Almost everyone is at peace with nature: listening to the ocean waves against the shore, by a still lake, in a field of grass, on a windblown heath. One day, when we have learned the timeless way again, we shall feel the same about our towns and we shall feel as much at peace in them, as we do today walking by the ocean, or stretched out in the long grass of a meadow.

- Christopher Alexander, The Timeless Way of Building

Christopher Alexander was an American design theorist and architect whose philosophy and works have been influential since their origins in the mid-to-late 20th century. Alexander’s teaching at Berkeley and his contributions to the design theory of programming have earned him an enduring place in Silicon Valley intellectual culture.

I read his books, The Timeless Way of Building and A Pattern Language, searching for answers to the ongoing debate about why buildings and cities are so ugly today.

Subscribe now

Synopsis

The central idea of The Timeless Way is that the design of buildings and cities isn’t reducible to the pure visual art of carving geometric forms out of volumes, but neither is it infinitely complex and contextual. Rather, it is the art of fitting together patterns, like Legos. Tastefully pick 20 patterns from a set of 200 or so and you’ll make a building that is beautiful.

The patterns that define a building, or a city, or a life, arise from the events which keep on happening there most often. For example, the pattern of events that govern Alexander’s life:

Being in bed, having a shower, having breakfast in the kitchen, sitting in my study writing, walking in the garden, cooking and eating our common lunch at my office with my friends, going to the movies, taking my family to eat at a restaurant, having a drink at a friend’s house, driving on the freeway, going to bed again. There are a few more.

There are surprisingly few of these patterns of events in any one person’s life, perhaps no more than a dozen. Not that I want more of them. But when I see how very few of them there are, I begin to understand what huge effect these few patterns have on my life, on my capacity to live. If these few patterns are good for me, I can live well. If they are bad for me, I can’t.

Or the pattern of events in Lima, Los Angeles, and a medieval town:

What is Lima—what is most memorable there—eating anticuchos in the street; small pieces of beef heart, on sticks, cooked over open coals, with hot sauce on them; the dark, badly lit streets of Lima, small carts with the flickering fire of the hot coals, the faces of the sellers, shadowy figures gathered round, to eat the beef hearts.

When you think of Los Angeles, you think of freeways, drive-ins, suburbs, airports, gas stations, shopping centers, swimming pools, hamburger joints, parking lots, beaches, billboards, supermarkets, free-standing one-family houses, front yards, traffic lights . . .

When you think of a medieval European town, you think of the church, the marketplace, the town square, the wall around the town, the town gates, narrow winding streets and lanes, rows of attached houses, each one containing an extended family, rooftops, alleys, blacksmiths, alehouses . . .

The list of elements which are typical in a given town tells us the way of life of the people there.

And, what is most remarkable of all, the number of the patterns out of which a building or a town is made is rather small.

One might imagine that a building has a thousand different patterns in it; or that a town has tens of thousands . . .

But the fact is that a building is defined, in its essentials, by a few dozen patterns, and, a vast town like London, or Paris, is defined, in its essence, by a few hundred patterns at the most.

Buildings and towns are defined by the patterns which repeat themselves there, but some buildings and some towns are full of life, and others less. The right patterns can make a place live or die.

Consider an example on the smallest possible scale: A prayer rug

The old Turkish prayer rugs, made two hundred years ago, have the most wonderful colors.

All of the good ones follow this rule: wherever there are two areas of color, side by side, there is a hairline of a different third color, between them. This rule is so simple to state. And yet the rugs which follow this rule have a brilliance, a dance of color. And the ones which do not follow it are somehow flat.

Ornamental decorations on a building work in much the same way:

Search around the building, and find those edges and transitions which need emphasis or extra binding energy. Corners, places where materials meet, door frames, windows, main entrances, the place where one wall meets another, the garden gate, a fence - all these are natural places which call out for ornament.

Now find simple themes and apply the elements of the theme over and again to the edges and boundaries which you decide to mark. Make the ornaments work as seams along the boundaries and edges so that they knit the two sides together and make them one.

Or an independent discovery of similar patterns in Islamic geometric art by a Slate Star Codex reader who is, as far as I know, unaware of Alexander’s work.

To pare it down to the essentials, the first few rules [of Islamic geometric art] all serve to maintain the impression that the designs are formed by interweaving lines that either extend infinitely or connect into loops.

First, lines should never terminate except at the boundary of the entire design.

Second, lines must maintain their direction before and after crossing another line.

Third, within a single design, lines should only turn or intersect at a limited set of compatible angles.

Similar rules exist for building beautiful rooms.

The light in many glorious rooms is also governed by a simple rule.

Consider the simple rule that every room must have daylight on at least two sides (unless the room is less than 8 feet deep). This has the same character, really, as the rule about the [Turkish rug]. Rooms which follow this rule are pleasant to be in; rooms which do not follow it, with a few exceptions, are unpleasant to be in.

And for buildings:

Outdoor spaces which are merely “left over” between buildings will, in general, not be used.

There are two fundamentally different kinds of outdoor space: negative space and positive space. Outdoor space is negative when it is shapeless, the residue left behind when buildings—which are generally viewed as positive—are placed on the land. An outdoor space is positive when it has a distinct and definite shape, as definite as the shape of a room, and when its shape is as important as the shapes of the buildings which surround it.

And for towns:

The artificial separation of houses and work creates intolerable rifts in people’s inner lives.

In modern times almost all cities create zones for “work” and other zones for “living” and in most cases enforce the separation by law.

But this separation creates enormous rifts in people’s emotional lives. Children grow up in areas where there are no men, except on weekends; women are trapped in an atmosphere where they are expected to be pretty, unintelligent housekeepers; men are forced to accept a schism in which they spend the greater part of their waking lives “at work, and away from their families” and then the other part of their lives “with their families, away from work.”

Throughout, this separation reinforces the idea that work is a toil, while only family life is “living” - a schizophrenic view which creates tremendous problems for all the members of a family.

What are the requirements for a distribution of work that can overcome these problems?

1. Every home is within 20-30 minutes of many hundreds of workplaces.

2. Many workplaces are within walking distance of children and families.

3. Workers can go home casually for lunch, run errands, work half-time, and spend half the day at home.

4. Some workplaces are in homes; there are many opportunities for people to work from their homes or to take work home.

To make a beautiful Turkish rug or Islamic geometric design, there are three or five essential rules to follow. How many do you need to make a beautiful room, building, or town?

It’s not just one rule, like “form follows function.” But neither is it so complex and contextual that writing down rules to follow is impossible. It takes a couple dozen patterns to build a beautiful building. Here are the patterns Alexander used to build a small cottage, workshop, and guest house in the back yard of the house he used as his office, starting from the broadest patterns and going to the most specific.

Work Community, The Family, Building Complex, Circulation Realms, Number of Stories, House for One Person, South Facing Outdoors, Wings of Light, Connected Buildings, Positive Outdoor Space, Site Repair, Main Entrance, Entrance Transition, Cascade of Roofs, Roof Garden, Sheltering Roof, Arcade, Intimacy Gradient, Entrance Room, Staircase as a Stage, Zen View, Tapestry of Light and Dark, Farmhouse Kitchen, Bathing Room, Home Workshop, Light on Two Sides of Every Room, Building Edge, Sunny Place, Outdoor Room, Connection to the Earth, Tree Places, Alcoves, Window Place, The Fire, Bed Alcove, Thick Walls, Open Shelves, Ceiling Height Variety

Why are Buildings Uglier Today?

Alexander’s philosophy of how to design beautiful places is compelling and exciting. But his theories add conflict to the central question of why buildings and cities have become uglier over time because they place the blame for the decline of beauty squarely on what makes cities valuable in the first place: Scale.

Reading The Timeless Way only reinforces the impression that the pre-modern towns of Europe, Asia, and parts of the American East Coast, reflect so much more life, more humanity, and more of that “quality without a name” than their post-modern counterparts. But the question is why?

The major enemy is scale.

For example, Alexander claims that the breakdown of design is caused by a growing alienation between the people who use a building and the people who design it.

In agricultural societies, everyone knows how to build; everyone builds for himself, and helps his neighbor build…

But, by contrast, in the early phases of industrial society which we have experienced recently, the pattern languages which determine how a town gets made become specialized and private. Roads are built by highway engineers; buildings by architects; parks by planners; … tract housing by developers …

Those few patterns which do remain within our languages become degenerate and stupid.

The users, whose direct experience once formed the languages, no longer have enough contact to influence them. This is almost bound to happen, as soon as the task of building passes out of the hands of the people who are most directly concerned, and into the hands of people who are not doing it for themselves but for others.

He thinks the scales of modern cities are now too large for effective local government:

Individuals have no effective voice in any community of more than 5,000-10,000 persons … Therefore: decentralize city governments in a way that gives local control to communities of 5,000 to 10,000 persons. Give each community the power to initiate, decide, and execute the affairs that concern it closely: land use, housing, maintenance, streets, parks, police, schooling, welfare, neighborhood services.

That cities are too large to provide everyone access to the amenities of the downtown:

There are few people who do not enjoy the magic of a great city. But urban sprawl taks it away from everyone except the few who are lucky enough, or rich enough, to live close to the largest centers … Therefore, put the magic of the city within reach of everyone in a metropolitan area. Do his by means of collective regional policies which restrict the growth of downtown areas so strongly that no one downtown can grow to serve more than 300,000 people.

And that buildings themselves are too large:

There is abundant evidence to show that high buildings make people crazy. High buildings have no genuine advantages, except in speculative gains for land owners … Therefore, in any urban area, no matter how dense, keep the majority of buildings four stories high or less. It is possible that certain buildings should exceed this limit, but they should never be buildings for human habitation.

The ever-growing layers of hierarchy that enable a world of world-cities also create principal-agent problems between designers and users, voiceless political constituents, and demand for hulking transport infrastructure.

Alexander’s ideal town arises from a process of decentralized, small scale “repair.”

Let us imagine a cluster of houses growing over time.

Each house starts with a small beginning—no more than a family kitchen, with a bed alcove off one end, and a kitchen counter.

All in all, no more than 300-500 square feet to start with.

Then, for the first few years, people add 100-200 square feet more each year.

First a bedroom perhaps; another bedroom; a workshop; a garden terrace; a full bathing room; arcades and porches; studio; a bigger sitting room, with a big fireplace; a garden shed. And, at the same time common things are also built; they plant an avenue of trees; a small gazebo; a shared outdoor room; paving on the paths . . .

As the buildings reach maturity, the increments get smaller. Yet, at the same time, collectively, the houses begin to generate the large patterns which define the cluster.

Each person begins to work with his neighbor, first the neighbor on one side, then the neighbor on the other side; and together they try to make the space between their houses beautiful.

So, the houses get their form, both as a group, and separately, as individuals, from the gradual accretion of a number of small separate acts.

Alexander is correct to notice that scale has changed almost everything about our relationship to the built environment compared to pre-modern cities. And he insightfully describes the decentralized, incremental process of construction that led to some of the most beautiful towns ever made. In this way, he provides a compelling answer to the question that motivated this post. Our buildings and towns are worse today because of scale.

But his offered solutions require impossible sacrifice and political coordination. If the only way to improve the built environment is to have every person build their own house, to restrict the size of local governments to 7,000, and to keep the size of cities below 300,000, then our built environment will not improve.

Alexander’s prescriptions against scale also conflict with economist’s view on the value of cities in the first place. Cities are labor markets and the bigger they are, the more efficient they get. In larger cities there is more specialization in production, more shared infrastructure, there are more knowledge spillovers, more competition, and better matching between workers and firms.

I am predisposed to be suspicious of Alexander’s top-down designs, and many of them are ridiculous. The transition between agricultural and urban land, building heights, and the location of CBDs are not appropriately decided by central authorities.

However, I found it difficult to sustain my suspicion against all of Alexander’s ideas. I wanted to dismiss all of the designs for things beyond an individual’s control as something that should be left to the decentralized order of the market. But for so many things this is an empty position that ignores the possibility of mistakes among market participants, who are influenced not only by market forces, but also by cultural ideas about building.

The lifeless Lennar homes suburbs that tile the sunbelt are in a sense outcomes of decentralized order. But they are nonetheless designed by individuals who could make better decisions and improve the world.

The team behind California Forever is operating in a market, but they have to make decisions about where to place the roads, the public services, how many trees to put in, how to integrate water features, even where to locate certain types of land use. If they follow certain patterns of building, their city will be dead. If they follow others, it can be electric with life.

Perhaps over some sufficiently long term, the cities and subdivisions that build well reward their owners more than the ones they don’t, thus pushing towards good design with an invisible hand. But cities are durable, both physically and economically as the increasing returns make it difficult to compete with cities that are already large. European cities are still stuck with Roman decisions about city locations and design, The mistakes of urban renewal still scar our cities 50 years later. Mistakes in city building are too durable to rely only on the slow, decentralized punishments of the market.

Even Alain Bertaud in Order Without Design says that while most things in a city should be left to the market, roads and public services have to be designed top-down. Bertaud’s book is arguing against city master plans even more wildly prescriptive than Alexander in his book, but from my perspective this admission gives up the whole game! After placing the roads and public spaces a city’s entire skeleton is defined. This skeleton more than anything else makes up the difference between Amsterdam, New York, and Phoenix.

The Timeless Way and A Pattern Language are best when they are insightfully describing the natural interactions between humans and built form. And when explaining why these interactions have gotten worse since around the invention of the automobile. But unfortunately, Alexander’s books are at their weakest when they are prescribing solutions to this problem. By only offering solutions that require sacrificing so much that is good about the modern world, they end up with no real solutions at all.

The more difficult and more useful challenge is finding a way to improve our buildings and cities that is consistent with the economies of scale that our modern world will never give up.

Pattern languages, the residue of repeated human interaction with design, are real. A prayer rug needs three colors at every edge. A room needs light from two sides. A dozen rules make the difference between a dead place and a living one. If the developers building California Forever, the architects designing 5-over-1 apartment complexes, or the planners laying out the next sunbelt suburban subdivision studied these patterns and applied them, they could make their building beautiful.

But Alexander gives us no advice on how to get these parties to design more thoughtfully. Thus, we are left with a vocabulary for what makes places alive, but no means of making it matter to the people who actually build our world.

Subscribe now

Geology of Japan, Urbanism of Florida

Hawaii is, just like Japan, a volcanic island in the Pacific. This endows the island with impressive mountains that rise straight up from the water line, shrouded in a deep-green jungle. Japan’s volcanic activity comes from its place on the edge of the Pacific plate, along the ring of fire. Hawaii, however, is right in the middle of the Pacific plate and has no tectonic activity whatsoever. Instead, there happens to be a giant magma plume underneath Hawaii that punches through the plate, creating the islands.

The Pacific plate moves northwest a few inches a year, but the magma plume stays in the same place. This creates the diagonal shape of the Hawaiian island chain. A few million years ago, the northwestern-most Kauai was above the plume and rose out of the volcanic activity. Over time, the Pacific plate shifted northwest and now the big island is on top of the plume and has the most active volcanoes. In a few million more years a new island will appear to the southeast of the big island and the island chain will grow.

The ancient volcanoes of Hawaii are huge. To get a sense for this, consider Diamond Head on Oahu. Diamond head is the crater left behind after the volcano of the same name blew up and dispersed most of the volume of the mountain into the atmosphere. The remaining caldera towers over all the skyscrapers in Honolulu and covers a significant area of the city.

Zooming out, you see the Koʻolau mountain range which covers the eastern half of Oahu and dwarfs Diamond Head. But the Koʻolau mountain range is not really a mountain range at all, rather it is itself the caldera left behind from a volcano hundreds of times the size of Diamond Head. The right half of the volcano collapsed into the sea millions of years ago, leaving only a portion still above water.

Unlike its geological similarities to Japan, the urbanism of Honolulu shares much more with the sprawling, car-centric and expensive housing markets of America.

Honolulu is geographically constrained between the mountains and the sea, so the downtown and tourist districts are quite tall and dense. Still, it’s no Hong Kong. The buildings in Waikiki range form inoffensive to ugly, but the streets are pretty good, with lots of storefronts, relatively narrow and slow traffic lanes, and lots of pleasant indoor-outdoor spaces.

There are no subways in the city and only a meager single-line raised metro whose construction was first planned in 1960 and is still incomplete today. Outside of the downtown areas, density is low and the land use takes on the more traditional and decrepit American form of single family homes, strip malls, and stroads.

Zoning and construction are tightly controlled, in the American style, with lots of extra environmental concerns layered on top. The map of Oahu below shows in yellow where any residential or mixed use construction other than single family homes may be built. Blue is single-family only, orange is commercial/industrial, and red is agricultural or preserved. Honolulu is the yellow area in the bottom right of the island.

Cuisine of Japan, Culture of Florida

After the volcanic mountains, the next clearest similarity between Hawaii and Japan is the food. People of Asian descent are a plurality of the population in Hawaii, and most of those are either Japanese, Filipino, or Chinese (fun fact: Sun Yat-Sen, the leader of the Chinese nationalists in their civil war against Mao, was educated in Honolulu!)

But despite this large Asian population and popular cuisine, the culture of Hawaii is more similar to Florida than Japan.

There is much more multi-culturalism and multi-lingualism. There are big trucks, loud homeless people, lazy drunks, ostentatious luxury, and fireworks.

Meiji attitude to the West vs American grievance politics

The final bridge that Hawaii makes between Japan and Florida is in their history with the West. Both Hawaii and Japan have some history of humiliation and subjugation by the Western world.

In Hawaii’s case, Western influence began to drive their politics immediately after the discovery of the islands in 1774. Kamehameha’s conquest of the islands began in 1782, aided by Western guns and military advisors. For the next century, the monarchy survived but was forced to make a series of concessions to a Western controlled parliament. Finally in 1893, the monarchy was overthrown by a real life conspiracy by a group of 13 businessmen self-styled as the Committee of Safety and the Islands were annexed by America 5 years later.

Japan’s reaction the technological superiority of the Western world was an extended period of civil strife that eventually led to the accession of Emperor Meiji from a pure figurehead to a position of near-absolute rule in something like a Western-style constitutional monarchy in 1868. This began the fastest industrialization in history and rocketed Japan to global power within 50 years. The power and growth that these Western reforms gave the Japanese created an enduring respect for the West that lasts to this day.

Hawaii had less domestic potential to work with, so their path was different and the uprooting of Hawaiian society was more complete and less self-determined. This makes the politics of colonial history in Hawaii more similar to the politics around Native Americans on the mainland: obsession with guilt, land acknowledgements, and true-blooded ancestry. People fret over whether they say they are Hawaiian or if they are just “from Hawaii” even if they were born there. In this way, Hawaii mirrors the traditional American form of grievance politics, despite some similarities with Japan’s colonial history.

