Hello, LessWrong! Long time reader, first-time commenter.
I think that your description of the counter-maze tendency is wrong, and you’ve misunderstood some aspects of Zvi’s model while being distracted by superficialities in the other. To wit:
Startups employ a trivial percentage of the workforce and do not contribute much to the economy. Startups that get big occupy more economic space, but by that point they’re no longer startups. So the attributes of startup culture are not really relevant to the economy at large.
It’s understood by everyone that being a startup is a kind of corporate childhood—romanticised, but temporary. It is implicitly accepted by everyone, including startup founders, that growing up and becoming a mature, “adult” company requires becoming a maze with multiple levels of hierarchy.
Tech culture is anti-maze only insofar as it consists of startups. All of FAANGM are fully converted to mazes. Consequently, most actual tech workers work within mazes.
The change away from suits etc. is a change of fashion with no impact on the underlying dynamics. Arguably it actually helps out the sociopaths because it replaces a fixed, legible standard of dress with an unclear and illegible question of “culture fit”, which creates more room for maze games.
The move away from “company men” was not a move away from large firms, but rather a move away from a vertical system to a stratified one. In the old system (prior to 1970) you could expect to work your entire life for the same company, and middle and upper management was typically promoted from the rank-and-file. In the newer system, middle and upper management are hired from people with MBAs and other credentials, and they move freely between industries. As a consequence, the maze-nature is transmitted quickly from company to company, and to a certain extent all management everywhere is joined in a super-maze, as all management shares the same culture and experiences which is completely separate from the culture and experiences of the workers.
Actually, this last point deserves more elaboration: according to Zvi, the main thing that mitigates against mazes is direct engagement with the object-level reality. This engagement is present in the rank-and-file, and to a certain extent at the very top, but is absent in the middle. However, in a vertical system of advancement, where management hires are made from within, the middle ranks will at least have a memory of working on the object-level concerns of the firm. The rise of a permanent managerial class means that many middle managers are of a type which has never worked directly on the actual product that their company makes, and whose entire education and experience is in the context of immoral mazes.
So I find it unpersuasive to think that any of the cultural changes of the previous decades have done anything to reverse the advance of mazes as the normative corporate structure, and some of the things that you mention as inhibiting maze structures (such as frequently changing companies over the course of a career) have probably actually accelerated them.
An additional, unrelated note: the model of The Dictator’s Handbook suggests that incentives push away from the middle, towards total democracy (when there are already a large number of key supporters) or total autocracy (where the number of key supporters approaches one). But don’t other models suggest that the middle state of oligarchy is actually the default, and that both democracy and monarchy tend to decay towards oligarchy over time? And aren’t examples of this widespread? I notice that I am confused.
However, in a vertical system of advancement, where management hires are made from within, the middle ranks will at least have a memory of working on the object-level concerns of the firm.
I think this is part of why the startup nature of tech firms is important, even if all of the important firms are no longer in childhood; it means some fraction of the executive class are people who were around from the beginning, or who built the systems themselves, and so do have the relevant memory.
That said, my sense is that FAANGM are somewhere between 50% and 80% as mazy as corporations as a whole; enough less to be noticeable, but it’s still probably the best lens through which to view the situation.
I think this is part of why the startup nature of tech firms is important, even if all of the important firms are no longer in childhood; it means some fraction of the executive class are people who were around from the beginning, or who built the systems themselves, and so do have the relevant memory.
Which startups get big, while keeping their founders in charge, isn’t random. The founders that do that will tend to be the ones who can attract investment and partnerships and have a network of (personal or professional or familial or whatever) connections to draw on. They’re the ones who already know to (and know how to) play maze games.
Startups employ a trivial percentage of the workforce and do not contribute much to the economy. Startups that get big occupy more economic space, but by that point they’re no longer startups. So the attributes of startup culture are not really relevant to the economy at large.
It’s understood by everyone that being a startup is a kind of corporate childhood—romanticised, but temporary. It is implicitly accepted by everyone, including startup founders, that growing up and becoming a mature, “adult” company requires becoming a maze with multiple levels of hierarchy.
Tech culture is anti-maze only insofar as it consists of startups. All of FAANGM are fully converted to mazes. Consequently, most actual tech workers work within mazes.
