The Epistemic Pain of Prop 22

Prop 22 is a California ballot proposition that would exempt app-based drivers from Assembly Bill 5. If it fails, Uber and associates will likely have to reclassify their drivers as employees.

This is the most difficult post I’ve written. Not because it’s deeply personal or technically complex, but because every point has an equally valid counterpoint, and I’m no longer sure what I believe.

I wish this was simple cognitive dissonance. In that case, you’re confused in the moment, but driven to find resolution. Do you love your pets but still eat meat? No problem, just convince yourself that dogs matter more than chickens.

In contrast, epistemic pain only drives you further into nuance. The more you seek to understand, the deeper you fall into the abyss. Every statement is qualified by context, each conclusion bracketed by caveats. Each new insight gives way to more questions, fractally expanding into an unkillable hydra. Instead of a frontier of knowledge, you’re faced with a coastline of infinite perimeter.

That’s epistemic pain. [1]


You might object that Prop 22 is simple. Price floors are socially inefficient and lead to job loss. Uber drivers are overwhelmingly in favor of remaining independent contractors. And besides, why should the government interfere with two consenting adults exchanging capital for labor?

Even if we don’t engage with the object-level arguments, every smart person seems to endorse Prop 22. Scott Alexander is voting yes, Yudkowsky is mocking the opposition, Mike Solana’s Substack (endorsed by Patrick Collison and Ben Thompson) says “A YES vote on 22 will save literally hundreds of thousands of jobs.” Tyler Cowen has consistently (1, 2, 3) expressed support for Uber, including an argument against drivers-as-employees.

So it seems simple enough. Liberals are trying to regulate every industry to death. Prop 22 is our last defense, and a necessary measure to save jobs and bolster the economy. Easy choice right?

Oh you sweet summer child.

1. The Price Floor Question

First of all, price floor efficiencies only occur under a precise set of conditions. The model doesn’t account for failures of collective bargaining, and is frequently disputed by empirical evidence.

SlateStarCodex summarizes the studies for and against minimum wage:

…take the minimum wage question (please). We all know about the Krueger and Card study in New Jersey that found no evidence that high minimum wages hurt the economy. We probably also know the counterclaims that it was completely debunked as despicable dishonest statistical malpractice. Maybe some of us know Card and Krueger wrote a pretty convincing rebuttal of those claims. Or that a bunch of large and methodologically advanced studies have come out since then, some finding no effect like Dube, others finding strong effects like Rubinstein and Wither. These are just examples; there are at least dozens and probably hundreds of studies on both sides.

Okay, so surely we just need to look at all available evidence? SSC goes on to provide this chart from a comprehensive meta-analysis:

So does minimum wage cause unemployment? Yes? No? Maybe?

Obviously it depends on the extent of the price floor. A minimum wage of $100/hour would probably have some detrimental effects. But $15/hour? It’s unclear.

We can sanity check SSC’s armchair analysis by asking economists directly. Chicago Booth’s Initiative on Global Markets regularly conducts expert surveys, including a relevant one on minimum wage:

Okay, so 26% agree or strongly agree that a higher minimum wage could cause unemployment, and 24% disagree. Weight by confidence, it’s 34 to 29. So there’s a very slight skew against minimum wage, but the biggest group is uncertain.

(Maybe it’s the nature of economists to vacillate on every issue? I don’t think so. The 4 latest IGM surveys on Auction Theory, Economic Recovery, Tax Proposals and Foreign Competition show much stronger consensus.)

Note that Prop 22 isn’t actually about minimum wage. A shift from drivers-as-contractors to drivers-as-employees wouldn’t just be a raise, it would impose a minimum number of hours worked. Are there adjustments we can make to better apply our model?

Again, there are infinite possible adjustments, but here’s a key one: Uber et al. are nearly as monopsony. There are a few app-based employers, but as demonstrated in their coordinated support for this bill, they’re sufficiently capable of collaboration. In the Econ 101 world, monopsonies produce deadweight loss, meaning that they’re socially inefficient. Intuitively, it makes sense that an imbalance of bargaining power would lead to unfair negotiations.

But that’s just one caveat. Probably the more important adjustment is in the shift from flexible employment to mandated full-time employment, which isn’t clearly captured in the simple models or data presented here.

So yes, my arguments are insufficient, but we’re already much more sophisticated than everyone tweeting “PRICE FLOOR” and calling it a day.