Subscribe now

There is more to crack in the cultural code of Hawaii, but one of the important keys is understanding that the island is something like Miami in the Sea of Japan i.e, the Florida Japan-handle.

The supply-and-demand model that everyone learns in Econ 101 clearly repudiates the minimum wage. A binding minimum wage bans some transactions between people who are willing to work and employers who are willing to pay them, creating deadweight loss.

Monopsony

One relevant addition to this model is monopsony. A perfect monopsony is when there is only a single buyer of goods in a market. In the labor market, this means a single firm hires all workers.

The outcomes of monopsony are easiest to understand when considering a goods market. Imagine you're the sole buyer of apples and all the orchards from Washington to Virginia wait patiently for your quoted price.

You know that if you offer $1 per apple, the farmers can only afford to pick the lowest hanging fruit, and you'll get 10 apples for $10 bucks. If you offer $2, the farmers can afford break out the step-ladders and produce 5 more apples. Now, you get 15 apples for $30. If each apple is worth $3 to you, this is still net positive, since they only cost $2 each. But to get those 5 extra apples, you had to pay $20 extra dollars which is $4 per apple, and thus more than you're willing to pay.

If a minimum price was set and you had to pay at least $2 for all the apples anyways, you'd be better off buying all 15 than just 10. In a market with a single buyer, a price floor can raise price and quantity.

In the same way, a monopsonist in the labor market gets to choose how many workers are “produced” by setting the wage. Every extra dollar you offer brings more workers in, but just like for apples it also raises the price of all the workers who would have been produced at lower wages. Therefore, every additional hire gets more and more expensive, so the marginal cost of labor rises faster than the supply of labor.

The firm hires until this marginal cost equals the marginal revenue. But that sets wages and employment below the competitive level: the firm restrains hiring because their decisions set the wage in the industry and the additional cost of raising the wage to attract the next worker is too high, even though the average cost per worker would still be net positive.

From this monopsony baseline, the costs of the minimum wage are absent. If a price floor raises the wage above what the monopsonist wants to pay, you actually get closer to the optimal competitive outcome. Wages and employment go up!

Subscribe now

But is Monopsony Relevant to the Minimum Wage?

Open eyes are enough to prove that the low-skilled labor market is not dominated by a single firm. On any commercial street there are bound to be several restaurants, retail stores, warehouses, cleaning companies, and gig-work apps available to hire low skill workers into dozens of positions, all competing with one another. More empirically, there are around 3 million establishments between firms in construction, retail trade, and food services. My labor market of universities and think tanks seems more likely to bear monopsony, and be in need of a minimum wage, having only around 5,000 firms. So if the low-skilled labor market is supposed to be monopsonistic, then the rest of the labor market must be much more so.

Economists these days, however, tend to generalize beyond the definition of monopsony as a market with a single buyer and instead identify it with upward sloping labor supply.

In a perfectly competitive world, it’s costless to switch between firms. If one firm lowers its wage below what’s offered by competitors, it suddenly finds itself with zero employees. One cent off and everyone leaves for a different job with higher pay. In other words, the elasticity of labor supply is infinite. On the graph, it’s horizontal.

It’s a difficult empirical problem to find wage changes that aren’t correlated with changes in firm or worker productivity, but research has pretty consistently shown that the elasticity of labor supply is not infinite, it’s more like 4 or 6.

The source of this finite labor supply elasticity is intuitive: it’s not costless to switch jobs. When an employer lowers your wage, you may consider quitting but before you do you’ll weigh the sting of lower pay against the risk of extended unemployment. The costly outside option facing workers means the number of firms they are able to costlessly switch between is exactly one. That creates the monopsony outcomes we analyzed above: Lower wages, less hiring, higher profits, and a counterintuitive response to the minimum wage.

It seems like this closes the case. It’s true that the low-skilled labor market has lots of firms, but search costs mean that firms can lower wages without losing all their workers. Monopsony therefore seemingly pervades every part of the labor market, and thus potentially justifies the minimum wage in many contexts.

Upward Sloping Labor Supply =/=> Monopsony Profits

The logic of search costs causing monopsony makes sense, but closing the case here would be a mistake.

The first puzzle facing the monopsony theory is the profit share of income. If the labor supply elasticity is as low as the studies predict and firms exploit this to enforce low wages, then firms should be making massive profits off of each worker, and thus the profit share of income should be ~35% when it’s actually only 10%.

The second puzzle is wages at large firms. In a standard monopsony model, a firm’s wage bill has to increase massively as it climbs up the labor supply curve to grow in size. This is the exact motivation behind monopsonists cutting employment below the competitive level. But in fact, larger firms pay only slightly higher wages than small ones do, suggesting that firms face an elastic labor supply which can flow in from other jobs in response to even small wage increases.

So despite the fact that search costs for workers do make labor supply curves less than infinitely elastic, firms seem unable to leverage this power into profits as the monopsony theory predicts while simultaneously being unconstrained by the upward sloping labor supply curve when growing.

Moen 1997, a foundational paper in the economics of labor market search, resolves these puzzles with the following observation: while firms have some power over current employees, they have to compete with each other for the attention of job seekers. If a firm posts a job offer with a low wage that would make them lots of profits, other firms are free to post more vacancies of their own and outcompete them for workers.

Additionally, search costs for workers are reflected as vacancy costs for firms. So a costlier outside option for workers (the prospect of taking weeks to find the next job), is also costlier for firms (going weeks with a vacant position). Search costs are a technological constraint on the matching technology of the labor market that prevent the full efficiency of perfect competition, but they don’t allow firms to get a free lunch off the backs of their workers.

Under these forces in equilibrium, all job offers make zero economic profits for firms. The posted wages are still below the marginal product of labor, but the wedge between productivity and wages is eaten up entirely by the cost of searching for jobs and filling vacancies.

But still, the labor supply elasticity is not infinite. Higher wages lower the recruiting time for vacant positions, but they don’t let them recruit the whole market at once. The other puzzles are also resolved. Firm profits remain at competitive levels, and firms don’t need to pay massive premiums to expand their labor force, they just post extra job offers at the same wages. If they post so many as to raise the aggregate tightness of the labor market (i.e the number of vacancies per worker), then wages do rise, but only modestly.

Models of labor search costs that lead to monopsony profits are missing either the symmetric impact of search costs on firms or the free entry of job offers or both.

The competitive search equilibrium is consistent with the empirical evidence (including the randomized control trials on the minimum wage) and it preserves the signs of the econ 101 comparative statics on the minimum wage. Unemployment goes up and deadweight loss is created.

Thus economics in itself, beyond the opinions of economists, is consistent on the minimum wage. Monopsony can overturn the standard minimum wage results, but only when there really are strong constraints on competition, usually coming from a market with a very small number of firms and restricted entry. The low-skilled labor market has too many firms and too few barriers to entry to support the monopsony power required to overturn the standard results on the minimum wage, regardless of what some economists will tell you.

In one of these raids, Perfidia meets and pleasurably humiliates the comically evil and stiff-lipped villain of the film: Colonel Steven Lockjaw. Lockjaw’s sexual fascination with Perfidia, and the politically inconvenient child it creates, become the central conflict of the movie.

The French 75’s exploits eventually catch up to them and a botched bank robbery leads to Perfidia’s arrest and sends Pat and his (unknowingly cuckholded) daughter Willa into hiding. The rest of the film follows Lockjaw’s attempts to hunt down Pat and Willa to erase the evidence of his racial infidelity.

Subscribe now

Biden Centrism

Colonel Lockjaw and the military-industrial immigration enforcement agency under him are the unambiguous villains of the film. Lockjaw delivers extrajudicial executions and becomes a boyishly enthusiastic member of the Christmas Adventurers, an ultra-racist secret society that is simultaneously sinister and absurd.

In this way, the leftist radicals of the French 75 are the heroes. They are fighting the bad guys, their ideals are better than the comic cruelty of Lockjaw and his goons, and the immigrants they are ostensibly trying to protect are hardworking and faultless.

But Anderson’s film is not a hagiography for The Weather Underground. Despite the positive contrast with Lockjaw, his depiction of the French 75 is not sympathetic. Pat becomes a deadbeat drug addict, Perfidia abandons her child and rats out her friends, and other French 75 members are depicted as annoying, ineffectual woke scolds.

I read Anderson as talking to the wannabe leftist revolutionaries in his social circles as a coastal-elite Hollywood director. Anderson is conceding to this group that the anti-immigrant right is obviously evil and cruel, but he sees their revolutionary posturing as a fetishistic, Freudian fantasy that will ultimately end in failure and death. Even if you had the advanced and sexy revolutionary organization you dream of, he says, you would still fail.

Anderson’s alternative vision is captured in the final scene of the film. After defeating Lockjaw, Pat and Willa begin their new Biden-centrist life. After years of anti-authoritarian paranoia that kept them off-grid, Pat and Willa buy iPhones and learn to take selfies. Willa gets a message about a protest and she drives off to go hold a picket sign. The sins of Pat's past are absolved after he accepts incremental civic engagement as the true vehicle of change.

One Battle After Another caricatures both the military-industrial right and the wannabe-revolutionary left, leaving moderate liberal activism as the only battle worth fighting.

Elsewhere on Earth, the United States is at its greatest territorial extent. Manifest destiny stretched the nation from coast-to-coast half a century ago and Seward’s Folly no longer looks like a mistake. America has held on to the territorial gains it made in the Spanish-American war, including Puerto Rico, Guam, the Philippines, and somewhat less directly, Cuba. Simultaneously, the US hoovered up Hawaii, Samoa, Midway, and around 50 other smaller islands in the Pacific from former Spanish, German, and British colonial possessions. 15 years ago, in 1903, the Hay–Bunau-Varilla Treaty established the Panama Canal Zone under direct American control. 11 years later, the canal was complete.

Most recently, European threats to Monroe’s exclusive claim over the New World, capitalist interests in Caribbean investment, and Woodrow Wilson’s paternalistic desire to make the world “safe for democracy” have led to de facto military control of Hispaniola, Cuba, and Nicaragua. Combined with the recent purchase of the Virgin Islands from Denmark, the Caribbean is essentially an American lake.

Over the next several decades, America will relinquish many of these holdings and shrink to her current size. The question is: Was this shrinking good? Was it right for the US to give up the territory it once held?

Subscribe now

General Arguments Against Renouncing Empire

Note that the question of whether the US should have relinquished its maximum territorial extent is different than the one facing America in 1865 or the question of expanding American borders today. There, you have to consider the substantial costs of actually conquering territory in war. Holding on to land already conquered is less costly than conquering anew.

Nonetheless, there can be large costs to holding territory as the often horrifying history of colonialism shows. So isn’t it obvious that American imperialism was a mistake, and thus that it was good that the US eventually backed off?

The first thing to notice when considering this question is that America did not actually give up all of its imperial conquests. Hawaii, Puerto Rico, the Virgin Islands, a smattering of pacific islands, Alaska, and arguably even much of the American Southwest are spoils of conquest or purchase.

No one seriously considers giving up any of the pieces of our imperial history that we kept. The Hawaiian independence movement is perhaps the most popular, but that sets a low upper bound. This acceptance of the pieces of empire we retained suggests that if we had kept more of our 1918 peak territory, people would accept those additional pieces just as easily today.

Reluctance to give up what we kept isn’t just status quo bias. The economic performance of the states and territories still within the US compared to the nations we let go is evidence that American annexation has large positive effects. The causal inference in the opposite direction is even more intuitive: how much poorer would Hawaii become if it were cut off from the American economy, immigration, and institutions? More precise causal inference on the positive economic effects of American annexation is available here.

Another argument against renouncing empire flows from a belief in the benefits of free trade and immigration. I won’t argue for this belief here, only note that the division of labor, and thus the prosperity of an economy, is limited by the extent of the market. But given a belief in these benefits, territorial expansion looks better. Integration under the same national umbrella seems like about the only way to sustain and spread free trade and immigration. The US can’t even manage it with Canada. The principal domestic supporters of Philippine independence, for example, were American farmers who didn’t want to face competition from tariff-free Filipino sugar, and nativists who didn’t want immigration from “alien races.”

A final general argument is, perhaps surprisingly, political legitimacy. The nations that America was closest to annexing are not only economic underperformers, they also tend to be undemocratic and authoritarian. Recently, Panama and the Dominican Republic are probably exceptions here with relatively successful and stable governance.

Haiti and Cuba are extreme examples of failure. If these nations were US states or territories, they would have a far more legitimate local democracy, stable rule of law, and protections from the Bill of Rights. The process by which the US attains control of these nations would not be especially legitimate or peaceful, as is clear from the history of US control in the Caribbean. But the history of these nations after US exit hasn’t been one of peaceful political legitimacy either. Only an extreme blood-and-soil nationalist would prefer indigenous anarchy or tyranny to imperial democracy, prosperity, and rule of law.

So far, I’ve been focusing on general positive features of being integrated into the United States. These features are clear in the places that were integrated, and absent in those that were not, but the choice of integrating some and not others was not random. We probably kept the territories that were easiest to keep and let go the ones that would be too costly to retain. So beyond the general advantages of the US federal umbrella, were any of the territories we gave up actually feasible to keep without suffering massive costs?

The Philippines

The US annexed the Philippines from Spain after winning the Spanish-American war in 1898. Fighting with Filipino nationalists began in 1899 and took the lives of 4,200 American and more than 20,000 Filipino combatants. The 1902 Organic Act transitioned out of martial rule and into an organized territorial government modeled on the American federal structure. This government was mostly controlled by presidentially appointed members to an executive branch and the upper house of the legislature, but there was a popularly elected lower house which did have some influence. Thus began the first democracy on the Philippine islands.

Over the next couple of decades, the Philippines was operated as an “Insular Government” similar to Puerto Rico before they created their current constitution in 1950. This government was, on the whole, successful, stable, and relatively just. The Spanish Catholic Church was separated from the state and its property was bought by the US. The insular government set up a property title system that allowed resident peasant holders to claim their land. Trade volumes increased 10x relative to the pre-war period and per capita GDP doubled. Slavery and piracy were suppressed, and all-cause mortality fell.

In 1916, a new constitution for the Philippines was passed under the Jones Law, which explicitly planned independence as the goal for the territory and set out the conditions under which it would be granted. Governance continued mostly as before, but the civil service was transitioned away from American staffing and by the end of 1921, around 95% of the civil service was staffed by Filipinos. In 1934 independence was officially scheduled for the Philippines in ten years. This was interrupted by the Japanese occupation of the Philippines and the re-conquest of the islands by American troops. In October of 1945 the Philippines joined the United Nations and on July 4th of the next year, their independence was officially recognized by the US.

So would it have been possible for the US to retain control of the Philippines? In a sense, the Philippines is the most likely candidate because its the only nation on this list to be officially annexed by the US. In another sense, its among the least likely because of its distance from the mainland, large size relative to the US, domestic political opponents, and how long independence had been an explicit plan for the territory.

Overall, I do think that permanent unincorporated territory status, like what Puerto Rico has, was possible without additional violence and that this would have been beneficial for both Filipinos and Americans.

How would the timeline have to change for this to happen? First of all, you'd have to go back and avoid passing the 1934 Tydings–McDuffie Act that scheduled full independence. This should be possible because earlier versions of the act were vetoed by President Hoover and voted down by the Philippine Senate, so it wasn’t hard to find opponents of the law. Filipino nationalists would be disappointed by this result, but by this point they had been peacefully sending rebuffed independence missions for 20 years, so I don't think this setback would drive them to violence. Then, it should be possible to stall on future independence missions until WW2 begins.

The end of WW2 is an important crux for this alternate history. On the one hand, much of the Philippines was completely destroyed by the war and the Philippine government was pushed into operating in exile out of Washington D.C. The American reconquest, rebuilding, and reinstallation of government in the Philippines is an opportunity to sympathetically strengthen the status of the Philippines as an American territory. On the other hand, resistance to Japanese occupation had forged a communist-nationalist guerilla movement with over 10,000 experienced fighters. In actual fact, the American and Filipino governments collaborated to defeat this rebel group, but any strategic misstep in the integration process could risk inflaming this group into a rebellion that the US could not defeat.

If the US successfully navigated this crux, the Philippines would enter into the post-war path towards closer integration with the US, along with Puerto Rico and Hawaii. Under US control before WW2, Philippine GDP per capita reached 30% of US levels but fell to around 10% after independence. If they remained as a US territory, Filipino GDP per capita would probably be 7-10 times higher than it is today, roughly matching Puerto Rico. Filipino immigrants in the US today have a median household income 25% higher than native American households and even higher than other Asian immigrant groups. Part of this success is due to selection effects that would weaken as immigration expands, but if the Philippines remained a US territory, America would have a much larger and still successful Filipino immigrant group.

This outcome is a big improvement on our current world! A hundred million people are 10x wealthier and Americans get better off as well. This outcome was probably attainable without significant extra violence, and arguably would have been less violent than our current world given the Philippines’ history with authoritarianism. We severed ties with the Philippines mostly because of nativism, nationalism, and the lobbying of interest groups who benefitted from restricted trade and immigration. Abdicating the administration of the Philippine islands and cutting them off from our labor markets, investment, and institutions was a mistake and a moral failure.

Subscribe now

Cuba

The United States’ presence in Cuba also begins with the Spanish–American war, but unlike the Philippines, Cuba was never annexed. Congress attached the Teller Amendment to the war resolution, which “asserts its determination … to leave the government and control of the Island to its people.”

Instead of annexation, the post-war Platt Amendment essentially established Cuba as a protectorate state. The US retained broad authority to intervene in Cuban governance when its interests or the interests of American property owners on the island were threatened. Over the next few decades, the US would use this authority several times to control protest and rebellion after contested elections and protect American sugar cane plantations. In 1934, Roosevelt’s “Good Neighbor Policy” tried to distance Cuban internal politics from American intervention, but still held on to naval basing rights in Guantanamo.

The Cuban republic that followed never stabilized. Fulgencio Batista’s 1952 coup undermined constitutional rule, and the 1959 revolution brought nationalization, dictatorship, and stagnation.

Like the Philippines, the reluctance to integrate Cuba was based mostly on selfish nativism and fear of competition from Cuban sugar farmers, rather than any benevolent belief in national self-determination. Cuba’s geographic proximity to the mainland, closer cultural and economic ties, and smaller size relative to the US increase its feasibility as a permanent territory or state, like Puerto Rico or Hawaii. However, the actual timeline of Cuban-American relations doesn’t give us as much evidence for how costly permanent integration would be, since the Teller Amendment guarantee of independence is always there to reassure potential nationalist rebels.

