rate compensation and is a better incentive than even the lottery that is stock-options. At start-ups, you’re likely to see languages that are “fun” to work with, often dynamically typed, even though these languages are often considered less inherently comprehensible than their more “enterprise-friendly” statically typed brethren.

It’s with languages like Java (or its Microsoft clone, C#) and C++ that companies like Google and Amazon build the underlying infrastructure that powers large tracts of the internet. Among the big pure software companies, Facebook is the odd one out for using PHP (and this choice required them to rewrite the code underlying the language from scratch to make it performant enough for their large load).

It’s also at larger companies where team work, design documents, and comprehensibility start to be very important (although there’s room for super-stars at all of the big “tech” companies still; it’s only in companies more removed from tech and therefore outside a lot of the competition for top talent where being a good team player and writing comprehensible code might top brilliance as a qualifier). This isn’t to say that no one hiring for top talent appreciates things like good documentation, or comprehensibility. Merely that it is easy for a culture that esteems individual brilliance to ignore these things are a mark of competence.

Here the logic goes that anyone smart enough for the job will be smart enough to untangle the code of their predecessors. As anyone who’s been involved in the untangling can tell you, there’s a big difference between “smart enough to untangle this mess” and “inclined to wade through this genius’s spaghetti code to get to the part that needs fixing”.

No doubt there exist countless other examples in fields I know nothing about.

The point of gathering all these examples and shoving them into my metaphor is this: I think there exist two important transitions that can occur when a society needs to focus a lot of energy on a problem. The transition from conscripts to soldiers isn’t very interesting, as it’s basically the outcome of a process of continuous improvement.

But the transition from warriors to soldiers is. It’s amazing that we can often get better results by replacing a few highly skilled generalists who apply a lot of hard fought decision making, with a veritable army of less well trained, but highly regimented and organized specialists. It’s a powerful testament to the usefulness of group intelligence. Of course, sometimes (e.g. Google, or the Mongols) you get both, but these are rare happy accidents.

Being able to understand where this transition is occurring helps you understand where we’re putting effort. Understanding when it’s happening within your own sphere of influence can help you weather it.

Also note that this transition doesn’t only go in one direction. As manufacturing becomes less and less prevalent in North America, we may return to the distant past, when manufacturing stuff was only undertaken by very skilled artisans making unique objects.

Footnotes:

[1] Note the past tense throughout much of this essay; when I speak about soldiers and warriors, I’m referring only to times before the 1900s. I know comparatively little about how modern armies are set up. ^

[2] Best of all were the Mongols, who combined the lifelong training of warriors with the discipline and organization of soldiers. When Mongols clashed with European knights in Hungary, their “dishonourable” tactics (feints, followed by feigned retreats and skirmishing) easily took the day. This was all possible through a system of signal flags that allowed Subutai to command the whole battle from a promontory. European leaders were expected to show their bravery by being in the thick of fighting, which gave them no overall control over their lines. ^

[3] Historically, professional armies with good logistical support could somewhat pay for themselves by expanding an empire, which brought in booty and slaves. This is distinct from raiding (which does not seek to incorporate other territories) and has its own disadvantages (rebellion, over-extension, corruption, massive unemployment among unskilled labourers, etc.). ^

Data Science, Literature, Model

Two Ideas Worth Sharing From ‘Weapons of Math Destruction’

Posted on by Zach

Recently, I talked about what I didn’t like in Dr. Cathy O’Neil’s book, Weapons of Math Destruction. This time around, I’d like to mention two parts of it I really liked. I wish Dr. O’Neil put more effort into naming the concepts she covered; I don’t have names for them from WMD, but in my head, I’ve been calling them Hidden Value Encodings and Axiomatic Judgements.

Hidden Value Encodings

Dr. O’Neil opens the book with a description of the model she uses to cook for her family. After going into a lot of detail about it, she makes this excellent observation:

Here we see that models, despite their reputation for impartiality, reflect goals and ideology. When I removed the possibility of eating Pop-Tarts at every meal, I was imposing my ideology on the meals model. It’s something we do without a second thought. Our own values and desires influence our choices, from the data we choose to collect to the questions we ask. Models are opinions embedded in mathematics.

It is far too easy to view models as entirely empirical, as math made form and therefore blind to values judgements. But that couldn’t be further from the truth. It’s value judgements all the way down.

Imagine a model that tries to determine when a credit card transaction is fraudulent. Fraudulent credit cards transactions cost the credit card company money, because they must refund the stolen amount to the customer. Incorrectly identifying credit card transactions also costs a company money, either through customer support time, or if the customer gets so fed up by constant false positives that they switch to a different credit card provider.

If you were tasked with building a model to predict which credit card transactions were fraudulent by one of the major credit card companies, you would probably build into your model a variable cost for failing to catch fraudulent transactions (equivalent to the cost the company must bear if the transaction is fraudulent) and a fixed cost for labelling innocuous transactions as fraudulent (equivalent to the average cost of a customer support call plus the average chance of a false positive pushing someone over the edge into switching cards multiplied by the cost of their lost business over the next few years).

From this encoding, we can already see that our model would want to automatically approve all transactions below the fixed cost of dealing with false positives [1], while applying increasing scrutiny to more expensive items, especially expensive items with big resale value or items more expensive than the cardholder normally buys (as both of these point strongly toward fraud).

This seems innocuous and logical. It is also encoding at least two sets of values. First, it encodes the values associated with capitalism. At the most basic level, this algorithm “believes” that profit is good and losses are bad. It is aimed to maximize profit for the bank and while we may hold this as a default assumption for most algorithms associated with companies, that does not mean it is devoid of values; instead it encodes all of the values associated with capitalism [2]. Second, the algorithm encodes some notion that customers have freedom to choose between alternatives (even more so than is encoded by default in accepting capitalism).

By applying a cost to false positives (and likely it would be a cost that rises with each previous false positive), you are tacitly acknowledging that customers could take their business elsewhere. If customers instead had no freedom to choose who they did business with, you could merely encode as your loss from false positives the fixed cost of fielding support calls. Since outsourced phone support is very cheap, your algorithm would care much less about false positives if there was no consumer choice.

As far as I can tell, there is no “value-free” place to stand. An algorithm in the service of a hospital that helps diagnose patients or focus resources on the most ill encodes the value that “it is better to be healthy than sick; better to be alive than dead”. These values might be (almost-)universal, but they still exist, they are still encoded, and they still deserve to be interrogated when we put functions of our society in the hands of software governed by them.

Axiomatic Judgements

One of the most annoying parts of being a child is the occasional requirement to accept an imposition on your time or preferences with the explanation “because I say so”. “Because I say so” isn’t an argument, it’s a request that you acknowledge adults’ overwhelming physical, earning, and social power as giving them a right to set arbitrary rules for you. Some algorithms, forced onto unwelcoming and less powerful populations (teachers, job-seekers, etc.) have adopted this MO as well. Instead of having to prove that they have beneficial effects or that their outputs are legitimate, they define things such that their outputs are always correct and brook no criticism.

Here’s Dr. O’Neil talking about a value-added teaching model in Washington State:

When Mathematica’s scoring system tags Sarah Wysocki and 205 other teachers as failures, the district fires them. But how does it ever learn if it was right? It doesn’t. The system itself has determined that they were failures, and that is how they are viewed. Two hundred and six “bad” teachers are gone. That fact alone appears to demonstrate how effective the value-added model is. It is cleansing the district of underperforming teachers. Instead of searching for the truth, the score comes to embody it.

She contrasts this with how Amazon operates: “if Amazon.​com, through a faulty correlation, started recommending lawn care books to teenage girls, the clicks would plummet, and the algorithm would be tweaked until it got it right.” On the other hand, the teacher rating algorithm doesn’t update, doesn’t look check if it is firing good teachers, and doesn’t take an accounting of its own costs. It holds it as axiomatic ­–a basic fact beyond questioning– that its results are the right results.

