Urban planning: a response to dath ilan
In a 2014 Tumblr post framed as an “April Fool’s confession,” the rationalist writer Eliezer Yudkowsky described “dath ilan,” a fictional society he claimed to remember from a parallel world. A central feature of dath ilan is that almost everyone who isn’t tied to farmland lives in a single planet-sized “great city,” built on the largest stretch of good-weather land on Earth, with all roads buried underground for full grade-separation, mass-manufactured tree-camouflaged houses, moving sidewalks at peak hours, an annual blackout to look at the stars, and food delivered by drone. Alyssa — whose work on housing and infrastructure economics had given her strong views on what cities can and can’t physically be — replies with a long, numerate object-level critique. The CC’d second recipient is Tomer Kagan, then advising Google’s Sidewalk Labs project to build a new city. The proposal Alyssa is responding to lives at yudkowsky.tumblr.com ; passages quoted below are from that post.
The first, and most important, urban planning principle is that the purpose of cities is to help humans interact. Like all tools, cities have to be designed around their purpose, which imposes constraints on their form. The most obvious constraint is the length of human interactions, which typically vary from a few minutes to around ten hours. This length limits the distance, as measured by travel time, that two people can be from one another without making interaction prohibitive. For shorter interactions, long travel times are very costly, because they mean that traveling time exceeds interaction time; and for longer ones they are very costly because they start to bump up against the natural sleep cycle. Therefore, the demand for interaction drops off sharply after an hour or two of travel time distance. It’s easy to spot this in real estate prices — Stockton, a bit under two hours from San Francisco, has decent single-family houses available for $100K. Ditto, eg., Allentown and New York City. The demand for 100-minute and longer trips is so small that it doesn’t make these places more desirable than random parts of the Midwest. And since cities exist to fulfill demand for interaction, this limits their size to one where any person in the city can easily interact with any other person.
This leads into the first problem with the proposal, as written; like most of the other issues, it might not be obvious because the proposal contains essentially no numbers. There’s a 2007 Eliezer post that I recommend ( The Simple Math of Everything ) about why it’s important to know the simple math of fields, even ones where you aren’t a professional expert; even knowing a little math is a major advantage over knowing no math at all. The proposal states: “…there was coordination to pick the largest stretch of land on Earth which had good year-round weather. And then that became ‘the great city’ and pretty much everyone who wasn’t farming, mining, or otherwise tied to a particular location moved house there.”
It’s made clear elsewhere that, in this proposal, a typical household lives in a detached, single-family residence, which is common in American cities but much less so in other countries. At this level of density, fitting the world population requires a truly enormous amount of land. 7 billion people divided by 1,400 people per square kilometer (Houston’s density) equals 5 million square kilometers ( map of the area required ).
This is called a “city”, but it isn’t really a city in any meaningful sense. A median pair of people in the “city” are a day or two of driving, or several hours of flying, away from each other. This kind of thing is usually called a “megalopolis” , where many different metropolitan areas overlap. The value of creating such a megalopolis for the world population isn’t going to be that large, since long distance trips are so rare; for most possible interaction pairs, all that will happen is that a six-hour or twelve-hour flight might become a two-hour flight. But the mean person only flies at all once every two and a half years, per IATA data; that figure includes many frequent fliers, so I’m sure the median is much lower still. (All of this ignores that ~1/3 of the global labor force still works in agriculture, per CIA World Factbook, and they and the infrastructure supporting them are tied to the land anyway.)
On the other hand, the costs of constructing such a megalopolis would be very large. Civil engineering projects (roads, power, water, sewer, bridges, tunnels, railroads, etc.) are extremely expensive, but that’s normally OK because they last forever, with a mean lifetime typically in the centuries range. Buildings are a bit more transient, but still last on the order of decades or more. The large capital cost is amortized over an equally large number of years of use. In this scenario, almost everything would have to be reconstructed from scratch. If we ignore all the expansion proposals, and assume the new infrastructure is the same as the old infrastructure in India, China, Indonesia, etc. (that is, in some cases very bad), the total reconstruction cost is very roughly on the order of 10x global GDP, or around $1 quadrillion US. Buildings and infrastructure are a quite large part of GDP, at a very rough guess around a third, and they’re typically amortized over a large number of years, so that results in a very high number. (All of this ignores that the proposed new infrastructure is specifically designed around 2015-era technology, which forces either locking in 2015-style equipment permanently, or paying to rebuild everything yet again when something new comes out.)