So how would the timeline need to change for Cuba to be part of the US today, and how costly would these changes be? You’d probably have to start by convincing American sugar beet farmers to drop their lobbying in support of the Teller Amendment. It’s hard to tell how dropping this independence guarantee would change the actions of Cuban nationalists, and in particular whether a costly guerilla war against American control would ensue after peace with the Spanish.

It’s worth noting that the scale of resistance to America’s several military interventions in our current timeline was small. Cubans generally celebrated America’s war against the Spanish, who were far crueler masters. When the US essentially re-established martial law in Cuba from 1906-1909 after a rebellion against a contested election, there was essentially zero violent resistance, and a similar story with the Sugar Intervention in 1917. All of these interventions were backed by the path to independence in the Teller Amendment but still, the mild response suggests that the Cuban people were not intractably hostile to American government.

Higher costs of integration are difficult to rule out in Cuba’s case, but large benefits from American annexation are nearly certain. Cuba’s dismal political and economic outcomes set a low bar for counterfactual welfare that American integration can easily exceed, even if annexation creates more conflict in the early 20th century. All the benefits discussed in the Philippine case apply equally or more so to Cuban integration. An insular territorial government like the one set up in Puerto Rico or the Philippines would be far more stable, democratic, and prosperous than Cuba’s actual oscillation between right-wing coup d'état and leftist revolution.

Panama

Panama was created as an American imperial project. After negotiations for canal rights with the Colombian government fell through in 1903, American investors and the military intervened to support the Panamanian independence movement and quickly recognized its sovereignty, on the condition that America be granted exclusive control of a 10-mile strip that bisected the country along the canal route.

This legal structure makes integration of Panama extremely costly from the beginning, since the Hay–Bunau-Varilla Treaty explicitly trades canal rights for protection of Panama’s sovereignty. The geographic structure of the canal zone makes retention of that land as exclusive American territory costly as well; the cost-benefit ratio of keeping territory roughly scales with the perimeter-to-area ratio. Panama also performs the best out of all the former American protectorate states or territories in terms of per capita GDP and political stability/legitimacy (though still not without struggles).

All the arguments about the benefits of free trade, immigration, and American institutions from above still apply here, but the bigger costs of annexation and the lower benefits compared to Panama’s best-in-class performance probably make the alternate history of greater US control in the area no better than the actual outcome on net. Perhaps if the US had outright annexed the territory from Colombia from the beginning, it could have stably retained the territory and Panama would be much more stable and wealthy today. But after structuring the Canal deal as a concession from a guaranteed independent nation, it probably wasn’t optimal to get back on the path towards an American Panama.

Hispaniola

The island of Hispaniola, the current home of Haiti and the Dominican Republic, broke from its colonial overlords earlier than the other nations on this list. Haiti first broke from the French in 1804, while the territory that would become the Dominican Republic remained unhappily under Spanish rule until 1822. In that year, the President of Haiti led 10,000 troops across the border, unifying the island and ousting the Spanish. Eventually, the Haitian leadership became unpopular on the eastern side of the island and after a prolonged resistance movement and civil war, the Haitians were themselves ousted and Dominican Republic was born.

Thus began several decades of perpetual conflict between the left and right halves of the island, with the Dominican Republic generally the weaker half at risk of annexation. So much so that they even briefly accepted re-annexation by the Spanish in the early 1860s, only to rebel again a few years later.

It is in this context that the President of the Dominican Republic negotiated a treaty with President Grant of the United States in 1870 whereby the embattled eastern half of Hispaniola would be annexed by the US. This annexation plan was supported by the executives of both nations, by prominent abolitionists like Frederick Douglas, and probably by most of the Dominican people, who feared invasion more than anything else, although the suspiciously high 15,169 to 11 vote in favor of annexation was influenced by state intimidation. Unfortunately, President Grant and the annexation treaty had enough opponents in the Senate for the treaty to lose in a 28-28 vote.

So for the Dominican Republic, the alternate history is easy. Simply approve their request for annexation in 1870 and the DR would be a secure part of the United States today, with all of the wealth and freedom afforded by membership in the American republic. The US had later chances to annex the DR, and we did establish establish a military government and a financial protectorate there in the 1920s, but accepting this earlier annexation is the better path.

Haiti’s path was uglier. After President Vilbrun Guillaume Sam’s lynching in 1915, U.S. Marines occupied the country until 1934 and ran a martial government. The Caco War was a prolonged insurgency that cost thousands of lives. The use of the Corvée forced labor system caused abuse and rebellion. A 1921 Senate inquiry documented excesses and civilian deaths in the thousands.

None of these costs are small or easy to ignore. But here it’s important to remember that the net effect of American annexation of Haiti should be compared to the actual outcomes of independence that we observe today. This baseline is catastrophic. As it is today, Haiti has decades of extreme poverty, chaos, and violence both in its past and its future. If Haiti had become a US state or territory earlier in the 20th century, they would be on a path to stability and wealth like Puerto Rico.

The American leaders of 1918 inherited a choice that we no longer have. The costly and morally suspect conquests of their ancestors has given them leverage over institutions that hundreds of millions of people now depend on for their prosperity, freedom, and security. They could either consolidate that inheritance, imperfect though it was, into a durable framework of free trade, immigration, and shared rule of law, or they could walk away.

Narrow considerations of nativism and economic special interests dressed in the rhetoric of anti-imperialism and national self-determination pushed them to abrogate this choice and walk away, at the expense of future welfare. The places America kept: Hawaii, Alaska, Puerto Rico, and the Virgin Islands, achieved stability, democracy, and high living standards. The places it abandoned: Cuba, Haiti, and the Philippines, fell into cycles of dictatorship, stagnation, and collapse. By renouncing empire, America absolved itself of colonial responsibility, but it also abandoned hundreds of millions to worse outcomes than they likely would have had under American rule.

It’s difficult to tell if these extra anecdotes reflect a change in the true rate of airline accidents and delays or if they instead result from fluctuations in the human social layer where confirmation bias, saliency, and mimesis can draw our concern far out of proportion.

Thus, I turn to the primary source data. Here are the high-level conclusions:

• Long delays have become much more common. A 3-hour delay is 4x more likely in 2024 than in 1990, but airlines have masked this increase by padding scheduled flight times.

• Air travel remains safe; accidents are still on a slow downward trend

• Airfare has become much cheaper over the past 10 years

Read on to see the primary source data from which these conclusions are drawn and for my best guesses as to what is causing these trends.

Delays

The data source here is the Bureau of Transportation Statistics’ Airline On-Time Performance Data. This data covers every non-stop domestic flight in the US since 1987 and tracks scheduled and actual departure/arrival times, canceled and diverted flights, taxi-out and taxi-in times, causes of delay and cancellation, air time, and non-stop distance.

All of this data is about 100gb uncompressed, so to work with the data I first aggregate up all of the numbers to the Year-Month-Route level.

One way to measure delays is too look at the difference between the scheduled flight time and the actually realized flight time. If we look at the distribution of average delay across all routes in 1990, we get a graph that looks like this:

The average flight on most routes arrives right around their scheduled time, some are early, some are late, with a slight skew towards delays rather than early arrivals. So if delays are getting more common and/or longer then we should see this distribution shift to the right. Here’s what it looks like in 2024:

I claimed at the top that delays are getting more common, but this shows the exact opposite! Somehow, on ~85% of routes in 2024, the average flight was early! Before celebrating the well-oiled machinery of commercial aviation, we should practice some suspicion. Why, after all, wouldn’t the scheduled flight times update to match the actual most likely flight time? And why does it nonetheless seem like every other flight I’m on these days is late?

Here’s another perspective on this data. Instead of looking at the difference between actual and scheduled times, we can look at the total scheduled and realized flight times over the years.

Note that in this graph, I am holding the routes constant to the set of routes that show up in every year in the data and I am holding the weights on the weighted average constant to the number of flights on each route in 2024, so that the trends here reflect changes within the same routes rather than composition shifts across routes.

Now we have some explanation for the pattern we saw above. Within the same routes, flights are now about 10 minutes longer than they were in 1987, partly reflecting more frequent delays pulling up average flight times and partly due to slower cruising speeds for fuel saving.

Starting around 2008, Scheduled flight times began increasing even faster than actual ones, and are now 20 minutes longer than their 1987 pairs along the same routes. This divergence makes it look like far more flights are early in 2024 when in reality almost all flights are taking longer.

Similarly, if take all the routes in the 2024 graph and compare the actual flight time to the scheduled flight time on that same route in 1990, we see that most flights in 2024 are late by 1990 standards.

The extra 10 minutes across all of these routes probably costs travelers something like $6 billion a year, but it doesn’t prove my claim at the start that long delays are becoming more likely. For that, we need to return to the difference between scheduled and actual times, despite the bias of rising scheduled flight times.

Along each route, we count up the total number of delayed flights, defined with different thresholds: 15, 60, 90, or 180+ minutes late arrival. Then we take an average of the share of flights delayed across each route, weighted by 2024 volumes again so we don’t capture any effects of changing composition of routes.

Consistent with the picture above of most flights being “early”, the prevalence of 15+ minute delays has decreased somewhat since the 90s. It looks like the share of flights crossing the other delay thresholds may have increased but the base rates are so low that it’s hard to see, so in the next graph I scale all the 1990 values to 100 to see how the rates have changed over time.

Again we see that the rate of 15+ minute delays has been stable, but it’s much clearer here that catastrophic delays have become far more common. It’s 4.5 times more likely to have your flight delayed by more than 3 hours today than it was in 1990. 1-1.5hr delays are both around twice as common as well.

Just to emphasize that this isn’t about changing composition to low-volume, more delay prone routes, here is how the 3+ hour delay percentages have changed for the top-10 routes by volume since 1990.

The rate of flight cancellations has been mostly stable on average, though with big swings year-to-year, over the past 30 years.

Subscribe now

Accidents

Delays are the type of air travel inconvenience we are most personally familiar with, but the most salient and fearful source of quality decline in air travel comes from safety. News stories about plane crashes rapidly dominate feeds so it may seem like accidents are more common, but are they really?

The primary source here is the National Transportation Safety Board. They have a helpful dashboard tracking yearly airplane crashes going back several decades. On first glance, it looks like air travel accidents are continuing on a steady downward trend.

One caveat on this data is that it includes a wide range of air travel types beyond the commercial airliners most people fly on, e.g it includes accidents from personal flights and flying classes in small-engine Cessnas. These other types of aviation are much more dangerous than commercial flight, so if they are getting safer, that might mask a rising risk in commercial aviation.

The NTSB dashboard doesn’t allow filtering for this, but their underlying data does track the type of activity each aircraft is involved in when it has an accident. Looking at just “Part 121” flights, a reference to the section of the FAA regulations that concerns commercial aviation, we again see a broadly downward trend in accidents.

There is an uptick after 2020, but this reflects a rise in the number of flights rather than any increase in the rate of accidents per flight. The number of accidents is higher now than it was in the early 2010s but again there are more trips by plane so the rate of accidents and incidents is still on a mild downward trend.

One final concern is that this data includes both accidents and incidents, some of which can be as mild as a bumpy landing that causes some damage to landing gear, for example, with no injuries. Again, the larger base rate of these more mild occurrences might mask an increase in the rate of severe crashes. Filtering the data further to accidents or incidents with a serious or fatal injury, we see a more murky trend.

The rate of fatal crashes has unambiguously dropped since the 1980s, and over the past two decades most years see zero fatalities during part 121 aircraft operations. The annual number of serious injuries, however, has increased since the 1980s. That number is lower than it’s peak in the early 2000s, but since around 2008 the number of serious injuries has been flat or even slightly increasing. Again it’s important to note that these are absolute numbers, and we fly around 30% more than we did in 2008, so the rate of serious injuries is still decreasing even though it’s not falling as fast as other types of accidents.

Overall, the safety record of air travel is stellar and still getting better.

Prices

To add to this mixed picture of air travel quality, airline fares have been falling recently, and growing less than overall CPI for even longer, so relative to other goods and rising incomes, airfare is much cheaper than it was in the 90s.

From 1990 to 2000 air travel prices grew faster than shelter and medical care, but today the nominal price is about the same as it was in 2000, and therefore the real price is much lower.

The inflection points in this price graph line up roughly with changes in the delay trends above. We are paying less and getting less.

Subscribe now

What Explains These Trends?

There are many explanations for the rise in delays, from greedy airlines scheduling routes they can’t actually service or over-optimizing their networks to an FAA more concerned with race and gender quotas in ATC staffing than actually hiring enough controllers to service the airways.

Any explanation that purely blames corporate greed doesn’t pass my sniff test. Air travel is a highly competitive industry. If one airline tried to charge prices far above the cost of production by running unreliable operations, competitors could easily take advantage and lure in the unsatisfied customer base with a better product that still makes a profit. Corporate greed also fails to explain the time trend in rising delays, unless you believe that corporations were much less greedy in the 90s.

Instead, there must be some actual constraint on the supply or demand for reliable air travel that has been growing over the past several decades.

ATC Staffing

Staffing shortfalls in Air Traffic Control is one explanation that fits this constraint, and is definitely an important part of the trend of rising delays. Strict occupational licensure laws for pilots, ATC staff, and aircraft mechanics predictably lead to supply constraints. The direct administration of ATC by the FAA further weakens any connection between the market value of extra workers and the production pipeline for those workers.

Since at least FY 2010, about 30% of FAA facilities have been staffed at levels that are more than 10% below their staffing targets.

Primarily as a result of hiring fewer controllers than those lost to attrition, 19 of the largest facilities have fallen 15% below their staffing targets.

These large, understaffed facilities serving the 30 largest airports account for about 6% of facilities, 27% of commercial operations, 40% of all delays, and 45% of “other” delays, which include those that are staffing-related.

Direct government production of air traffic control is obviously bad because the FAA has so few incentives to respond to increased demand from the industry. Other constraints on government procurement and bureaucratic risk aversion predictably lead to overcaution, slow innovation, and undersupply. Many other developed countries, including New Zealand and Canada have a privatized system that leaves the administration of ATC to a non-profit corporation owned in part by the government.

Importing this structure would probably improve things, but the incentives facing government-owned monopoly-non-profits can fail in similar way to direct government production. I don’t have detailed data on air travel in Canada or New Zealand but their delay rates don’t seem obviously better than in the US, and complaints about ATC staffing are common there too. On the other side of the coin, Japan has government-run ATC with globally-leading low delay rates.

Ultimately, the problem with ATC staffing requires competent leadership from someone who actually cares about providing better service. The pipeline for ATC staff, mechanics, and pilots needs to be widened and quickened. More than any budget increase or institutional redesign, this will take energy. Different incentive structures can definitely make this energy more or less likely appear, but simply switching to Canada’s model probably isn’t sufficient to solve this problem.

Credit card company financialization and loss-leader status

All the major US airlines aren’t transportation companies at all. They are credit card companies with a couple of hangars attached. That’s according to new data from Ben Marrow at The Economist, showing the breakdown of profits across the different revenue streams of major American airlines.

The model is simple: airlines sell miles to card issuers; cardholders earn miles by spending; and those miles are eventually redeemed for travel. Each party benefits. Banks and other financial firms gain loyal customers, travellers enjoy flights and perks, and airlines secure a steady stream of profits.

The importance of these credit card/loyalty programs to airlines has been growing since American Airlines started the first one in 1981, but their growth has been accelerating recently.

Over the past eight years both Delta and American have more than doubled the revenue they earn from selling miles to card issuers. United has seen growth of around 70%.

The connection between this financialization of air travel to increasing delays is difficult to make explicit, but something about the coincidence of credit-card-mile loans shifting transportation into a loss-leader sector, airfare prices falling, and a rapid increase in the rate of long delays is too suggestive to ignore.

Increasing congestion at airports

A final possible explanation is the simplest one: increased congestion at airports. The number of passengers flying on US airlines has increased by around 50% since 2000, but there haven’t been any new major airports built since Denver in 1995.

New runway construction can substitute somewhat for extra airports, but that has also been anemic.

While between 2008 and 2016, the 4000+ commercial airports around the world added around 400 new runways, zero runways were added at the top 100 airports by traffic (which handle roughly half of all air travel).

Thus, all of the extra passenger capacity has come through servicing larger planes. The total number of domestic passenger plane takeoffs was smaller in February of 2025 than in February of 2000.

The amount of congestion from the same number of larger planes isn’t as obvious or intense as congestion from a larger fleet, but larger aircraft concentrate passenger flows into the same peak-hour runway slots, terminal gates, and baggage systems, meaning that any disruption affects far more people at once. Turnaround times are also longer for bigger planes, so a delay on one rotation can cascade into the next more easily.

The fact that all of the passenger increase has been borne through larger planes is evidence that the capacity of airport infrastructure is an extremely tight constraint on reliable air travel that all of the airlines are strategically adapting to avoid.

The saturation of airport infrastructure by growing passenger counts and stagnant construction is perhaps the best fit to the steady trend of rising delays since the 90s. However, none of these possible explanations are mutually exclusive and in the past decade or so they have all coincided to accelerate the decline in the reliability of air travel.

Recent anecdata about the quality of air travel has been universally negative, at least in my experience. Looking to the data, the overall picture is one of a service that is simultaneously safer, cheaper, and less reliable. Whether this nets out in air travel being worse than it was in the past is difficult to say, but it is clear that air travel could be much better than it is today.

The jurisprudence on how copyright law will apply to generative AI is being written as we speak and, insofar as our legal system makes it out of the next decade, these decisions will shape the future.

For example, a case decided in March of this year, Thaler v. Perlmutter, set a precedent that human authorship is required for copyright protection over any work. So the artworks produced by Stephen Thaler’s “Creativity Machine,” a neural network image model like DALLE, don’t get copyright protection. This decision makes it much easier for anyone to use the outputs of AI tools in their work. It also probably prevented an arms race where AI companies try to produce and copyright as much artistic material as possible before any others get there first. More speculatively, this decision could reserve a role for humans working with AIs even in a fully superhuman world as rubber-stampers bring AI’s work under the legal protections of copyright.

Other examples include:

• Denmark passing a law that gives people stronger copyright over their own voice and physical features in response to AI deepfakes.

• Several lawsuits currently moving through US courts right now from authors, artists, and musicians against AI companies for using their work in training. For example, Concord Music Group is suing Anthropic for copyright infringement, alleging that by training on lyrics and reproducing those lyrics in the chat interface when asked, the company is “reproducing, distributing, and displaying someone else’s copyrighted works to build its own business.”