Try reading The Refragmentation and see if you still think the same, if you haven’t yet.
One choice quote:
But change was coming soon. And when the Duplo economy started to disintegrate, it disintegrated in several different ways at once. Vertically integrated companies literally dis-integrated because it was more efficient to. Incumbents faced new competitors as (a) markets went global and (b) technical innovation started to trump economies of scale, turning size from an asset into a liability. Smaller companies were increasingly able to survive as formerly narrow channels to consumers broadened. Markets themselves started to change faster, as whole new categories of products appeared. And last but not least, the federal government, which had previously smiled upon J. P. Morgan’s world as the natural state of things, began to realize it wasn’t the last word after all.
What J. P. Morgan was to the horizontal axis, Henry Ford was to the vertical. He wanted to do everything himself. The giant plant he built at River Rouge between 1917 and 1928 literally took in iron ore at one end and sent cars out the other. 100,000 people worked there. At the time it seemed the future. But that is not how car companies operate today. Now much of the design and manufacturing happens in a long supply chain, whose products the car companies ultimately assemble and sell. The reason car companies operate this way is that it works better. Each company in the supply chain focuses on what they know best. And they each have to do it well or they can be swapped out for another supplier.
In other words, it’s not just about isolated startups which employ a tiny percentage. According to Paul Graham, across the board, things broke up. Horizontal and vertical integration became less common as the market increased its churn rate. This means smaller organizations doing fewer things well, which is one piece of advice Zvi gives for combating mazes.
(For those not familiar, “vertical integration” means taking care of more parts of your supply chain, EG if a steel company started directly doing some of the mining which produced its raw materials; and “horizontal integration” means branching out into other businesses (perhaps with some supply chain overlap, but not so directly related).)
Now, obviously, the new companies (FAANGM as you put it) have umbrellad outwards horizontally and in some cases vertically. But these are very new organizations. So unless the rate of mazification has itself increased, they should be at lower maze levels than these older organizations that were around in the 1950s. The rate of mazification may have increased, I’m not saying it hasn’t, but that still requires some other driving factor.
The move away from “company men” was not a move away from large firms, but rather a move away from a vertical system to a stratified one. In the old system (prior to 1970) you could expect to work your entire life for the same company, and middle and upper management was typically promoted from the rank-and-file. In the newer system, middle and upper management are hired from people with MBAs and other credentials, and they move freely between industries. As a consequence, the maze-nature is transmitted quickly from company to company, and to a certain extent all management everywhere is joined in a super-maze, as all management shares the same culture and experiences which is completely separate from the culture and experiences of the workers.
Actually, this last point deserves more elaboration: according to Zvi, the main thing that mitigates against mazes is direct engagement with the object-level reality. This engagement is present in the rank-and-file, and to a certain extent at the very top, but is absent in the middle. However, in a vertical system of advancement, where management hires are made from within, the middle ranks will at least have a memory of working on the object-level concerns of the firm. The rise of a permanent managerial class means that many middle managers are of a type which has never worked directly on the actual product that their company makes, and whose entire education and experience is in the context of immoral mazes.
This point seems very significant, and pushes more “this is what’s really going on” type buttons in my brain.
“Company men” have more cooperative incentives. People who switch careers many times have more adversarial incentives. At the risk of muddying our spatial metaphors, this is vertical vs horizontal transmission all over again.
I still have a picture where some key maze factors have reduced over time, but others (including this creation of a separate middle management career track) have increased, plausibly resulting in an overall increase.
An additional, unrelated note: the model of The Dictator’s Handbook suggests that incentives push away from the middle, towards total democracy (when there are already a large number of key supporters) or total autocracy (where the number of key supporters approaches one). But don’t other models suggest that the middle state of oligarchy is actually the default, and that both democracy and monarchy tend to decay towards oligarchy over time? And aren’t examples of this widespread? I notice that I am confused.
Frankly, the argument that there are two attractive states with instability in the middle is a bit fuzzy, and I wouldn’t be surprised if it somewhat breaks down with better models.
