I finally got around to reading (most of) Zvi’s sequence Immoral Mazes. I may want to turn some of the following ideas into longer posts, but I thought I’d try to get out some quick thoughts.
Connections to Dictator’s Handbook
I’ve also recently finished reading Dictator’s Handbook. The two share some similarities, but offer very different analyses of the problem and very different directions for hope/improvement.
You can think of government as a game played between rulers and ruled. Rulers who don’t play to maximize the amount of time they stay in power don’t rule for long, so we can model rulers’ utility function as just maximizing time in power. This fits most cases pretty well, and explains a lot of political phenomena.
We can ignore most of the details of a given political system, and just put systems on a spectrum from “perfect dictatorship” to “perfect democracy” by how many key supporters a ruler needs to please to stay in power. In a realistic dictatorship, the answer is a handful of people. In a realistic democracy, the answer is a significant fraction of the population.
The ruler’s incentive is always to minimize the number of key supporters needed, and then make those supporters satisfied.
The key supporters sometimes have incentives aligned with the ruler, IE, also want to minimize the number of key supporters (to get more rewards per key supporter). This is especially true when there are few key supporters already. This will lead to culls and movement further and further toward dictatorship.
At other times, the key supporters may want to expand the number of key supporters. This is because (1) they want to protect themselves against a cull, and (2) they may want to incentivise the ruler to produce public goods (rather than special benefits to supporters). This is especially true when the number of key supporters is already very large, so special benefits to supporters are small, and won’t be hurt much by expanding the coalition (and are outweighed by benefits from public goods). Expanding the coalition further provides insulation from culls. This leads to further democratization.
Immoral Mazes seem sort of like the Dictator’s Handbook model but with more levels. The Dictator’s Handbook did mention (I think?) that the key supporters are usually playing the same game one level down, but didn’t make very much of this.
I think the Dictator’s Handbook model is clearer and more precise than Immoral Mazes, so it would be cool to try and meld them together further.
Some big differences between the models:
Dictator’s Handbook sees everyone as just following their incentives. Moloch is very strong in organizations closer to the dictatorship end of the spectrum, and relatively weak in organizations close to the democracy end of the spectrum.
Immoral Mazes sees everyone as following perceived incentives, but strongly holds that it isn’t worth it. People in Mazes are making a mistake.
Both models would agree that flatter hierarchies result in better outcomes (for the majority of people). But for different reasons:
Immoral Mazes holds that hierarchy itself is the problem. Deeper hierarchy means increasingly warped incentives.
Dictator’s Handbook holds that small branching numbers in hierarchy is the problem. Smaller branching numbers naturally lead to deeper hierarchy, but the more important point according to Dictator’s Handbook is that each boss has to satisfy fewer supporters.
Dictator’s Handbook doesn’t really care what the executive hierarchy looks like, so long as it is answerable to a large number of people. A president may set up a deep hierarchical government, but since the president is ultimately answerable to the people, you’re in a democratic (and therefore relatively benevolent) regime.
Immoral Mazes predicts that the deep hierarchy in the government is as big a problem as deep hierarchy anywhere else; it matters only a little that the president answers to the people.
Here’s my summary, biasing toward connections to the other two models:
Like the other two models, Gervais Principle emphasizes that large organizations are fundamentally dysfunctional, IE, not very aligned with carrying out their purported function.
Like Immoral Mazes, Gervais Principle gives a special role to middle management.
In the Gervais Principle, the top of an organization is a collection of sociopaths. The middle management are called the clueless. The bottom rungs of an organization are called the losers.
Sociopaths are the really ambitious people who create the warped moral frame that middle management buys into. Sociopaths rise to the top and then proceed to extract profit, like the rulers in Dictator’s Handbook.
Clueless are the people who sociopaths want beneath them. Ideally (for the sociopath) there would only be one sociopath; the key supporters would all be clueless loyalists who buy into the narrative. Unfortunately (for the top sociopath), other sociopaths will manipulate their way into the highest positions. So the top management realistically consists of power-hungry sociopaths who weave a web of moral confusion. Middle management lives in this web of confusion.
Losers are at the bottom. They work for their paycheck and little more. The sociopaths don’t need to fool them, so they are usually more aware of the dysfunctional nature of the organization, but can’t do anything about it.
The Gervais Principle takes an almost anthropological approach to these three groups of people, describing their different cultures and norms. It offers no hope for building better institutions.
I like the way The Gervais Principle factors loyalty into the model, highlighting the way that rulers select key supporters based on loyalty in a way the more game-theoretic model of The Dictator’s Handbook can’t really explain. Loyalty in a game-theoretic model just means that dictators make supporters dependent on them and rule through fear. Loyalty in real life includes more than this; as Immoral Mazes put it, giving up your soul.
The Great Fragmentation
In Immoral Mazes, Zvi asserts that mazes are on the rise. Corporations and other organizations with large mazes have been around for a long time in America, slowly but surely increasing their internal maze levels, and seeping out maze culture into the general cultural mixing pot.
Now, I’m not saying Zvi is wrong, but I see a lot of things which are discordant with this picture.