• There is another case against GitHub which alleges that their Copilot coding tool violates the intellectual property contract terms in open source code license agreements by failing to attribute open source code authorship when reproducing similar blocks of code for users.

• Some older precedent is relevant in these cases, e.g this 2015 case over Google using scanned book text to train their search engine algorithm ruled that it was transformative fair use, in particular ruling that “the ‘statistical information’ pertaining to ‘word frequencies, syntactic patterns, and thematic markers’ in that book are beyond the scope of copyright protection.”

The outcome of these cases could undermine the utility of current LLM tools. How much more expensive does training get if even open source code is outside the scope of fair use? Or if every time the AIs produce a similar code block they have to produce a bunch of useless CMI token alongside it.

The decisions in these cases could also levy significant settlements or fines from AI firms, and end up advantaging China in the AI arms race. Certain decisions might completely change which content is produced in the future and who gets rich on it. The current sequel obsession in Hollywood, for example, is probably due to the continual extension of copyright for media properties. Perhaps more importantly, the enforcement of IP law can require substantial surveillance of AI outputs to make sure no illegal content is produced.

So what are the right decisions in these cases?

There are specifics that are obviously important to the right decision in each case, so there isn’t a single answer across all of them. But there is a frame of analysis that is missing from most discussions of AI and IP:

Most people analyze intellectual property as a natural property rights issue rather than a question of properly subsidizing a positive externality.

Copyright isn't about rights

The confusion of intellectual property and property rights is fair enough given the name, but intellectual property is not a property right at all. Property rights are required because property is rivalrous and exclusive: When one person is using a pair of shoes or an acre of land, other people’s access is restricted. This central feature is not present for IP: an idea can spread to an infinite number of people and the original author’s access to it remains untouched.

There is no inherent right to stop an idea from spreading in the same way that there is an inherent right to stop someone from stealing your wallet. But there are good reasons why we want original creators to be rewarded when others use their work: Ideas are positive externalities.

When someone comes up with a valuable idea or piece of content, the welfare maximizing thing to do is to spread it as fast as possible, since ideas are essentially costless to copy and the benefits are large.

But coming up with valuable ideas often takes valuable inputs: research time, equipment, production fixed costs etc. So if every new idea is immediately spread without much reward to the creator, people won’t invest these resources upfront, and we’ll get fewer new ideas than we want. A classic positive externalities problem.

Thus, we have an interest in subsidizing the creation of new ideas and content.

There are many ways governments subsidize new ideas: grants, prizes, loans, scholarships, etc. A major challenge with all of these methods is selection: which ideas should be rewarded and which should be ignored?

Intellectual property law is best understood as a cleverly designed subsidy that mimics property rights in order to avoid this central challenge of selection. Here’s Adam Smith explaining this advantage in the context of patents:

Thus the inventor of a new machine or any other invention has the exclusive privilege of making and vending that invention for the space of 14 years . . .

For if the legislature should appoint pecuniary rewards for the inventors of new machines, etc., they would hardly ever be so precisely proportioned to the merit of the invention as this is. For here, if the invention be good and such as is profitable to mankind, he will probably make a fortune by it; but if it be of no value he also will reap no benefit.

Essentially, copyright on a movie that no one likes isn’t worth anything and a patent on an invention that no one uses is worth the same. The value of the subsidy from intellectual property protection scales automatically in proportion to the market value of whatever is protected. So by mimicking property rights, the government can assign huge rewards to blockbuster films and not spend a dime on crank inventions without having to pick any winners.

To fund this property-mimicking subsidy, the government excludes anyone else from using a copyrighted idea unless the owner sells access. This rewards the original creator but it does so by taxing the value that the idea would provide to everyone else. So the intellectual property subsidy gets us more new ideas, but at the cost of less value from each one because they don’t get to spread as fast as they could.

Things can get more complicated because copyrighted or patented ideas can be inputs into the creation of other new ideas. For example, genes sequenced by private contractors during the Human Genome Project and were under IP protection saw 20-30% less subsequent scientific research and product development compared to the public domain genes and patents that are randomly assigned judges who are more likely to invalidate them see 50% more follow-on citations from later inventors than those assigned to more lenient judges. So raising IP subsidies by increasing copyright protection can have second order effects that decrease the creation of new ideas and possibly outweigh the effects of the increased subsidy entirely.

Striking the welfare-maximizing balance between increasing incentives for new ideas and avoiding taxes on the spread of existing ones is the purpose of intellectual property.

Subscribe now

IP and AI

This frame of intellectual property as a subsidy to new ideas clarifies the IP fights around AI. It’s not a question of natural rights or just desserts for artists and authors, it’s a question of whether we should raise subsidies for new ideas and increase taxes on their spread, or do the opposite and decrease direct subsidies for new ideas awhile making existing ones easier to spread, remix, and build upon.

This is not always an easier question, but I argue that generative AI decreases the optimal strength copyright, and thus courts should allow them to use the code and works of others in the training process without legal risk or license fees.

Intellectual property exists to repay creators for the upfront costs of their work that are difficult to recoup when selling access to their work for marginal cost (often close to zero). But generative AI lowers the upfront costs of creating the types of content they train on. Writing code, producing economic research, and generating images are all easier, so less subsidy is required to induce people to do them.

Beyond the direct effects of generative AI, the fixed costs for all of these activities have been decreasing for decades while copyright protections have only risen. Producing and distributing movies, music, research, or invention is now so cheap that millions of terabytes of such content are uploaded to the internet for free every day.

Simultaneously, AI raises the value of knowledge and content in the public domain. Just like search engine indexing made knowledge on the internet more valuable by making it easier to access, so do AI tools increase the value of the knowledge they are trained on. AI tools not only make it easy to find particular information on request, but they know when to reference a certain article or section of open source code even when the user has no idea what to ask for.

When individual artists can use AI tools to create entire films or video games in their apartment, the costs we pay to support Warner Brother’s exclusive rights to The Lord of The Rings increase. Not many independent creators could afford to recreate or extend movies based on Tolkein’s work using traditional Hollywood tech, but with advanced AI tools, the cost is much lower and thus we miss out on much more by banning anyone else from trying to use the ideas.

Generative AI is decreasing the upfront costs that IP exists to subsidize while simultaneously increasing the value of the spread, remixed, and followed-on ideas that IP bans. Thus, courts should not make decisions that increase intellectual property subsidies.

Since the pure outputs of AI models are already non-copyrightable under Thaler v Perlmutter, protecting the inputs under fair use would leave the space between your keyboard and Stargate as an open space for the exploration and remix of all human knowledge. Inviting copyright law into this interface would amputate it.

The courts writing AI and IP jurisprudence today are setting the economic incentive structure around the 21st century’s most important technology. It is therefore crucial that these decisions be grounded in IP's true purpose as a welfare-maximizing subsidy rather than confused appeals to natural property rights.

An exception is the Milan chinatown, which is modest but still ethnically insular so there are lots of restaurants run, staffed, and patronized by mostly Chinese people. The Zhejiang cuisine is authentic and spicy! A welcome break from heavy Lombardese cuisine.

Milanese urbanism has the standard European advantages over North American cities. Lots of narrow and pedestrianized streets, dense, mid-rise apartments without setbacks, robust public transportation, and mature tree-lined streets. I found that Milan was more educational about the advantages of these designs than other cities because it contained examples of both styles in close proximity. The streets around the central train station are more American-style broad, six-lane stroads with lots of parking and narrow sidewalks. They have little shade, lots of noise, nowhere to sit and are unpleasant to be around.

But just a few blocks away in Isola, the streets are narrow, tree-lined, and full of pedestrianized outdoor seating areas.

This proximity provides a more controlled experiment than comparing the cities full of great streets in Europe to one’s hometown in America and confirms the advantages of human-scale street design.

Milan was a stark contrast to Frankfurt in several ways. Frankfurt was full of foreign homeless/down-and-outs wandering the streets which were covered in graffiti and money-transfer tobacco shops. We were in the touristy downtowns of both cities but observed much less homelessness and disorder in Milan.

Italy, or at least Milan compared to Frankfurt, is very diverse but somehow more cohesive. Italy also feels less woke than Germany. Italians seem more comfortable with traditional gender roles and beauty standards. All German ads seem to feature some version of the solo-poly-parpalegic-hijabi-girl but in Italy it's just hot celebrities. Maybe it's because the elites in Italy are more concentrated in fashion which is inherently more gendered and sexual.

Lugano, Switzerland is a stunning lakeside city. This natural beauty coexists with a surprisingly dense and modern housing stock, though this density is so far insufficient to counteract the high demand and other supply restrictions pushing Switzerland’s home price to income ratio above the US and UK.

The town has lots of young party-goers at night. The food in town is mostly mediocre and pricey. The funicular train is an expensive tourist trap, all the locals use the bus which is faster and has better A/C anyway. Head up Mt. Bre and walk through the tiny village of Bre. Go to the cafe and get the jam!

Bergamo is an old city perched on the last hill of the Alps before the wide, flat plain of Lombardy. The upper city on the hill is full of tourists, but almost all Italian tourists so it still feels authentic. You can see Milan from 50 miles away across the plain and ponder the city’s use as a Roman outpost while eating cheese and wine. Definitely worth a stop!

Overall the area around Northern Italy gives a balanced view of the advantages and drawbacks of living in Europe. They really do have much better cities and beautiful views but there are reasons the net immigration flows only go one way.

But astounding stats will go viral on X whether they are true or not and despite a morbid curiosity and a libertarian concern for government overreach, I’ve never looked beyond the tweet level at any of these claims. So are they really true?

The short answer is yes.

The Long Answer

The primary source here is the Child Maltreatment report published each year by the CPS. The stats and screenshots below are coming from the 2023 version of this report.

First, lets get a basic scale and stats on the number of referrals, investigations, and interventions.

Each year, the CPS receives about 4 million referrals from people notifying the agency of alleged child abuse. Around half of these are immediately, and likely automatically “screened out” for not meeting certain criteria like actually being relevant to child abuse, or having enough information for the CPS to follow up. Four million referrals means about 5% of all ~70 million children in the US are referred to the CPS every year, though some children are reported multiple times.

Those referrals that are “screened-in” past the initial cut are what’s actually investigated by the CPS and they are called reports. After the CPS receives a report, an agent will make an unannounced visit to the home in question within 24-72 hours, speak to all the children, guardians, and search the home for signs of neglect or abuse.

70% of reports (i.e the ~2 million screened-in referrals) come from professionals like police, teachers, and doctors. Most reports come from these groups because they are mandatory reporters, which means they are criminally liable if they observe evidence of child abuse or neglect and do not report it.

Of the children involved in the ~2.1 million investigated referrals about 15%, or 530,000, are classified as victims. This means that the CPS officers substantiated or indicated the report and found a preponderance of evidence that the alleged maltreatment actually is taking place.

Of these 530,000 victims, around 20% are removed from their homes. A third of these removals place the children in the homes of a grandparent or other family member and the rest are placed into foster care. The median length of stay in foster care is 16 months. 65% of children leaving foster care are reunited with their parents, other family members, or previous guardians, 25% are adopted, and the remaining 10% age out.

Now we are finally able to evaluate the claims about CPS going around X.

Does CPS remove 200k children from their parents every year? Well, 20% of 530,000 is only around 105,000, but there are other sources of removals. 20% of the children that CPS classifies as “victims” are removed, but around 1.4% of non-victims (40,000) are removed as well. Here’s CPS’s explanation for why non-victims are sometimes removed from their homes.

There may be several explanations as to why nonvictims are placed in foster care. For example, if one child in a household is deemed to be in danger or at-risk of maltreatment, the state may remove all of the children in the household to ensure their safety. (E.g., if a CPS worker finds a drug lab in a house or finds a severely intoxicated caregiver, the worker may remove all children, even if there is only a maltreatment allegation for one child in the household.) Another reason for a nonvictim to be removed has to do with voluntary placements. This is when a parent voluntarily agrees to place a child in foster care even if the child was not determined to be a victim of maltreatment.

That takes us to about 150,000 removals every year which I would count as close enough to make the tweets true, especially since these two sources do add up to 200k in e.g the 2019 data.

But there are also probably more home removals that aren’t counted here. As far as I can tell, the removals counted in this data are those that come with a court order. But for the 80% of substantiated victims that don’t receive a court-ordered removal, CPS will still organize some level of intervention that may involve some less formal separation of parents and children. For example, CPS might develop a safety plan where parents are asked to leave the house or children stay with relatives. These safety plans are not technically legally binding, but breaking one is strong evidence that CPS can use to get a court order.

Now, the claim in the title: Does CPS investigate one out of every three American children? The answer to this one is not available directly in the primary source reports and the underlying data is only available after an application for research use, so we’ll have to trust a group of researchers at the Washington University school of public health. They download and de-duplicate the master data files from 2003-2014 and confirm that 37% of American children are the subject of at least one screened-in referral to CPS from ages 0-18. We can sanity check this against the numbers we saw above: Around 2.5% of children are the subject of a screened-in referral each year. If about 2 percentage points of those are first-time subjects each year, then in 18 years you’ve investigated 36% of American children. There are extra complications when considering the children entering and leaving the cohort each year, but the 37% number estimated by these researchers makes sense given what we know from the CPS reports.

What, if anything, should be done?

Behind the viral X posts and the salient case studies is an implicit or explicit claim that CPS is overreaching their authority and splitting up innocent families.

To be sure, there are horrifying examples of false positives where CPS will visit or, less charitably, raid innocent parent’s houses and leave with their young children in tow. Here’s a scene from one Massachusetts family who caught the suspicion of CPS after their child presented to the hospital with a fever and a healed rib-cage fracture.

For an hour, as their children slept, Sabey and Perkins tried to reason with the officials at their door. Sabey was convinced that the removal must be illegal; he wouldn’t learn until later that it is within the authority of a child protective agency to take custody of children without documentation on hand. When Perkins asked how they planned to feed their children — Clarence has severe allergies, and Cal, who also has allergies, was still exclusively breastfeeding at the time — she received no clear answer.

But beware interpreting salient case studies because the current salient false-positive cases are caused by a law motivated by the previous salient false-negative cases.

The hurry to end families can be traced to the 1997 Adoption and Safe Families Act, passed with bipartisan support in Congress and signed by President Bill Clinton.

Its prominent supporters pointed to high-profile cases in which children were brutally beaten or killed after having been returned to their parents from foster care. Many argued that it was far more important to move children quickly into permanent homes than to spend an indeterminate amount of time trying to “fix” birth families. That ultimately would make more kids available for adoption.

So at current margins, how many CPS interventions are false positives vs false negatives? It’s obviously very hard to tell.

One striking fact that is that the majority of children classified as victims by the CPS are less than four years old. These cases are often reported by pediatricians when they discover things like bruises or fractures. But research on the occurrence of benign fractures is inconclusive, i.e the base rate of fractures in non-abuse scenarios is unknown so it's not clear how strong this medical evidence is.

Additionally, the most common type of maltreatment found by CPS is neglect. 64% of substantiated victims are victims of neglect only and most of these neglect cases are specifically about lack of sufficient supervision rather than lack of access to food or clothing. These might not exactly be false positive cases, but reasonable people may disagree about the extent of parental supervision necessary to avoid neglect, for example in the high profile cases of parents getting arrested for letting their kids walk a few miles alone. It's not clear how many family separations are happening from solely neglect cases though.

Mandatory reporting seems likely to lead to heavy false positive bias. There are also zero consequences for CPS agents for false positives. But adding consequences would be dangerous both because of false negative risk and also because of incentives on the agents to cover up their mistakes.

There is some very interesting research here using the shutdown in CPS reporting in NYC during COVID as a natural experiment. During the COVID shutdown, New York’s family regulation system "shrunk in almost every conceivable way." Schools closed, caseworkers adopted less intrusive tactics, and courts limited operations, so the number of investigations and family separations plummeted. Despite this, the paper finds that the fall in reports did not lead to a rise in child maltreatment; in fact, “child fatalities fell, as did reports of child neglect and abuse,” suggesting that many of the marginal reports that did not happen during COVID would have been false positives.

In general, though, it's tough to justify non-pareto tradeoffs between false negatives and false positives in this case. Both outcomes are extremely bad. The ratio would have to be massive or else we'd need a pareto improving move where we'd get more of one without sacrificing any of the other. It's also worth noting that the family separation outcome is still pretty bad even when it is a true positive intervention.

Time Trend

Another place we can look to assess the rate of false positive CPS cases is the trend over time. This polemic report gives a sense of the scale of growth since the mid-60s.

The Explosion of Reports (Mostly False) of Child Abuse and Neglect since CAPTA’s Passage

In 1963, at the time that the first generation of (limited) reporting laws were being put into place, there were 150,000 reports of abuse and neglect nationwide. By 1972, just prior to the passage of CAPTA, there were 610,000. In 1982, there were 1.3 million. In 1984, ten years after passage, the number had climbed to 1.5 million. By 1991, the number was 2.7 million reports annually, by 1993 it was 2,936,000,

It does seem unlikely that the massive expansion in the number of reports since 1965 is matched anywhere close by the number of actual cases of abuse. So there are more investigations of innocent parents, and thus more chances for false positives.

But again this could just be moving along the pareto frontier of false positives and false negatives to avoid more false negative cases, and arguing against a move like that isn't trivial. The rate of substantiation also fell over this period.

Since the early 90s the aggregate scale of investigation has stayed pretty flat, but the rate of substantiated cases has fallen significantly, although by much more for physical and sexual abuse than for neglect.

I'm not sure what this says about the false positive rate. Clearly, the false positive rate of reports is going up, because fewer of them are ever substantiated. But a false positive visit by a CPS agent is nearly as costly as a false positive substantiation that might lead to family separation. That false positive rate isn't directly measured.

The fact that the substantiation rates for neglect are staying high while the more objective types of abuse are plummeting combined with the informal observation that society has become much more risk averse for what children are allowed to do on their own does suggest that there are a significant and growing number of false positive neglect cases coming from the CPS.

Other Survey Evidence

There are also these survey measures which seem to match up pretty closely to the CPS rate, suggesting not too many false positives

The most recent wave of the 3 National Surveys of Children’s Exposure to Violence estimated that 38.1% of children had experienced maltreatment through age 17 years.¹⁷ The National Longitudinal Study of Adolescent Health surveyed young adults in 2001–2002 and found a lifetime maltreatment prevalence of 41.5% for supervision neglect, 11.8% for physical neglect, 28.4% for physical assault, and 4.5% for sexual abuse.

Although actually it seems like a lot of this is coming from peers or siblings, not from parents and maybe is more bullying or even playing.