However, to defend it a little: iirc the book does give historical examples of governments in inbetween states being very unstable. And it sort of makes sense. The larger the number of key supporters, the more competition you have as leader. Democracies are relatively stable because they’ve settled on a stable way to overturn leaders. Dictatorships are relatively stable because they don’t overturn leaders. But somewhere in the middle, you expect leadership to be overturned moderately often, as there is steep competition for the top (just not as steep as in a full blown democracy). Unless the government sets up a regular way for that to happen, you’d expect governmental instability, meaning things can easily transition into a more full-blown democracy or a more full-blown dictatorship.
I might be wrong here, but I have a stereotype of long-lived dictators but short-lived royals. Monarchies of the late middle ages were often somewhere in the middle, with unelected rulers but large courts. (But I’m no historian.)
I had read The Refragmentation before, but I reread it shortly after reading your article to make sure I hadn’t missed something. I definitely think that Graham is onto something, but I’m just not sure that it actually cashes out into a lower maze level overall. In particular, deregulation seems to have reduced the size of moats around incumbents, but doesn’t seem to have resulted in an overall reduction in firm size; instead, what happened is that incumbents were merged or reorganized, and in some cases upstarts replaced them and grew bigger. But this does not necessarily mean that the replacements were less maze-like. Microsoft is now bigger than IBM, but does it actually have less middle-management maze behavior?
I seem to recall stats to the effect that the largest N corporations employ a steadily-increasing portion of the population and the economy, which would support this analysis. Unfortunately, I can’t find a data set online that shows this, though I did come up with https://www.bls.gov/web/cewbd/table_g.txt, which has some interesting data in it. (For example, I didn’t realize that the number of public sector firms was actually that small.)
I also noticed this, tucked away in a footnote inThe Refragmentation:
More precisely, there was a bimodal economy consisting, in Galbraith’s words, of “the world of the technically dynamic, massively capitalized and highly organized corporations on the one hand and the hundreds of thousands of small and traditional proprietors on the other.” Money, prestige, and power were concentrated in the former, and there was near zero crossover.
The tiny mom-and-pop store is now much rarer than it used to be. Decades ago, your groceries, home goods, clothes, and gas might all have been bought from retailers that had less than 5 employees, even if they were manufactured by much larger corps. These days you probably get all of them from large organizations.
I seem to recall stats to the effect that the largest N corporations employ a steadily-increasing portion of the population and the economy, which would support this analysis. Unfortunately, I can’t find a data set online that shows this, though I did come up with https://www.bls.gov/web/cewbd/table_g.txt, which has some interesting data in it. (For example, I didn’t realize that the number of public sector firms was actually that small.)
Ahhh yes, this would negate most of my confusion.
The tiny mom-and-pop store is now much rarer than it used to be.
Root comment is excellent. Makes some but not all of the points I would make in response. I think it would mostly be more enlightening for me not to say much more on the subject here and let the conversation unfold, so that’s the plan.
I’m currently left with the feeling that I raised a few points plausibly against mazes, and comments pointed out how those points could plausibly be pro-maze instead. I don’t have enough gears to evaluate which we should really expect. So my current models fail to be predictive.
Factor 1: more people moving between companies / between different careers.
On the one hand, having more autonomy means you can opt out of toxic environments. To my eye, this severely weakens your story about super-perfect competition.
On the other hand, moving between companies is an additional way to cover your tracks, promoting shorter memories and less skin in the game. Taking a biological analogy, this is horizontal transfer, which promotes parasitism.
The answer might be “both”, but this leaves the question of which effect is larger.
It also leaves the question of whether super-perfect competition is really important to moral mazes and whether it’s really prevalent in today’s middle management (despite a lot of apparent freedom to move).
Factor 2: more fragmentation of big companies into more highly specialized companies.
On the one hand, you named fragmentation as a way to help reduce maze levels.
On the other hand, maze levels seem to be increasing with fragmentation.
Fragmentation in the modern context seems to increase maze level, as more fragmentation somehow means more managers. You get companies splitting into specialized entities, but continuing to have deep management hierarchies below the top.
My best explanation for this at the moment is simply increasing background expectation that you need a deep management hierarchy to run things.
Startups employ a trivial percentage of the workforce and do not contribute much to the economy. Startups that get big occupy more economic space, but by that point they’re no longer startups. So the attributes of startup culture are not really relevant to the economy at large.
