Comments
International Relations; States, Rational Actors, and Other Approaches (Policy and International Relations Primer Part 4)
19 min read · Jan 22, 2020
27
This is the fourth post in my Primer on Policy and International Relations series. Previous posts were the introduction, explaining who should bother reading this, the second post explaining what is different about the different academic areas, and the third post discussing political science and approaches generally. This post talks about some key concepts in international relations, as well as discussing realism and the rational actor approach in International Relations. Finally, it introduces some of the other key perspectives, as an introduction to the next post.
International Relations is the study of how countries and political units interact politically, economically, and legally. Political units include countries, but also international organizations, non-government organizations, and (typically to a lesser extent), sub-national groups like political parties, military organizations, and corporations. (This was discussed and compared to other approaches in an earlier post.)
The way in which international relations theorists approach the subject differs somewhat from the approaches in political science - partly for historical reasons, and partly because the best approaches are different. I won't spend much time reviewing conceptual approaches, because the line up with the approaches discussed in political science generally in the previous post. (If you haven't read that, you should do that before reading this one.)
Our focus here is on what IR theory says that's useful for Effective Altruism. It's easy to see why it could be critical. The most critical EA-Cause areas, including global poverty, global health, and existential risk reduction, all depend heavily on the actions of countries.
States, also called countries, are the units most commonly discussed in international relations. (The name gives it away; "inter-national.") For international relations, this can be a useful level of abstraction. Sometimes IR-theory gets more fined grained and talks about sub-national units, and sometimes it goes higher, to discuss national blocks or alliances. On the other hand, sometimes the idea of a state is not really appropriate. The key question is whether the definitions and assumptions are useful, and what level of abstraction is appropriate for the question at hand.
The notion of states in modern times in the west is fairly clear, and nations have been a coherent unit of analysis for a long time. While there are lots of disagreements, a "state" meant a fixed population and territory with a government. In Europe, you need to go back a thousand years or more to when the idea of country wasn't a category for important international actors. (Before this point there were many independent groups, ungoverned territory, and so on.) European countries were basically modern since the middle ages. And since political science is a western conception, most of it focused on that notion of a state.
But the rest of the world was different. In most of the rest of the world, including The Middle East, Africa, Eastern Europe, South America, and southeast Asia, colonial powers displaced a mix of kingdoms and empires that didn't follow the pattern of centralized governments and long-lasting territorial boundaries. The colonial powers mostly imposed states on the native populations (where they didn't displace them,) and that creates lots of important issues I'll touch on a bit below.
Back in Europe, the idea of states and their territorial integrity and right to self-determination led to the Westphalian State, where each country got to rule itself, and invading other countries or interfering in their domestic political affairs was at least discouraged. The treaty of Westphalia was in 1648, which made de-facto sovereignty more clearly legally established. And at the time, this was simple - as Louis XIV (never actually) said, L'État, c'est moi ("I am the state.") Each country was more-or-less under the direction of a single individual. That individual might need to negotiate internal power-struggles, keep nobles in line, or avoid having his (or more rarely, her) head cut off. To the extent that they worried about other countries, it was to make sure they didn't invade. While economic concerns were important, the idea of a Westphalian state is one that can decide to fight a war, or have an act of war committed against it.
But over the last 100 years, post-WWI and as colonialism ended, the non-military concerns became far more complex - trade, taxation, international political movements, domestic political groups with differing ideas of how the country should treat other countries, groups that were for or against wars, and so on. The idea of Westphalian states has also become less relevant with the growth of globalization, more tightly integrated unions like the EU, the willingness of the international community to allow humanitarian or other interventions in sovereign states. Now, wars can be between NATO and individual countries, so that only one side is a Westphalian state, or between ISIS and an international coalition, neither of which is a state.
Not only that, but post-colonialism, many states failed to meet the qualifications for a different reason; lack of control. The definition of state is disputed, but ungoverned areas usually don't qualify. The lack of central government control is extremely important when Effective Altruists are thinking about the developing world, because often the areas with the least developed governance are those with the poorest and worst-off people.