One other obvious large cost of consolidation is weather. In the diagram above, one might notice that one part of the “city” was in Phoenix, Arizona (average high June to August: 45 degrees C), and another part of the “city” was in International Falls, Minnesota (average low December to February: minus 20 degrees C). This isn’t because the US has bad weather; a global climate map shows the US has the best weather of almost anywhere. Rather, this is an inherent problem with putting everyone in one large, contiguous land mass. Water has a very high heat capacity, by far the highest of any common material, and so bodies of water greatly moderate climate by absorbing heat in summer and releasing it in winter. This, combined with the very low cost of water transportation (depending on who you ask, around 2% to 15% of the cost of truck transportation in cents per ton-mile), is why so many cities are concentrated on the coasts. Any map of Mediterranean (Bay Area-like) climates , widely considered the best kind of climate, will show that they’re scattered in various small pockets close to large water bodies. There’s an inherent contradiction between the goal of good weather, and the goal of putting everyone on the same continent at low or modest density.
Getting back to city size, everyone agrees that the obvious implication of cities being bounded by travel time is that faster travel could make cities larger, with major economic benefits. Unfortunately, fast travel in urban areas is hard. There’s been technology since the 1950s to accelerate people to very high speeds, but there’s an inherent tradeoff in transportation between mode speed and mode flexibility, which creates a “mode hierarchy” similar to the “memory hierarchy” in computer systems. One sees this hierarchy in every long-distance drive: you walk to the car, then drive on local streets, then arterial streets, then freeways, then back down to arterials, locals, and walking. Similarly, with planes, one walks or bikes, then takes road or urban rail, then gets on the plane, and then back to road/rail and back to walking/biking.
This tradeoff ultimately stems from problems of acceleration and deceleration, which in turn stem from what human physiology finds comfortable. In mass transit, such as planes and railways, stops that are close together mean that the vehicle has to spend most of its time speeding up and slowing down, which limits average speed on long trips. California’s high speed rail, at ~200 kph average speed, will have only four stops in the whole Bay Area. Elon Musk’s Hyperloop plan, at ~1,000 kph, has no stops at all between SF and LA. Jets, which fly at similar speeds, have no stops for the same reason. For roads, all vehicles in a given lane must travel at similar speeds to not run into each other. Freeways can’t have entrances to houses and side streets, because that would force the freeway into having either low speed (so that cars could accelerate from a stop quickly), or very low capacity (so cars would be spaced far apart, and could accelerate to high speed before crashing into other cars). A Toyota Corolla, a fairly typical car, has a 0–100 kph speed of 9 seconds; allowing 3 seconds for safety buffer, a freeway that had entrances and exits on random houses would only have a capacity of 3600 / (9 + 3) = 300 vehicles per lane per hour, around 1/7th of a modern typical capacity of 2,000. (Central lanes might have higher capacity, but in an urban setting, the road would still be severely restricted by the capacity of people to enter or exit.) This is (one of the reasons) why short trips aren’t done at freeway speed.
The proposal includes a plan to solve this problem by putting all roads underground, with full grade separation to eliminate traffic lights, and enough road capacity to eliminate all traffic jams and allow freeway speeds even for short trips. This… is not physically impossible, but would be enormously difficult, and is orders of magnitude beyond contemporary engineering capacity. In the current US, mean vehicle miles traveled (VMT) per capita per day is in the 40–50 km range. All of this is done in manually driven cars, with expensive gasoline and expensive insurance; and much of it is done at slow speeds, in heavy traffic, in difficult bad weather, and so on. Creating an enormous new underground road network, with cheap self-driving cars, no traffic, and freeway speeds everywhere, would result in huge levels of induced demand for highway travel. Standard back-of-the-envelope models predict the induced demand would be literally infinite, making the goal of freeway speeds everywhere impossible; this is unrealistic, but I think a 4–5x expansion in demand (~200 km VMT/capita/day) is quite reasonable. If this demand were evenly spread through the day, any one hour would have 1/24 = 4.2% of the total, but of course it won’t be even because of diurnal cycles. If we estimate that 15% of demand occurs during the peak hour, and we again use Houston’s population density (1,400 people/km^2), a capacity of 3,000 cars per lane per hour (arguably a bit pessimistic for fully self-driving cars, but there will always be some capacity cost to everything being at freeway speed; see detailed discussion ) gives us: 1,400 people/km^2 × 200 km/day × 0.15 hour/day / 3,000 lane/hour = 14 lane km/km^2 This is a tremendous amount of freeway to build. Houston, a very car-oriented city, currently has these freeways . At the same map scale, assuming that the freeways are a mix of six lanes and eight lanes wide, building freeways that densely would give us this .