When I look at recent American history, I get the following impression:
The 50s and 60s saw “peak square”—the socioeconomic landscape was dominated by large companies and “the company man”.
The 60s and 70s saw a huge cultural revolution, which began a major decline in squares. Men stopped wearing suits and ties, and alternative hairstyles (IE long hair) became much more socially acceptable for men. Women were less and less pressured to live traditionally as housewives.
The “company man” became less and less a thing, as the average number of careers in one’s lifetime increased. Today it’s hardly realistic to expect to work for one company for your whole life.
In the 90s, start-up culture started to be a major thing. We saw the rise of “the tech industry” as it exists today, which Zvi himself admits is somewhat less prone to immoral mazes.
In the 00s, hipster culture (like hippy culture before it) created commercial pressure against homogenized goods. Large record labels became much less important in the music industry. The same thing happened across many other areas of media, continuing into the 2010s, due to the new possibilities opened up by the internet. We saw the rise of indie everything, which by Zvi’s account, should greatly reduce mazes.
All of these impressions are greatly reinforced by Paul Graham’s essay, The Refragmentation.
All of these forces push against large organizations with deep hierarchies. The fall of “square” culture should be a major blow to maze culture—americans were no longer interested in becoming cookie-cuttout people (as required by mazes).
Yet Zvi claims that maze culture has been on the rise!
What could be the cause, if many drivers of mazes are in decline?
One hypothesis could be that although the root causes are in some ways decreasing, the damage has been done—like someone exposed to the common cold at a party, and who gradually gets worse once they get home. America has contracted the maze disease, and it continues to fester. In other words, even if large corporations with deep hierarchies are actually less prevalent than they once were, the maze cultures in those that exist are far, far more developed.
I’m somewhat suspicious of this explanation, because in a culture where non-maze opportunities are opening up more and more, it seems like mazes should die. I think people’s behavior is more dependent on context than it is on these subtle cultural influences, so I expect to see larger effect sizes from incentive structures changing than from negative cultures slowly festering over time. But perhaps it’s a factor.
Another hypothesis is slow strangulation via red tape. As regulations and restrictions increase over time, it gets harder and harder to escape mazes, and mazes get deeper and deeper. More generally, it could be that although some causes of mazes have diminished, other more important causes have increased, making mazes worse overall.
It’s also possible that Zvi is wrong, mazes as such are not on the rise after all, and the creeping insanity which Zvi is noticing has some other source.
Quick Thoughts on Immoral Mazes
I finally got around to reading (most of) Zvi’s sequence Immoral Mazes. I may want to turn some of the following ideas into longer posts, but I thought I’d try to get out some quick thoughts.
Connections to Dictator’s Handbook
I’ve also recently finished reading Dictator’s Handbook. The two share some similarities, but offer very different analyses of the problem and very different directions for hope/improvement.
Dictator’s handbook in brief (see also the excellent summary by CPG grey on youtube):
You can think of government as a game played between rulers and ruled. Rulers who don’t play to maximize the amount of time they stay in power don’t rule for long, so we can model rulers’ utility function as just maximizing time in power. This fits most cases pretty well, and explains a lot of political phenomena.
We can ignore most of the details of a given political system, and just put systems on a spectrum from “perfect dictatorship” to “perfect democracy” by how many key supporters a ruler needs to please to stay in power. In a realistic dictatorship, the answer is a handful of people. In a realistic democracy, the answer is a significant fraction of the population.
The ruler’s incentive is always to minimize the number of key supporters needed, and then make those supporters satisfied.
The key supporters sometimes have incentives aligned with the ruler, IE, also want to minimize the number of key supporters (to get more rewards per key supporter). This is especially true when there are few key supporters already. This will lead to culls and movement further and further toward dictatorship.
At other times, the key supporters may want to expand the number of key supporters. This is because (1) they want to protect themselves against a cull, and (2) they may want to incentivise the ruler to produce public goods (rather than special benefits to supporters). This is especially true when the number of key supporters is already very large, so special benefits to supporters are small, and won’t be hurt much by expanding the coalition (and are outweighed by benefits from public goods). Expanding the coalition further provides insulation from culls. This leads to further democratization.
Immoral Mazes seem sort of like the Dictator’s Handbook model but with more levels. The Dictator’s Handbook did mention (I think?) that the key supporters are usually playing the same game one level down, but didn’t make very much of this.
I think the Dictator’s Handbook model is clearer and more precise than Immoral Mazes, so it would be cool to try and meld them together further.
Some big differences between the models:
Dictator’s Handbook sees everyone as just following their incentives. Moloch is very strong in organizations closer to the dictatorship end of the spectrum, and relatively weak in organizations close to the democracy end of the spectrum.
Immoral Mazes sees everyone as following perceived incentives, but strongly holds that it isn’t worth it. People in Mazes are making a mistake.
Both models would agree that flatter hierarchies result in better outcomes (for the majority of people). But for different reasons:
Immoral Mazes holds that hierarchy itself is the problem. Deeper hierarchy means increasingly warped incentives.