I wonder what fraction of GDP in 2060 will be accounted for by the work that startups are doing today. US GDP in 1975 was about 9% of what it was in 2015. Do we think that it was the biggest companies that existed in 1975 that were primarily responsible for driving this growth? Or was it startups that grew over time?
I’m willing to accept that the large companies that exist today are primarily responsible for generating the lion’s share of GDP and employment today. But it seems plausible that startups are more important to the future economy, and thus to absolute long-term levels of production, than large companies that exist today.
This is just pure speculation; I’d be curious to hear what a real economist had to say about this.
I looked up the founding dates of the 30 companies that currently make up the Dow. 19 of them (making up 56% of the index) were founded before 1975, and the other 11 (44%) afterwards.
The DOW itself was only 4.5% of it’s 2015 (nominal) value in 1975
About 10⁄30 of the companies that were in the DOW in 1976 are still major economic drivers today
So, overall, I’d guess 50% of GDP in 2060 will be from companies that exist on January 1, 2022 (using 2020 as a baseline year is all sorts of fraught)
An additional, unrelated note: the model of The Dictator’s Handbook suggests that incentives push away from the middle, towards total democracy (when there are already a large number of key supporters) or total autocracy (where the number of key supporters approaches one). But don’t other models suggest that the middle state of oligarchy is actually the default, and that both democracy and monarchy tend to decay towards oligarchy over time? And aren’t examples of this widespread? I notice that I am confused.
I haven’t read The Dictator’s Handbook or know what models there are already, but an autocrat could choose to convert to oligarchy to ensure a stable succession plan (assuming they felt no other autocrat could successfully wield power other than them), and a democracy could become an oligarchy if no individual could seize enough power to directly become an autocrat, but a group working together could. Under the right circumstances these incentives could overwhelm the ones going in the opposite direction.
The common strategy for autocrats is to name successors, typically their own children.
If the dictator’s handbook is right about the forces, oligarchy isn’t a “stable succession plan” (because it’s less stable), so this wouldn’t make oligarchy an equilibrium, it would just make nothing an equilibrium. (IE these forces don’t balance out, they just keep transitioning other states into the least stable state, which would result in continuing transitions back and forth.)
Hello, LessWrong! Long time reader, first-time commenter.
I think that your description of the counter-maze tendency is wrong, and you’ve misunderstood some aspects of Zvi’s model while being distracted by superficialities in the other. To wit:
Startups employ a trivial percentage of the workforce and do not contribute much to the economy. Startups that get big occupy more economic space, but by that point they’re no longer startups. So the attributes of startup culture are not really relevant to the economy at large.
It’s understood by everyone that being a startup is a kind of corporate childhood—romanticised, but temporary. It is implicitly accepted by everyone, including startup founders, that growing up and becoming a mature, “adult” company requires becoming a maze with multiple levels of hierarchy.
Tech culture is anti-maze only insofar as it consists of startups. All of FAANGM are fully converted to mazes. Consequently, most actual tech workers work within mazes.
The change away from suits etc. is a change of fashion with no impact on the underlying dynamics. Arguably it actually helps out the sociopaths because it replaces a fixed, legible standard of dress with an unclear and illegible question of “culture fit”, which creates more room for maze games.
The move away from “company men” was not a move away from large firms, but rather a move away from a vertical system to a stratified one. In the old system (prior to 1970) you could expect to work your entire life for the same company, and middle and upper management was typically promoted from the rank-and-file. In the newer system, middle and upper management are hired from people with MBAs and other credentials, and they move freely between industries. As a consequence, the maze-nature is transmitted quickly from company to company, and to a certain extent all management everywhere is joined in a super-maze, as all management shares the same culture and experiences which is completely separate from the culture and experiences of the workers.
Actually, this last point deserves more elaboration: according to Zvi, the main thing that mitigates against mazes is direct engagement with the object-level reality. This engagement is present in the rank-and-file, and to a certain extent at the very top, but is absent in the middle. However, in a vertical system of advancement, where management hires are made from within, the middle ranks will at least have a memory of working on the object-level concerns of the firm. The rise of a permanent managerial class means that many middle managers are of a type which has never worked directly on the actual product that their company makes, and whose entire education and experience is in the context of immoral mazes.
