divine curation

Invisible Hand Fail Modes

January 29, 2021

I’ve been browsing through the archives of the newly reanimated Slate Star Codex, and came across a post reviewing Eliezer Yudkowsky’s book Inadequate Equilibria. I have not read it myself, but Scott Alexander’s summary relates a text which (among other things) examines the ways in which free markets can lead to undesirable outcomes. What is interesting about this for me is that it is written by someone who is broadly pro-capitalist and believes in the general benevolence of the Invisible Hand. As Alexander explains, the book is fundamentally a work of theodicy—an attempt to understand how evil can arise in a world ruled by a loving God.

Yudkowsky identifies three Invisible Hand fail modes, scenarios in which we can expect the market to fail to incentivise a solution to a real problem:

  1. Cases where the decision lies in the hands of people who would gain little personally, or lose out personally, if they did what was necessary to help someone else.
  2. Cases where decision-makers can’t reliably learn the information they need to make decisions, even though someone else has that information.
  3. Systems that are broken in multiple places so that no one actor can make them better, even though, in principle, some magically coordinated action could move to a new stable state.

An example of the first is when prestige hungry central bankers do the respectable but useless thing instead of the thing which might actually fix the economy, but is sure to raise eyebrows.

An example of the second is when you’re trying really hard to make an informed decision about your child’s nutrition given their rare allergies, and you know that somewhere out there is a team of crack scientists shouting the right answer, but whenever you google it all you find is hysterical blog posts and pay-walled journals and Alex Jones and you’d have to be literally superhuman to make sense of this mess.

Shitty Nash Equilibria

The third is the one I want to talk about, mainly because it links up with some patterns in Marxist thinking.

As Alexander writes, these situations represent bad Nash equilibria:

Everyone hates Facebook. It records all your private data, it screws with the order of your timeline, it works to be as addictive and time-wasting as possible. So why don’t we just stop using Facebook? More to the point, why doesn’t some entrepreneur create a much better social network which doesn’t do any of those things, and then we all switch to her site, and she becomes really rich, and we’re all happy?

The obvious answer: all our friends are on Facebook. We want to be where our friends are. None of us expect our friends to leave, so we all stay. Even if every single one of our friends hated Facebook, none of us would have common knowledge that we would all leave at once; it’s hard to organize a mass exodus. Something like an assurance contract might help, but those are pretty hard to organize. And even a few people who genuinely like Facebook and are really loud about it could ruin that for everybody. In the end, we all know we all hate Facebook and we all know we’re all going to keep using it.

Another example he gives is Uber:

Or: instead of one undifferentiated mass of people, you have two masses of people, each working off the other’s decision. Suppose there was no such thing as Lyft – it was Uber or take the bus. And suppose we got tired of this and wanted to invent Lyft. Could we do it at this late stage? Maybe not. The best part of Uber for passengers is that there’s almost always a driver within a few minutes of you. And the best part of Uber for drivers is that there’s almost always a passenger within a few minute of you. So you, the entrepreneur trying to start Lyft in AD 2017, hire twenty drivers. That means maybe passengers will get a driver…within an hour…if they’re lucky? So no passenger will ever switch to Lyft, and that means your twenty drivers will get bored and give up.

If you were in the business of defending the Invisible Hand, then you would need to argue that these situations are either edge cases or involve some kind of distortion of smooth market functioning. If so, you could maintain that avoiding them is just a matter of becoming better at capitalism.

Against this, a general anti-capitalist point could be made by arguing that these scenarios are predictable, ubiquitous, or in some way necessitated by the internal logic of capitalism. Come to think of it, I don’t think it would be too much of a stretch to gloss Marx’s central claim as “capitalism literally optimises for bad Nash equilibria you bunch of clowns this is a fucking terrible idea.” (See, for example, Ian Wright’s essay Prolegomena to a demonology of capitalism. Wright does not talk about Nash equilibria, he talks about negative feedback control loops—but Nash equilibria provide a way of thinking about how negative feedback loops can be implemented on existing social software.)