2. Expertise as Authority, Expertise as Bias

The second tension is between bias and authority.

Uber’s own analysis finds a strong need for Prop 22. Without it, they’ll lose 76% of their drivers (158 thousand), 23-59% of their trips, and be forced to increase prices 25-111%.

Should we believe them?

The paradox is that Uber is both incredibly biased, and also the only firm with the requisite data and understanding to conduct a thorough analysis. Maybe the ideal situation is that they share data with an impartial third party, but even then the choice of third party won’t be random.

Along those lines, Uber commissioned a third party administered survey showing that 72% of drivers support Prop 22. But then another third party site (not funded by Uber) administered its own survey, and found that 60% of CA drivers were in favor of Prop 22. What happened to the other 12%?

The Uber-funded survey is light on methodological details. It wouldn’t have been hard to find a non-representative set of drivers that disproportionately support Prop 22. But of course, there’s no guarantee that the non-Uber-funded survey has a representative sample either!

How about a third source? The subreddit for uber drivers has a number of top posts this month about Prop 22, all of which are against it (1, 2, 3, 4, 5).

So what’s the actual rate of drive approval? 72%? 60%? 0%? For that matter, how seriously should we actually take driver opinions, given that they’re a captive audience to any propaganda their apps choose to throw at them (1, 2, 3)

3. Heuristics For and Against Underdogs

Cash contributions to Prop 22 are currently at $20 million against, and $203 million for. That’s the highest total of any ballot proposition in California ever. The against donors are mostly labor unions, the for donors are Uber, DoorDash, Lyft, Instacart and Postmates (which was acquired by Uber earlier this year).

At first glance, this seems suspicious right? The 2020 presidential candidates raised $1.57 billion and $1.51 billion, for a pretty even ratio of around 25:24. In contrast, Prop 22 spending is at 10:1! Maybe suspicious isn’t the right term, but it certainly doesn’t feel very democratic.

On the other hand, doesn’t this demonstrate that Uber et al. are getting more value out of Prop 22 than their opponents stand to lose? Isn’t this a genuine expression of their willingness to pay?

I worry that the discrepancy is mostly a result of coordination costs. An Uber driver opposed to Prop 22 knows their small donation has a tiny chance of influencing the outcome, so it’s better to just sit back, hold on to their money, and hope for the best. [2] The companies in favor of Prop 22 don’t have these problems. They still have to coordinate, but only with a couple other entities, all of which have legal teams.

There’s also the issue of time horizons. The Diff has an applicable anecdote:

Containerization reduced the amount of labor needed by ports, but longshoremen could shut down ports if they worried about job security, so the process was slow. However, since corporations are immortal and workers are not, port operators and shipping companies could engage in time arbitrage, paying high wages to current union members without replacing their jobs when they retired.

Similarly, we might worry that drivers and unions are only willing to pay what they think they’ll stand to win back in this generation. In contrast, companies are really good at not discounting the future, and Uber is the world champion of time arbitrage. They will always be willing to lose money today to make more money in the future. The kind of company willing to burn $1 billion a year to compete in China has no problem taking a $50 million hit to operate in California.

So once again, I have no idea how to interpret this. Is it a gross violation of democratic principles? A fair and genuine expression of willingness to pay? Should we start modeling elections as auctions?

4. Varieties of Accelerationist Argument

If Marx were alive today, he would probably not say “we need to strike down Prop 22 to ensure workers’ rights.”

More likely, it would be something like “Yes, let the conditions worsen, let the class consciousness grow! Only then will we transcend late capitalism and emerge free to engage in an infinite revolution by the masses!”

I’m exaggerating, but not by that much, here’s the actual quote, in which Marx comes out against economic regulation and in favor of free trade:

But, in general, the protective system of our day is conservative, while the free trade system is destructive. It breaks up old nationalities and pushes the antagonism of the proletariat and the bourgeoisie to the extreme point. In a word, the free trade system hastens the social revolution. It is in this revolutionary sense alone, gentlemen, that I vote in favor of free trade.

I’m sympathetic to this line of thinking. The basic logic makes sense. But it’s also just such a dangerous path to tread. If you accept accelerationist thinking, you can find yourself endorsing any number of arbitrarily evil views. Should we let the government murder people? Perpetuate the war on drugs? Incarcerate dissidents without a trial? Sure! So long as it “pushes the antagonism of the proletariat and the bourgeoisie to the extreme point”, it’s fine by Marx.