Dictator’s Handbook holds that small branching numbers in hierarchy is the problem. Smaller branching numbers naturally lead to deeper hierarchy, but the more important point according to Dictator’s Handbook is that each boss has to satisfy fewer supporters.
Dictator’s Handbook doesn’t really care what the executive hierarchy looks like, so long as it is answerable to a large number of people. A president may set up a deep hierarchical government, but since the president is ultimately answerable to the people, you’re in a democratic (and therefore relatively benevolent) regime.
Immoral Mazes predicts that the deep hierarchy in the government is as big a problem as deep hierarchy anywhere else; it matters only a little that the president answers to the people.
Connections to Gervais Principle
Another model with striking similarities is The Gervais Principle.
Here’s my summary, biasing toward connections to the other two models:
Like the other two models, Gervais Principle emphasizes that large organizations are fundamentally dysfunctional, IE, not very aligned with carrying out their purported function.
Like Immoral Mazes, Gervais Principle gives a special role to middle management.
In the Gervais Principle, the top of an organization is a collection of sociopaths. The middle management are called the clueless. The bottom rungs of an organization are called the losers.
Sociopaths are the really ambitious people who create the warped moral frame that middle management buys into. Sociopaths rise to the top and then proceed to extract profit, like the rulers in Dictator’s Handbook.
Clueless are the people who sociopaths want beneath them. Ideally (for the sociopath) there would only be one sociopath; the key supporters would all be clueless loyalists who buy into the narrative. Unfortunately (for the top sociopath), other sociopaths will manipulate their way into the highest positions. So the top management realistically consists of power-hungry sociopaths who weave a web of moral confusion. Middle management lives in this web of confusion.
Losers are at the bottom. They work for their paycheck and little more. The sociopaths don’t need to fool them, so they are usually more aware of the dysfunctional nature of the organization, but can’t do anything about it.
The Gervais Principle takes an almost anthropological approach to these three groups of people, describing their different cultures and norms. It offers no hope for building better institutions.
I like the way The Gervais Principle factors loyalty into the model, highlighting the way that rulers select key supporters based on loyalty in a way the more game-theoretic model of The Dictator’s Handbook can’t really explain. Loyalty in a game-theoretic model just means that dictators make supporters dependent on them and rule through fear. Loyalty in real life includes more than this; as Immoral Mazes put it, giving up your soul.
The Great Fragmentation
In Immoral Mazes, Zvi asserts that mazes are on the rise. Corporations and other organizations with large mazes have been around for a long time in America, slowly but surely increasing their internal maze levels, and seeping out maze culture into the general cultural mixing pot.
Now, I’m not saying Zvi is wrong, but I see a lot of things which are discordant with this picture.
When I look at recent American history, I get the following impression:
The 50s and 60s saw “peak square”—the socioeconomic landscape was dominated by large companies and “the company man”.
The 60s and 70s saw a huge cultural revolution, which began a major decline in squares. Men stopped wearing suits and ties, and alternative hairstyles (IE long hair) became much more socially acceptable for men. Women were less and less pressured to live traditionally as housewives.
The “company man” became less and less a thing, as the average number of careers in one’s lifetime increased. Today it’s hardly realistic to expect to work for one company for your whole life.
In the 90s, start-up culture started to be a major thing. We saw the rise of “the tech industry” as it exists today, which Zvi himself admits is somewhat less prone to immoral mazes.
In the 00s, hipster culture (like hippy culture before it) created commercial pressure against homogenized goods. Large record labels became much less important in the music industry. The same thing happened across many other areas of media, continuing into the 2010s, due to the new possibilities opened up by the internet. We saw the rise of indie everything, which by Zvi’s account, should greatly reduce mazes.
All of these impressions are greatly reinforced by Paul Graham’s essay, The Refragmentation.
All of these forces push against large organizations with deep hierarchies. The fall of “square” culture should be a major blow to maze culture—americans were no longer interested in becoming cookie-cuttout people (as required by mazes).
Yet Zvi claims that maze culture has been on the rise!
What could be the cause, if many drivers of mazes are in decline?
One hypothesis could be that although the root causes are in some ways decreasing, the damage has been done—like someone exposed to the common cold at a party, and who gradually gets worse once they get home. America has contracted the maze disease, and it continues to fester. In other words, even if large corporations with deep hierarchies are actually less prevalent than they once were, the maze cultures in those that exist are far, far more developed.
I’m somewhat suspicious of this explanation, because in a culture where non-maze opportunities are opening up more and more, it seems like mazes should die. I think people’s behavior is more dependent on context than it is on these subtle cultural influences, so I expect to see larger effect sizes from incentive structures changing than from negative cultures slowly festering over time. But perhaps it’s a factor.
Another hypothesis is slow strangulation via red tape. As regulations and restrictions increase over time, it gets harder and harder to escape mazes, and mazes get deeper and deeper. More generally, it could be that although some causes of mazes have diminished, other more important causes have increased, making mazes worse overall.
It’s also possible that Zvi is wrong, mazes as such are not on the rise after all, and the creeping insanity which Zvi is noticing has some other source.