Anselm Jappe has distinguished between what he calls the “exoteric” and the “esoteric” threads in Marx. The rest of this post will consider exoteric and exoteric arguments based on bad Nash equilibria. (And as disclaimer, I’ll be making zero effort to be faithful to Marx’s actual arguments here, just to capture the spirit of Jappe’s distinction as it pertains to the rationalist market theodicy.)

An Exoteric Argument

The exoteric argument begins with the observation that Nash equilibria, like those in the examples, can themselves be exploited for profits. Once most people are on Facebook the execs will know that their user base is self-maintaining, and that they no longer have a financial incentive to provide a platform that anyone actually likes. In fact, since they can almost certainly profit by producing a worse user experience (more ads, covertly appropriating your bits, implementing sneaky attention capturing mechanisms, etc.) without losing users, they have good financial reasons to do exactly that. Even if Facebook was founded with the best of intentions, once it achieves a certain critical mass an incentive forms that works against those intentions.

Uber have pretty much always been straightforwardly evil and proud of it, and represent a more aggressive version of the same pattern. If you know that a Nash equilibrium can be exploited, then as long as you are backed by enough capital you can make it part of your business plan to create one. Uber are good examples of this, deliberately operating at a loss to undercut local competition and mop up market share. Once a new equilibrium has been produced, they are free to exploit it as they like.

So the argument is that these bad Nash equilibria are not edge case fail modes, but scenarios the Invisible Hand will actively select for. We should expect them to be common, not rare.

There’s a couple of responses that can be made to this argument. The first is that it is sensitive to non-economic factors, like Reason and Virtue. Mark Zuckerberg’s sincere goal might well be to act in accordance with Virtue, but if he believes in the naïve, pre-Yudkowskian benevolence of the Invisible Hand then Reason will be telling him that the best way to satisfy Virtue is to obey the profit principle. Keep laying on those ads, surely they’re optimising distribution of something or other to someone or other. But if he’s read Inadequate Equilibria he will know that he’s faced with a fail mode, and Reason will be warning him that profit is not a good guide to Virtue in this case. In short, this problem can be solved in the case of a Virtuous producer by improving their Reason.

What about in the Uber scenario, where Reason is acting purely on the profit principle and recognises no further Virtue? In this situation the solution lies in smarter consumers—by understanding the Invisible Hand’s fail modes better, they can see Uber’s strategy for what it is from the beginning and choose not to use it. By cultivating their Reason, consumers will better be able to understand that short-term benefits in price and convenience can sometimes be traps leading to long-term misalignments of Virtue. To make this response is to develop the market theodicy in what we could (somewhat liberally) describe as a “Molinist” direction, ultimately locating blame for the presence of evil in the world in failures of human free will.

A second response is that this argument only applies in situations where our wanting something is dependent on what everyone else is doing or wanting. This might be the case for social networks, but it probably is not the case for other things, like hammers and nails. The extent to which this argument targets markets in general is then a question of how many of our wants fall under similar dependency patterns. If you’re buying some clothes as a social marker, then the value you place on them presumably depends on your perception of other people’s perceptions of these clothes. How much do consumption patterns in general reflect our independent wants as individuals, and how much do they reflect our dependent, socially mediated wants? If most of what we label “the economy” is about satisfying independent wants then the exoteric argument applies only to its edge cases, and fails in the general case.

An Esoteric Argument

Now, what Jappe calls the “esoteric” in Marx is all the stuff in the notoriously baffling first bit of Capital: the critique of value and its exposé of the janus-faced nature of commodities (Marx, 1976). I’m not going to get into all that. Rather, what I want to do here is think a bit about the form of this line of analysis, contrast it with the exoteric line, then consider how it responds to a Molinist market theodicy.

According to Jappe, the key difference is that esoteric arguments question the foundational categories of capitalist society, whereas exoteric arguments are concerned only with distribution. The exoteric line of thought above takes the Invisible Hand at face value, then argues that by its own lights it is more or less guaranteed to produce loads of shitty Nash equilibria. These arguments are tempting because they play by the rules, which is to say they accept the dominant categories used to frame the question of economic distribution. But a market theodicy can easily absorb these arguments by appealing to those very categories. If the Invisible Hand has fail modes, then it is still the best it could do given its inputs: individual consumers and producers with their desires, reasons, and knowledge. If something has gone wrong at the level of economic distribution, then this just implies a non-economic solution, like more Reason and/or Virtue.