I am in full agreement with Dr. O’Neil’s criticism here. Not only does it push past the bounds of fairness to make important decisions, like hiring and firing, through opaque formulae that are not explained to those who are being judged and lack basic accountability, but it’s a professional black mark on all of the statisticians involved.

Whenever you train a model, you hold some data back. This is your test data and you will use it to assess how well your model did. That gets you through to “production” – to having your model out in the field. This is an exciting milestone, not only because your model is now making decisions and (hopefully) making them well, but because now you’ll have way more data. You can see how your new fraud detection algorithm does by the volume of payouts and customer support calls. You can see how your new leak detection algorithm does by customers replying to your emails and telling you if you got it right or not.

A friend of mine who worked in FinTech once told me that they approved 1.5% of everyone who applied for their financial product, no matter what. They’d keep the score their model gave to that person on record, then see how the person fared in reality. If they used the product responsibly despite a low score, or used it recklessly despite a high score, it was viewed as valuable information that helped the team make their model that much better. I can imagine a team of data scientists, heads together around a monitor, looking through features and asking each other “huh, do any of you see what we missed here?” and it’s a pleasant image [3].

Value added teaching models, or psychological pre-screens for hiring do nothing of the sort (even though it would be trivial for them to!). They give results and those results are defined as the ground truth. There’s no room for messy reality to work its way back into the cycle. There’s no room for the creators to learn. The algorithm will be flawed and imperfect, like all products of human hands. That is inevitable. But it will be far less perfect than it could be. Absent feedback, it is doomed to always be flawed, in ways both subtle and gross, and in ways unknown to its creators and victims.

Like most Canadian engineering students, I made a solemn vow:

…in the presence of these my betters and my equals in my calling, [I] bind myself upon my honour and cold iron, that, to the best of my knowledge and power, I will not henceforward suffer or pass, or be privy to the passing of, bad workmanship or faulty material in aught that concerns my works before mankind as an engineer…

Sloppy work, like that value-added teacher model is the very definition of bad workmanship. Would that I never suffer something like that to leave my hands and take life in the world! It is no Quebec Bridge, but the value-added teaching model and other doomed to fail algorithms like it represent a slow-motion accident, steadily stealing jobs and happiness from people with no appeal or remorse.

I can accept stains on the honour of my chosen profession. Those are inevitable. But in a way, stains on our competence are so much worse. Models that take in no feedback are both, but the second really stings me.

Footnotes

[1] This first approximation isn’t correct in practice, because certain patterns of small transactions are consistent with fraud. I found this out the hard way, when a certain Bitcoin exchange’s credit card verification procedure (withdrawing less than a dollar, then refunding it a few days later, after you tell them how much they withdrew) triggered the fraud detection software at my bank. Apparently credit card thieves will often do a similar thing (minus the whole “ask the cardholder how much was withdrawn” step), as a means of checking if the card is good without cluing in the cardholder. ^

[2] I don’t mean this as a criticism of capitalism. I seek merely to point out (that like all other economic systems) capitalism is neither value neutral, nor inevitable. “Capitalism” encodes values like “people are largely rational”, “people often act to maximize their gains” and “choice is fundamentally good and useful”. ^

If socialist banks had ever made it to the point of deploying algorithms (instead of collapsing under the weight of their flawed economic system), those algorithms would also encode values (like “people will work hard for the good of the whole” and “people are inherently altruistic” and “it is worth it to sacrifice efficiency in the name of fairness”).

[3] Dulce et decorum est… get the fucking data science right. ^

Data Science, Literature, Model

Two Fallacies From ‘Weapons of Math Destruction’

Posted on by Zach

Much thanks to Cody Wild for providing editing and feedback. That said, I would like to remind my readers that I deserve full credit for all errors and that all opinions expressed here are only guaranteed to be mine.

[12 minute read]

I recently read Weapons of Math Destruction by Dr. Cathy O’Neil and found it an enormously frustrating book. It’s not that whole book was rubbish ­– that would have made things easy. No, the real problem with this book is that the crap and the pearls were so closely mixed that I had to stare at every sentence very, very carefully in hopes of figuring out which one each was. There’s some good stuff in here. But much of Dr. O’Neil’s argumentation relies on two new (to me) fallacies. It’s these fallacies (which I’ve dubbed the Ought-Is Fallacy and the Availability Bait-and-Switch) that I want to explore today.

Ought-Is Fallacy

It’s a commonly repeated truism that “correlation doesn’t imply causation”. People who’ve been around the statistics block a bit longer might echo Randall Monroe and retort that “correlation doesn’t imply causation, but it does waggle its eyebrows suggestively and gesture furtively while mouthing ‘look over there'”. Understanding why a graph like this:

In addition to this graph obviously being anchored, using it is obviously fair use.
Image Copyright The New York Times, 2017. Used here for purposes of commentary and criticism.

Is utter horsecrap [1], despite how suggestive it looks is the work of a decent education in statistics. Here correlation doesn’t imply causation. On the other hand, it’s not hard to find excellent examples where correlation really does mean causation:

This would be a risky graph to use if echo chambers didn't mean that I know literally no one who doesn't believe in global warming
Source: The National Centers for Environmental Administration. Having to spell “centre” wrong and use inferior units is a small price to pay for the fact that the American government immediately releases everything it creates into the public domain.

When trying to understand the ground truth, it’s important that you don’t confuse correlation with causation. But not every human endeavour is aimed at determining the ground truth. Some endeavours really do just need to understand which activities and results are correlated. Principal among these is insurance.

Let’s say I wanted to sell you “punched in the face” insurance. You’d pay a small premium every month and if you were ever punched in the face hard enough to require dental work, I’d pay you enough to cover it [2]. I’d probably charge you more if you were male, because men are much, much more likely to be seriously injured in an assault than women are.

I’m just interested in pricing my product. It doesn’t actually matter if being a man is causal of more assaults or just correlated with it. It doesn’t matter if men aren’t inherently more likely to assault and be assaulted compared to women (for a biological definition of “inherently”). It doesn’t matter what assault rates would be like in a society without toxic masculinity. One thing and one thing alone matters: on average, I will have to pay out more often for men. Therefore, I charge men more.

If you were to claim that because there may be nothing inherent in maleness that causes assault and being assaulted, therefore men shouldn’t have to pay more, you are making a moral argument, not an empirical one. You are also committing the ought-is fallacy. Just because your beliefs tell you that some aspect of the world should be a certain way, or that it would be more moral for the world to be a certain way, does not mean the world actually is that way or that everyone must agree to order the world as if that were true.

This doesn’t prevent you from making a moral argument that we should ignore certain correlates in certain cases in the interest of fairness, merely that you should not be making an empirical argument about what is ultimately values.

The ought-is fallacy came up literally whenever Weapons of Math Destruction talked about insurance, as well as when it talked about sentencing disparities. Here’s one example:

But as the questions continue, delving deeper into the person’s life, it’s easy to imagine how inmates from a privileged background would answer one way and those from tough inner-city streets another. Ask a criminal who grew up in comfortable suburbs about “the first time you were ever involved with the police,” and he might not have a single incident to report other than the one that brought him to prison. Young black males, by contrast, are likely to have been stopped by police dozens of times, even when they’ve done nothing wrong. A 2013 study by the New York Civil Liberties Union found that while black and Latino males between the ages of fourteen and twenty-four made up only 4.7 percent of the city’s population, they accounted for 40.6 percent of the stop-and-frisk checks by police. More than 90 percent of those stopped were innocent. Some of the others might have been drinking underage or carrying a joint. And unlike most rich kids, they got in trouble for it. So if early “involvement” with the police signals recidivism, poor people and racial minorities look far riskier.