So I find it unpersuasive to think that any of the cultural changes of the previous decades have done anything to reverse the advance of mazes as the normative corporate structure, and some of the things that you mention as inhibiting maze structures (such as frequently changing companies over the course of a career) have probably actually accelerated them.
An additional, unrelated note: the model of The Dictator’s Handbook suggests that incentives push away from the middle, towards total democracy (when there are already a large number of key supporters) or total autocracy (where the number of key supporters approaches one). But don’t other models suggest that the middle state of oligarchy is actually the default, and that both democracy and monarchy tend to decay towards oligarchy over time? And aren’t examples of this widespread? I notice that I am confused.
I think this is part of why the startup nature of tech firms is important, even if all of the important firms are no longer in childhood; it means some fraction of the executive class are people who were around from the beginning, or who built the systems themselves, and so do have the relevant memory.
That said, my sense is that FAANGM are somewhere between 50% and 80% as mazy as corporations as a whole; enough less to be noticeable, but it’s still probably the best lens through which to view the situation.
Which startups get big, while keeping their founders in charge, isn’t random. The founders that do that will tend to be the ones who can attract investment and partnerships and have a network of (personal or professional or familial or whatever) connections to draw on. They’re the ones who already know to (and know how to) play maze games.
Welcome! :)
Try reading The Refragmentation and see if you still think the same, if you haven’t yet.
One choice quote:
In other words, it’s not just about isolated startups which employ a tiny percentage. According to Paul Graham, across the board, things broke up. Horizontal and vertical integration became less common as the market increased its churn rate. This means smaller organizations doing fewer things well, which is one piece of advice Zvi gives for combating mazes.
(For those not familiar, “vertical integration” means taking care of more parts of your supply chain, EG if a steel company started directly doing some of the mining which produced its raw materials; and “horizontal integration” means branching out into other businesses (perhaps with some supply chain overlap, but not so directly related).)
Now, obviously, the new companies (FAANGM as you put it) have umbrellad outwards horizontally and in some cases vertically. But these are very new organizations. So unless the rate of mazification has itself increased, they should be at lower maze levels than these older organizations that were around in the 1950s. The rate of mazification may have increased, I’m not saying it hasn’t, but that still requires some other driving factor.
This point seems very significant, and pushes more “this is what’s really going on” type buttons in my brain.
“Company men” have more cooperative incentives. People who switch careers many times have more adversarial incentives. At the risk of muddying our spatial metaphors, this is vertical vs horizontal transmission all over again.
It also smacks of the rise of meritocracy. See this book and this podcast episode.
I still have a picture where some key maze factors have reduced over time, but others (including this creation of a separate middle management career track) have increased, plausibly resulting in an overall increase.
Frankly, the argument that there are two attractive states with instability in the middle is a bit fuzzy, and I wouldn’t be surprised if it somewhat breaks down with better models.
However, to defend it a little: iirc the book does give historical examples of governments in inbetween states being very unstable. And it sort of makes sense. The larger the number of key supporters, the more competition you have as leader. Democracies are relatively stable because they’ve settled on a stable way to overturn leaders. Dictatorships are relatively stable because they don’t overturn leaders. But somewhere in the middle, you expect leadership to be overturned moderately often, as there is steep competition for the top (just not as steep as in a full blown democracy). Unless the government sets up a regular way for that to happen, you’d expect governmental instability, meaning things can easily transition into a more full-blown democracy or a more full-blown dictatorship.
I might be wrong here, but I have a stereotype of long-lived dictators but short-lived royals. Monarchies of the late middle ages were often somewhere in the middle, with unelected rulers but large courts. (But I’m no historian.)
Thanks in turn :).
I had read The Refragmentation before, but I reread it shortly after reading your article to make sure I hadn’t missed something. I definitely think that Graham is onto something, but I’m just not sure that it actually cashes out into a lower maze level overall. In particular, deregulation seems to have reduced the size of moats around incumbents, but doesn’t seem to have resulted in an overall reduction in firm size; instead, what happened is that incumbents were merged or reorganized, and in some cases upstarts replaced them and grew bigger. But this does not necessarily mean that the replacements were less maze-like. Microsoft is now bigger than IBM, but does it actually have less middle-management maze behavior?