Notably, this map doesn’t include all of the intersections between all these freeways. There are two major types of four-way intersection with full grade separation, the cloverleaf interchange and the stack interchange. At the proposed car speed of 125 kph, a typical highway turning radius might be a bit under 1 km. Compared to eg. San Francisco, a cloverleaf intersection with that turning radius would be about this size, to scale (the lanes are scaled up proportionally and would not really be that wide, but the problem is pretty obvious).
Even at current car speeds, these mostly aren’t used in urban areas, because they take up way too much land. The stack interchange, which uses gentler curves, would scale like this . OK, that looks a bit more reasonable (again, the lanes look thicker than they’d really be). The catch — there’s always a catch — is that this type of intersection requires four different vertical levels of road, one stacked on top of the other. This doesn’t include all of the huge number of access roads for cars to get to individual houses. And the proposed highway density is still high enough that, with an intersection between every pair of roads, all of the intersection ramps would still overlap with each other. This kind of network would wind up being enormously complex, and the entire city would essentially become a big maze of intersecting elevated concrete, with a few people squeezed in somewhere in between. So the original proposal calls for all these roads to be put underground.
This is a handy chart of the depth of various pieces of infrastructure in London . The shallow Tube lines at 5–6 m are built using “cut-and-cover”; the reason they’re so shallow is that there’s mostly nothing above them. They’re built under surface streets, and the whole street is dug up, the train tunnels are put in, and then the road goes back over on top. Obviously, this won’t work if the point is to get rid of the surface streets. And the Tube tunnels rarely have complicated intersections.
This London Underground track map shows each line of track and each train platform.
There are places where one pair of tracks crosses over or under another, but almost no complicated interchanges. To change direction, passengers just get off at one platform at a station where lines intersect, and walk over to the other one. If everything is built with freeways and cars, then to accommodate having four vertical levels of roadway at each intersection, plus roads to access actual above-ground buildings, plus things like power systems, drainage systems, ventilation, maintenance, emergency access tunnels, and so on, you’re essentially just going to have to dig up everything to a depth of around 50 meters, and then build the infrastructure as you fill the hole back in. And this is over the whole megalopolis, with an area of five million square kilometers.
During the 1950s, the US and Soviets proposed to use nuclear weapons for large civil engineering projects. Of course, this is a bad idea, because mixing dirt and nukes leads to tons of dangerous fallout. But just for fun, suppose you were going to do all of this excavation for your megacity with 50s-style “peaceful nuclear explosions”. The Sedan nuclear test had a yield of about 100 kilotons, and dug a crater with an average depth of around 50 meters, which is conveniently around the same depth we’d need to excavate to (of course deeper at the center and less at the sides, but hey, this is a thought exercise anyway). The crater had an area of 0.12 km^2, so to do excavation for the entire city, we’d need nuclear weapons with a total yield of: 105 kilotons/weapon × 5,000,000 km^2 / 0.12 km^2/weapon = 4.4 billion kilotons = 4.4 million megatons The combined global nuclear arsenal has a yield of ~6,400 megatons, so to do this, we’d need to use every nuclear weapon in the world, 700 times over. (Assuming the craters don’t overlap.) Of course, the fallout would make the Earth uninhabitable.
More realistically, the engineering capabilities for this are orders of magnitude out from anything humans currently have.
This is Steve Crane , the CEO of our friends at LightSail Energy, standing in front of the tunnel-boring machine used to dig the Eurasia Tunnel under the Bosphorus. That photo actually makes the machine look smaller than it is, because it’s much longer than it is wide (130 meters long, weighing 1,500 tons). This machine will be used to dig one 5 km roadway tunnel, with two lanes in each direction at a speed of 80 kph, and with no intersections except at the ends. The longest road tunnel in the entire state of California is 1.3 km long, one lane each direction, no intersections, with a maximum speed of 55 kph, and in the middle of Yosemite National Park so there’s nothing else around to get in the way (and built in 1933 so probably much less safe than modern construction; see Wawona Tunnel ). The type of freeway intersection which is proposed to replace every other road intersection in the world — fully underground and fully grade-separated — has, as far as I can tell, never been built, anywhere, ever. The closest thing that exists appears to be Boston’s Big Dig (cost $22 billion), which created a grade-separated interchange that was mostly underground, but was really a merger of two roads into one rather than a full intersection.