Now I happen to agree with Dr. O’Neil that we should not allow race to end up playing a role in prison sentence length. There are plenty of good things to include in a sentence length: seriousness of crime, remorse, etc. I don’t think race should be one of these criteria and since the sequence of events that Dr. O’Neil mentions make this far from the default in the criminal justice system, I think doing more to ensure race stays out of sentencing is an important moral responsibility we have as a society.

But Dr. O’Neil’s empirical criticism of recidivism models is entirely off base. In this specific example, she is claiming that some characteristics that correlate with recidivism should not be used in recidivism models even though they improve the accuracy, because they are not per se causative of crime.

Because of systematic racism and discrimination in policing [3], the recidivism rate among black Americans is higher. If the only thing you care about is maximizing the prison sentence of people who are most likely to re-offend, then your model will tag black people for longer sentences. It does not matter what the “cause” of this is! Your accuracy will still be higher if you take race into account.

To say “black Americans seem to have a higher rate of recidivism, therefore we should punish them more heavily” is almost to commit the opposite fallacy, the is-ought. Instead, we should say “yes, empirically there’s a high rate of recidivism among black Americans, but this is probably caused by social factors and regardless, if we don’t want to create a population of permanently incarcerated people, with all of the vicious cycle of discrimination that this creates, we should aim for racial parity in sentencing”. This is a very strong (and I think persuasive) moral claim [4].

It certainly is more work to make a complicated moral claim that mentions the trade-offs we must make between punishment and fairness (or between what is morally right and what is expedient) than it is to make a claim that makes no reference to these subtleties. When we admit that we are sacrificing accuracy in the name of fairness, we do open up an avenue for people to attack us.

Despite this disadvantage, I think keeping our moral and empirical claims separate is very important. When you make the empirical claim that “being black isn’t causative of higher rates of recidivism, therefore the models are wrong when they rank black Americans as more likely to reoffend”, instead of the corresponding ethical claim, then you are making two mistakes. First, there’s lots of room to quibble about what “causative” even means, beyond simple genetic causation. Because you took an empirical and not ethical position, you may have to fight any future evidence to the contrary of your empirical position, even if the evidence is true; in essence, you risk becoming an enemy of the truth. If the truth becomes particularly obvious (and contrary to your claims) you risk looking risible and any gains you achieved will be at risk of reversal.

Second, I would argue that it is ridiculous to claim that universal human rights must rest on claims of genetic identicalness between all groups of people (and trying to make the empirical claim above, rather than a moral claim implicitly embraces this premise). Ashkenazi Jews are (on average) about 15 IQ points ahead of other groups. Should we give them any different moral worth because of this? I would argue no [5]. The only criteria for full moral worth as a human and all universal rights that all humans are entitled to is being human.

As genetic engineering becomes possible, it will be especially problematic to have a norm that moral worth of humans can be modified by their genetic predisposition to pro-social behaviour. Everyone, but most especially the left, which views diversity and flourishing as some of its most important projects should push back against both the is-ought and ought-is fallacies and fight for an expansive definition of universal human rights.

Availability Bait-and-Switch

Imagine someone told you the following story:

The Fair Housing Act has been an absolute disaster for my family! My brother was trying to sublet his apartment to a friend for the summer. Unfortunately, one of the fair housing inspectors caught wind of this and forced him to put up notices that it was for rent. He had to spend a week showing random people around it and some snot-nosed five-year-old broke one of his vases while he was showing that kid’s mother around. I know there were problems before, but is the Fair Housing Act really worth it if it can cause this?

Most people would say the answer to the above is “yes, it really was worth it, oh my God, what is wrong with you?”

But it’s actually hard to think that. Because you just read a long, vivid, easily imaginable example of what exactly was wrong with the current regime and a quick throw away reference to there being problems with the old way things were done. Some people might say that it’s better to at least mention that the other way of doing things had its problems too. I disagree strenuously.

When you make a throw-away reference to problems with another way of doing things, while focusing all of your descriptive effort on the problems of the current way (or vice-versa), you are committing the Availability Bait-and-Switch. And you are giving a very false illusion of balance; people will remember that you mentioned both had problems, but they will not take this away as their impression. You will have tricked your readers into thinking you gave a balanced treatment (or at least paved the way for a defence against claims that you didn’t give a balanced treatment) while doing nothing of the sort!

We are all running corrupted hardware. One of the most notable cognitive biases we have is the availability heuristic. We judge probabilities based on what we can easily recall, not on any empirical basis. If you were asked “are there more words in the average English language book that start with k, or have k as the third letter?”, you’d probably say “start with k!” [6]. In fact, words with “k” as the third letter show up more often. But these words are harder to recall and therefore much less available to your brain.

If I were to give you a bunch of very vivid examples of how algorithms can ruin your life (as Dr. O’Neil repeatedly does, most egregiously in chapters 1, 5, and 8) and then mention off-hand that human decision making also used to ruin a lot of people’s lives, you’d probably come out of our talk much more concerned with algorithms than with human decision making. This was a thing I had to deliberately fight against while reading Weapons of Math Destruction.

Because for a book about how algorithms are destroying everything, there was a remarkable paucity of data on this destruction. I cannot recall seeing any comparative analysis (backed up by statistics, not anecdotes) of the costs and benefits of human decision making and algorithmic decision making, as it applied to Dr. O’Neil’s areas of focus. The book was all the costs of one and a vague allusion to the potential costs of the other.

If you want to give your readers an accurate snapshot of the ground truth, your examples must be representative of the ground truth. If algorithms cause twice as much damage as human decision making in certain circumstances (and again, I’ve seen zero proof that this is the case) then you should interleave every two examples of algorithmic destruction with one of human pettiness. As long as you aren’t doing this, you are lying to your readers. If you’re committed to lying, perhaps for reasons of pithiness or flow, then drop the vague allusions to the costs of the other way of doing things. Make it clear you’re writing a hatchet job, instead of trying to claim epistemic virtue points for “telling both sides of the story”. At least doing things that way is honest [7].

Footnotes

[1] This is a classic example of “anchoring”, a phenomenon where you appear to have a strong correlation in a certain direction because of a single extreme point. When you have anchoring, it’s unclear how generalizable your conclusion is – as the whole direction of the fit could be the result of the single extreme point.

Here’s a toy example:

Note that the thing that makes me suspicious of anchoring here is that we have a big hole with no data and no way of knowing what sort of data goes there (it’s not likely we can randomly generate a bunch of new countries and plot their gun ownership and rate of mass shootings). If we did some more readings (ignoring the fact that in this case we can’t) and got something like this:

I would no longer be worried about anchoring. It really isn’t enough just to look at the correlation coefficient either. The image labelled “Also Not Anchored” has a marginally lower correlation coefficient than the anchored image, even though (I would argue) it is FAR more likely to represent a true positive correlation. Note also we have no way to tell that more data will necessarily give us a graph like the third. We could also get something like this:

In which we have a fairly clear trend of noisy data with an average of 2.5 irrespective of our x-value and a pair of outliers driving a slight positive correlation.

Also, the NYT graph isn’t normalized to population, which is kind of a WTF level mistake. They include another graph that is normalized later on, but the graph I show is the preview image on Facebook. I was very annoyed with the smug liberals in the comments of the NYT article, crowing about how conservatives are too stupid to understand statistics. But that’s a rant for another day…  ^

[2] I’d very quickly go out of business because of the moral hazard and adverse selection built into this product, but that isn’t germane to the example. ^

[3] Or at least, this is my guess as to the most plausible factors in the recidivism rate discrepancy. I think social factors ­– especially when social gaps are so clear and pervasive – seem much more likely than biological ones. The simplest example of the disparity in policing – and its effects – is the relative rates of being stopped by police during Stop and Frisk given above by Dr. O’Neil. ^

[4] It’s possible that variations in Monoamine oxidase A or some other gene amongst populations might make some populations more predisposed (in a biological sense) to violence or other antisocial behaviour. Given that violence and antisocial behaviour are relatively uncommon (e.g. about six in every one thousand Canadian adults are incarcerated or under community supervision on any given day), any genetic effect that increases them would both be small on a social level and lead to a relatively large skew in terms of supervised populations.