In terms of political science and International Relations, places were there is no coherent government are in some sense not actually states. For example, one criteria for control is the monopoly on the use of violence; only the government is allowed to use or license the use of force. But this is a continuum. Afghanistan is a mix of a central government and quasi-independent tribal areas. Pakistan's border is in many ways under control of neither government, and while the border may be fixed on maps, the actual control is very different. This is the reality in many places in formerly colonized areas, especially those that were never fully brought under control of a colonial power.
On the other hand, the most influential countries are precisely those where there is a strong government - which gives it the ability to act as a coherent entity internationally. A second point in favor of this model is that even when states are not actually in control of their territory, or are falling apart, they are treated as states for the purposes of international treaties, trade negotiations, and membership in bodies like the UN. (For example, when rebel groups or ISIS controlled far more than half of Syria, Assad's representative at the UN wasn't removed.) So we can treat unity and coherence of a Nation's international relationships as a relatively common state of affairs, and it's often true - especially in places with the most international power.
Such unity is particularly likely when discussing a Nation-State, an autonomous country where most or all people share a common culture and language. In such a Nation-State, most people in the country will agree about what their national interest is, and strong central governments are a near-certainty. But this is also less and less common as globalization becomes more pervasive. Even when immigration and population differences are small or nonexistent, national identities have become weaker.
Of course, people who don't spend time reading international relations textbooks often confuse the terms. But people who do spend their time reading such textbooks argue about the terms, rather than confuse them with one another. And so it's important to note that there are a bunch of international relations theorists who have started using the term nation-state to mean country, to differentiate it from states in the United States or similar country. This is confusing. I'm sorry.
So to clarify, there are three different ideas in the above section, and not all "countries" are states, few modern states are nation-states in the original sense, and while all Westphalian states are countries, the reverse isn't true. First, we have States, meaning a political body with a government that controls its territory and can operate independently Second, we have Westphalian States, meaning a state that has the right to national sovereignty and can set its own rules. Third, we have Nation States, which are states that have relatively uniform populations that agree on overall priorities. (Except that, again, sometimes "Nation State" just means state.)
As noted earlier, some countries lack centralized control, or do not control their own territory. These may be failed or 'fragile' states like Syria, Sudan, or Somlia, nascent states like Palestine, contested states like Taiwan, or ungoverned areas. Other nations are states, but not Westphalian states. For example, countries that were in the Soviet Bloc. This is also different than the United States, where the individual states are not nations, but are instead clearly subnational units.
Another unclear case that breaks many of the rules traditionally assumed for states are members of the EU, where the countries ceded control of particular sovereign decisions to some other unit, but are clearly sovereign, and could secede - if they every manage to decide what they want. But EU countries are unusual because they cannot restrict entry into their territory from other states, (mostly) do not control their own currencies, cannot make trade agreements on their own, and are required to come to each other's defense, so that in some sense they cannot be attacked individually.
Before talking about specific approaches, there's an elephant in the room - realism. As with Institutionalism in political science, this has deep roots, and is in many ways the default, and is a baseline for understanding how other approaches differ. That's especially true for rational-actor theories, so it's useful to discuss it first.
A dominant school of thought in International Relations is realism, which is closely related to the Realpolitik-approach. This looks at the world through a somewhat cynical lens; states do what they want to get the things they want, they do so in whatever way works best, i.e. what we'd now call rationally, and the only reason they follow rules or abide by norms is because they benefit from them, or don't want to pay the costs of breaking them and getting caught.
As with the approaches discussed in the previous post on political science, a non-dogmatic and reasonably hedged version of this is roughly consistent with most other approaches. It's also far from sufficient as a theory to predict anything - for that, we need data or analysis of what states want, how they can get those things, and how their relationships work.
It also looks at the world through the lens of states, and only states. This means that it assumes the correct level of analysis for the questions it addresses is the state. This isn't the only approach, though it's a really important one - and again, it isn't enough by itself. Even with all of those assumptions, we don't have much of a theory without looking at how it plays out in the world.
At this point, I'll depart from traditional categories, and point out that essentially everyone, whether or not they embrace realism, agrees that international relations needs to learn from data and history. In my view, there are three main approaches for doing so.