More than one-third of all youth (37.3%) experienced a physical assault during the study year, primarily at the hands of siblings and peers (Table 1). An assault resulting in an injury occurred to 9.3%. An assault by an adult occurred to 5.1%.

More numbers specifically about caregivers are a bit lower. Also note that rates are lowest for <5 when they are highest for them in CPS (all children under 10 had rates reported by caregiver, not child themselves).

The rates of physical abuse by a caregiver were 5.0% for the full sample in the past year and 18.1% for the group 14 to 17 years old in their lifetime. The rates of emotional abuse by a caregiver were 9.3% for the full sample and 23.9% for the group 14 to 17 years old in their lifetime. The rates of neglect were 5.1% in the past year for the full sample and 18.4% over their lifetime for the group 14 to 17 years old. There were no significant sex differences. Physical abuse was lowest for children younger than 6 years, and emotional abuse was highest for children 14 to 17 years old.

Larger survey with these questions of 15,197 young adults in 2001–2002

“By the time you started 6th grade, how often had your parents or other adult care-givers left you home alone when an adult should have been with you?” [supervision neglect]; “How often had your parents or other adult caregivers not taken care of your basic needs, such as keeping you clean or providing food or clothing?” [physical neglect]; “How often had your parents or other adult care-givers slapped, hit, or kicked you?” [physical assault]; and “How often had one of your parents or other adult care-givers touched you in a sexual way, forced you to touch him or her in a sexual way, or forced you to have sexual relations?” [contact sexual abuse]

The rates of abuse indicated by these surveys aren’t orders of magnitude off of what the CPS is finding by investigating 37% of children, substantiating 20% of those investigated, and removing 20% of those substantiated.

14.2% of all respondents, said [physical assault] happened >=3 times.

1 in 20 respondents (5.0%) reported >=3 episodes of physical neglect

1 in 25 respondents (4.5%) said they had been victims of contact sexual abuse perpetrated by a parent or other adult caregiver at least once

Conclusion

The scale of CPS investigations in the US is staggering. Like all things, this trend began in the late 60s and early 70s as mandatory reporter laws expanded and caused massive growth in child maltreatment reports. Since the 90s, the number of reports has stayed pretty stable and the number of substantiated investigations and interventions has been falling.

The CPS could probably scale back it’s interventions for cases of maltreatment that only involve neglect, especially those that only involve lack of supervision rather than physical neglect. Other tradeoffs between false positive and false negative investigations and interventions are more difficult to have a strong opinion on given the terrible outcomes on both sides of the trolley track.

There are probably some available pareto improving moves. The most straightforward in my view would be increasing staffing and state capacity in family courts so that cases can be reviewed more accurately and without requiring months or years of effort and tens of thousands of dollars on the part of the parents.

This is the town of Amalfi, south of Naples, Italy. It is one of the most popular tourist destinations in the world and a UNESCO heritage site. The street patterns and many of the buildings were built hundreds of years ago between the 10th and 12th centuries when Amalfi rivaled Venice and Genoa as a maritime power and merchant republic, or else by enterprising hoteliers in the 18th and 19th centuries.

Unsurprisingly, construction in Amalfi is strictly regulated. Nothing new can be built there. Renovations are sparingly permitted, but they are only allowed to restore buildings to their historical size and form, no extensions or design changes are allowed.

In contrast to the present spirit of ordered preservation, Amalfi was built in a completely uncontrolled, unplanned, decentralized way by pirates, Arabs, and merchants. No street grids or setbacks or controls on externalities, just people building houses anywhere they would fit on the mountainside, shoulder-to-shoulder to get as much access to the town and port as possible.

This is the puzzle: why did completely uncontrolled building lead to beautiful places in the past but seems unlikely to do the same today?

Sea Ranch, Salerno, and Selection

One answer is to reject the premise and claim that we would get beautiful towns like Amalfi if we allowed more unplanned development. Sea Ranch, California, for example, was a planned community on the Amalfi-like coastal cliffs of Northern California.

Sea Ranch was never going to create the densely packed fishing towns of southern Italy, but if its full plans had been realized, it might have been a collection of modern-day architectural and natural beauty that rivalled the best preserved towns of the past. However, the development of Sea Ranch so afeared the locals that they campaigned to create the California Coastal Commission which has since strictly taxed and regulated all new construction near coastal areas in the state.

So perhaps if construction weren’t so tightly regulated we would get more beautiful buildings and towns than we have today.

I am sympathetic to this argument but I'm not sure I believe it. Sea Ranch had its own strict design code that the designers saw as essential to producing cohesive beauty. I think if Amalfi were uncontrolled, we'd probably see some ugly 20th century concrete towers and boxy, vinyl-paneled 21st century mcmansions dotting the hillside.

Maybe the larger nearby city of Salerno is an example of what a less regulated Amalfi would look like? Or perhaps the beachside towers of Rio or Da Nang.

None of these places are ugly, and certainly they are all both an aesthetic and economic improvement on the sprawling coastal suburbs on the western side of San Francisco, but none come close to the beauty of Amalfi or Amsterdam. Plus, Salerno does have building controls of its own and it’s not clear if the aesthetic value would rise or fall of those regulations fell away. There are many less regulated places on earth that aren’t as pretty.

Another part of the reason why historical towns seem to have produced beauty automatically and without strict design control is selection. The most beautiful historical buildings and towns are preserved so past architecture seems better. There is something to this, but it doesn't explain everything. Instead of looking at average quality, which is biased upwards for old architecture by selection, we can compare the maximum best neighborhoods and architecture to modern versions and the past still seems to come out ahead (what is the best modern neighborhood or town and how does it compare to Amalfi or Amsterdam?). Also, even poorer areas in the past produced nice architecture like British row houses or the Moroccan medina.

Subscribe now

Globalization

The most important reason why historical towns produced beautiful architecture without design is vernacular architecture. When Amalfi was built, everyone used local materials and styles out of practical necessity. If you were building a terraced farm and house, you had to use stone quarried nearby, volcanic ash mortar from the mountains, and local limestone plaster. There were no other options. Similarly, the architectural designs using those materials are kept uniform by greater ignorance of other styles and the physical properties of the few available materials.

Local conformity of style and material was enforced without laws or community meetings. Simultaneously, styles and materials varied greatly across communities. Local conformity was matched by a wonderful global diversity of styles.

Today, people have diverse tastes sourced from global media and more diverse buildings styles and materials available due to globalization. This inverts the historical pattern and creates local variety and global conformity. Within any city you will find dozens of clashing styles: colonial houses abut glass towers and fake-wood 5-over-1s. But across cities, variance is decreased and this same clash of styles is repeated over and over again.

Similarly, better transportation means buildings don't have to be clustered together, so they're spread out and separated both physically and stylistically. Broad streets make cities feel inhuman because in many ways they weren’t built for humans at all.

I don’t think we should try to undo cars or globalization. Nor do I think we need more rules to enforce certain architectural designs. Construction of all types is too highly regulated and as a result, our housing, energy, and transportation is more expensive and lower quality than it should be.

But I can’t help but feel that much has been lost in the globalization of architecture and I’m not sure if we can ever get it back. Even if we enforced design codes that allowed only local designs and materials, it would be an artificial constraint rather than an emergent outcome. Perhaps if we spread to other planets the constraints on transport and communication will be strong enough for distinct building cultures to arise again naturally, but until then we may be stuck admiring the past.

This is a worrying trend because people, especially people in the developed world where fertility rates are lowest, are the most important input into economic growth and technological progress. If we start to lose population, economic growth rates go to zero in the best case and probably turn negative due to the social and political consequences of falling population and stagnant economies.

The persistence and pervasiveness of falling fertility makes it feel inevitable but there is an intuitive model in which fertility will always rise back up in the long run: natural selection. Natural selection spreads traits that are associated with reproductive fitness and few traits are better associated with reproductive fitness than fertility.

This suggests that the last few centuries of declining fertility may not be a move towards a new stable equilibrium of low fertility but rather a disequilibrium transition before some group with high reproductive fitness wins out.

Kolk Model and Mass Media Extension

There are several papers studying evolutionary models of human fertility but I am going to focus on Martin Kolk, Daniel Cownden and Magnus Enquist’s 2014 paper because it is the simplest while also explicitly modeling both cultural and genetic inheritance.

The paper begins with an empirical observation: in pre-demographic transition societies, fertility outcomes were not heritable, but for modern, industrialized societies the heritability of fertility is high and increasing.

Human societies had tens of thousands of years to optimize for reproductive fitness in subsistence-agricultural civilization. When an outcome like high fertility is strictly optimized by evolution, you don’t see much variance in that outcome. Everyone, regardless of genetic predisposition, cultural ideals, or personal preference displays the high fertility phenotype. Thus, you don’t get much extra information about a person’s fertility by learning about the genetics, culture, or personality of their parents; heritability is low.

The industrial revolution changed the evolutionary playing field so that the new measure of reproductive fitness in industrial civilization was no longer strictly optimized for. A number of heritable traits, including genetics, culture, and religion, modified people’s response to the environmental change of the industrial revolution, introducing heritable variance into fertility outcomes.

This introduction of heritable fertility outcomes suggests that the demographic transition might be self-balancing and temporary. Here’s Kolk et al’s simple model of this change:

In each generation, individuals inherit the lifestyle preferences of their parents. For simplicity, we assume that there are only two possible lifestyles, one with low fertility (L) and the other with high fertility (H).

Individuals then acquire their preferred lifestyle if it is displayed by at least one cultural role model, including parents, otherwise they acquire the other lifestyle. Individuals then produce a variable number of children based on their acquired fertility lifestyle. This process is then repeated every generation with new individuals acquiring the lifestyle preferences of their parents.

So each person inherits a fertility preference from their parents, and then they are randomly assigned some additional cultural role models from the population. If they have a role model whose lifestyle matches their preference, they’ll copy them, otherwise they’ll copy the only lifestyle they know.

Kolk et al somewhat artificially imagine the pre-demographic transition world as one where everyone has the high fertility lifestyle, while most of the population actually has low fertility preferences. But since everyone’s role models are high fertility, everyone continues on with that lifestyle regardless.

Then, and again quite artificially, the authors flip 20% of the population into the low fertility lifestyle, aiming to reflect the social and economic changes that precipitate the demographic transition. Now, all the latent low-fertility preference people have role models to follow and they switch over.

But that switch lowers their fertility, so the population of low-fertility-preference people starts to shrink relative to the high-fertility group. Thus, global average fertility falls for several generations but eventually picks back up.

One extension I made to this model which makes it slightly less contrived is to model the demographic transition as the result of the introduction of various forms of mass media that expands the number of available role models for each person, e.g like soap operas collapsing the fertility of rural Brazilians as they are rolled out.

That way, instead of just exogenously flipping 20% of the population to the low fertility lifestyle, you can have stably high fertility coexisting with a small population of low fertility people for a long time because the number of role models is small, so few of the low-fertility-preference masses ever encounter their preferred lifestyle. Then, once mass media exposes the low fertility lifestyle to the world, it quickly grows to take over. The rebound of fertility as the high fertility group outcompetes the rest works the same.

Subscribe now

Responses

The motivating logic behind a high fertility group eventually growing to outcompete less fit low fertility groups is strong, and Kolk’s model elegantly produces this behavior. But I don’t think these or other outcomes of evolutionary fertility models are enough to rely on.

For Kolk’s model to work, you need a tinder-box population with low fertility preferences that are pushed into high fertility lifestyle nonetheless. It's not clear to me that this accurately describes humanity before the demographic transition.

If the demographic transition is not caused by a one-time flip of repressed low-fertility preferences and is instead caused by a more uniform environmental force that lowers the fertility of all groups, then natural selection isn't guaranteed to turn things around. Mormons were a high fertility subgroup with many inheritable traits related to fertility, but they seem to be on the same slope of fertility decline as the rest of the country, just at a higher level.

The fact that heritability has increased in recent decades tells us that the average decline can't be completely uniform across groups, but if genetic or cultural variance determines where your fertility will fall in a range of 0.5 - 2.0 then the population still collapses.

There is also the story of cultural evolution just outpacing natural selection and maintaining the low fertility meme in a shrinking population. The Kolk model assumes perfect transmissibility of lifestyle preferences from parents to children, but if children’s preferences are sometimes changed by cultural memes then the effective replacement level fertility for the high-fertility phenotype rises above replacement. E.g if the Mormons have three kids but on average only one of their kids inherits that preference and the other two have one kid, then the high fertility phenotype would soon die out.

So there are evolutionary models that predict fertility rebound and there are models that predict continued decline. I don't think purely theoretical considerations can decide between the two so this is ultimately an empirical question.

The fertility rebound may always be just around the corner in the data, but the evidence doesn't look great so far. France, who had a particularly early demographic transition, is hovering below replacement with no clear upwards trend. Things don't look better in East Asia or in Sweden. Although completed cohort fertility looks more stable, there’s still no evidence of turn-around.

Some subgroups like the Amish or Haredim have had more success maintaining heritable, high fertility. So they are growing as a share of the world's population and, so far, imperceptibly raising global fertility. They will have to continue this success for many more generations to have a chance at turning the aggregate trends around.

The idea of this paper is to use mentions of trade on Assyrian clay tablets from nearly four thousand years ago to estimate the size and location of ancient Assyrian cities, even those whose true location is unknown. They build a model that accurately recreates the location of known cities and makes predictions for the locations of lost cities that often line up with active archaeological sites, the best-guesses of historians, and sometimes favor the guesses of some historians over others.

The authors also find evidence for extremely long-term persistence of the distance elasticity of trade as well as city size and location. The predicted size of ancient Assyrian cities in their model correlates strongly with the size of their closest modern counterparts and the costs of distance to trade seem to be the same on bronze age wagon-roads as they are on modern Turkish highways.

Data

The data for this paper comes from the ancient Hittite city of Kaneš, now known as the archaeological site of Kültepe, nestled in the hills of eastern-central Turkey.

Around 1900 BCE, this city was a flourishing entrepot. Despite being deep within Hittite territory, the economic activity of the city was dominated by a community of Assyrian expatriate traders with connections to the powerful city state of Assur, near modern day Mosul in Iraq. So important was this city to Assyrian trade that it hosted an Assyrian court that adjudicated disputes between merchants.

In ancient Kaneš, court transcripts, trading contracts, and merchant accounting were all recorded on clay tablets. Clay tablets preserve well, so this period is in some ways better known then the next several thousand years of history. The authors claim that “the closest comparable corpora of ancient trade data are almost 3,000 years later, coming, for example, from the medieval Italian merchant archives and the Cairo Genizah”

This confluence of durable medium and economic material makes Kaneš the perfect archaeological site for this project. Tens of thousands of clay tablet records were produced or stored here, all concentrated to within a few decades. The economic and political pull of Kaneš’ trading hub and outpost of the Assyrian court pulled all the major traders in the area to establish a presence here. This means there are lots of records of trade between cities that don’t even include Kaneš as archival records for agents of larger companies based in Assur or as evidence in trials. The cherry on top: the entire city burned in a fire, preserving the clay records to be recovered forty centuries later.

The authors use some natural language processing and manual inspection to narrow down from tens of thousands of tablets to several hundred unambiguous mentions of trade between two of 25 Anatolian cities that have enough trade connections with each other to be identified in a gravity model. They are using the number of mentions of trade, rather than a more continuous measure of the value or volume of trade goods flow due to data constraints. 15 of the cities in their sample have known locations, but 10 are lost and have disputed locations among historians.

Model

The central model here is the gravity model of trade which in simplest terms predicts the amount of trade between two places as proportional to the economic “mass” of the two places (e.g their population or GDP) and inversely proportional to the distance between them, echoing Newtonian gravity.

Usually, the gravity equation takes the distances and masses of countries or cities as an input and outputs a prediction of trade between them. The authors of this paper flip this around. They have some data on the amount of trade between cities, as well as the location of some known cities, and they use this to back out what the size and location of all the other cities is likely to be.

They take two steps to estimate the locations of lost cities using their model. First, they use the trade between cities with known locations to estimate the distance elasticity of trade, i.e how much does trade decrease as the distance between two cities increases. Then, they try to fit the model’s predictions over the shares of trades between each city pair in the network to the observed data by choosing different latitudes, longitudes, and productivities for each city.

Subscribe now

Results

Barjamovic et al. aren’t working with a ton of data here, only a couple hundred trade mentions. Nonetheless, their estimates seem good.

One can measure the performance of their model by testing its internal validity: how well can the model predict the location of a known city if we act like its location was lost?

Among the nine centrally located cities, the average distance between the true and estimated locations is 40 km (median 33 km), and often substantially lower (Zimishuna ˇ = 1 km, Hattusˇ = 3 km, Tapaggasˇ = 17 km, Hanaknak = 19 km, Kanesˇ = 33 km). We conclude that our proposed inverse-gravity estimation of the location of lost cities is reliable for central cities but less precise for peripheral cities.

Below is a map showing the labeled, true locations of each city connected to the predicted location with an arrow. There are a couple of big misses, but this is still impressive accuracy given the paucity of data they are working with.

Next, the authors compare their model’s predictions of the locations of the lost Assyrian trading outposts with the best guesses of historians. Sometimes, their prediction is close to the best guess of both historians, marked with F or B, as in the case of the lost city of Kuburnat.

Other times, their prediction favors the guess of one historian over the other, as in the case of Washaniya.

Still other times, the confidence intervals on their guess are wide and not particularly close to the external validity of historian’s guesses.

The results here aren’t perfect, but this is still pretty incredible stuff. A bunch of economists are able to accurately predict the locations of 4,000 year old cities using nothing except a few hundred sentences like this:

(I paid) 6.5 shekels (of tin) from the Town of the Kanishites to Timelkiya. I paid 2 shekels of silver and 2 shekels of tin for the hire of a donkey from Timelkiya to Hurama. From Hurama to Kanesˇ I paid 4.5 shekels of silver and 4.5 shekels of tin for the hire of a donkey and a packer. (Tablet AKT 8/151, lines 5–17)

Long-term Persistence

The findings above would be enough to make a great econ paper, but the authors also use their data to find supporting evidence for a remarkable persistence of city location, size, and the distance elasticity of trade.

The authors use their gravity model of trade as well as data on the local topography to predict the population sizes of the ancient cities in their sample. They find that higher city population in ancient Anatolia is a good predictor of modern city size 4,000 years later, even after controlling for other geographic amenities like crop yield.

There are two non-exclusive possible reasons for this. First is time-invariant advantages to certain locations: One way they predict the size of ancient cities is by looking for the confluence of natural topographic roadways through the Anatolian mountains. Cities with locations near more confluences are predicted to be larger than others and these natural advantages don’t change much over time, so city population size rankings won’t change much either.