This would occur in the same way that repeat offenders tend to be about one standard deviation below median societal IQ but the correlation between IQ and crime explains very little of the variation in crime. This effect exists because crime is so rare.

It is unfortunately easy for people to take things like “Group X is 5% more likely to be violent”, and believe that people in Group X are something like 5% likely to assault them. This obviously isn’t true. Given that there are about 7.5 assaults for every 1000 Canadians each year, a population that was instead 100% Group X (with their presumed 5% higher assault rate) would see about 7.875 assaults per 1000 people, a difference of about one additional assault per 3500 people.

Unfortunately, if society took its normal course, we could expect to see Group X very overrepresented in prison. As soon as Group X gets a reputation for violence, juries would be more likely to convict, bail would be less likely, sentences might be longer (out of fear of recidivism), etc. Because many jobs (and in America, social benefits and rights) are withdrawn after you’ve been sentenced to jail, formerly incarcerated members of Group X would see fewer legal avenues to make a living. This could become even worse if even non-criminal members of Group X would denied some jobs due to fear of future criminality, leaving Group X members with few overall options but the black and grey economies and further tightening the spiral of incarceration and discrimination.

In this case, I think the moral thing to do as a society is to ignore any evidence we have about between-group differences in genetic propensities to violence. Ignoring results isn’t the same thing as pretending they are false or banning research; we aren’t fighting against truth, simply saying that some small extra predictive power into violence is not worth the social cost that Group X would face in a society that is entirely unable to productively reason about statistics.  ^

[5] Although we should be ever vigilant against people who seek to do the opposite and use genetic differences between Ashkenazi Jews and other populations as a basis for their Nazi ideology. As Hannah Arendt said, the Holocaust was a crime against humanity perpetrated on the body of the Jewish people. It was a crime against humanity (rather than “merely” a crime against Jews) because Jews are human. ^

[6] Or at least, you would if I hadn’t warned you that I was about to talk about biases. ^

[7] My next blog post is going to be devoted to what I did like about the book, because I don’t want to commit the mistakes I’ve just railed against (and because I think there was some good stuff in the book that bears reviewing). ^

Model, Politics

The Graph Model of Conflict Resolution – Sensitivity Analysis

Posted on by Zach

[10 minute read]

Last week, I used the Graph Model of Conflict Resolution to find a set of stable equilibria in the present conflict between North Korea and the USA. They were:

  • The tense status quo (s. 0)
  • An American troop withdrawal, paired with North Korea giving up its nuclear weapons (s.10)
  • All out conventional warfare on the Korean Peninsula (s. 4)
  • All out nuclear warfare on the Korean Peninsula (s. 5)

But how much can we trust these results? How much to they depend on my subjective ranking of the belligerent’s preferences? How much do they depend on the stability metrics I used?

To get a sense of this, I’m going to add another stability metric into the mix, come up with three new preference vectors, and look at how the original results change when we consider a North Korean invasion to be irreversible. After these eight new stability calculations, we’ll have nine slightly different ways of looking at the conflict; this should help us guess which equilibria are robust to my subjective choices and which might exist only because of how I framed the problem.

Alternative Stability Metrics

Previously we assessed stable states using Nash Stability and Sequential Stability. Sequential Stability allowed us to see what would happen if the decision makers were looking two moves ahead and assuming that their opponents wouldn’t “cut off the nose to spite the face” – it assumes, in essence, that people will only sanction by moving to states that they like more, not states they like less.

Maybe that’s a bad assumption dealing with Trump and Kim Jong-un. In this case, wouldn’t it be better to use Symmetric Metarationality? With Symmetric Metarationality, all sanctioning unilateral moves are on the table. Symmetric Metarationality also allows decision makers to respond to sanctioning. In effect, it lets them look three moves ahead, instead of the two allowed by Sequential Stability.

Before we see how this new metric changes things, let’s review our states, preference vectors, and stability analysis from last time.

The states are:

Click for a copyable version

Or in plain English:

State Explanation
0 Status quo
1 Nuclear strike by the US, NK keeps nuclear weapons
2 Unilateral US troop withdrawal
4 North Korean invasion with only conventional US responses
5 North Korean invasion with US nuclear strike
6 US withdrawal and North Korean Invasion
8 Unilateral North Korean abandonment of nuclear weapons
9 US strike and North Korean abandonment of nuclear weapons
10 Coordinated US withdrawal and NK abandonment of nuclear weapons
12 NK invasion after abandoning nuclear weapons; conventional US response
13 NK invasion after abandoning nuclear weapons; US nuclear strike
14 US withdrawal paired with NK nuclear weapons abandonment and invasion

From these states, we saw the following equilibria and unilateral improvements:

Click for copyable version

When dealing with Symmetric Metarationality, I find it very helpful to modify the chart above so that it also includes unilateral moves. After we make this change and blank out our results, we get the following:
Click for copyable version

From here, we use a simple algorithm. First, all states without unilateral improvements are Nash Stable. Next, we check each unilateral improvement in the remaining states against the opponent’s unilateral actions, then against the original actors best unilateral action from each of the resulting states. If there are no results lower than the original actor started, the move is unstable. Otherwise it’s stable by Symmetric Metarationality (and we’ll mark it with “S”). Like Sequential Stability, you can’t truly call this done until you check for states that are simultaneously sanctioned (this is often easy because simultaneous sanctioning is only a risk when both sides are unstable).

An example: There exist a unilateral improvement for America from s. 4 to s. 5. From s. 5, North Korea can move to s. 1, 13, or 9. America disprefers both s. 1 and s. 13 to s. 4 and has no moves out of them, so the threat of North Korea taking either of those actions is an effective sanction and makes s. 4 stable on the American side.

Once we repeat this for all states across both sides, we get the following:

Click for copyable version

We’ve kept all of our old equilibria and gained a new one in s. 12: “NK invasion after abandoning nuclear weapons; conventional US response”.

Previously, s. 12 wasn’t stable because North Korea preferred the status quo (s. 0) to it and the US had no UIs from the status quo. North Korea moving from s. 12 to s. 0 is sanctioned in Symmetric Metarationality by the US unilateral move from s. 0 to s. 1, which leaves North Korea with only the option of moving from s. 1 to s. 5. State 5 is dispreferred to s. 12 by North Korea, so it can’t risk leaving s. 12 for s. 0. State 12 was always Nash Stable for the US, so it becoming stable for North Korea makes it an equilibrium point.

To put this another way (and to put an example on what I said above), using Symmetric Metarationality allows us to model a world where the adversaries see each other as less rational and more spiteful. In this world. NK doesn’t trust the US to remain at s. 0 if it were to call for a truce after an invasion, so any invasion that starts doesn’t really end.

It was heartening to see all of our existing equilibria remain where they were. Note that I did all of the work in this post without knowing what the results would be and fully prepared to publish even if my initial equilibria never turned up again; that they showed up here made me somewhat relieved.

Irreversible Invasions

Previously we modelled invasions as reversible. But is this a realistic assumption? It’s very possible that the bad will from an invasion could last for quite a while, making other strategies very difficult to try out. It’s also likely that America wouldn’t just let North Korean troops give up and slink away without reprisal. If this is the case, maybe we should model a North Korean invasion as irreversible. This will mean that there can be no unilateral improvements for North Korea from s. 4, 5, or 6 to s. 0, 1, 2, 8, 9, or 10.

