Stripping Away the Protections

post by Zvi · 2020-01-08T13:10:00.912Z · LW · GW · 6 comments

Contents

6 comments

Previously in sequence: Moloch Hasn’t Won [LW · GW], Perfect Competition [LW · GW], Imperfect Competition [LW · GW], Does Big Business Hate Your Family? [LW · GW], What is Life in an Immoral Maze? [? · GW]

The previous post painted a bleak picture of life as a middle manager in an Immoral Maze. Not every middle manager faces the high maze levels described in Moral Mazes, but I am confident that many do.

How did things get so bad?

We previously discussed Meditations on Moloch. That has a lot of gear-level analysis of how easy it is for undifferentiated high stakes competition along a single variable to destroy everything of value, and how much worse it is to do all those things at once than would be the sum of their parts.

Perfect Competition summarized and fleshed out that model so we could work with it, and defined Super-Perfect Competition as perfect competition without free exit, which when that restriction is meaningful results in less than zero economic profits.

Imperfect Competition illustrated how different idealized perfect competition is from the conditions we typically experience. Even when we use markets that seem highly competitive. Even in a market used as a canonical example of perfect competition. 

The difference between the situations discussed there, and the situation in Moral Mazes, is that the conditions of super-perfect competition really do apply. The protections against this happening have been stripped away.

An immoral maze is not as extreme as the imagined possible future em world where any simulated human caught doing something suboptimal is erased and replaced. But it is closer than one would first think. In a maze, people can’t yet be copied, but there are lots of aspirants to choose from, and any opportunity to rule one out is pounced upon.

Sources of positive differentiation between managers are systematically destroyed. As we saw above, managers do not believe they exist. So they don’t matter. This leaves only negative differentiation and selection, plus politics. 

Sources of meaningful object-level concerns and detail, and the relevance of any long term consequences of actions, are stripped away as well. Again, managers do not believe they exist. So they don’t matter.

This leaves a homogenous product that managers must become, the production of which is subject to perfect competition. 

Then the resulting politics runs amok. The system self-reinforces the reinforcement of conformity with these ideals.

Let’s go over the details that keep the rest of the world safe from such conditions. The bold terms are copied from the previous post, Imperfect Competition. 

No boss. Not only is there a boss. There are bosses and underlings as far as the eye can see. Everything is reduced to simple metrics and impressions that can be run up or down the chain. Everyone is judged on everything. Based on how they look and cause others to look, and what it says about the person being judged. There are only better and worse judgments, and ways to get them. Avoiding standing out in a negative way (e.g. ‘a thousand atta boys are wiped out by one oops’) is all that matters. The richness of real situations falls away. 

Skin in the game. Immoral mazes actively fight against anyone having skin in the object-level or long term game. They don’t track or remember anything, deliberately destroying or avoiding records to avoid future scapegoating. That leaves no skin in the real game, only skin in the game of short term appearances and politics, which is what skin in the game is supposed to defend against. Even if the system wanted to preserve skin in the game and distribute it usefully, rather than destroy it, there would be a severe shortage of it. Major corporations are too large and consequences too diffuse. There is not enough potential skin in the game to distribute to dozens of levels of management, even in theory under the best of circumstances, even if the company is fully privately held. Skin in the object level game is restricted to the very top and bottom of the hierarchy, if they use it at all. The CEO and board can own stock or options, and the people directly ‘on the line’ can perhaps be judged on individual performance.

Products and producers are not homogenized, and information about them is costly. This is the big one that is easiest to miss or be confused about. It’s hard not to notice that there’s a lot of bosses and a lack of skin in the game. Noticing that managers are effectively homogenized is hard, because they aren’t actually homogenized at all – rather it is the perception that they are, which causes them to act and be treated as if they were. Throughout the process of reaching middle management, aspiring managers become homogenized products, and learn to view themselves and their competition in this fashion, and that within managerial circles this is common knowledge. Any deviations are career suicide and severely punished. The idea that one could be better doesn’t parse – see skills differ, below. There’s also the double-think that if you were somehow meaningfully better, that (through what seems to them like some instinctive but unknown mechanism) likely makes you a threat and means you need to be taken out, or that if you’re better that means they are worse and being worse is death, so again, take you out. 