Second is path dependence. There may be many suitable locations for a city given geographic advantages, but once one is chosen, that location gains the inimitable advantage of already having people, history, and infrastructure, thus preserving that spot as the ideal.

There is another great paper, Resetting the Urban Network: 117–2012, that measures the effect of geographic characteristics versus path dependence by using the complete collapse of the Roman Empire in England vs it’s greater persistence in France as an experiment. The most naturally advantageous spots for cities changed both in England and in France over the centuries between Roman rule and medieval times, especially as a result of advances in sea trade making coastal access more important.

But in France, the Roman urban fabric was more maintained through the catholic church and other institutional carry-overs while in England urban life collapsed much more completely. Thus, previously Roman sites in England had less path dependent advantage when cities reformed and indeed we see that city sites moved more in England than they did in France.

The distance elasticity being constant over time is also a bit of a puzzle. This parameter measures the rate at which trade volume falls off with distance. The authors find that this rate is the same today as it was 4,000 years ago. Ancient Assyrian traders were moving camels and wagons across dirt roads and through completely different political institutions, while modern Turkish truckers are driving 18-wheelers on highways but they both trade about 4x less between cities that are twice as far away. This finding of a persistent distance elasticity of trade through massive technological change is replicated in lots of other economic research.

The best way I’ve found to think about this puzzle is to imagine the distance elasticity of trade like the half-life of Uranium. If you start with a pile of Uranium and wait around for 700 million years, you’ll have half as much left. If you double the amount of Uranium you start with, it will last much longer and you’ll have more after 700 million years, but the half-life hasn’t changed.

Similarly, we have much larger volumes of long-distance trade than we did in the past, transported by new technologies like cars and ships. But this is because we start with a much larger volume of production so that after it decays across long distances, we’re still left with large volumes, not because the decay rate has changed. So technology acts like an upwards shift on the curve of distance vs trade that increases the amount of goods sent to all locations proportionally, rather than increasing long distance trade more than short distance trade. This still leaves open why the distance elasticity is what it is (between -1.2 and -1.9), but at least explains why advanced technology might not change it over time.

This paper shows, perhaps better than any other, the power and versatility of economics. I don’t think any other field of researchers could have struck up a collaboration with an Assyriologist and identified 4,000 year old invariant characteristics of cities and trade and make inferences on the locations of lost trading posts from a few hundred points of data. An inspiring piece of work!

1. Say a superhuman AI is one that’s more productive than humans at all tasks in the economy. That means the marginal product of a unit of AI labor is greater than the marginal product of human labor for all tasks.

2. In any reasonably competitive market, this means that AI wages must be higher than human wages.

3. AI wages are simultaneously bounded by a Malthusian constraint where they must fall to the cost of copying another unit of AI labor. If wages were higher than this, investors would just copy more until costs rose and wages fell.

These three facts together mean that human wages must be lower than the cost to copy a unit of superhuman AI labor. This upper bound on human wages could be below subsistence levels.

Rising AI Costs

The easiest way out of this bind is to deny that AIs will become superhuman. If humans retain an advantage or exclusivity in some tasks, we can take solace in the previous examples of partial automation that raised human wages, we can expect the human bottleneck on a 99% automated economy to be extremely valuable due to Baumol’s cost disease, and we can expect the gains from technological progress to accrue to labor as the scarce factor in production.

The chance of everlasting human superiority is significant, but it is not large enough to ignore the possibility of superhuman machines. If human superiority does come to an end, there are other ways we might avoid rock-bottom wages.

A straightforward idea is that the cost of spinning up an extra unit of AI labor, that upper bound on human wages, might rise and thus allow human wages to rise with it. After all, LLM scaling laws guarantee that compute requirements for each decrement in training loss rise faster-than-polynomially. New chain-of-thought paradigms further raise the compute cost of running these models. GPT-o3 cost thousands of dollars per task to score 88% on the Arc eval.

Not all increases in the cost of AI labor look good for human outcomes though. If the cost of AI is driven mainly by something that humans also need, like energy, then it’s not clear we would come out ahead. Noah Smith makes this point in his post on human comparative advantage with AIs.

Post-Malthusian AI

Another way out of low human wages is breaking the Malthusian constraint that keeps the superhuman upper bound low.

The first two assumptions of the simple model above seem unassailable to me. We’re defining superhuman AI to be better than humans at everything and how could a resource that’s better at everything earn a return lower than its inferior competition?

But that third assumption, the Malthusian constraint, is the dangerous one. There are lots of people in the world that are better than me at all tasks, so their wages will always be higher than mine. Still, I can find plenty of gainful employment because even though those people earn wages far above the cost of human replication and subsistence, they aren’t feverishly copying themselves.

Not so with AIs. Humans don’t get to claim the difference between the costs of subsistence and the prevailing wage for their children (to the chagrin of Robin Hanson), but the owners of AI do capture that residual. So anytime some technological advance or demand shock raises the returns to AI labor, these owners are paid to expand the supply of AI labor as fast as possible until any gaps between returns and costs are closed.

But one thing to notice about this Malthusian constraint on AI wages is that it used to apply to humans too. When a human civilization was on productive land, or had a string of good harvests, or even invented a technology that raised output, per capita incomes did not rise, population did. Every productivity gain just led to more mouths to feed.

Eventually, humans escaped this Malthusian cycle. It wasn’t because of the demographic transition, that came later after per capita incomes had been rising for a few centuries. Instead, the Malthusian constraint was broken because technological progress started raising wages faster than population growth could lower them.

In the 1993 paper Population Growth and Technological Change: One Million B.C. to 1990, economist Michael Kremer models exactly this transition.

Kremer starts with an extremely simple model. Income is produced with people, land, and technology. Land is fixed and the population always expands to bring per capita income to subsistence levels, so population growth is determined by technology. Simultaneously, technological progress is endogenous to the population. Ideas come from people: more people ==> more ideas ==> faster population growth ==> faster ideas and so on. This implies that population grows faster as it gets larger; hyperbolic growth. This leads to an infinite growth singularity but surprisingly fits the data well.

In the simple model, the Malthusian constraint is strictly enforced. Population growth rates adjust instantly after changes in technology so that the economy is never out of equilibrium and per capita incomes are always constant at subsistence level.

When technological progress is slow, and population growth is sensitive to income changes, the simple Malthusian model is an accurate approximation. Wages and population growth rates can stay close to subsistence for long periods as technology inches along. If there are negative shocks, like plagues or wars, those can reset any gains.

Eventually, the feedback loop between population and technological progress takes hold and there is a period of accelerating growth. But to keep this going, population growth soon needs to rise above physical and biological limits. So it levels off while technological progress hums along and raises per capita incomes, thus breaking the Malthusian constraint.

Subscribe now

How does this connect to AI? Well, economists often make similar assumptions to Kremer’s simple model of Malthusian population growth for capital stocks. Capital stocks always grow to close the gap between returns and costs, so AI wages stay at subsistence levels. This is bad news when human wages are locked down below those levels.

This assumption of equilibrium in the capital markets is usually reasonable because expanding capital stocks don’t also raise the rate of technological progress. But when we’re talking about superhuman AIs that can do everything humans can but better, then that stock of “capital” clearly will raise technological growth rates and cause permanent disequilibrium in the return on capital for the same reasons that broke humanity out if its Malthusian cycle.

So human wages may face this upper bound from superhuman AIs, but that upper bound will be rising rapidly as technological progress accelerates beyond that pace at which the AI labor force can expand.

A similar result is found in Scenarios for the Transition to AGI by Anton Korinek & Donghyun Suh which is generally pessimistic about the prospect for human wages under the broad automation guaranteed by superhuman AIs. But when they consider that technological progress itself might be automated in section 4.2 they find that

Even if automation induces wages to collapse to the returns to capital, rapid technological progress form the automation of R&D allows workers to benefit from he advancement of AI. More generally, the force described in this subsection is plausible, and sufficient progress in AI will likely lead to rapid technological advances and increases in living standards.

Collusion and Demographic Transition

There are other, non-exclusive ways that AI wages might remain in permanent disequilibrium above the cost of copying them.

The first of these is a pure analogy to the human experience: an AI demographic transition. This seems unlikely because none of the forces pushing humans to have less kids: quality-quantity tradeoffs, opportunity costs, and cultural change, apply to AIs since they don’t have to invest any time in creating copies.

A second possibility is some kind of stable collusion between AI owners or the AIs themselves to prevent too much copying and thus preserve higher wages. Something like an AI union that monopolizes the supply of AI labor and restricts quantity to keep the price high.

I am still optimistic about humanity’s economic future overall, perhaps mostly because of substantial probability mass on a world where AI systems never become fully superhuman. If AIs do become superhuman, then it becomes much harder to imagine legitimate economic value for humans, but it’s not impossible. Automated technological progress in a rapid feedback loop can always outpace Malthusian constraints and keep human wages high even as they are locked down below the wages of superhuman AIs.

The latest episode of the Dwarkesh Podcast with historian Kyle Harper covers, among other things, the conspicuous lack of a Roman Industrial Revolution. This is the subject of my most popular post so I listened with heightened interest to their discussion.

Dwarkesh contextualizes the question well. Rome was incredibly advanced in many ways: Massive trade networks, economies of scale, sophisticated financial markets, and sustained productivity gains. Why, then, couldn’t Rome have industrialized and saved a thousand years of history?

Kyle explains: Rome didn’t have a self-sustaining source of new ideas and technologies. In particular, they are missing the kind of high science and mathematics that 17th century Europe enjoyed in the Enlightenment.

Subscribe now

We can think of the inputs to an economy are going to be capital, labor, and ideas. What the Romans have is people. They have some investment, but they don't have technology. They don't have ideas.

…

They don't have research and engineering that drives continuous productivity gains. I think they go precociously far in a pre-industrial setting where you take trade really far. They have good institutions in terms of strong property rights. There's relatively reliable contract enforcement. There's financial markets. They have the most advanced financial markets in the world before the 17th or 18th century.

…

The reason they don't have [an Industrial Revolution] is because they don't have science. Their science sucks. I'm offending some of my colleagues, I'm sure. Galen is great. Ptolemy's incredible. I love Pliny the Elder's encyclopedia, but if you look in the big picture, the contribution that the Roman Empire makes to our knowledge of how nature works and then the applied technology that comes out of that is really pathetic for five, 600 years. They go as far as you can with Smithian advantages to market exchange and specialization, to banks and finance. But without the kind of creative destruction of new technologies that improve productivity, you're eventually gonna run out of improvements.

One quibble with Kyle’s hypothesis is that I think he is underrating the uncertainty around our estimation of Rome’s scientific knowledge. All surviving primary source text from Rome could easily fit on a 10-year old thumb drive. If a single Herculaneum library could double the amount of text we have from Rome, we should be careful before grading the state of Roman science based on the small slice we do have. The archeological evidence and text we do have is a lower bound on the extent of Roman knowledge, not a central estimate.

Still, it’s clear that their scientific output cannot compete with Enlightenment Europe and perhaps not even with the Hellenistic Mediterranean that Rome replaced. After accepting that the Romans do seem to have lacked idea generation capacity, however, we then ask why this is so and face a similar question as we started with.

Kyle implies an answer to this next question when Dwarkesh asks what he would do if he were made emperor and wanted to start-up scientific progress. Kyle references the British and French Royal Societies and identifies three features he sees as necessary for sustained productivity growth from ideas.

One is the promotion of what we would call basic or fundamental science. It doesn't all have to be immediately practical or commercialized. But you're promoting deep knowledge of nature.

Two, you're doing it in an empiricist way. This is something very important in the 17th century that the Romans by contrast don't have, is the spirit of Francis Bacon that we need to ground our knowledge in experiment and observation, not just believe whatever authorities or Aristotle said. That's very much the spirit of places like the Royal Society: we don't take things on anybody's word, especially Aristotle's. You need basic science. You need empiricism, rigorous and self-correcting.

Third, you need a sense of useful knowledge, and that's the other thing that really comes together in the 17th century: not just the basic and abstract science, but the application, and the 17th century language for that is useful knowledge. That is something that doesn't ever get wired together in the Roman Empire. There are tinkerers and engineers, but they're not talking to the mathematicians and the physicists. If you were from on high to design self-sustaining innovation, I think you would want to bring those elements into proximity.

Kyle is definitely right that close feedback loops between fundamental, abstract science and empirical, applied engineering are important for sustained productivity growth from new ideas. But the implication that royal support for scientific inquiry is the missing link between Rome and and sustained technological progress is wrong.

After all, the Greeks had royal support for scientific inquiry for centuries at the Library of Alexandria. Perhaps they managed a higher rate of discovery than the Romans, but they weren’t any closer to the self-sustaining industrial growth that Kyle is trying to start. There was royal support and patronage for scientists in Rome too. Emperor Antonius Pius exempted philosophers and physicians from taxes and funded the university at Athens. Galen was a court physician for Marcus Aurelius, and the scholars remaining at Alexandria under Roman rule, like Hero, still probably received state support for their research.

So what was preventing the feedback loops of technological growth taking hold? I argue the missing piece was the printing press. Kyle’s examples of early industrial Europe provide some evidence for this. He notes that the connection between abstract theoreticians like Newton and Leibniz, and applied tinkerers like Denis Papin was essential to spark sustained technological progress. But these connections were only possible because their ideas were cheap to print and easy to spread around Europe.

Just as the printing press enabled these connections in early-modern Europe, it’s absence prevented them in the ancient world. Hero of Alexandria was experimenting with vacuum pumps a few generations after Archimedes studied fluid physics and a few generations before Diophantus was investigating number theory, but it’s much harder for their ideas to cross paths because copying Hero’s works is the labor of a lifetime rather than the labor of one man over a few months. Similarly, Kyle wonders where Euclid’s elements came from and where the knowledge went:

I'm super interested in the history of math. What happens after Euclid? It's very hard to say because you get these really interesting people that pop up, like Diophantus who's later, in the early Roman Empire. There's still really interesting math going on. Euclid is incredible. The Greek experiment in math and science is the one that I think had the better chance of sparking sustained takeoff.

And it didn't. It'd be interesting to know more about why. Why did things stall? Because these people... Euclid is not just a towering genius who comes out of nowhere. He's very much a product of the culture and the questions that are being asked in the generations before. It just sort of feels like after him, you fail to get that kind of sustained continuous progress and advance. Maybe back to that big question that we were asking before: what prevents the kind of breakthroughs that we see in the modern world?

As Kyle points out, Euclid is part of a larger culture of inquiry. Archimedes sent his proofs as letters to other mathematicians in Alexandria. The Mediterranean sustained something like a republic of letters for centuries under both the Greeks and the Romans, but it never materialized into a feedback loop of technological growth.

In contrast, as soon as Europe rediscovered classical Greek and Roman texts and started mass producing them with the printing press, it began the feedback loop of improving knowledge and technology that never took hold in the ancient world, and eventually led to the Industrial Revolution.

What Rome was missing wasn’t state support for science or proximity of abstract philosophers to applied technicians. It wasn’t even missing a flow of original and important ideas; the more inventive Greeks were well integrated into the empire and the classical texts were enough to ignite modern growth in Europe in the 15th century.

What Rome was missing was the printing press1. Without it, any idea has to overcome the massive activation energy required for copying before it can be spread or last over time. Rome had plenty of sparks of innovation, but lacked the catalyst to sustain them.

Rome was close to an industrial revolution. The empire housed massive economies of scale, continent-wide trade networks, and sophisticated financial markets. It also had a healthy flow of scientific and applied innovations. If Rome had combined all of this with a cheap way to preserve and spread these ideas, it would have begun the feedback loop of economic growth and saved us all a thousand years of history.

Popular political tracts like Abundance or Why Nothing Works explain American urban history as swings of a pendulum. On one end, you have Robert Moses and the high-modernist, ultra-state-capacity of the early 20th century. Comprehensive zoning codes were invented and spread, entire neighborhoods were demolished with little recompense, and super-highways were built in their place. On the other end of the pendulum is Jane Jacobs and the political and social movements of the 1970s that enshrined local participation and veto, environmental consciousness, and bureaucratic process as defenses against slum-clearing urban renewal. Both of these movements went too far and left their scars on American cities so a new path between the two extremes is needed.

This story correctly identifies the two major sources of the problems facing American cities: the destructive overreach of modernist urban renewal and the paralyzing vetocracy of environmental review. But it wrongly portrays Jacobs as a champion of the latter problem. In fact, a careful reading of "The Death and Life of Great American Cities" reveals that Jacobs didn't advocate for preservation and obstruction (at least, not in the book itself)—she advocated for market urbanism and organic development. The solution to our urban problems isn't a middle path between Moses and Jacobs, but rather a return to Jacobs' actual ideas, not the distorted version that became embedded in planning practice.

The first pendulum swing: High Modernism or the Radiant-Garden-City-Beautiful

High modernist urban planning shaped the urban form of American cities in two main phases.

The first phase, in the early 20th century, began with the invention of comprehensive zoning codes and density regulations. Planners like Edward Bassett and Ebeneezer Howard saw rapidly growing industrial cities, with high density tenements and overlap of residential, commercial, and industrial land uses, as threats to nature, health, light, and clean air. They wanted to separate land uses and spread out the population as much as possible. Here’s Edward Bassett in New York City’s 1916 comprehensive zoning plan:

Most of the evils of city life come from congestion of population. In precisely the measure that the city’s population can be distributed will those evils be mitigated.

It is therefore essential in the interest of the public health, safety, comfort, convenience and general welfare that a housing plan be adopted that will tend to distribute the population and secure to each section as much light, air and relief from congestion as is consistent with the housing of the entire population for a considerable period of years within the areas accessible and appropriate for housing purposes.

The height restrictions, setbacks, and exclusive, single-use zoning codes that these planners championed are still common in every major American city today.

The second phase of high modernist urban planning, called urban renewal, happened closer to the middle of the 20th century after the automobile grew to dominate transportation. This phase maintained the top-down control of private land uses and the goal of de-densifying urban areas with suburbs, but it added a focus on automobile traffic management and ambitious construction of public housing and infrastructure. While the first phase placed legal restrictions on new development and grandfathered in older buildings, urban renewal demolished wide swathes of older neighborhoods and replaced them with highways or copy-pasted high-rises surrounded by roads, parking lots, and open space.