In practical terms, modelling an invasion as irreversible costs North Korea one unilateral improvement, from s. 4 to s. 0. Let’s see if this changes the results at all (we’re back to sequential stability):

Click for copyable version

We end up losing the simultaneous sanctioning that made s. 4 a stable state, leaving us with only three stable states: the status quo, a trade of American withdrawal for the North Korean nuclear program, and all out nuclear war on the Korean Peninsula.

We’ve now tried three different ways of looking at this problem. Three equilibria (s. 0, 10, 5) showed up in all cases, one in two cases (s. 4), and one in one case (s. 12). We’re starting to get a sense for which equilibria are particularly stable and which are more liable to only pop up under certain conditions. But how will our equilibria fare when faced with a different preference vectors?

Bloodthirsty Belligerents

What if we’ve underestimated how much North Korea and the United States care about getting what they want and overestimated how much they care about looking reasonable? I’m going to try ranking the states so that North Korea always prefers invading and the US always prefers first that North Korea doesn’t invade the South and second that they have no nuclear weapons program.

This gives us the following preference vectors:

US: 8, 9, 0, 10, 13, 12, 5, 4, 1, 2, 14, 6
NK: 6, 14, 4, 12, 5, 13, 2, 0, 10, 1, 9, 8

Since we’re modelling the actors as more belligerent, let’s also assume for the purposes of these analyses that invasions are irreversible.

Here are the preferences vectors we’ll use to find equilibria:

Click for copyable version

Sequential Stability

Click for copyable version

Here we have only two stable states, s. 5 and 12. Both of these involve war on the Korean Peninsula; not even the status quo is stable. State 2 is at risk of simultaneous sanctioning, but the resulting states (4, 12, 5, 13) aren’t dispreferred, to s. 2 for either actor, so no simultaneous sanctioning occurs. There really are just two equilibria.

Symmetric Metarationality

Click for copyable version

Symmetric Metarationality gives us the exact same result. Only s. 5 and s. 12 are stable. This is suspicious, as the conflict has managed to stay in s. 0 for quite some time. If these preferences were correct, North Korea would have already invaded South Korea and been met with a nuclear response.

What if these preferences are substantially correct and both sides are more aggressive than we initially suspected, but North Korea disprefers being attacked by nuclear weapons below s. 0 and s. 10? That state of affairs is perhaps more reasonable than the blatantly suicidal North Korea we just imagined. How does a modicum of self-preservation change the results?

Nuclear Deterrence

If we’re assuming that North Korea has broadly similar preferences to our last variation, but doesn’t want to get attacked by nuclear weapons, we get the following preference vectors:

US: 8, 9, 0, 10, 13, 12, 5, 4, 1, 2, 14, 6
NK: 6, 14, 4, 12, 0, 10, 5, 13, 2, 1, 9, 8

Here are the annotated preferences vectors we’ll use to assess stability with Sequential Stability and Symmetric Metarationality. Since we’re leaving the belligerency of the United States the same, we’ll continue to view invading as an irreversible action.

Click for copyable version

Sequential Stability

Click for copyable version

One “minor” change – deciding that North Korea really doesn’t want to be nuked – and we again have the status quo and a negotiated settlement (in addition to two types of war) as stable equilibria. Does this hold when we’re using Symmetric Metarationality?

Symmetric Metarationality

Click for copyable version

Again, we have s. 0, 5, 10, and 12 as our equilibria.

As we’ve seen throughout, Symmetric Metarationality tends to give very similar answers to Sequential Stability. It’s still worth doing – it helps reassure us that our results are robust, but I hope by now you’re beginning to see why I could feel comfortable making an initial analysis based just off of just Sequential Stability.

Pacifistic People

What instead of underestimating the bloodthirstiness of our belligerents, we’ve been overestimating it? It’s entirely possible that both sides strongly disprefer all options that involve violence (and the more violence an option involves, the more they disprefer it) but talk up their position in hopes of receiving concessions. In this case, let’s give our actors these preference vectors:

US: 8, 0, 10, 2, 9, 12, 4, 5, 14, 13, 6, 1
NK: 6, 14, 2, 10, 0, 8, 4, 12, 5, 1, 9, 13

(Note that I’m only extending “peacefulness” to these two actors; I’m assuming that North Korea would happily try and annex South Korea if there was no need to fight America to do so)

There are fewer unilateral improvements in this array than in many of the previous ones.

Click for copyable version

Sequential Stability

Click for copyable version

This is perhaps the most surprising result we’ve seen so far. If both powers are all talk with nothing behind it and both powers know and understand this, then they’ll stick in the current high-tension equilibria or fight a war. The only stable states here are s. 0, 4, and 5. State 10, the “negotiated settlement” state is entirely absent. We’ll revisit this scenario with hypergame analysis later, to see what happens if the bluff is believed.

Symmetric Metarationality

Click for copyable version

Here we see more equilibria than we’ve seen in any of the other examples. States 2 (unilateral US withdrawal) and 8 (North Korea unilaterally abandoning its nuclear weapons program) make their debut and s. 0, 4, 5, 10, and 12 appear again.

Remember, Symmetric Metarationality is very risk averse; it considers not just opponents’ unilateral improvements, but all of their unilateral moves as fair game. The fact that s. 0 has unilateral moves for either side that are aggressive leaves the actors too scared to move to it, even from states that they disprefer. This explains the presence of s. 2 and s. 8 in the equilibrium for the first time; they’re here because in this model both sides are so scared of war that if they blink first, they’ll be more relieved at the end of tension than they will be annoyed at moving away from their preferences.

I think in general this is a poor assumption, which is why I tend to find Sequential Stability a more useful concept than Symmetric Metarationality. That said, I don’t think this is impossible as a state of affairs, so I’m glad that I observed it. In general, this is actually one of my favourite things about the Graph Model of Conflict Resolution: using it you can very quickly answer “what ifs”, often in ways that are easily bent to understandable narratives.

Why Sensitivity Analysis?

The cool thing about sensitivity analysis is that it shows you the equilibria a conflict can fall into and how sensitivity those equilibria are to your judgement calls. There are 12 possible states in this conflict, but only 7 of them showed up in any stability analysis at all. Within those seven, only 5 showed up more than once.

Here’s a full accounting of the states that showed up (counting our first model, there were nine possible simulations for each equilibrium to show up in):

State Explanation #
0 Status quo 7
2 Unilateral US troop withdrawal 1
4 North Korean invasion with only conventional US responses 4
5 North Korean invasion with US nuclear strike 9
8 Unilateral North Korean abandonment of nuclear weapons 1
10 Coordinated US withdrawal and NK abandonment of nuclear weapons 6
12 NK invasion after abandoning nuclear weapons; conventional US response 6

Of the five that showed up more than once, four showed up more than half the time. These then are the most robust equilibria; equilibria that half of the reasonable changes we attempted couldn’t dislodge.

Note “most robust” is not necessarily equivalent to “most likely”. To get actual probabilities on outcomes, we’d have to put probabilities on the initial conditions. Even then, the Graph Model of Conflict Resolution as we’ve currently talked about it does little to explain how decision makers move between equilibria; because this scenario starts in equilibrium, it’s hard to see how it makes it to any of the other equilibria.

Hopefully I’ll be able to explain one way we can model changes in states in my next post, which will cover Hypergame Analysis – the tool we use when actors lack a perfect understanding of one another’s preferences.

Model, Politics

The Graph Model of Conflict Resolution – Introduction

Posted on by Zach

[10 minute read]

Why do things happen the way they do?