The first empirical approach is to focus on qualitative in-depth explorations, without trying to quantify things. This is the approach of historical analyses, geopolitics, and comparative politics. It's also the approach of realism before rational actor models were considered. The second approach is to see what can be found from data analysis, without making lots of additional assumptions. At times, the facts are clear enough that we can say something meaningful. The conclusions may seem obvious, in retrospect - but that's hindsight bias, and the data collection and analysis is still probably valuable. Other times, political scientists use econometric analyses, despite the fact that the statistical assumptions are difficult to assess, and rarely hold in practice. This can be insightful, but is unlikely to provide sufficiently precise answers to most EA-relevant questions.
We'll get back to these first two approaches. The third approach, however, is to build simplified models of how states interact, make assumptions, and see what we think will happen - and possibly use data or qualitative methods to calibrate those models or estimate the inputs. This use of models is a typical approach in non-social science areas, and similar approaches are becoming more common in social science as well. But given the absurd number of researcher degrees of freedom, the complexity of the models, the fact that the models are being built post-hoc, i.e. with foreknowledge of the data, and contain many simplifications, this is absurdly under-determined. And yet...
As mentioned in the previous post, rational actor approaches are a big deal in International Relations. That's because they give analysts a clear type of theory to work with, which is far better than ungrounded theorizing, or theory-free data analysis. (To be less charitable, it's also probably partly physics- and economics-envy on the part of international relations theorists. But we're here to understand international relations, not to praise it.)
The more mathematical class of rational actor models are explicit game theory, and the obvious example of this is also the most famous - Mutually Assured Destruction during the cold war. To set this model up, we assume that the USSR and the United States are the only two relevant actors, and each has a very simple set of goals relevant to the analysis - they want to survive, and , given survival, they'd prefer to the defeat the other side. Given that, it's straightforward to set up the payoffs for investing in nuclear weapons, and for actually attacking.
I'm not going to explain the basics of Game Theory here, (I'll leave that to Scott Alexander's 10 part sequence on Lesswrong,) but I'll assume you're either familiar with the basics already, or you decided to read Scott's very clear, non-technical, and highly recommended sequence. Given that basic understanding, the dynamics are incredibly simple; either build nuclear weapons that can kill your opponent, or surrender, then either have a nuclear war, or don't. The model of investment and an iterated prisoner's dilemma is enough to show why actually fighting a thermonuclear war is a bad idea, but preparing for one will still be useful. And at that point the only winning move is... not to reason from fictional evidence.
This model works particularly well because it's simple enough that the game theoretic model gives a clear answer, and the individual actors were mostly willing to try to act rationally. Not only that, but the two countries were focused on this one issue to the extent that anything else paled in comparison, and so annoying side-issues didn't invalidate the assumptions about not having other motives. And finally, the two actors were coherent nation-states, each with a clear decision making apparatus that, if not centralized, was at least in agreement about the goals.
Deterrence theory is a big deal in international relations, and there is extensive literature on the topic. Not only that, but deterrence theory relates to a number critical questions for a number of questions in existential risk policy research and mitigation. These issues are worth noting here, and discussing further. But other than noting that they are issues, I'm not getting into a discussion here about how AI affects deterrence, or how deterrence works differently for biological weapons - but those are good questions for EAs to continue to think about. (And have already been the focus of a fair amount of work at FHI and other places.)
Rational choice is famous for being used in deterrence theory, but it's also a much more general approach. Given that, it's reasonable to ask if the rational actor model would still work in places where the payoffs are less stark, the motives were less unified, or there were more actors. And the answer there is more complicated. Above, I noted that game theory was the more mathematical way of applying the rational actor model.
At this point, recall that the rational-actor approach was originally grounded in realism. It doesn't need to embrace all of the same assumptions, however, and is often extended to allow consideration of non-state actors, sub-national groups, and more complex relationships. This can still be explicitly game-theoretic, and to get some idea of what this looks like, I have a surprisingly topical example - Iran.
A modern-style agent-based rational actor model of Iran needs to look at the relative influence of the president, the supreme leader, the revolutionary guard, the reform movement, and so on. Each has some amount of influence on decisions in a given domain, and each has preferences about the outcomes. These can all be elicited from experts, and combined with some extra simplistic assumptions about the utility functions of each actor. (They need to be single peaked, and one-dimensional, so we can do the math.) It's a lot of work, but it's tractable as long as you have a couple years to do the work as a dissertation, you focus on a single country, treat issues individually, and you have access to experts.