The exoteric stance can’t really reply to this, because it restricts itself to talking about distribution, implicitly accepting the categories that frame it. This is where esoteric Marx swoops in to remind us all that those categories—you know, “consumer,” “producer,” “individual desire,” etc—are actually all just superstructural effects of a value-form that establishes relations of production on the basis of abstract labour power.

Wait, what? What the fuck does that mean? See—there’s a problem that arises at this point, and it’s one that the rationalsphere is particularly sensitive to. Rationality enthusiasts like Yudkowsky and Alexander place a premium on something called Clarity. This is fair enough. But the thing about Clarity is that it is achievable only against a background of stable descriptive categories. Making one’s point Clear is about making it in a way that conforms to familiar descriptive norms and uses familiar conceptual schemes and vocabularies. There is a lot to be said for Clarity, and Scott Alexander in particular is extremely good at it.

But elevating it to a high Virtue is sure to create blindspots, particularly when those familiar background categories are exactly the thing you want to call into question. Somewhere on Twitter Pete Wolfendale described dialectics as what happens when a line of enquiry leads to a place which forces it to modify the concepts it used to frame the question in the first place. If Clarity requires stable background categories, then dialectical points which deliberately probe and destabilise those categories can’t fail to be Unclear. In a space whose discursive norms demand Clarity these points are rendered systematically disreputable, even when they are good ones.

Esoteric Marx wants to make a dialectical point. He wants to say that to even conceive the economic problem as a matter of optimally satisfying individual needs—to adopt this conceptual scheme as part of our descriptive vocabulary—is to have already deprived ourselves of the resources required to avoid those shitty Nash equilibria. Why so? To conceive things this way is to separate the economic sphere from the domain of subjective value. First there are people with their needs and preferences, then economics asks how to organise distribution to reach maximal compromise between those preferences. The important thing about this framing is that since it treats individual preference-bearers as ontologically prior to economic factors, those factors cannot be envisaged as acting on those preferences.

How might we disagree with this? Let’s return to the difference between preferences which are socially mediated and those which are not. While everyone can grant that sometimes we want things because our friends want them, this is rarely taken as evidence that economics is misconceived in its fundamentals. But consider the case of buying clothes as a status marker. We might argue that desire is operating at two levels here:

  1. I want social status. (unmediated)
  2. I want these clothes because they signal status. (mediated)

The desire for status is the primary desire, while the pursuit of clothes is instrumental to it. This is an unmediated primary desire—wanting social status is not, in itself, dependent on other people’s wants. These only come into play at the instrumental stage, when the question becomes how to satisfy the primary desire within a particular social context.

In this situation the instrumental mediated desire bottoms out in a primary unmediated desire. On another day we might just want some clothes, without this desire being sensitive to what others want or have, in which case the desire for clothes is an unmediated primary desire. If we were to reckon that desires can always be understood under this pattern, this would give:

What does UPD have to say about the Facebook equilibrium? Well, it says that our socially mediated desire for Facebook (as registered by the fact that we continue to use it) ultimately answers to an unmediated desire—something like a desire for a Good Social Network, which is currently unsatisfied. Practically speaking this makes no difference to anything. We’re still stuck in the shitty Nash equilibrium, and escaping it is still a coordination problem. But it does lend credence to the idea that the way to avoid (if not solve) these kind of problems is to level up our faculties of Reason. If back in the early 2000’s we’d better appreciated that getting caught in Facebook’s net would ultimately work against our desire for a Good Social Network, we could have taken steps to avoid it. If in the future we find ourselves faced with a similar situation, we’ll have a better understanding of what not to do. In short, UPD complements and supports the Molinist market theodicy.

A problem with this view is that it seems to depend on our having information that we could not possibly have had access to at the time of decision making. For example, it presumes that the desire motivating the original uptake of Facebook was the same desire for a Good Social Network that we have now that we know Facebook is rubbish. Yet the content of this desire is something that can only have been shaped through the experience of using Facebook, since prior to this experience we didn’t even have much sense of what a social network was. To ascribe this desire to those earlier versions of ourselves is to retroject, and to suggest that we could have reasoned on its basis would seem to require superhuman powers of foresight. Why not just say that the original desire to use Facebook was just the desire to use Facebook, full stop?