Every day, there are conflicts between decision makers. These occur on the international scale (think the Cuban Missile Crisis), the provincial level (Ontario’s sex-ed curriculum anyone?) and the local level (Toronto’s bike lane kerfuffle). Conflict is inevitable. Understanding it, regrettably, is not.

The final results of many conflicts can look baffling from the outside. Why did the Soviet Union retreat in the Cuban missile crisis? Why do some laws pass and others die on the table?

The most powerful tool I have for understanding the ebb and flow of conflict is the Graph Model of Conflict Resolution (GMCR). I had the immense pleasure of learning about it under the tutelage of Professor Keith Hipel, one of its creators. Over the next few weeks, I’d like to share it with you.

GMCR is done in two stages, modelling and analysis.

Modelling

To model a problem, there are four steps:

  • Select a point in time for the model
  • Make a list of the players and their options
  • Remove outcomes that don’t make sense
  • Create preference vectors for all players

The easiest way to understand this is to see it done.

Let’s look at the current nuclear stand-off on the Korean peninsula. I wrote this on Sunday, October 29th, 2017, so that’s the point in time we’ll use. To keep things from getting truly out of hand in our first example, let’s just focus on the US and North Korea (I’ll add in South Korea and China in a later post). What options does each side have?

US:

  • Nuclear strike on North Korea
  • Withdraw troops and normalize relations
  • Status quo

North Korea:

  • Invasion of South Korea
  • Abandon nuclear program and submit to inspections
  • Status quo

I went through a few iterations here. I originally wrote the US option “Nuclear strike” as “Pre-emptive strike”. I changed it to be more general. A nuclear strike could be pre-emptive, but it also could be in response to North Korea invading South Korea.

It’s pretty easy to make a chart of all these states:

Click for a copyable version

If you treat each action that the belligerents can make as a binary variable (yes=1 or no=0), the states will have a natural ordering based off of the binary sum of the actions taken and not taken. This specific ordering isn’t mandatory – you can use any ordering scheme you want – but I find it useful.

You may also notice that “Status quo” appears nowhere on this chart. That’s an interesting consequence of how actions are represented in the GMCR. Status quo is simply neither striking nor withdrawing for the US, or neither invading nor abandoning their nuclear program for North Korea. Adding an extra row for it would just result in us having to do more work in the next step, where we remove states that can’t exist.

I’ve colour coded some of the cells to help with this step. Removing nonsensical outcomes always requires a bit of judgement. Here we aren’t removing any outcomes that are highly dispreferred. We are supposed to restrict ourselves solely to removing outcomes that seem like they could never ever happen.

To that end, I’ve highlighted all cases where America withdraws troops and strikes North Korea. I’m interpreting “withdraw” here to mean more than just withdrawing troops – I think it would mean that the US would be withdrawing all forms of protection to South Korea. Given that, it wouldn’t make sense for the US to get involved in a nuclear war with North Korea while all the while loudly proclaiming that they don’t care what happens on the Korean peninsula. Not even Nixon’s “madman” diplomacy could encompass that.

On the other hand, I don’t think it’s necessarily impossible for North Korea to give up its nuclear weapons program and invade South Korea. There are a number of gambits where this might make sense – for example, it might believe that if they attacked South Korea after renouncing nuclear weapons, China might back them or the US would be unable to respond with nuclear missiles. Ultimately, I think these should be left in.

Here’s the revised state-space, with the twelve remaining states:

Click for a copyable version

The next step is to figure out how each decision maker prioritizes the states. I’ve found it’s helpful at this point to tag each state with a short plain language explanation.

State Explanation
0 Status quo
1 Nuclear strike by the US, NK keeps nuclear weapons
2 Unilateral US troop withdrawal
4 North Korean invasion with only conventional US responses
5 North Korean invasion with US nuclear strike
6 US withdrawal and North Korean Invasion
8 Unilateral North Korean abandonment of nuclear weapons
9 US strike and North Korean abandonment of nuclear weapons
10 Coordinated US withdrawal and NK abandonment of nuclear weapons
12 NK invasion after abandoning nuclear weapons; conventional US response
13 NK invasion after abandoning nuclear weapons; US nuclear strike
14 US withdrawal paired with NK nuclear weapons abandonment and invasion

While describing these, I’ve tried to avoid talking about causality. I didn’t describe s. 5 as “North Korean invasion in response to US nuclear strike” or “US nuclear strike in response to North Korean invasion”. Both of these are valid and would depend on which states preceded s. 5.

Looking at all of these states, here’s how I think both decision makers would order them (in order of most preferred to least preferred):

US: 8, 0, 9, 10, 12, 5, 4, 13, 14, 1, 2, 6
NK: 6, 14, 2, 10, 0, 4, 12, 5, 1, 13, 8, 9

The US prefers North Korea give up its nuclear program and wants to keep protecting South Korea. Its secondary objective is to seem like a reasonable actor on the world stage – which means that it has some preference against using pre-emptive strikes or nuclear weapons on non-nuclear states.

North Korea wants to unify the Korean peninsula under its banner, protect itself against regime change, and end the sanctions its nuclear program has brought. Based on the Agreed Framework, I do think Korea would be willing to give up nuclear weapons in exchange for a normalization of relations with the US and sanctions relief.

Once we have preference vectors, we’ve modelled the problem. Now it’s time for stability analysis.

Stability

A state is stable for a player if it isn’t advantageous for the player to shift states. A state is globally stable if it is not advantageous for any player to shift states. When a player can move to a state they prefer over the current state without any input from their opponent, this is a “unilateral improvement” (UI).

There are a variety of ways we can define “advantageous”, which lead to various definitions of stability:

Nash Stability (R): Stable if the actor has no unilateral improvements. States that are Nash stable tend to be pretty bad; these include both sides attacking in a nuclear war or both prisoners defecting in the prisoner’s dilemma. Nash stability ignores the concept of risk; it will never move to a less preferred state in the hopes of making it to a more preferred state.

General Metarationality (GMR): Stable if the actor has no unilateral improvements that aren’t sanctioned by unilateral moves by others. This tends to lead to less confusing results than Nash stability; Cooperation in the prisoner’s dilemma is stable in General Metarationality. General Metarationality accepts the existence of risk, but refuses to take any.

Symmetric Metarationality (SMR): Stable if an actor has no unilateral improvements that aren’t sanctioned by opponents’ unilateral moves after it has a chance to respond to them. This is equivalent to GMR, but with a chance to respond. Here we start to see the capacity to take on some risk.

Sequential Stability (SEQ): Stable if the actor has no unilateral improvements that aren’t sanctioned by opponents’ unilateral improvements. This basically assumes fairly reasonable opponents, the type who won’t cut off their nose to spite their face. Your mileage may vary as to how appropriate this assumption is. Like SMR, this system takes on some risk.

Limited Move Stability (LS): A state is stable if after N moves and countermoves (with both sides acting optimally), there exists no improvement. This is obviously fairly risky as any assumptions you make about your opponents’ optimal actions may turn out to be wrong (or wishful thinking).

Non-myopic Stability (NM): Equivalent to Ls with N set equal to infinity. This predicts stable states where there’s no improvements after any amount of posturing and state changes, as long as both players act entirely optimally.

The two stability metrics most important to the GMCR (at least as I was taught it) are Nash Stability (denoted with r) and Sequential Stability (denoted with s). These have the advantage of being simple enough to calculate by hand while still explaining most real-world equilibria quite well.

To do stability analysis, you write out the preference vectors of both sides, along with any unilateral improvements that they can make. You then use this to decide the stability of each state for each player. If both players are stable at a state by any of the chosen stability metrics, the state overall is stable. A state can also be stable if both players have unilateral improvements from it that result in both ending up in a dispreferred state if taken simultaneously. This is called simultaneous sanctioning and is denoted with u.