Reputation and experience matter a lot. Reputation inside a maze boils down to whether someone has negative marks that doom them, and how much momentum their career has, and what kind of political allies they’ve made. Your commitment and time spent are also tracked, but that rapidly ceases to differentiate between any of the survivors. None of that is the kind of reputation that does the work of keeping competition imperfect. Experience beyond generic management experience is not only not considered relevant, it is a liability. It means that you lack career momentum.

Market information is costly for consumers. Consumers are the bosses and fellow managers, who already understand the market in question all too well. They have no choice but to learn, even if the cost of doing so is high. The protection from expensive information comes from consumers choosing not to purchase all of it. 

Market information is costly for producers. Producers are also the bosses and fellow managers, since no one is close to the object level. So again, they already understand the market in question all too well. No one who matters lacks market information.

Skills differ. Complete skill stacks are rare. It is presumed that once a certain level is reached, everyone has a complete managerial skills stack. Everyone’s skill levels are assumed to be equal, aside from skill in politics. The conventional wisdom claims there is common knowledge that there is a generic ‘ability to run a thing’ skill stack. Everyone who has made it this far has that stack. Skills that aren’t generic or politics are considered to effectively not exist, or to be things you should be delegating regardless. Having them yourself is, as noted above, a liability rather than an asset. So there’s no reason to pick particular managers with a specific set of skills for particular jobs that require those skills. There’s no way to positively differentiate yourself using your skills. 

Unknown unknown information exists and matters. Managers are presumed to all be the same. There aren’t meaningful unknown unknowns. The products for which they coordinate production have unknown unknowns that will eventually matter to the company. They also have aspects that are hard to measure, or hard for someone outside the department (or anyone other than the manager themselves) to notice at all. But that is all a long term problem, the same way we were concerned with long term nutritional effects or the long term reliability of car models. No one cares if a manager was producing something with such long term issues. By the time they come to light, all association between that manager and the product has been long forgotten. Memories don’t last long enough. If one did notice, a middle manager would primarily consider this praiseworthy, because the person successfully got away with something, unless they had an opportunity to somehow use it against them. Management nirvana is described in Moral Mazes as getting promoted and then blaming your successor for your own “mistakes.” Which, of course, are not mistakes from your perspective. The manager is responsible for making sure things don’t go to hell in an observable way while they are still in charge, which would be really bad and likely end one’s career.

Fixed costs exist. Fixed costs matter for proper decision making only when they can be avoided (including the option of avoiding them by avoiding the whole enterprise). Otherwise, the ship has sailed, so they are sunk costs, which don’t count and don’t matter, except insofar as sunk cost fallacy [LW · GW] is thing in context. If they try to pass those costs along, they just dig the hole deeper. Managers have paid huge fixed costs to become managers. At each step they have invested everything they have in their career trajectory. At many steps, they may regret having started down this path, but don’t realize all they are doing is digging a deeper hole for themselves. Or they realize it, but can’t figure out a way to stop. 

Production costs are asymmetric. There are massive asymmetric sunk costs. Sunk costs don’t count. Everyone has already given everything to be able to produce. Everyone is producing the same thing as a result, with the same marginal costs. Anyone not willing to visibly bear those costs is wiped out for that alone. Effectively, everyone has the same production costs aside from political battles. 

Economies of scale exist. Each manager is only one person. While the companies themselves have economies of scale, the competition between employees has no such concept.

Producer preferences differ. Producer preferences (aka employee preferences) in theory still exist, but letting them noticeably impact decisions is fatal. Doing so in secret still means sacrificing position in the game for whatever else you care about. Successful managers learn not to do that.

Location matters. Location is expected not to be a limiting factor. If a manager is asked to go manage a plant or office in another state or even abroad, refusing to do is a sign that they are insufficiently committed. They might try to steer decisions a non-zero amount to avoid the costs involved, since those costs do interfere with one’s ability to do the job, but this mostly isn’t a thing. This prevents differentiation based on location, which was how this made the original list, and it also wrecks the lives of the managers involved since they cannot choose where to live or predict where they will be.  

Consumers care about the individual producers. It is considered a grave sin in middle management culture to care about your underlings, also known as your producers. You protect them while they are loyal and competent and not the scapegoat at this time, because that is part of the job. You need to be seen as the type of person who does that. But actually sacrificing value for them would be seen as terribly weak. As you do not care about them, the people above you do not care about you either. Loyalty until such loyalty is no longer useful is a thing, but that is very much not the same thing nor does it fill the same role.  