A better version of this argument comes from SlateStarCodex, writing again about minimum wage:

This strikes me as the strongest argument for the minimum wage and other job-killing labor regulations: that they are turning otherwise-miserably-employed people into unemployed welfare recipients. “Too many people are unemployed and receiving welfare” seems more like a problem society will actually try to solve than “too many people are miserably employed”, and maybe the solution will actually do us some good.

That’s still a flavor of accelerationism, but with the caveat that it works not through antagonism, but through incentive alignment. The purpose of allowing and even endorsing short-term harms isn’t to drive class divides, but to force an internalization of costs by the appropriate parties.


In conclusion, politics is impossible and I have no idea what to believe anymore.

We started out debating the intricacies of labor market elasticities. This is a perfectly fine economics question. Except that there isn’t a generic answer, and in this particular case, the only qualified expert with access to good data is the would-be subject of regulation!

We can’t appeal to drivers themselves because results vary drastically, are all skewed, and also influenced by a perpetual barrage of in-app propaganda.

Campaign contributions are shady, but on the other hand, they’re a fair expression of willingness to pay, and on the third hand, modulated by coordination costs and discrepancies in time horizons.

And we can’t even appeal to the basic goodness of one side or the other since there is a perfectly valid argument for letting things go to shit in the short term.

In the long term, everything is a messy choice over which local maximum you want to be stuck in, and how long you’re willing to go through a valley of pain to get there.

If I was an authoritarian Tzar, I would kill Uber and improve public transit. In my utopia, we’re all riding bikes and taking the subway.

But in the messy coordination-problem-governed world that we actually live in, I don’t know how we get there.

One good objection might be: San Francisco has had decades to build good public transit, and never did. Uber wouldn’t exist unless it was necessary.

That’s fine, except that as bad as things are, they can always get worse. Another controversial ballot measure wants to impose a new tax so that Caltrain can continue operations. BART, the other major Bay Area system, notes that it “is facing about $600 million in budget deficit when combining this current fiscal year and next fiscal year.” SFMTA is also bleeding money, expects to lose $568 million over the next 4 years, and may end service for 40 (out of 68 total) lines.

Voting down Prop 22 won’t fix any of this. Nothing will fix any of these problems, not even the short term stimulus bills the agencies are asking for.

After hours pouring over these issues, I have a pretty good sense of where we’re at now. And after years of being an armchair urbanist, I have a pretty good sense of where I’d like to go. But the path between those two points isn’t a straight line, it’s not even close.

The tools we have available for analyzing these questions (economics, political science, social theory), and the tools we have for resolving them (assembly bills, ballot propositions, stimulus packages) are just woefully insufficient.

There is work on complex systems, and ideas on overcoming local maxima, and even project management skills designed for tackling wicked problems, but none of that tells us how to vote.

And so I’m in epistemic pain, a neurotic investigation into an infinite fractal frontier. And as much as it sucks, it doesn’t even compare to the work required to push towards actual solutions.


[1] Yes, “epistemic pain” is a made up term, which means I’m now at the stage of inventing new language to describe the phenomenology of my intellectual journey. This is what peak neuroticism looks like.

[2] In economic language, a free rider.

Are the Neurotypicals Okay?

I got new clothes last night, which means I spent this morning meticulously removing every tag with a seam ripper

I’ve done this for years, but this time I had a new reason to feel ashamed. In It’s Bayes All The Way Up, SlateStarCodex argues that:

The autistic version works differently. The top-down model tries to predict the feeling of the shirt on my skin, but tiny changes in the position of the shirt change the feeling somewhat; bottom-up data does not quite match top-down prediction. In a neurotypical with wide confidence intervals, the brain would shrug off such a tiny difference, declare it good enough for government work, and (correctly) ignore it. In an autistic person, the confidence intervals are very narrow; the top-down systems expect the feeling of shirt-on-skin, but the bottom-up systems report a slightly different feeling of shirt-on-skin. These fail to snap together, the perceptual handshake fails, and the brain flags it as important; the autistic person is startled, upset, and feels like stopping what they’re doing in order to attend to it.

And from the original paper Scott’s referencing:

In other words, in autism there may be a failure to attenuate sensory precision and contextualize sensory information in an optimal fashion. For example, an individual who always expects a high sensory precision would, to some extent, be a slave to their senses, affording sensations disproportionate weight in driving beliefs about their world.