The choice of stability metrics will determine which states are stable. If you only use Nash stability, you’ll get a different result than if you combine Sequential Stability and Nash Stability.

Here’s the stability analysis for this conflict (using Nash Stability and Sequential Stability):

Click for copyable version

Before talking about the outcome, I want to mention a few things.

Look at s. 9 for the US. They prefer s. 8 to s. 9 and the two differ only on a US move. Despite this, s. 8 isn’t a unilateral improvement over s. 9 for the US. This system is called the Graph Model of Conflict Resolution for a reason. States can be viewed as nodes on a directed graph, which implies that some nodes may not have a connection. Or, to put it in simpler terms, some actions can’t be taken back. Once the US has launched a nuclear strike, it cannot un-launch it.

This holds less true for abandoning a nuclear program or withdrawing troops; both of those are fairly easy to undo (as we found out after the collapse of the Agreed Framework). Invasions on the other hand are in a tricky category. They’re somewhat reversible (you can stop and pull out), but the consequences linger. Ultimately I’ll call them reversible, but note that this is debatable and the analysis could change if you change this assumption.

In a perfect world, I’d go through this exercise four or five different times, each time with different assumptions about preferences or the reversibility of certain states or with different stability metrics and see how each factor changes the results. My next blog post will go through this in detail.

The other thing to note here is the existence of simultaneous sanctioning. Both sides have a UI from s. 4; NK to s. 0 and the US to s. 5. Unfortunately, if you take these together, you get s. 1, which both sides disprefer to s. 4. This means that once a war starts the US will be hesitant to launch a nuclear strike and North Korea would be hesitant to withdraw – in case they withdrew just as a strike happened. In reality, we get around double binds like this with negotiated truces – or unilateral ultimatums (e.g. “withdraw by 08:00 tomorrow or we will use nuclear weapons”).

There are four stable equilibria in this conflict:

  • The status quo
  • A coordinated US withdrawal of troops (but not a complete withdrawal of US interest) and North Korean renouncement of nuclear weapons
  • All out conventional war on the Korean Peninsula
  • All out nuclear war on the Korean Peninsula

I don’t think these equilibria are particularly controversial. The status quo has held for a long time, which would be impossible if it wasn’t a stable equilibrium. Meanwhile, s. 10 looks kind of similar to the Iran deal, with the US removing sanctions and doing some amount of normalization in exchange for the end of Iran’s nuclear program. State 5 is the worst-case scenario that we all know is possible.

Because we’re currently in a stable state, it seems unlikely that we’ll shift to one of the other states that could exist. In actuality, there are a few ways this could happen. A third party could intervene with its own preference vectors and shake up the equilibrium. For example, China could use the threat of economic sanctions (or the threat of ending economic sanctions) to try and get North Korea and the US to come to a détente. There also could be an error in judgement on the part of one of the parties. A false alarm could quickly turn into a very real conflict. It’s also possible that one party could mistake the others preferences, leading to them taking a course of action that they incorrectly believe isn’t sanctioned.

In future posts, I plan to show how these can all be taken into account, using the GMCR framework for Third Party Intervention and Coalitional Analysis, Strength of Preferences, and Hypergame Analysis.

Even without those additions, the GMCR is a powerful tool. I encourage you to try it out for other conflicts and see what the results are. I certainly found that the best way to really understand it was to run it a few times.

Note: I know it’s hard to play around with the charts when they’re embedded as images. You can see copyable versions of them here.

Politics

Nuclear Weapons Explained Backwards

Posted on by Zach

[10 minute read]

The following is the annotated speakers notes for a talk I gave on nuclear weapons today. I’d like to claim that it was a transcript, but after practicing from these notes for almost a week, I ended up giving the talk mostly ex tempore. Like I always do. 

Note: The uncredited photos were created by the US government and therefore have no copyright attached. All other images are either original (and therefore covered by the same license as the rest of the blog) or are credited and subject to the original license (normally CC-BY of some sort).

Hi I’m Zach.

This will be a backwards explanation of nuclear weapons; I don’t have time to cover it all so instead of covering the boring stuff like how fission works, I’m going to talk about the strategic realities surrounding the use of nuclear weapons.

Let’s actually do this thing like you’re a bunch of kids; I’m going to assume you’re always asking me “why?”. So at the highest level: this is a presentation about nuclear weapons.

Why am I doing this?

This got the laughs I was hoping for

Like maybe a lot of you, I’ve been worried about nuclear weapons of late. My worrying actually started in September 2016. I don’t know if you remember, but that was the first time it seemed like Trump might really win. And then I think a lot of us had to grapple with what that meant.

And the biggest question there was “could this mean the end of the world?”

I was worried about the end of the world because I knew Trump might end up with the nuclear launch codes and all I really knew about nuclear weapons was that they were really dangerous. At this point I was very much in the pop culture mode of “these are the things that end the world in blockbuster movies”.

Of course before I could really take this fear seriously, I had to think about why Trump might actually use nuclear weapons. Like I was pretty sure he wasn’t going to nuke Tuvalu just for fun.

Here’s what the payoff matrix looks like for nuclear war between major powers. Everyone is pretty happy doing nothing, although they’d be happier if they could wipe out their pesky rivals [1]. Unfortunately, their rivals want to avenge themselves if they’re going to die.

The decision-making algorithm that tells us we’re going to stick to doing nothing is General Metarationality. We know how our opponents will act in response to our actions and we avoid actions that will cause strong sanctioning. And I don’t know of any sanctions stronger than getting nuked.

This whole thing works because everyone understands it. The logic is so inescapable and the probable actions of your enemies are so obvious that the whole edifice survives, even though doing nothing isn’t technically even the Nash equilibrium.

But this is just theory. How do you ensure mutually assured destruction in practice?

That’s a question people have been asking since the 1950s. By now everyone’s agreed that there’s a right way to do it and a wrong way to do it. The wrong way is to stick a bunch of missiles in a desert and call it a day. The right way is to come up with three separate ways of delivering your warheads, spend billions and billions of dollars on them, and call it a day.

Here we have the three methods that everyone’s chosen – and I want to make it clear that this is arbitrary; three others would work just as well [2]. As it stands though, the conventional nuclear triad is Nuclear armed bombers, like the US B-52 or B-2, Intercontinental Ballistic Missiles (ICBMs) like the US Minuteman III, and submarine launched ballistic missiles (SLBMs), like the US Trident II. The idea with this triad is that it’s impossible for an enemy to launch a first strike so devastating that they take out your whole ability to respond.

Planners are always vaguely trying to build up their capacity for a “first strike” (remember the payoff matrix before; all nuclear powers would like it best if they could win a nuclear war). The first strike idea is this pernicious thought that maybe if you nuke someone else hard enough, you’ll take out all their nukes and just win. No one has ever felt confident in their ability to pull off a first strike, which is good because if someone ever was, nuclear war would become inevitable.

But why do we care about first strikes and MAD?

Because MAD has made civilization destroying nuclear war the default form of nuclear war, at least as far as all of the non-regional nuclear powers are concerned. With respect to Trump, it means that any nuclear war he starts with China or Russia, America’s traditional nuclear adversaries, would be really bad.

Now we all know that Trump is basically in Putin’s pocket. Because of this, I wasn’t very worried about a nuclear war with Russia; I always figured that if things got heated with Russia, Trump would fold.

At the time I first did this research – remember, this was September 2016, before we found out that Xi Jinping was more than a match for Trump – I thought nuclear war with China would become a lot more likely if Trump was president. So I looked into the Chinese nuclear arsenal and there I found the question that unlocked my understanding of nuclear weapons.

China’s premier missile is the silo-based Dongfeng 5. It has a range of about 12,000km and is tipped with a 5 Mt warhead.