Okay, those are probably the biggest parts of the civil engineering problem. Getting into the other aspects of city design, there have been numerous attempts in the 20th century to master plan new cities on “greenfield” or previously unused land. The most famous are probably Brasilia, built by the government to be the capital of Brazil, and Canberra, built by the government to be the capital of Australia. Numerous towns have also been built by companies , though usually on a much smaller scale. There have been few catastrophic failures, but as a generalization, it mostly hasn’t gone all that great.
GaWC, which ranks the world’s cities by the degree to which they serve as global economic centers , ranks both Brasilia and Canberra as one-hundred-somethingth. Cities are complicated enough that no one really knows why, but to hazard a guess… in order to master-plan a new large city you have to be very powerful, and very powerful people can ignore any advice they’re given, even when the advice is useful and important. Eliezer might also say that none of the people involved were personally paid more or less depending on how well the city did, which causes obvious incentive problems.
In terms of immediate cause-and-effect, the biggest problem with these cities was most likely that the purpose of a city is to help people interact, and these cities made essentially all of their outdoor space useless for interaction. Of course, people can interact inside buildings, but in a well-designed city the outside areas are interactive too. For example, this is a street in Tokyo . This is a city square in Tallinn, Estonia . This is Central Park in New York City .
All of these places are designed for human interaction. For comparison, this is a satellite view of Newark, Delaware . Almost all of the outdoor area is designed for cars, and cars are so different from people — ~30 times larger, ~30 times heavier, and preferring hard concrete surfaces to greenery — that people basically can’t use this land at all. (Unless they’re inside cars themselves, but virtually no interaction happens inside cars, they aren’t really designed for it.) Even most of the green bits of that image aren’t really designed for people to use, but are just some landscaping to make all the concrete a little less ugly. (Eg., no one ever picnics in the grass in the highway median.) So in terms of fulfilling the city’s purpose, the large majority of the land area goes to waste.
This image illustrates the problem nicely.
Eliezer’s proposal describes the outside of the city, with all the roads and infrastructure buried, like so: “You stepped outside and you saw a green landscape full of trees, dotted with houses decorated as vegetation (which was a rule where we lived, and easy to do because our houses were centrally manufactured so we could have nice things). At night there were no street lights, just soft red lights that glowed along the walkways. Except for a 45-minute span, after all the last vestiges of sunlight had disappeared from the sky, when even the red lights went out so everyone could see the stars perfectly . Once a year on the winter solstice, on the Night of Stars, every light stayed out all night, so that those who wished could clearly see the constellations that appear before dawn.”
This is certainly an improvement over modern American cities — but then, almost anything is an improvement over modern American cities, given comparable GDP per capita. (It’s widely agreed in the urbanist community that America is terrible at this, and that bragging about beating Americans at urban planning is like bragging about beating a seven-year-old in a boxing match.) At the low densities this plan describes, the largest problem is that there is no real opportunity for spontaneous interaction. You might run into people, but only people who you’ve deliberately chosen to be around. You might go places, but only places you’ve deliberately chosen to go. You might see things, but only things you’ve deliberately chosen to see, and so on. Higher density allows for spontaneity.
This blurb also mentions mass-manufactured housing, which of course many people have tried to do before . Almost universally, the problem with mass-manufactured buildings is that they’re ugly, and they’re especially ugly when there are many of them in a row right next to each other . There’s no law of physics that requires this, and so a lot of people think that of course it will be easy to solve. You should be able to just build stuff that’s not ugly. But the same line of reasoning predicts that any competent manufacturer of computers, phones, or tablets should easily be able to compete with Apple. They have 40% gross margins, after all, and surely the cost of making stuff not be ugly shouldn’t almost double the price. And yet, every year, Apple sells even more stuff than they did the year before, and right now their main competitor is that they’ve saturated the market and have to compete against their own used stuff. When there’s this much data, you have to take the outside view, and measure the difficulty of a task by the number of apparently-competent people who keep trying and failing. (FWIW, this also goes for things like dieting and quitting smoking.)
“A lot of nerd tastes they share with the creative class in general. For example, they like well-preserved old neighborhoods instead of cookie-cutter suburbs, and locally-owned shops and restaurants instead of national chains. Like the rest of the creative class, they want to live somewhere with personality.