So this morning I felt embarrassed, and a bit sad. No one wants to be “a slave to their senses”. No one wants to be a slave to anything! And yet there I was, engaging in neurotic behavior, contextualizing my sensory information in a suboptimal fashion.


Okay, so I’m not a neuroscientist, and I’m biased in this matter, but I have a lot of doubts about this model.

As far as I’m concerned, tags on clothing really are uncomfortable. It’s a failure of society that we continue to put itchy tags on clothes that have to be painstakingly removed.

You might object that the vast majority of people are fine with the tags, so consensus reality dictates that they’re not itchy. In this view, it’s unreasonable for me to shift the burden onto society.

That’s a fair point, except that neurotypical fashion is consistently horrible. Women’s pants have tiny pockets, if they have pockets at all. Not to even mention skirts, dresses and yoga pants. High heels not only hurt in the moment, but actually cause long term damage. Contact lenses are associated with corneal ulcers (2), not a particularly high rate, but enough that one might ask why we don’t just wear glasses which are both easier and safer.

And all of that’s just the harm actively done, not even considering what we might be missing out on. Why do we have to wear restrictive dress shoes instead of comfortable sneakers? Why can’t we use velcro or slip ons, instead of unnecessarily burdensome laces?

In fact, it seems like neurotypical fashion is deliberately designed to be uncomfortable. Here’s my theoretical model: Fashion is primarily about signaling that you’re willing to undergo great personal discomfort to conform to societal expectations. As a result, convenience and utility are highly anti-correlated with what’s fashionable.

We can confirm this by looking at the extremes. It’s no secret that high fashion is absurd, so what about the other end of the spectrum? If I ask you to think about the single least fashionable piece of clothing, what jumps to mind?

If you’re a journalist at Buzzfeed or Business Insider or Yahoo, the answer is unequivocally cargo shorts. It’s not a coincidence that cargo shorts are also the article of clothing with singularly high utility. They are a miracle of convenience with 6 giant pockets, and no restrictions on mobility. And so of course, society has chosen to demonize them.

But go on Karl Firston, tell me more about autists are the ones failing to “contextualize sensory information in an optimal fashion”. The jokes on you: we have the most optimal fashion.


To be clear, I don’t actually think Karl Fristion or Scott Alexander are at all bigoted. Scott is also a tag cutter, and Karl’s closest thing to an autobiography is titled Am I austistic? [1]

But getting back to fashion, can’t we fault society and still coherently claim that there’s something wrong with autistic people?

Imagine a small minority of women, let’s call them Hautists, who hate wearing high heels. Their feet feel cramped, their calves get sore, they trip and stumble. In this scenario, we could acknowledge that heels are at fault, but still diagnose the Hautists with a particular medical condition.

I think this is a fair point, but we also have to consider long term utility:

  • “Normal” people can tolerate high heels, so they continue wearing them despite the mild pain, and experience discomfort for the rest of their life.
  • Hautists identify high heels as unbearably painful, stop wearing them entirely, and are free from discomfort forever.

I’m particularly sensitive to tags, and it’s an annoying problem to have. But with a small upfront investment, I can cut the tags, and obviate the problem forever. A neurotypical feels the tag, is slightly uncomfortable, contextualizes the sensory information, and goes on to be slightly uncomfortable for the rest of their life, but not so much that they’ll ever actually take steps to resolve it.

If an environment is over-stimulating or loud or distracting in some other way, an autistic person can leave. Neurotypicals might be more likely to “tough it out” and endure the hardship, even when there’s no good reason to.


You might further object that there’s no reasonable model of behavior where this actually happens. In this view, each scenario is a cost/benefit tradeoff, and if a person decides to remain in their environment, it’s the result of a rational choice based on personal preferences. Neurotypicals leave their tags on, autists cut them off. All things considered, it’s still better if you can just ignore a situation that doesn’t deeply affect you.

But again, I have my doubts.

There are countless examples of people failing to alter their environment, even when it would benefit them. The easiest proximate cause to point to is temporal discounting. [2]

The point is, people are lousy at making investments in the future, even when it’s likely to pay off. You procrastinate now even though it makes you busier later. You drink now even though you’ll be hungover later. You avoid the gym now, even though you want to look and feel good in the future.

And similarly, maybe you don’t cut the tags off your clothes because it requires upfront investment, you don’t have a seam ripper handy, and the tag is only mildly uncomfortable.