A brief digression: when we talk about nuclear weapons and say “ton”, “kiloton”, or “megaton”, we’re referring to the explosion that would be created by a given mass of TNT. So, the warhead on the DF5 explodes with the same force as you’d expect from 5 million tonnes of TNT. Everyone always compares yields to Hiroshima and Nagasaki. I hate this and think it’s stupid – for reasons I’ll get into in just one minute – but I’ll do it anyway. The DF5 releases 250 times as much energy as the 20 kt bomb that destroyed Nagasaki.

Anyway, the DF5 is 5Mt. The premier missile used by the US is the Trident II. It also has a range of 12,000km, it’s launched from a ballistic missile submarine, and it is armed with eight W88 warheads, each of which has a yield of 475kt.

And this was confusing, because we generally think of the US as more advanced than China when it comes to military technology – and here it definitely is! So why does China have bigger nukes?

That’s our key question right there.

So I just told you I hate the Hiroshima comparison. Here’s why: it assumes that nuclear weapons scale linearly. Get twice the yield and you should get twice the destruction, right? But very few things in the real world are linear. Nuclear weapons certainly aren’t.

There are actually 5 or 6 ways a nuclear weapon can kill you. There’s the shockwave, which knocks over buildings. There’s the gamma ray burst, which make death inevitable even as you appear to recover. There’s the thermal radiation, which can give you third degree burns, even if you’re kilometers distant. There’s the central fireball, which rips apart everything it touches. And then there’s the neutron burst and the X-rays and all the other ionizing radiation sources.

Each of these scales differently, but all of them are sublinear. This means that as a nuclear weapon gets bigger, it gets less efficient. The number of people you kill per additional ton of yield is much higher when your yield is 20kt than when it is 5Mt.

Some of these scaling effects are really complicated because of interactions with the ground or the air, but two are simple enough that I can give you a quick explanation of how to calculate them.

When it comes to shockwave, I want you to imagine a sphere. The amount of stuff in that sphere is proportional to the radius of that sphere, r, to the third power. Energy is just the capacity to do work, in this case, move stuff. If you want to figure out the amount of stuff energy can move – say move in a city destroying shockwave – you move this equation around a bit and you end up with the cube root of energy. To double the range of the shockwave, you need eight times as much energy.

For thermal radiation, I want you to think about the surface of a sphere. The size of the surface is proportional to r to the second power. Now we have a set amount of thermal radiation at the start that gets spread evenly around the whole surface of the sphere as flux, even as the sphere grows. So, you get ten meters out and the energy is spread out one hundred times as much as it was at one metre. You multiplied the radius by ten and saw the energy go down by a factor of one hundred. This also means that if you add in 100 times as much energy, the radius with a given flux will only grow ten-fold. The destructive radius (for any given destructive radiation effect) is proportional to the square root of the initial energy.

In both these cases, this means you’re facing severe diminishing returns. The 5 Mt Chinese warhead isn’t 10x as powerful as the 475kt American bomb. It’s between 2 and 3 times more powerful.

It gets worse for the Chinese warhead. It’s error radius is 800m, about 10x the 70m error radius of the Trident II. When aimed at a specific hardened target, like say a silo or a fortification, a target that needs to be hit with a certain amount of energy, the Chinese weapon is actually between 3x to 5x less likely to damage it than the American one, even though it’s much bigger. That’s not even to mention that there are eight American warheads on each missile.

They’re on these things called Multiple Independent Re-entry Vehicles, or MIRVs for short. Each one can pick its own target. Add all this up and the Trident II missile is something like 24x to 40x more dangerous than the DF5, despite looking less powerful at first glance.

The Chinese warhead is big because they haven’t mastered accuracy or MIRVs. With those, size matters much less.

24x or 40x or whatever is nice and all, but why doesn’t America go for broke and pack their thing full of 5 Mt warheads too? Wouldn’t that be best?

Well that’s because space and especially weight is at a premium on a rocket. The heavier it is, the shorter its range. There’s this whole laborious process called “miniaturization” that all nuclear weapons programs have to master. You detonate your test bomb in a big fixed installation, but then you need to make it small enough that you can fit it on a missile. That’s hard.

If you look at the real experts – not the pundits on CNN, but the brilliant folks at 38North or Ploughshares – you’ll see that there’s a lot of anxiety about North Korea “miniaturizing” their nuclear weapons. Jong-un say they have. We don’t know if he’s telling the truth or not. Miniaturization is the difference between some scary seismic readings and a crater where Tokyo used to be. If North Korea can get their physics package (the nuke part of the warhead) down to 400, 500kg, then they’ll have room to put on a heat shield. Then they’ll have an ICBM.

Not a triad. So, there’s still time for a first strike. But they’re working on SLBMs. Soon, maybe in a decade, they’ll be a “real” nuclear power. That’s bad for the US. But it’s really bad for China. Right now, China is actually more at risk from North Korea than the US, according to many analysts. It’s actually gotten so bad that China has set up missile defenses between North Korea and Beijing.

These probably won’t work if push comes to shove, but that’s a story for another day.

So to summarize:

  • The major nuclear powers are China, Russia, and the USA
  • Mutually Assured Destruction is guaranteed by a nuclear triad and has kept these powers from nuking each other.
  • As long as the triad lasts, first strikes will bring massive retaliation
  • Retaliation means that you have to do a certain amount of damage to certain targets. You can achieve this with really big nukes, or really precise nukes.
  • Scaling means that 10x the yield does not bring 10x the destructive power. Conversely, accuracy gives a lot of bang for your buck. 10x accuracy means 100x or 1000x damage to a specific target.
  • Don’t use Hiroshima as a unit of measure, because people will assume that destruction is linear and overestimate how bad things will be
  • North Korea can’t do anything until they miniaturize a nuke. It’s unclear if they have yet.

If you want to learn more, either about nuclear weapons in general or North Korea’s nuclear weapon program in particular, you can go to my blog at socraticfm.com. Also Wikipedia exists.

Questions

  • In response to a question about the risk of Pakistani nuclear weapons falling into the wrong hands, I explained that this would be locally really bad, but drew the distinction between events that are bad for a localized group of people (like the Taliban nuking Karachi) and events that are bad for the human race (a MAD-level nuclear exchange between China and the US). If you’re worrying about the existential risk posed by nuclear weapons, the first is really just noise, except insofar as it can make the second more likely by increasing tensions all around.
  • In response to a question about disarmament, I talked about the New START treaty and the need to distinguish between warheads that are stored (most of them, at least for Russia and America) and warheads ready to go (1,550 for both the US and Russia, if they’re sticking to their treaty obligations). I stressed the need for further treaties like New START to slowly reduce the number of active (and therefore existentially dangerous) nuclear weapons in the arsenals of major powers.

Footnotes

[1] I was questioned pretty heavily on this pay-off matrix. Several people thought that Do Nothing should be preferred to Attack. I have two things to say to this:

  1. In an iterated game with this sort of matrix, the highest payoff comes when people cooperate the most. So while at any given point in time attacking might be preferred, once you take into account that real life is iterated, doing nothing is a better long term strategy.
  2. All of us born after the Cold War, or even born after the 60s, cannot adequately understand what it was like to live in a world where it really did seem like the Soviets might “bury us”. Faced with that kind of existential threat, a first strike seemed like an appealing option. In this globalist age, it does seem much worse to launch a first strike, especially because major powers do major mutual trade.

^

[2] If questioned here, I was going to mention carrier based bombers (France tried this for a while) and nuclear tipped cruise missiles (the US may move in this direction).  ^

Using this presentation

The slides are available here. This content, like everything else original on my blog, is covered by the CC-BY-NC-SA v4 International license. If you present this, please reference this blog and include a link. If you are a student and using this presentation is against the academic policies of your institution, I’d ask that you please refrain from plagiarizing it.