On the other hand, the questions you are interested in might be more complex, say, to tell you something about how Iran will react to foreign aggression. Responses to this might involve many parts. The Revolutionary guard might work with Hezbollah to get the Syrian government to pressure Iranian groups to support the Ayatollah, student groups might receive tacit or explicit support from outside sources, or an outside country like Saudi Arabia or Israel might (secretly) fund domestic reformers in Iran to help them oppose Iran's nuclear program, or have them push for more radical change.
To model any of this, you need to embed the domestic model in a regional model. This can be incredibly messy, since coalitions can form not only within a country, but across them. This also requires understanding of not only what each group wants, but their relative preference for outcomes across domains; will the IRGC accept a slight increase in domestic stability in exchange for not funding parts of their international network, or will the Ayatollah make concessions on religious matters to appease protesters? Some of this can be addressed with more information, but as I'll discuss below, that's limited as well.
Hopefully, this gives an idea about what explicitly game-theoretic rational choice conversations would involve. But as mentioned earlier, countries themselves change over time, and the relevant actors shift. That makes extracting useful results from expert opinions and historical cases far harder. And while people are fairly obviously a coherent category, but as discussed above, what counts as a state, and how much each place qualifies, is a bit fuzzier.
Before getting to key drawbacks, there are a number poorly-considered criticisms that should be addressed. For example, states with “irrational” goals are easily incorporated in these models because rationality does not dictate values. Iran or Saudi Arabia may have "irrational" goals relating to religion, but that doesn't mean they don't pursue those goals rationally. Similarly, the idea that humans are simply too complex to model mathematically is unreasonable - it's either claiming that we can't see regularity in behavior at all, which is simply false, claiming that math can't represent what our minds can, which is less simply but still empirically false, or claiming that the system is too chaotic to predict - which is partially true, but not restricted to rational actor approaches.
But as the example shows, there are some actually critical issues with the approach as well - it's often unwieldy due to the number of highly uncertain and subjective factors that need to be input, it will be very sensitive to some of the assumptions, it is fragile to omitting potential options, groups, and outside factors, and it doesn't usually tell you which of many possible stable outcomes (Nash equilibria) should be expected. Not only that, but there are lots of places that game theory doesn't provide an understandable or useful way to talk about many things that are actually discussed in international relations - norms, rules, psychological factors that affect decisions, and institutions.
And we need inputs for these models - but there just aren't very many historical data points or enough information to guess what groups' goals are, since reality is often underpowered. We also have trouble generalizing, since while getting sufficient samples sizes for social science is always hard because it's costly, for countries it is likely that to answer even basic questions with a high degree of confidence, we need more data than could possibly exist. And this is particularly difficult when we're looking at rational actor models that are very complex, and can require exponentially many inputs for even fairly simple problems.
Some of this, especially the last concern, can be fixed by simply backing off of the explicitly mathematical approach, while keeping the rational actor assumptions. This works less well than many assume, however, since all of the problems with rational actor models (such as non-predictability, sensitivity to assumptions, fragility to changes or missing factors,) remain even when the concerns are not explicit.
Given the issues with rational actor models, we need to consider other approaches, and whether they help address these issues. As I said above, perhaps the easiest way to do this is to return to realism, suitably modified - but there are a spectrum of approaches for how far to depart from the assumptions.
Neorealism is mostly based around an understanding that the most important factor in game theory isn't the specifics of the motivations, it's the structure of the game, and the strategies available. And the prevalent neorealist view is that since there is no-one making binding rules on states, the rules of the game are anarchy. The structure is therefore just about economic and military power, and threatening or actually using it.
Another approach is Rational choice institutionalism, which both disagrees about the fundamental idea that international relations are anarchic, and following the view of institutionalism in political science, pays attention to institutions. This may think of structures as pieces of the broader game, but pays less attention to norms and values, and more attention to how states use institutions. This is also a version of realism that departs form just looking at states, and uses rational actor methods to understand how states relate to their constituents, and how international institutions relate to states. (Don't worry, it's principal-agent problems all the way down.)
But there are obviously more radical departures from realism, and I'm hoping to discuss them in future posts. To introduce them, however, I wanted to outline the alternatives to realism, rational actors, and neorealism.