I seem to recall stats to the effect that the largest N corporations employ a steadily-increasing portion of the population and the economy, which would support this analysis. Unfortunately, I can’t find a data set online that shows this, though I did come up with https://www.bls.gov/web/cewbd/table_g.txt, which has some interesting data in it. (For example, I didn’t realize that the number of public sector firms was actually that small.)
I also noticed this, tucked away in a footnote inThe Refragmentation:
The tiny mom-and-pop store is now much rarer than it used to be. Decades ago, your groceries, home goods, clothes, and gas might all have been bought from retailers that had less than 5 employees, even if they were manufactured by much larger corps. These days you probably get all of them from large organizations.
Ahhh yes, this would negate most of my confusion.
Yep, a very good point.
Root comment is excellent. Makes some but not all of the points I would make in response. I think it would mostly be more enlightening for me not to say much more on the subject here and let the conversation unfold, so that’s the plan.
I’m currently left with the feeling that I raised a few points plausibly against mazes, and comments pointed out how those points could plausibly be pro-maze instead. I don’t have enough gears to evaluate which we should really expect. So my current models fail to be predictive.
Factor 1: more people moving between companies / between different careers.
On the one hand, having more autonomy means you can opt out of toxic environments. To my eye, this severely weakens your story about super-perfect competition.
On the other hand, moving between companies is an additional way to cover your tracks, promoting shorter memories and less skin in the game. Taking a biological analogy, this is horizontal transfer, which promotes parasitism.
The answer might be “both”, but this leaves the question of which effect is larger.
It also leaves the question of whether super-perfect competition is really important to moral mazes and whether it’s really prevalent in today’s middle management (despite a lot of apparent freedom to move).
Factor 2: more fragmentation of big companies into more highly specialized companies.
On the one hand, you named fragmentation as a way to help reduce maze levels.
On the other hand, maze levels seem to be increasing with fragmentation.
Fragmentation in the modern context seems to increase maze level, as more fragmentation somehow means more managers. You get companies splitting into specialized entities, but continuing to have deep management hierarchies below the top.
My best explanation for this at the moment is simply increasing background expectation that you need a deep management hierarchy to run things.
Congrats on a great first comment! :)
+1
I wonder what fraction of GDP in 2060 will be accounted for by the work that startups are doing today. US GDP in 1975 was about 9% of what it was in 2015. Do we think that it was the biggest companies that existed in 1975 that were primarily responsible for driving this growth? Or was it startups that grew over time?
I’m willing to accept that the large companies that exist today are primarily responsible for generating the lion’s share of GDP and employment today. But it seems plausible that startups are more important to the future economy, and thus to absolute long-term levels of production, than large companies that exist today.
This is just pure speculation; I’d be curious to hear what a real economist had to say about this.
I looked up the founding dates of the 30 companies that currently make up the Dow. 19 of them (making up 56% of the index) were founded before 1975, and the other 11 (44%) afterwards.
The DOW itself was only 4.5% of it’s 2015 (nominal) value in 1975
About 10⁄30 of the companies that were in the DOW in 1976 are still major economic drivers today
So, overall, I’d guess 50% of GDP in 2060 will be from companies that exist on January 1, 2022 (using 2020 as a baseline year is all sorts of fraught)
I haven’t read The Dictator’s Handbook or know what models there are already, but an autocrat could choose to convert to oligarchy to ensure a stable succession plan (assuming they felt no other autocrat could successfully wield power other than them), and a democracy could become an oligarchy if no individual could seize enough power to directly become an autocrat, but a group working together could. Under the right circumstances these incentives could overwhelm the ones going in the opposite direction.
The common strategy for autocrats is to name successors, typically their own children.
If the dictator’s handbook is right about the forces, oligarchy isn’t a “stable succession plan” (because it’s less stable), so this wouldn’t make oligarchy an equilibrium, it would just make nothing an equilibrium. (IE these forces don’t balance out, they just keep transitioning other states into the least stable state, which would result in continuing transitions back and forth.)