Temporal discount is just one of many reasons people fail to take actions to improve their situation. Others include lack of self-awareness (you understand that you’re feeling irritated, but fail to identify the root cause) and learned helplessness (you understand all too well that some clothes aren’t designed to be comfortable, but just accept that you’ll have to wear them anyway).


To be clear, I don’t mean to glorify autism. Tags are easy to remove, but other situations might not be as amenable to alteration. I’m blessed with control over my physical space, the ability to afford a nice chair, noise canceling headphones and so on. In other words, I have a privileged existence. (Also, obligatory reminder that “autism severity” is probably not a useful concept.)

But I do think autists are maybe less likely to make the kind of mistake I’m describing. The mistake where you’re miserable, in a job you hate, wearing clothes that make you uncomfortable, scrunching your feet into small shoes, stuck in a distracting environment, and yet somehow, manage to convince yourself that everything is fine.

That’s neurotypical behavior, and maybe it’s an optimal contextualization of sensory information for some purposes, but it also just sucks. It’s a failure of self-awareness and self-efficacy. In this model, neurotypicality is better understood as a generalized inability to understand how your environment is impacting you, and then take actions to improve it.

See Also:
Friston et al., An aberrant precision account of autism
Friston, Am I autistic? An intellectual autobiography
SSC, It’s Bayes All The Way Up
Michelle Dawson on Autism and Atypicality
Unfeeling Malice, Review of Asperger’s Children: The Origins of Autism in Nazi Vienna

Footnotes
[1]  The answer ends up being no, but not before he opens with:

In looking back at my life, I can see some distinctly autistic traits in my childhood—and indeed current ways of engaging with the world. For example, I religiously avoid mobile phones and do not Skype. In fact, I find any disruption to my weekly routine rather nerve wracking. Unhappily, this means travelling to international conferences can be unsettling—where I spend most of the time avoiding other human beings; especially in the morning.

[2] I’m not going to throw the dozens of behavioral econ papers at you, if you’re really curious, you can just read the Wikipedia article on hyperbolic discounting and then skim its citations.

I Couldn't Read Papers Until I Learned The Greek Alphabet

I’ve always found it incredibly difficult to read papers. For years, I thought they were just written poorly or that academic writing is cryptic, but mostly I just thought I was stupid.

Here’s an example. The author starts out:

Some technologies save lives—new vaccines, new surgical techniques, safer highways.

Which is simple enough, and easy to parse. Then a paragraph later, he hits you with:

The agent is endowed with some initial stock of knowledge that generates a consumption level c and has a utility function u(c) = ū + [c1 – γ / (1 – γ)].

I want to clarify that I don’t have math anxiety. I would actually consider myself pretty decent, and am certainly comfortable with the basic math required to understand an exponential.

But for whatever reason, I just could not parse that formula.

Today, something finally something clicked and I realized it’s not a failure of reading, it’s a failure of over vocalization.

There are different varieties of inner speech and personally, I literally cannot read without sounding the words out loud in my head. It is equivalent to speaking silenty with my lips shut, or almost like whispering very very softly.

The problem is, I don’t know Greek! When I read “u(c) = ū + [c1 – γ / (1 – γ)]”, I can’t sound it out in my head because I don’t know how to pronounce γ.

That’s literally it. That’s the stupid and totally trivial problem behind so many of my failures. It’s unbelievably silly in retrospect, as so much of life is.

So to put it to the test, I downloaded Anki, found some Greek alphabet flash cards, and started learning. And now? It’s dramatically easier. It’s not even that I’m reading 2x faster, it’s that I’m able to read papers I would have previously given up on entirely.

This would all end up being really funny and stupid and embarassing in an “aw shucks” kind of way, except that this totally screwed me for years! I have no idea what else I’m missing, or what other low hanging fruit could be enriching my intellectual life. I also don’t know how many other people are suffering from this same problem, and just think they’re irreparably stupid.

If this describes you, or if you’re not sure and want to give it a try, here Anki, here’s the flash card deck, and best of luck!


[Edit: 03/23/21]: It turns out reading silently was not even normal until a couple hundred years ago, though there’s some debate over exactly when the shift happened, and if reading silently was considered difficitul, or just unusual at various times. Some monks were made to scribe silently in the 9th century. In the 4th century, Augustine of Hippo mentions Ambrose reading silenty, but seems to be surprised by his ability to do so. This Quartz article indicates that reading silently may have been unusual until the 18th century.