Neoliberalism is a slight modification of assumptions leading to a huge modification in approach. The key difference in assumptions is that neoliberals see things other than zero-sum power as the goal of states. And since positive sum interactions are possible, states can and will cooperate. This, in turn, both leads to emphasizing international organizations in a way very different from rational choice institutionalism, and leads to placing greater emphasis on norms and values.
Democratic peace theory is in many ways an extension of neoliberalism that focuses more on why democracies don't usually go to war, and they especially don't go to war with other democracies. That's partly because the population very rarely thinks the costs of war are worth the cost. It's also partly because so many of the ways wars start are shortcut by requiring public deliberation.
Departing even further from realism are Marxist approaches, which view the primary driver of decisions as class-driven, rather than state-driven. (As in the previous post, note that Marxist approaches are very different than advocating communism or socialism.) In this view, the United States goes to war because the upper class want power. They invade Iraq for oil to satisfy the economic needs of the Bourgeoisie, and don't care about the costs to the country or to the people at large. Similarly, international trade is driven by a tension between the public's desire for good jobs and cheap goods versus the upper class desire for profits and rent seeking. For EAs, this viewpoint seems particularly relevant to debates about the impacts of artificial intelligence, or universal basic income.
There are also a variety of approaches related to Constructivism, the third general approach after neorealism and neoliberalism . This approach looks at how international relations is dominated by human mental constructs. On this level states are just abstractions, or even shared delusions of the masses. This can include marxism, but can take these views even further. For example, some claim that most decisionmaking in international relations is following roles rather than pursuing goals - perhaps the equivalent of system 1 decision-making at the state and international level. Other strands related to constructivism include feminist theory, which focuses on identities (not necessarily those related to gender,) and geopolitics, which as discussed in the second post, focuses on conceptions of the world.
In the next post, I'm hoping to look at how each of these provides useful insight into international relations, and either in that post or the following one, I will talk more about the methods and tools used.
1) I will make no comment about what this says about the southern United States' gun culture and the seemingly legitimacy of the use of violence to defend property - though evidently there is something to be said for the claim that the private right to bear arms is something that makes America unique, at least among developed nations.
2)
3) An interesting story about the founding of the UN illustrates this. The non-USSR Soviet bloc, i.e. Poland, Romania, and other members of the Eastern Bloc, were independent states that were allied with the USSR - so each got a seat in the UN as an independent nation. But the USSR said that by itself, it should have 16 memberships in the General Assembly, one for each (now-former) soviet republic - the constituent republics of the USSR, which were not independent states at all. Supposedly, the United States countered this argument by saying that if so, it would be happy to have 50 seats, one for each of its states. This may seem ridiculous, but the final agreement was that the USSR was given 3 seats, with Ukraine and Byelorussia having their vote exercised by the USSR despite being fully part of the USSR, in exchange for the US getting the (never exercised) right to add an additional 2 seats for itself.
4) In the linked example, the book actually collected data about all the insurgencies in the past 50 years, and found perhaps unsurprising buy useful insights. For example, insurgencies that get help from outside great powers are more likely to succeed, longer insurgencies are less successful, and insurgencies that employ suicide bombers have never actually succeeded. But each case is unique, and making blanket statements ignores the differences.
5) Because as the movie evidently assumes, avoiding nuclear war is harder than coding an intelligent AI that can successfully apply meta-reasoning, despite the challenges.
6) I promise I actually wrote the entire next paragraph - and the next footnote - before Soleimani was killed. (The part after that was added more recently.)
7) Interestingly, it has now been long enough to see how well the projections from this model performed - and given the uncertainty, it did fairly well. (Full disclosure: I went to the same graduate school as the author of this dissertation, and discussed another iteration of this model with one of his advisers.)
8) These examples are plausible, but are not based on any knowledge of internal Iranian politics, which is notoriously opaque, or of any other government's involvement. (I don't work for Mossad.)
9) There's a really cool reason that this is hard to do. It's because we break the assumptions of single-dimensional preferences, and no analogue of the median exists in multidimensional space. (We need to splits the data into equal halves along all dimensions the way the median does in one dimension, making the median-voter result applicable. It doesn't work in more dimensions.) Instead, we need to account for logrolling, which is hard. (Not log-rolling, which is evidently also hard.)