OK—so we got the expensive dining table but are now worried about getting marks on it. We bought a tablecloth to protect it, but it kept slipping so we had to get an anti-slip mat too. It’s also, ugh, white so of course we had to get coasters, table mats, multiple doilys, and a new cupboard to put it all in. This ended up looking kind of bare so we bought some ornamental teacups to stack on its antique shelves. Why the hell did we buy a table so expensive that we can’t even put things on it without wigging out?

  1. We just really liked it.
  2. People like us should have an expensive dining table.

In essence, MPD is claiming that we can’t necessarily use introspection to distinguish between these two motivations. But they do have different behavioural implications. If you really do just have an unmediated desire for the table, then this desire will be indifferent to social context. But if your taste in tables mysteriously changes when you move to a new area or get a new job, then this would suggest that it is mediated by social perceptions. The claim of MPD is that this does not imply that there is some other, more primary desire lurking behind the desire for the table, like a desire for status or belonging or whatever. You might just have a socially mediated desire for the table. You might not be aware of the mediation, but this does not mean you are mistaken about the content of the desire.

The difference between UPD and MPD becomes clear when we start thinking about scenarios in which personal choices interact with highly social environments. Take Burning Man outfits, for instance. When questioned, a Burner taken at random might say that they have chosen their outfit because it expresses their deep, authentic individuality. But zoom out until you can see everyone as a group and it becomes obvious that there is a formula at work. It looks like what everyone has actually done is generate their own variation on a dominant code. What’s going on?


What does all this get us? Socially mediated desires are important because they are vulnerable to capture by bad Nash equilibria—Yudkowsky’s third fail mode. If many of our wants are socially mediated, then we should expect bad Nash equilibria to be common by the Invisible Hand’s own lights (because there is profit in them).

A Molinist market theodicy can reply to this by invoking UPD. It’s easy for them to do this, because it is implicit in the standard framing of economics as a matter of optimally distributing resources among an ontologically prior population of preference-bearing individuals. Since primary desire comes from within, it is unaffected by external factors that might be sensitive to details of economic distribution (though instrumental desires could well be and often are). So even if the Invisible Hand does have some fail modes, the solution is to better understand our own primary desires and get better at reasoning off them.

Exoteric Marx is unable to response to this, since he is implicitly signed up to UPD by virtue of accepting the standard economic framing. But esoteric Marx can respond by invoking MPD. If even primary desire is mediated by social perception, then it too can in principle be influenced by factors external to individuals—factors like communications media, advertising, PR, and so on. In these cases economic factors can act on and shape the very preferences they are supposed to be satisfying. No amount of “self-understanding” will help to avoid the fail modes, because there is no deeper layer to understand. Better reasoning won’t help either, because practical reasoning is only as good as the goals it aims at.

Of course the real question is to what extent our primary desires are socially mediated, if at all. My desire for a glass of water after a run is probably not particularly sensitive to whether water-drinking is in fashion this month. But if we do accept MPD in areas of consumption where social factors are clearly in play, then this implies not only that there are profits to made by engineering and exploiting Nash equilibria, but also that the exercise of greater Reason and Virtue is not going to help us dodge them.

Indeed we might wonder: how many domains of preference beyond those pertaining to basic material needs are not shot through with social significance? If we are the kind of creatures for whom MPD is the rule rather than the exception, then we should expect the engineering and exploitation of bad Nash equilibria not to be edge cases, but something the market systematically incentivises. Neither our reason nor our free will can help us, and what was supposed to be the Invisible Hand of a loving God turns out to look suspiciously like Moloch’s.


  1. Marx, K. (1976). Capital: A Critique of Political Economy (B. Fowkes, Tran.). Penguin.


Archibold Fannywrangler, 23 February 2021

Stonks go brrrrr.

Invisible Hand Fail Modes - January 29, 2021 - Divine Curation