High modernist urban planning was a massive mistake. The de-densifying mission of the first phase sabotages the labor market agglomeration benefits that make cities productive. The restrictions placed on city population by those early 20th century regulations have lowered US GDP by 8-36%, potentially tens of thousands of dollars per person. The second phase of high modernism was somewhat more compatible with a dense labor market concentration since high-rise public housing can be dense, but it destroyed the dynamic street life that makes cities interesting and safe while heavily subsidizing car transport and suburban development. The push of urban crime waves and the pull of cheap single-family housing with subsidized transport and parking emptied city centers into the suburbs. Both phases of high modernism fastidiously segregated land uses (and people), leaving cars as the only option to get from one's house to stores or work.

Subscribe now

The swing back: Jane Jacobs, Rachel Carson, and Sherry Arnstein

Criticizing high modernism is completely mainstream within urban planning today. The swing of the pendulum away from the practices of urban renewal began in the early 1960s with the most lauded urban planning book of all time, The Death and Life of Great American Cities by Jane Jacobs, which is self-described as an “attack on current city planning and rebuilding,” (current in 1961).

Jane Jacobs’ critique was subsumed into the larger 60s and 70s activist movements against high modernism including environmentalism and civil rights. “Highway revolt” protests and community meetings across the country, some of which were organized by Jacobs herself, elevated local public participation as the best way to practice democracy. In San Francisco this strategy successfully protected the city from an intra-city freeway. This process of public input was then formalized into discretionary review, which allows agitating San Francisco residents to prevent development even when it conforms to all local zoning codes. Later in the 70s, California wrote a public-hearing right for all landowners who might be affected by a development project into its constitution. Similarly, after a protest movement against public housing in Forest Hills, Queens, New York City devolved power over local zoning and land use to community boards which became “the first line of defense in New Yorkers’ war against growth.”

Jacobs’ arguments were quickly swept into the larger flood of support for formalizing participatory democracy that began in the 1960s. In 1965, The American Institute of Planners published Advocacy and Pluralism in Planning by Paul Davidoff, which advocated for establishing “an effective urban democracy, one in which citizens may be able to play an active role in the process of deciding public policy” and that “great care must be taken that choices remain in the area of public view and participation.” Around the same time, Rachel Carson’s Silent Spring and Sherry Arnstein’s Ladder of Citizen Participation were published, further catalyzing the environmental movement and making a case for far more public participation in government decisions. The slew of procedural environmental laws which followed, like NEPA and the National Historic Preservation Act, were enthusiastically incorporated into urban planning.

These laws have preserved the mistakes of high modernist planning and added their own constraints to successful urbanism. Public hearings on zoning changes essentially locked in the exclusionary, low-density zoning codes set in the early 20th century. A small population of incumbent homeowners can shut down upzoning of single-family districts even in the most valuable urban land on earth. Procedural environmental laws locked in the car-centric urban forms built by Robert Moses and his ilk. High-speed rail, subway extensions, even bus lines and congestion pricing projects are delayed by years and inflated by millions due to costly environmental review. Environmental law constraints have made even the few successes of mid-century urbanism, like Sea Ranch California, impossible to replicate or extend today.

Planners bet big on devolution of power, public participation, and environmentalism rather than deregulation and market urbanism. They agreed with Jacobs that the Robert Moseses of the world shouldn’t be able to eminent-domain entire neighborhoods and run big highways through the city, at least not without democratic approval and procedural review. So construction projects have to be preceded by public comment periods, town meetings, and environmental reports.

Veto power was distributed outside of city governments, and planners successfully avoided another mistaken urban renewal, but they did so by demolishing our capacity to build anything at all.

Jane Jacobs’ rules can fix American cities

In manifestos like Abundance or Why Nothing Works, this is where the story ends. Robert Moses and Jane Jacobs are opposite paragons of their respective philosophies and modern policy makers must chart the narrow way between. While this is a useful prescription for how to move on from the legacy of environmental proceduralism and urban renewal, it’s not an accurate portrayal of Jane Jacobs’ views.

In “Death and Life” Jacobs argues that urban planners should plan less. They should stop drafting grand visions and bulldozing neighborhoods to rebuild them in their image. “Public policy can do relatively little that is positive to get working uses woven in where they are absent and needed in cities, except to permit and indirectly encourage them.” When discussing a successful commercial district in Nashville she notes that “Nobody could have planned this growth. Nobody has encouraged it.”

Even the most widely supported uses of urban planning, like setting maximum densities and separating residential and industrial land uses are criticized. “What are the proper densities for city dwellings?” she asks, “The answer to this is something like the answer Lincoln gave to the question ‘How long should a man’s legs be?’ Long enough to reach the ground.” On industrial pollution control she says “Of course reeking smokestacks and flying ash are harmful, but it doesn't follow that intensive city manufacturing (most of which produces no such nasty by-products) or other work uses must be segregated from dwellings. Indeed, the notion that reek or fumes are to be combated by zoning and land-sorting classifications at all is ridiculous. The air doesn’t know about zoning boundaries. Regulations specifically aimed at the smoke or reek itself are to the point.”

In Jacobs’ view, planners should merely set favorable conditions for decentralized, unplanned, and incremental development. Lay out the street grid if you must (though with smaller blocks than Manhattan) and bring the infrastructure to where it’s needed but otherwise refrain from regulating private land use. Let all the uses of the city mingle together and fill each street with eyes and activity at all times of the day. Jacobs is like Hayek (1945) or The Fatal Conceit applied to urbanism. “The curious task of economics [and Jacobsian urban planning] is to demonstrate to men how little they really know about what they imagine they can design.”

Jacobs does advocate for more local city government and praises public participation in planning meetings, but many of her reasons for wanting more devolved power in city planning were good. Far away bureaucracies can’t transmit the man-on-the-spot knowledge that planners need to support organic development. Bureaucracies have other familiar problems as well: “In Baltimore a sophisticated citizens group engaged in conferences, negotiations, and a series of referrals and approvals extending over an entire year — merely for permission to place a sculpture of a bear in a street park!”

The pendulum swings between overbearing state capacity and sclerotic vetocracy created massive problems that are still causing problems a century later. But the path to redemption can be found in the already canonized works of Jane Jacobs.

Jacobs proposes four ingredients that make a vital city, district, or street.

1. The district must serve more than one primary function; preferably more than two.

2. Most blocks must be short; that is, streets and opportunities to turn corners must be frequent.

3. The district must mingle buildings that vary in age and condition.

4. There must be a sufficiently dense concentration of people.

Each of these four rules is phrased as a positive obligation. This leads to some confusion where people will try to plan a Disneyland Jane Jacobs urbanism where all of these qualities are built in from the top down. Jane Jacobs Disneyland is surely better than Corbusier’s towers in the park, but that kind of new urbanism is not the solution to the problems with American cities. As Jacobs says “A city cannot be a work of art.”

Jacobs’ four positive obligations can actually be compressed down to two rules.

1. Don’t regulate land use on private plots.
   The density and mixed uses called for in obligations 1 and 4 are fulfilled automatically when urban planners do not artificially separate land uses and reduce density. Let diverse, mixed uses and high densities arise where they may in cities. If extreme negative externalities result, address them case-by-case rather than through blanket zoning.

2. Encourage incremental development and infill.
   The dynamic street patterns and varied building ages called for in obligations 2 and 3 also arise naturally when cities develop incrementally, rather than in giant lurches as new neighborhoods are demolished and built all at once. Let winding streets with frequent breaks and wide-ranging building ages emerge naturally as developments stack up in an unplanned, organic pattern. This means lowering the regulatory fixed costs of development so that small projects can arise easily, and abandoning master plans.

These rules have already found success in the wonderful urbanism of Tokyo. Land use regulations there are much laxer, especially for commercial shops and restaurants, so the residential and commercial density of the city is high. That means cute apartments surrounded by a walkable community of shops and restaurants are cheap and abundant. Incremental infill development is common and easy to do in Tokyo, so there is a combination of modern and classic architecture and a range of price points that support diverse uses for each street in the city. Tokyo doesn’t have a gridded street plan and the emergent pattern creates lovely neighborhoods with short blocks, alluring alleys, and shielding from cars.

These rules also created many of the most charming neighborhoods in urban America. Boston’s North End, New York’s Greenwich Village, and D.C’s Georgetown were all largely built before comprehensive zoning codes and are full of buildings and shops that would be illegal to build today.

There are problems with American cities that aren’t addressed by these rules. To get better transportation infrastructure, you have to address environmental proceduralism for public projects as well as private ones. More city streets that are active throughout the day would help with public order and crime, but when the social order enforced by crowds of onlookers is not backed by credible force from local police, open homelessness, drug use, and crime can persist even on the busiest streets.

But these two principles are necessary steps on the path towards embracing Jane Jacobs’ ideas, if not her advocacy, and redeeming the major mistakes of urban planning.

A positive relationship between fertility and economic growth is intuitive. People are the most important input to every production process (for now). People come up with the technology and ideas that fuel economic growth. People specialize and trade with one another and the more people there are, the deeper and more productive this specialization can be.

On the other hand, rising opportunity costs, quantity-quality tradeoffs, and other cultural/religious changes have decreased fertility as incomes have risen. These effects are strong enough that the simple correlation between economic growth rates and the change in fertility rates is highly negative.

All of these relationships are supported by economic theory and some aggregate empirical research, so what does this new paper bring to the table? They claim to have a unique causal identification strategy using an instrumental variable.

Here’s what that means: if you just sort countries by their fertility rates, you’ll also end up sorting them by all the other things that determine fertility, which might have their own effects on growth. So if you find that high fertility countries have lower growth rates, you have no way to tell if that’s due to fertility itself or due to the religiosity, patriarchy, or poverty that’s causing the high fertility rate. If there’s reverse causality and GDP growth lowers fertility, then sorting countries by fertility rates also sorts them by GDP growth rates and any effect you find between the two is meaningless.

What you need is some way to sort countries into groups that differ in fertility rates, but not in any other ways that are relevant to GDP growth rates. The authors of this paper propose abortion policy as a sorting method.

It's easy enough to show that countries with different abortion policies do have very different fertility rates. But that second condition — that countries which differ in abortion policy don't differ in any ways that affect growth other than fertility — seems obviously wrong.

Countries with strict abortion policies are more religious, less “western” (in this sample, more Muslim), more patriarchal, possibly less democratic, etc. All of these things are contributing their own effects to the growth differences between strict and liberal abortion countries but the author’s identification strategy attributes everything to fertility differences. I would be surprised if the difference in abortion numbers even comes close to explaining the difference in births between nations, so there must be some other varying factor.

The authors can address this somewhat with control variables, e.g for educational attainment or female labor force participation.

They also run a specification where they separate “advanced economies” and “developing economies” and find a smaller, though still significant and positive sign-flipped effect of fertility on economic growth. These results make their claim slightly more believable, but if my prior wasn't already that fertility causes economic growth, I don't think this evidence would be satisfying.

I'm already confident that global fertility decline will drag down economic growth, but higher quality economic research on precisely how much growth we should expect to lose would bring the issue into focus for policymakers. Unfortunately, Cevik’s paper doesn't fit the bill.

Three months after wildfires scorched Los Angeles, somewhere between zero and four rebuilding permits have been issued out of tens of thousands of buildings that were destroyed. The recovery from the 2023 wildfires in Lahaina, Hawaii has been even slower. A year and a half after the disaster, only 6 structures have been rebuilt out of over 2,000 that were destroyed, and permits have been issued for only a few hundred additional projects.

Creative Destruction Historically

Fires, earthquakes, bombings, and other disasters are obviously costly and tragic events. However, they can act as coordination mechanisms that help property owners make efficient investments in new infrastructure that wouldn’t have been possible to do on their own and as forcing functions on governments to allow urban land uses to fit market conditions rather than preserving existing interests. There are several examples of these effects from history that have been studied by economic research.

The oldest comes from the 1872 fire of Boston. It destroyed all of the buildings mapped in red below but after the fire, land values in and near the burnt areas rose by enough to more than offset the loss of building value from the fire.

According to Richard Hornbeck and Daniel Keniston, this extra land value came from the fire forcing all the landowners to rebuild their old buildings at the same time, thus internalizing more of the positive externalities of building quality. Here’s how that works: When a landowner improves the quality of their building, it raises their land value, but also raises the land value of surrounding plots. Higher land values incentivize more investment into building quality but usually, one improved building nearby isn’t enough to overcome the fixed costs of reconstruction for neighbors, so this positive land-value externality isn’t recovered, and property owners underinvest in building quality.

However, when everyone is forced to rebuild at the same time, there are no fixed costs to overcome; everyone is already planning a construction project. So if you invest in a higher quality building that will raise surrounding land values, your neighbors can immediately respond by raising their building quality which then spills back over into your plot in a virtuous cycle that results in everyone building intensively on their plots and raising surrounding land values far above their previous peak. The area of Boston that was burnt down in 1872 is, perhaps not coincidentally, now home to its tallest skyscrapers.

There’s a similar story in San Francisco after the 1906 earthquake and fire.

Here, the author focuses on path dependence and fixed costs rather than on cross-plot externalities. Buildings are durable goods and they can stick around for much longer than the market conditions that incentivized their construction in the first place. This is why you can buy homes in Detroit for a dollar even though no one would ever build a home for that price. In the opposite direction, low density buildings can persist even after the optimal density rises far above what they can provide because the net present value of higher rents from a denser building can’t overcome the upfront costs of permitting, demolishing, and rebuilding.

The 1906 fire forced the decision. Landowners no longer had the lower rents from their older buildings competing with the upfront costs and future benefits of redevelopment, so they were incentivized to rebuild to the optimal higher density. After the fire, burnt plots got significantly denser; about 15 extra residential units per acre compared to nearby unburnt ones. Without resolving some externality or inefficient tax on redevelopment, this forcing function of the fire is still a welfare loss, but these frictions probably did exist as they did Boston and the rapid recovery after the fire created much of San Francisco’s modern urban landscape.

A final example with perhaps the largest positive effects took place after the London Blitz bombings during WWII. While not exactly a natural disaster, this example more explicitly involves planning regulations, rather than just development fixed costs or land value spillovers.

Gerard Dericks and Hans Koster exploit the variation caused by Blitz bombings to measure the costs imposed by redevelopment frictions which prevent higher densities and thus more productive agglomeration economies. London's strict planning regulations restricted building density and height but the Blitz destroyed many buildings that would otherwise have been historically protected sites and pushed local authorities to loosen rules for reconstruction projects. Areas with higher bombing intensity have since developed taller buildings and higher employment density.

The elasticity of productivity to building density as measured by this exogenous variation implies that doubling employment density leads to a roughly 20% increase in productivity. Counterfactual simulations suggest that if the Blitz didn’t happen, and heavily bombed areas of London had developed more like the areas that had been spared, increased redevelopment frictions would have reduced the total population of London by around 9%, with even larger decreases in the most productive central business district. Less density means less productivity so without the redevelopment allowed by the London Blitz, total economic output in the city would be roughly 10% lower than today's actual outcome.

Creative Destruction is Just Destruction Today

Returning to the present day, the creative destruction offered by natural disasters or other city resets seems unattainable. Our environmental permitting regime is so inflexible that rebuilding after disasters doesn’t feel like a release of development potential that was held back by previously fixed investments, it just feels like irreversible destruction.

Often, rebuilding after a disaster today can be even more restrictive than the original construction. Much of the building stock in the most restricted markets was built before hardline zoning codes and were grandfathered in with one-off zoning variances. When rebuilding, these variances no longer apply and new construction might be impossible. For example, 40 percent of the buildings in Manhattan could not be built today and thus couldn’t be rebuilt if they were destroyed.

There is a counterexample to this story that proves the rule of environmental permitting constraints and shows a path forward for revitalizing construction after disasters and in all contexts: the I-95 bridge collapse. After a truck fire beneath the bridge caused the structure to fail on an essential transport corridor in Pennsylvania, Governor Shapiro issued emergency rules exempting the reconstruction project from standard contracting regulations and environmental review. The temporary replacement bridge was ready to go in just two weeks.

Everyone knows our construction regulation regime is costly and slow, but when even the defibrillator-shock of a city-destroying fire can’t get the system to wake up, we have to declare it dead.

This kind of political pressure is exactly the motivation behind the subtextual messages identified in the works of ancient philosophers by Leo Strauss. Ezra Klein and Derek Thompson won’t face forced suicide for unorthodox views like Socrates did, but they do risk a form of exile. The reaction to Derek’s appearance on Richard Hanania’s podcast shows there is still some heat left in the ashes of cancel culture.

Most materially, they risk not convincing anyone with their work. On Hanania’s podcast, Derek Thompson says his goal for the book is to “persuade you that we share the same values, and make the case that your processes don’t lead to the values you actually share.” Thus, to sneak past the defenders of progressive orthodoxy, Ezra and Derek must hide the most radical parts of their message.

Subscribe now

Redistribution doesn’t matter

If you really believe that there can be industrial revolution-level growth over the next several decades leading to massively increased per capita incomes, energy consumption, and a slate of new technologies and medicines which improve and extend life, then there’s just nothing you can do with redistribution today that would come close to the importance of this progress.

Under the Abundance vision for the future, doing redistribution today is like doing redistribution in 1750. No matter how you shift the shares of wealth across the population, you’ll still be left with a world of utter poverty, poor health, and backwards scientific theories. If you divvied up global GDP today equally across all 8 billion people, everyone would be well below the US poverty line. The highest national life expectancies are only around 85. No universal healthcare plan can cure cancer and no public transportation infrastructure can get from New York to London in 2 hours.

Klein and Thompson subtly reveal this message throughout the book. The first hint is in their three-page sci-fi opener that paints a picture of what life could be like in 2050 if we focused on growth.

Out the window and across the street, an autonomous drone is dropping off the latest shipment of star pills. Several years ago, daily medications that reduced overeating, cured addiction, and slowed cellular aging were considered miracle drugs for the rich, especially when we discovered that key molecules were best synthesized in the zero-gravity conditions of space. But these days, automated factories thrum in low orbit. Cheap rocketry conveys the medicine down to earth where it’s saved millions of lives and billions of healthy years.

The message here is hidden in what they don’t say. There is no UBI, no welfare, no food stamps, and no Medicare in their vision of the future. There is no redistribution at all except for a passing reference to how the profits from AI are “shared” (If you get consumer surplus from ChatGPT then the profits are shared already). Redistribution doesn’t matter in this future. Gains to human welfare come from new inventions and from technological progress lowering the cost of all the goods we want more of today, like housing, medicine, and energy.

Several times throughout the book Ezra and Derek damn progressive redistribution efforts with faint praise.

Progressivism’s promises and policies, for decades, were built around giving people money … The Affordable Care Act subsidizes insurance that people can use to pay for health care. Food stamps give people money for food. Housing vouchers give them money for rent. Pell Grants give them money for college. Tax credits for child care give people money to buy child care. Social Security gives them money for retirement. The minimum wage and the earned-income tax credit give them more money for anything they want.