10) Note that neoconservatism is somewhat related to neorealism, with some different assumptions, but it's not an approach to understanding international relations, it's a set of strategies for actually playing the game.
11) This builds on liberalism, which was a rejection of realism. That rejection was about saying that sometimes states cooperate, they care about things other than power, and many institutions and public groups have influence despite not having real power. My view, which I don't claim is widely shared, is that neoliberalism gave this view a single unifying meta-assumption that justifies it, namely positive sum interaction.
12) I could argue that to the extent a country is willing to go to war, it's failing as a democracy. When the president of the US needs to go to Congress to declare war, it's no longer possible for it to be a surprise, and alternatives involving negotiations with the other country gains tremendous pressure from both sides seeing the other being willing to go to war, or leads one side to capitulate when it becomes clear that their population isn't willing to allow a war.
This is an interesting review. My family had a grad student from Kenya stay with us when he was working on his PhD in my city--on IR. It was not what i studied so he explained to me the various schools and people ---George Kennon, Hans Morgenthau, etc. (Noam Chomsky has many critiques of Kennon but I looked at what Kennon actually wrote, and Chomsky used selective quotes).
The 'median voter theorem' or 'single peaked /dimensional preferences' is the basic theorem in this area---if you model this mathematically. It also applies to US democracy, etc. It does not hold--in either politics or economics (K Arrow). So you have to deal with more complex situations.
The current world seems a bit too complex to easily model this kind of thing either nationally or internationally . Also, even if you did have a good model, it might be so complex most people could not understand it.
It's true that assuming single peaked preferences is usually really central to rational actor approaches, but there are a few different issues that exist which should be separated. Arrows theorem is in many cases, no voting system is Pareto-compatible, non-dictatorial, and allows independence of irrelevant alternatives.
First, as you noted, these classes of preference don't imply that there are coherent ranked preferences in a group (unless we also have only a single continuous preference dimension). If I prefer rice to beans to corn for dinner and you prefer beans to corn to rice, while our friend prefers corn to rice to beans, it's not a continuous system, and there's no way that voting will help - any alternative has 2/3rds of voters opposed. (Think this isn't a ever relevant issue? Remember Brexit?)
Second, even if the domain is continuous, if there is more than one dimension, it still can fail. For example, we need to order lunch and dinner together, I want 75% beans, and 25% rice for dinner, and 50% of each for lunch, and it's a monotonic and continuous preference - i.e. the farther away from my preferred split we get, the less I like it. If I take a bunch of similar types of preferences about these meals and need to make a single large order, arrow's theorem shows that there may be no voting system that allows people to agree on any particular combination for the two meals - there can be a majority opposed to any one order.
And third, it's sometimes simply incorrect as a description of people preferences. As an example, a voter might reasonably have preferences for either high taxes and strong regulation with a strong social safety net so that people can depend on the government, OR low taxes, little regulation, and no safety net so that people need to build social organizations to mutually support one another, and say that anything in between is a worse idea than either. These preferences are plausibly collapsible to a single dimension, but they still admit Arrow's problem because they are not single-peaked.
But in each case, it's not a problem for reality, it's a problem with our map. And if we're making decisions, we should want an accurate map - which is what the series of posts is hoping to help people build.
I didn't notice this reply to my comment, but today i did. (as an aside i notice i have -33 votes and +55 votes for my comments. it would be nice if EA world could tolerate diversity of opinions--so far i'm still above zero and i don't want to start a war. but if people cannot disagree agreeably i'm perfectly willing to go somewhere else--and it won't be silicon valley. )
While i dislike the term 'intelligent', i would call your comment or answer a very intelligent and well informed reply to my comment.
Since you mentioned Brexit, in USA we have a similar issue---- the democratic primaries (as you likely know, though my impression is you are not in USA) --there are not really any single peaked preferences here though it has been reduced to Biden vs Bernie and other --possibly alot of nonvoters.)
I agree that 'the map is not the territory'. One needs to revise the maps as you go through the territory. there is a tradeoff --- do you follow the map you have and keep going, or spend time revising the map and take a different path?