Everyone is judged on a combination of politics, and whether they can properly produce the homogenized product ‘successful middle manager.’ This centrally includes associating only with those also producing such products, rewarding such products, and punishing any deviations from such products. You are responsible for everyone and everything below you, so you must demand fealty from those below you, so they’ll serve whatever agenda is most helpful to you. You must also ensure that they are producing the ‘successful middle manager’ product.   

The Opt-Out Clause

The good news for managers is that once one reaches a certain level, as long as one continues to play by the rules and make others comfortable (comfortable is a key concept we don’t have space to discuss here, for more see section B of Quotes from Moral Mazes), one can typically ride things out while carving out some amount of outside value for yourself. One does this by visibly opting out of the game, and no longer being seen as a threat. Everyone knows that some people must stay in place to keep the wheels mostly turning. Without such people the whole system doesn’t collapse too fast and take current management down. Thus, there is an option for graceful retirement from the climb up the ladder.

More generally, those who accept immobility are unwilling to sacrifice family life or free-time activities to put in the extraordinarily long hours at the office required in the upper circles of their corporations. Or they have made a realistic assessment of the age structure, career paths, and power relationships above them and conclude that there is no longer real opportunity for them. They may see that there is an irreparable mismatch between their own personal styles and the kinds of social skills being cultivated in well-entrenched higher circles. In many cases, they decide that they do not wish to put up with the great stress of higher management work that they have witnessed. (Location 962, Quote 119, Moral Mazes)

Producing ‘successful middle manager’ is largely a pass/fail operation. You can’t produce it by halves. Thus, once you realize you are unable or unwilling to produce it properly, either you move somewhere where you can do so, you quit entirely, or you stop trying to produce ‘successful middle manager’ and instead produce ‘contented middle manager.’ Those are the level one damage mitigation strategies. 

Later sections will deal with such damage mitigation options more broadly.  

Is This Missing Something?

On one level, this robustly answers the question “how did things get so bad?”

On another level, this begs that question entirely. It does not explain why everyone involved ended up with such crazy beliefs, or allowed things to get so bad. It merely explains the bind the players are in within the game as it currently exists.

There is the implication that the whole thing is wrapped in conspiracy and malice that is left implicit, its motives and modes of operation unclear. Without that element, the explanation feels incomplete at best.

This gap is because the model only implicitly mentions Moloch’s Army. I still haven’t quite figured out how to take that on, slash we’re not ready yet. Hopefully you now have a better idea of what this represents.

Mazes Everywhere

These dynamics don’t only exist in major corporations. By default, at least right here and right now, they arise in any sufficiently large organization. Barring strong optimization pressure holding them back, things over time in any given organization with multiple levels of management will become increasingly maze-like.

Governments are presumed to be mazes. Armies. Political parties. Unions. Academia. Sufficiently large non-profits are almost certainly mazes – the profit motive is if anything fighting against the property of being a maze. Getting rid of it would only speed things along.

This is doubly true because mazes do not only support themselves being mazes. They tend to support and encourage maze behaviors in other places and organizations, favoring mazes over non-mazes. Our major corporations have largely been mazes for decades.

A maze need not even be a formal organization. You can get similar effective behaviors whenever there is an effective multi-level hierarchy.

Consider Paul Graham and Sam Altman’s description of the ecosystem surrounding Y-Combinator, as analyzed recently by Ben Hoffman. Or the model of the same system in my older post In a world… of venture capital, where each round of funding can be considered the boss of the investors in the previous round.

The Next Questions

There now exist three categories of next questions. 

The first category (taken from this comment [LW(p) · GW(p)] I made on Less Wrong) are the big global questions. To what extent are these dynamics the inevitable result of large organizations? If so, to what extent should we avoid creating large organizations? Has this dynamic ever been different in the past in other places and times, and if so why and can we duplicate those causes?

The second category are the big personal questions. Given the existence of these immoral mazes, what do I do? 

The third category is the one I’ve been struggling with, which is to finally get a good written model of the dynamics of anti-epistemic anti-virtue.

6 comments

Comments sorted by top scores.