What exactly is personality? I think it’s the feeling that each building is the work of a distinct group of people. A town with personality is one that doesn’t feel mass-produced. So if you want to make a startup hub — or any town to attract the ‘creative class’ — you probably have to ban large development projects. When a large tract has been developed by a single organization, you can always tell.
Most towns with personality are old, but they don’t have to be. Old towns have two advantages: they’re denser, because they were laid out before cars, and they’re more varied, because they were built one building at a time. You could have both now. Just have building codes that ensure density, and ban large scale developments.” — Paul Graham, “How to be Silicon Valley” Okay. Moving on to other things: “The economic friction involved in moving house — e.g. packing and unpacking all your things — was responsible for the economy being slow-to-adapt to shocks. That the friction costs of moving house were a primary reason why people wouldn’t move fast enough to places where the economy was booming, or leave old jobs that weren’t good for them.”
It’s true, and very important, that mobility is a major factor in economic growth. And of course, all the inhabitants of a newly built city will have to be mobile. However, the physical cost of moving (which the proposal goes to extraordinary lengths to minimize, including making a number of recommendations for entirely-new classes of large civil engineering projects) is pretty small, and at a guess might come in as the seventh or eighth most important. For my two-bedroom apartment, in the high cost Bay Area, hiring a team to do all the work of moving cost around $1,200. It might be more for a family with kids, or someone who is fifty and has accumulated tons of junk, but it’s not going to be in the five figure range.
By far the largest factor making it difficult to move is that most people spend thousands of hours during childhood forming relationships with their relatives. These relatives then have these relationships with their relatives, and those relatives with other relatives, and so on and so on, forming a web through whatever city they’re born in.
The New York Times provides a convenient percentile chart , showing that 40% of adult Americans live less than 10 km from their mothers.
Probably the second largest factor is that, before the Internet, it was very difficult to have close relationships with people not in the same city, and so there were many levels of encouragement to form valuable social capital by staying in the same place forever. Much of US policy (and I’m sure other countries’ policy too) is, to a greater or lesser extent, explicitly designed around this goal. For example, San Francisco’s rent control is largely supported on the grounds that it gives tenants big economic incentives to stay in the same apartment for decades. This has, of course, become much less important with the modern Internet, but (as I sometimes like to remind people) the average politician was already forty years old when the Web was first invented.
The third largest factor, at a guess, is that there are competing incentives between valuable land going unused, and transactions not happening because liquidity is low. (Not distinguishing between “land” and “houses” here, because it’s really the same either way.) If at any given time, 25% of plots of land are available on the market, then it’s very easy to find one that you want, but everything will be 33% more expensive to make up for the unused resources. If 0.1% of plots of land are available, then deadweight losses are low, but finding a suitable piece of land is annoying and slow and difficult.
Even among simple, well-understood, mundane speed bumps, there’s lots of more important things to worry about than moving costs. The median home price in the US is $200,000, and so buying and selling involves a $12,000 realtor monopoly transaction tax, as well as countless other private and public taxes. There’s no real central matching between landlords and tenants, and so renting or leasing an apartment is a long, tiring search process. Enforcement against defection on the Prisoner’s Dilemma is weak, and so landlords lie about their apartments, tenants lie about taking care of their apartments, and this adds hidden-information costs to every transaction. People who want to live nearby have different schedules, and need to move at different times.
In general, for building physical stuff, and especially for anything that’s mass produced or that has to last a long time, it’s very important to get a sense of what all the problems are, and about how large they are, before investing resources into solutions. This keeps biting battery companies in the ass, for example. Every year, there are announcements of new battery chemistries that try to use cheap materials to make the batteries cheap. But the large majority of the battery costs aren’t material costs. They’re manufacturing costs, and specifically capital costs that must be amortized over very large production runs to make economic sense. So the battery companies spend a bunch of time and effort solving the wrong problem, and in due course go bankrupt. As Alon Levy says, “organization before electronics before concrete”.
There are some points in the proposal which seem to be trying to solve the problem of houses not being designed for people’s needs. As above, the proposed system of redesigning the entire civilization around things being movable is crazy premature optimization, but the problem is real, is not talked about very much, and really should be addressed by new city designs. Some of the effect is that MIRI-sphere people are just weird and it will always be more difficult/expensive to satisfy weird preferences, but another part is that governments keep trying to use building codes to enforce utopian social engineering goals. In many places in the US, for example, it is illegal for unrelated people to live in the same house, because the local government disapproves of people living that way (and disapproves of things that correlate, such as being poor). Almost everywhere in the US has zoning codes that make it illegal or de facto illegal to build lots of small units instead of one big unit, because the local government thinks people should live in families with children, and thinks that families ought to have enough money to afford a big house (if they’re going to hang around these parts, anyway). (See, eg., this Seattle archive image .)