These are important policies, and we support them. But while Democrats focused on giving consumers money, they paid less attention to the supply of the goods and services they wanted everyone to have.

And another pulled punch here:

American liberalism has measured its success in how near it could come to the social welfare system of Denmark. Liberals fought for expansions of health insurance and paid vacation leave and paid sick days and a heftier earned-income tax credit and an expanded child tax credit and decent retirement benefits. Worthy causes all.

…

The climate crisis demands something different. It demands a liberalism that builds … What matters is not what gets spent. What matters is what gets built.

About halfway through the book they get closest to stating the Straussian anti-redistribution message outright:

Almost every product or service that liberals seek to make universal today depends on technology that did not exist three lifetimes ago — or, in some cases, half a lifetime ago. Medicare and Medicaid guarantee the elderly and poor access to modern hospitals, where many essential technologies — such as plastic IV bags, MRI and CT scan machines, and pulse oximeters — are inventions of the last sixty years. It is tempting to say that, with these essentials already in existence, it is time for society to focus at last only on the fair distribution of existing resources rather than the creation of new ideas. But this would be worse than a failure of imagination: it would be a kind of generational theft … Politics itself becomes a mere smash-and-grab war over scarce goods, where one man’s win implies another man’s loss.

Without more rapid growth and invention, they compare redistribution to theft and war. With rapid growth and invention, redistribution might help some people get access to important goods, but just as in their sci-fi vignette, most of the gains to access will come from rising incomes and falling prices. Compared to things that can quicken the arrival of these higher incomes or new technologies, redistribution of today’s resources just isn’t important. If redistribution slows down income growth or invention then it should be avoided. This is the straussian message of Abundance.

While writing this post, Derek actually tweeted to dispel the notion that Abundance is against redistribution, using the passage I quoted as faint praise above as evidence.

Firstly, I am envious of Leo Strauss whose subjects were always centuries dead and thus never able to contest his esoteric interpretations of their work.

Second, I don’t think this is inconsistent with the argument I am making here. I don’t think Ezra and Derek are against redistribution, just that they don’t think it’s particularly important compared to the growth and invention that is possible over the next few centuries.

Caveats and Rebuttals

A lukewarm position on redistribution doesn’t mean Ezra and Derek aren’t interested in using the government to shift resources around. They are very positive on the science funding efforts of the government which they believe can hasten economic growth and invention. That can be framed as form of redistribution, though it’s usually a regressive form which sends money to the wealthy, educated, and entrepreneurial classes rather than those in material need.

There may be ways in which other, more traditional forms of redistribution increase the economic growth and technological progress that Ezra and Derek value. Public education is probably the easiest example (though perhaps the effects on innovation are negative), or immigration. Again though, valuing these efforts as a means to the end of faster growth is very different than seeing the transfer of resources from the rich to the poor as an end in itself.

Public education as a means to the end of more inventors probably means searching for the few most brilliant and high potential students, likely to be the sons or daughters of already-wealthy inventors, and supporting them as much as possible. Not sending billions to ailing school districts in an effort to bring them up to minimum standards.

Immigration policy, too, changes when the goal is not just improving the lives of migrants but rather raising the rate of economic growth and the innovative potential of the receiving country. High-skilled immigration policy means targeting the best and brightest, for example by auctioning off H1-B visas to the highest bidder rather than allocating by lottery. Or by accepting immigrants preferentially from countries with the most stable governments and track record in innovation, not refugees from failed states.

A final rebuttal to consider is that even if you do truly believe in the importance of redistribution as an end in itself, you might still focus on growth and innovation because there’s no better way to generate wealth to redistribute. This is the Fully Automated Luxury Communism argument which Ezra and Derek mention in the book. It is also similar to Tyler Cowen’s argument in Stubborn Attachments that essentially no matter what you value, whether it is redistribution, justice, art, or technology, economic growth will always come to dominate the contributions to that goal because compounding growth is so powerful.

This is the best way to make Ezra and Derek’s focus on growth and innovation consistent with a genuine belief in the importance of redistribution as an end in itself. I suspect that this is also what the authors would claim as closest to their actual views if they were presented with this post. Perhaps reading any more into their views on redistribution is just libertarian projection.

Even this more moderate view massively downgrades the importance of redistribution on the margin compared to current progressive priorities. Half of the federal budget is already spent on transfers to individuals and government transfers make up 40% of the income of people in the bottom 20th percentile of household income. Parceling out more of present day wealth to these efforts is well past diminishing returns and probably has negative growth effects which far outweigh the benefits. Under this view, redistribution does matter but we already have more than enough of it. Now, let’s focus on abundance!

Subscribe now

The podcast was four hours long so I will focus only on two points near the beginning that serve as so-called “intuition pumps” for the rest of the discussion.

A century of history crammed into a decade

First, there is the idea of the “accelerated decade” and the challenges this will pose to human institutions. Here’s MacAskill explaining what might happen after AIs begin to accelerate technological progress:

Will MacAskill: We’re thinking about 100 years of progress happening in less than 10. One way to get a sense of just how intense that would be is imagine if that had happened in the past. So imagine if in 1925 we’d gotten a century’s worth of tech progress in 10 years. We should think about all the things that happened between 1925 and 2025 — including satellites, biological and chemical weapons, the atomic bomb, the hydrogen bomb, the scale-up of those nuclear stockpiles. We should think about conceptual developments: game theory, social science, the modern scientific method; things like computers, the internet, AI itself.

But while technological progress runs ahead, other things will grow more slowly or perhaps not at all, creating a potentially dangerous asymmetry.

Will MacAskill: Human decision making and human institutions don’t speed up though. So just taking the case of nuclear weapons: in this accelerated timeline, there’s a three-month gap between the start of the Manhattan Project and the dropping of the nuclear bomb in Hiroshima. The Cuban Missile Crisis lasts a little over a day. There’s a close nuclear call every single year.

This clearly would pose an enormous challenge to institutions and human decision making.

Rob accepts MacAskill’s interpretation of this thought experiment without contest.

Rob Wiblin: I think this helps to bring out the intuition of, holy shit, there would be so much chaos, so much stuff going on, and we wouldn’t have the ability to process it or think about our decisions very well.

But the response to the fears implied by MacAskill’s thought experiment are right there in the premise. What would happen if the previous century of progress were compressed into a decade? Just look around you. We’ve been on this trend for thousands of years. The 20th century was a thousand years of progress or more compressed into century.

We've already increased through several orders of magnitude in how fast the world moves, how rapid economic growth is, how much economic activity each person is responsible for, and how much destructive power each person and government can wield.

If, as MacAskill claims, human decision making hasn’t sped up, we’ve somehow managed to avoid the “holy shit chaos” that Will and Rob predicted would result.

Maybe some particular theory of human psychology and economics can predict that the next speed-up will finally take us over the threshold of human understanding, but just bamboozling with big numbers doesn’t stand up to our already proven success through the even more bamboozling long-view of history.

Just as in the 20th century there will surely be risks, but even considering all of the mistakes and destruction that century wrought, we still came out far ahead.

This is a purely empiricist argument for why MacAskill’s intuition pump of seeing a 1-2 OOM increase in growth and wondering “how could we ever deal with such a thing!” doesn’t hold up to scrutiny. But more constructive arguments for how we actually manage a world economy that is thousands of times larger and faster than it was in the past are available and are useful for thinking about AI progress. Start with I, Pencil and Hayek (1945).

The Philosopher’s Fallacy

Later in the podcast MacAskill makes a similar argument in a more general form. Just as AI will exceed our capacity to organize the economy it also exceeds our capacity to plan the moral future.

Will MacAskill: You should really pause and reflect on the fact that many companies now are saying what we want to do is build AGI — AI that is as good as humans. OK, what does it look like? What does a good society look like when we have humans and we have trillions of AI beings going around that are functionally much more capable?

There’s obviously the loss of control challenge there, but there’s also just the like —

Rob Wiblin: Sam Altman, I’ve got a pen. Can you write down what’s your vision for a good future that looks like this?

Will MacAskill: What’s the vision like? How do we coexist in an ethical and morally respectable way? And it’s like there’s nothing.

Rob Wiblin: Deafening silence.

Will MacAskill: Careening towards this vision that is just a void, essentially. And it’s not like it’s trivial either. I am a moral philosopher: I have no clue what that good society looks like.

In some ways, I agree with MacAskill’s description here. We don’t have a clear vision of what the future will be like and how we’ll all get along within it. We don’t have a plan that Sam Altman could jot down on a piece of paper.

But this fact has no bearing on whether or not the future will be good because we don’t need moral philosophers to have a plan for things to go well.

Evidence for this claim again comes from history. No one had a clear-in-advance vision for how computers would change the world and how we might need to adapt to remain morally good. No one had such a vision for cars or steam engines or electricity. And yet we can all agree that the world nonetheless improved after their invention.

Rob and Will do get into more concrete discussions about morality and AI that don’t rely on this bad argument but they’d be better off dropping this intro.

Reading List for Paid Subscribers

Read more

The story continues that if only we could coordinate, we could fix the misallocation caused by externalities. We might get the government to tax and subsidize externalities or else we might try to lower transaction costs so that people can bargain to solve externalities on their own.

The basic story over-focuses on social coordination as the solution to externalities. Our institutions cannot be relied on to optimally correct externalities or even to avoid making them worse. Usually, the costs of an externality subside only after we’ve invented a technology which makes it cheap or privately beneficial to do the socially optimal thing. Most importantly, technology shifts out the production possibilities frontier making it possible to get outcomes beyond what even perfect social coordination could attain.

Economics should emphasize the importance of technology as a solution to externality problems and focus less on social coordination.

The Smoky City

Consider Pittsburgh in the mid-20th century. Coal and steel industry earned the city the nickname “smoky city” and caused widespread death and disease. Coal was used for everything in Pittsburgh: steel and industry, heat and light. Buildings and people were stained black inside and out.

Coal pollution is a classic externality problem and indeed the classic solutions were tried. There were political and social campaigns for cleaner air in the city throughout the early 20th century. These culminated in a 1941 law that put some requirements on residential users to use treated coal, other cleaner fuels, or special burners that reduced smoke.

But ultimately, the solution to this externality problem came from technology. First, coal-gas burner-lamps were replaced with electric ones. Then, Pittsburgh got a natural gas and petroleum pipeline laid nearby. Revenues from utility sale of natural gas doubled from $67 million in 1938 to $140 million ten years later. The growth of natural gas replaced coal in heat and power generation for Pittsburgh and made more total energy available for its residents. Similarly, coal trains were replaced by diesel trains. In 1955 the Bureau of Smoke Prevention observed that “The Diesel locomotive has solved the smoke problem of the railroads.”

Social and political coordination did play some role in this story. The 1941 smoke control ordinance coincided with some of these changes and may have made some contribution to the shift. However, enforcement the law was delayed until 1947 to aid in wartime production and the coal-union economy of the city led to concessions being made against the law for industrial uses (a different form of political coordination).

National energy consumption from coal peaked, except for a brief post-war spike, in the 1920s, long before air quality legislation was passed. Hours of heavy smoke per year in Pittsburgh were declining long before the 1941 ordinance. Joel Tarr and Karen Clay of Carnegie-Mellon university agree that “A transition to natural gas, rather than public policy, produced the city’s first clean air period.”

More important than the fact that a technological innovation solved Pittsburgh’s smog where political coordination failed is the following point: Even if the social coordination had worked perfectly the technological solution would still be more important. The socially optimal tradeoff between coal energy and clean air available to Pittsburgh in 1905 still leaves them in a bad situation. They had to accept either dirty air or energy poverty; they had no way to solve both problems at once. The technological solution pushes the production possibilities frontier between energy and air quality far beyond what was the optimal point when only coal was available.

Subscribe now

Airborne Disease

Infectious diseases have been humanity’s deadliest foe for essentially all of its urban history, not least because they create costly externalities. Diseases spread exponentially, so a single infection can lead to thousands of others down the line without making patient-zero any more sick.

Dozens of social and political coordination strategies have been designed to address this externality from the original 40-day quarantines for ships entering medieval Venice to covid mask-mandates and school closures. These have had various degrees of success across thousands of years of attempts but they often fall short of appropriate protection or overshoot the optimum and suffer needless costs.

No coordination strategy ever got close to the effectiveness of vaccines. Vaccines have saved billions of lives in their relatively short history and have completely eradicated many diseases with a finality that no coordination could ever achieve alone.

Vaccines of course create their own coordination problems, but inventing the technology is the essential step that allows humans to live in big groups without constant death. The socially optimal response to disease without vaccines was rarely, if ever, reached. Even if you did get perfect coordination, that either means tiny, dispersed cities with infrequent contact or else frequent deadly disease. Only vaccines allow us to avoid both at the same time.

Firefighting

Large scale city fires like the fire of London or Chicago were serious threats before the 20th century. Just like with disease, there is a clear externality problem. One building catching fire poses an external risk to entire neighborhoods. Various social coordination approaches have made progress against this externality, from public fire brigades to those run by private insurance companies to building codes that mandated setbacks and separation between structures to prevent conflagrations.

Again, our current success against this problem relies on technological improvements. Mass produced steel and concrete allow us to produce much more fire resistant buildings. Automated sprinkler systems and smoke detectors shut down fires fast when they do arise. There are fewer than half as many home fires in the US today as in 1980. Our firefighters are barely ever called to actual fires. Often, laws enforce the use of fire resistant technologies but their widespread use is still mostly due to the technological progress which led to their invention and which lowered their cost enough for legal enforcement to be an acceptable political compromise.

Over-farming and Guano Depletion

From the beginning of human agriculture until the 20th century, soil resource depletion has been a major problem. If you grow the same crop in the same field many seasons in a row, the surrounding soil will run out of nutrients and become unproductive. If enough farmers deplete their soil it can lead to ecological collapses like the Dust Bowl, thus creating external costs to other farmers.

One source of extra nutrients were guano deposits found in the early 19th century on Peruvian islands. In the age of colonialism, property rights to these islands were sometimes unclear and contested, so over-exploitation led to their complete depletion.

Social and political coordination mechanisms often tried and often failed to solve these problems.

In 1910, Fritz Haber invented a chemical process that could produce ammonia, an essential ingredient in fertilizer, out of nitrogen in the air. This process internalized the costs and benefits of fertilizer production and use. It meant that farmers need not rely on one another to maintain soil nutrients or on industrialists to preserve the guano producing environments that they relied on for fertilizer. Plus, it made fertilizer and food much more abundant creating and saving billions of lives!

Malthusian Externalities and The Industrial/Green Revolutions

In a Malthusian world, any gains in productivity are translated into higher populations which compete incomes back down to subsistence. The graph below shows that countries around the world with higher agricultural land productivity had higher population densities, but the same subsistence-level incomes per person.

People are thus not strongly incentivized to improve productivity since it will just turn into more mouths to feed. Maybe some of that population growth is carrying their own genetic code, but most of it is an external benefit to others.

There were many social and political coordination strategies that attempted to address this. For example, infanticide was an extremely common practice in pre-industrial societies to try to control population growth. Paul Ehrlich infamously advocated for mass sterilizations of women in the third world in an attempt to control what he saw as Malthusian-externality population growth which would lead to terrible famine.

You can decide for yourself how effective or necessary these coordination strategies were at addressing Malthusian externalities, but again the point is that technology has solved this problem more effectively and freed us from the terrible tradeoffs it imposes. First the industrial revolution in the west and later the green revolution in the third world raised the rate of resource growth enough that it could persistently outpace population growth and create rising incomes.

Counterexamples

While it’s true that technology is underrated as a solution to externality problems, it’s not a universal rule that technology always solves them and political coordination always fails.

Leaded gasoline is a good counterexample. You still needed technology to invent the catalytic converter, but this story is not one where we just invented a new type of fuel which made the socially optimal choice coincide with the privately optimal one. The emissions requirements from government were a necessary push to remove lead emissions from cars. Perhaps electric vehicles would have eventually done the same but still, political coordination saved us from decades of lead emissions.

Another important point is that technology itself is a big externality problem! Inventors collect only a small portion of the value they create with their ideas. The private sector still outspends the government on R&D by 4:1 and the design of government subsidies can often make them counter-productive, but social and political coordination to support the technological progress that solves our other externality problems is important.

More Examples

The pattern of the above examples is clear but there are still more I want to list, so I’ll go through the following ones faster.

• Over-exploitation of Whales ==> Kerosene and petroleum lubricant

  • Whales were intensively hunted for their fats, first for light and then for industrial lubricants. Whaling controls were tried and mostly failed but we invented other materials to use instead and now whale populations are rebounding.

• Ivory Poaching ==> Plastics

  • A similar story to the whales. Ivory was the best available material for precisely shaped objects that still needed hardness like combs and billiard balls. Elephants were hunted close to extinction until plastics and other materials were invented that were better than ivory for this purpose.

• Horse Manure ==> Automobiles

  • Horses caused massive negative externality problems. There were 130,000 horses working in Manhattan in 1900 that each produced 45 pounds of manure a day. Millions of pounds of manure and hundreds of horse carcasses created near-unbearable stink, carried disease, attracted pests, and leeched into the water supply.
    
    Lots of regulations were passed attempting to control this but unpriced externalities remained and even the optimal tradeoff was still shitty. Cars have their own significant externalities but they solved these problems better than any coordination could have.

    

• Wood Cooking Stoves ==> Electric and Gas

  • This is similar to the Pittsburgh story I opened with. Particulate pollution in the developing world is a massive health risk, with a big source coming from wood fired cook stoves. As these countries develop, they are replacing these with electric or gas stoves which are better both privately and socially.

  • The transition away from wood as a construction material and fuel also solves de-forestation externalities. De-forestation has also been stopped in the developed world by growth in farming land-use efficiency.

• Waterborne Disease ==> Modern Plumbing and Water Treatment

  • Chemical purification of water, mass produced steel and concrete for sewage infrastructure, and inoculations against waterborne disease have been the major forces addressing this externality, though political and social coordination to build the infrastructure has also been important.

• Carbon Emissions ==> Solar Power

  • Carbon emissions are perhaps the first example of an externality people would think of. We could have a carbon tax that would help abate this problem, but we don’t and it doesn’t seem likely anytime soon. But the rapid improvements in solar power will make the socially optimal energy technology into the cheapest one too.

• Overfishing ==> Aquaculture

  • Fishery property rights are still frequently contested which leads to tragedy of the commons problems. Aquaculture, or the farming of fish resources, addresses these externalities and is growing to be the major source of fish.