It’s hard to identify what exactly in the proposal makes me think this, but I do also get the sense that it’s designed around the current lives of MIRI-sphere people much more than around average Americans or even upper-middle-class Americans, who would of course be an orders-of-magnitude larger customer base. The subgroups of the MIRI sphere overlap to a degree that’s very unusual everywhere else, except maybe in isolated small towns. The largest cause of long commutes, apart from zoning regulations and such, is that a typical American family forms Group #1 during the night, but during the day one parent joins Group #2, the other parent joins Group #3, and the kids join Group #4, and these groups have no relationship at all to each other and might be on opposite sides of the city. It’s essentially impossible for all of these groups to be close together at the same time, because each one has to optimize their own location based on an average of all of their members’ needs, and each member is also tied to several other groups, and the direction that one members’ groups are in and another members’ groups are in might well be polar opposites. I feel like this probably reduces to some kind of NP-complete graph theory problem, which you could have a computer come up with an approximation for, as long as everyone agreed to do what the computer said. But if you’re going to do that, you might as well propose building a Friendly AI that just solves all problems everywhere. :)
(There are also other biases like this; eg., I strongly suspect that most people don’t care about the night sky enough to create regular interruptions for it. I expect they’d appreciate the first time, and might even approve of the idea in theory because their mental image focuses on the first time, but that by the seventh week of it they’d just get mildly annoyed.)
There are a bunch of other small things that sound nice, but that don’t make any attempt to estimate capital costs, replacement cycles, weather-proofing costs, maintenance costs, etc., and weigh them vs. level of benefit: “The walkways were paved with bouncy material that people could run on without destroying their knees. There were central sidewalk-avenues, moving walkways that ran at morning and dusk when people were busiest.”
Just for fun, this estimate puts movable walkway cost at ~$400 per meter per year in combined maintenance and amortized capital costs. Assuming total walkway of 20 meters per household, this would be a total cost of $8,000 per household per year, not taking into account additional weatherproofing and maintenance costs from putting the walkways outdoors.
You might be able to do this part if you had sufficiently good and sufficiently cheap airborne drones: “Food in dath ilan was made by people who were very good at making a particular variety of food, and they’d pick a few dishes and make a huge amount of it on any given day. There’d be many places like that within 2 miles of you, and a small courier-carlike-thing would attach itself to another car and arrive with the food you liked within 2 minutes.”
Of course, tons of engineering challenges to make drones that aren’t crazy expensive, that won’t fall out of the sky and kill someone, that won’t crash into each other, that won’t get blown out from wherever they’re going (winds being a lot higher at altitude), that won’t run out of battery (li-ions are really not very energy dense), that won’t short out from rain, and so on, and so on; but I’m not an expert there and don’t really know how far along things are. As above, in this scenario the variety of food you can get will be directly proportional to density per square km. In a 3 km radius at Houston density, there are 40,000 people per restaurant target market, which is enough to support a decent variety of standard restaurants but will be constraining if food orders can’t be planned in advance. US average is 1 restaurant per 500 people, so 80 standard restaurants per area times 75 average menu items per restaurant equals 6,000 total options or ~7 people per option, and that would get shrunk by at least one and more likely two orders of magnitude if each option was an independent business in continuous production. (SpoonRocket tried a mass-produced food delivery model, but folded a few months ago .)
“Skyscrapers weren’t much built in dath ilan until we had extremely bright artificial light that could mostly substitute for sunlight” And of course, this isn’t going to happen anytime soon, because of the fairly extreme cooling requirements. Genuine full sunlight has a flux of ~1,000 W/m^2. A standard recommended air conditioner for a 100 m^2 space delivers ~5 kW of net cooling. Assuming perfect lighting efficiency, or if the roof is transparent and the sunlight just comes in naturally through glass, the air conditioning requirement is 100 m^2 × 1000 W/m^2 = 100 kW, or twenty times as large. It’s actually pretty easy to cook food this way — just make a box, put glass over the top to stop heat escaping through convection, maybe add a sheet or two of tinfoil, and watch it heat up to 100–150 degrees C ( solar oven, Portugal ).
I think that about wraps it up — there’s plenty of other stuff in the original post, but those are the major aspects in terms of urban planning and civil engineering.