In the videogame Zero Parades for Dead Spies, you play as Hershel Wilk, a middle-aged woman and a spy for Opera, the Communist Superbloc’s spy agency. You’re sent to Portofiro to do…something. You spend roughly half the game trying to figure it out. All you really know is that your enemies are the EMTERR bankers and the fascist Luzians. Whatever they want to do in Portofiro, your job is to accomplish the opposite.

The overwhelming emotion through Zero Parades is one of pointlessness. Like its predecessor, Disco Elysium, Zero Parades takes pains to emphasize that clinging to ideology is what people do when they don’t have anything left. If Hershel wasn’t a communist spy foiling the plans of the fascists, she’d have to face the reality that she’s spent her life in shitty hotel rooms, occasionally shooting other middle aged people in the head, and occasionally watching her friends and allies get shot instead.

Zero Parades, like Disco Elysium, is made by ZA/UM, a leftist artist collective/videogame company out of Estonia1. ZA/UM’s guiding mission seems to be to communicate what it’s like to be post-Soviet: no longer trusting in communism, but no longer trusting in anything else, either. Disco Elysium communicated this by being a postmodern detective game set in a fictionalized former Soviet bloc country. Zero Parades, as mentioned, communicates this by being a postmodern spy thriller set in a fictionalized Latin American country.

I prefer the Zero Parades approach myself. Not only because spying and counterspying was such an integral part of the Soviet and post-Soviet world, but because spying is the ultimate negative sum game. And that’s what I think Zero Parades is all about. It’s about how it feels to be a person whose goal is to just stop other people from doing stuff, and who’s willing to risk any amount of damage to themselves or those they care about to do it. It’s about how it feels to play the ultimate negative sum game: two spy agencies trying with all their cleverness and craft to enact exactly counteracting plans, with nothing gained and a lot of dead bodies at the end as a result.

And it feels bad, man. Alcohol, drugs, and cigarettes are a big part of both Zero Parades and Disco Elysium because it’s accepted that’s how you cope with how depressing and stressful the life is. If you’re not working towards building something, if you can’t even accept that something can be built, well, you might as well get drunk. Maybe that will make it better.

I’ve kept thinking about this game for the last week or so since I finished it because it’s really crystallized for me a thought I’ve had for a while. Mainly, my thought is about how zero sum belief is self-fulfilling. And, conversely, people can’t believe that the world isn’t zero sum unless they’ve seen it be positive sum.

This might sound trite. But I think it matters. In fact, I think if I had to point to a single reason why America is in the weird place where it is right now, I’d point to this.

And that’s weird to say, because America isn’t a zero sum country. Our economy has grown almost every year since its founding. We are not Russia or Venezuela, it is entirely possible for our neighbor to get rich and for us to get rich, too.

But we can’t figure out how to build things anymore. I keep going back in my mind to COVID2, right when we realized that things were serious. Everyone hunkered down for the long term. Toilet paper was nowhere to be found.

My dad, holed up in Florida, panicked, as he sometimes did. He took COVID seriously, in his own way. He resolved never to expose himself unnecessarily; to take up smoking again, seizing upon a half understood early study suggesting smoking reduced the risk of COVID; and to hoard any necessary supplies. He filled my parents’ house with paper towels, toilet paper, masks, gloves, and zinc (based off another vaguely understood early study about zinc maybe reducing risk of COVID infection). I’m not quite sure what the masks and gloves were for, given my parents didn’t go anywhere for the length of COVID, but whatever.

My dad’s hoarding wasn’t entirely rational, as I mentioned. He didn’t have definitive plans for the supplies. But it was somewhat rational. And, if I were to ask him, I think the rational part of his brain would say something like:

I’m not sure why or how these products make their way to my store shelves. All I know is that they likely come from China, which is having its own serious issues with COVID. In the meantime, the longer they sit on store shelves, the more likely it is they will be snapped up by one of my fellow geriatric Floridians and it won’t be available when we need it. Your mom and I can’t afford to take that chance.

And this isn’t wrong. It’s a little selfish, but it’s not wrong. Our manufacturing relationship with China isn’t just one of convenience or cost anymore. It’s one of necessity. We struggle to create key parts of our economy, including, actually, nitrile gloves, not to mention antibiotics.

So then let’s look at my parents, their wealth honestly accrued over a lifetime of highly paid professional services and capital accumulation3. They faced a world in which they were honestly afraid that their supplies of a key item might be finite. So, my dad grabbed what he could and sat on it, like a monkey at the zoo who’s trying to monopolize the zoo keeper’s bananas. If you don’t know where the banana tree is, you have to fight for whatever bananas are on the ground.

It’s such an ugly mentality, this scarcity thinking, this hoarding. It does bring out the worst in people. Right now, in August 2026, I can go to Whole Foods and rationally choose a small package of paper towels, just enough for me and my wife because we’re going to London soon. I don’t need to elbow anyone over it. If there’s a long line, I won’t go, and I definitely won’t go if I’m sick. If there’s only one package left, and it’s a big package, I won’t take it. It’s not that I can’t afford it, but that it feels entirely wasteful for two people to use one roll of a 16 roll package if a family could make much better use of it.

But we’re going to London soon, and we’ll have to scramble for an apartment (a “flat”). We’ll have to choose whatever apartment is convenient and livable, and we won’t have time to think twice about it. We’ll email a million brokers, and call a million more, and I’ll use my flexible schedule to elbow out other applicants who have less flexible ones. And if someone offers me a 4 bedroom for the same price as a one bedroom, my God I’ll take it. Better that than no bedroom.

You might say that’s just how apartment hunting is. But it doesn’t have to be. It’s that way because there are a lot of people looking for flats in London and not a lot being built, at least not in the near future. Each person like me who comes to London is faced with a small pool of available places. We’ve got to fight for the ones that exist.

And then, conversely, everyone who came out on top of this struggle over scarce resources accepts that as the way things are. In fact, they fight to defend it. They don’t accept building new apartment buildings because that causes disruption to their property values, “neighborhood character”, or morning routine; or building new nitrile glove factories because that would require land and utility negotiations. And, most importantly, they don’t really believe that either the new apartment building or the new glove factory would work, and it certainly wouldn’t benefit them. They’d rather fight over the limited supply that exists than trust in increasing the supply, because they haven’t ever seen the benefits of increasing the supply.

It’s ugly. It warps the soul. And that brings me back to America and to Portofiro. I watched UFC at the White House, recently, God forgive me. These were fights, set on the White House lawn, presided over by Trump and the UFC President Dana White. I’d call it trashy, but I watched it, so maybe I’m trashy too.

Anyways, one of the things that was remarkable about the event was that it was so closely tied to Trump that the advertisers on this event were half appealing to the UFC audience, half appealing to Trump personally. This meant that the normal advertisers, like Modelo beer (“brewed for those with a fighting spirit”) and Toyo Tires (“the official tire of the UFC”) were nowhere to be found.

Instead, the advertisers were those more ideologically aligned with the Trump vision. Which meant that the advertisers were Meta, Anduril, a bunch of gambling apps/prediction markets, the Trumpcoin project, and a separate Trump gold coin project.

It felt like the most zero sum vision in existence. Every single advertiser4 came from a world where no growth exists, where it’s impossible to invest in the future and grow prosperous. The only “positive” future, according to these advertisers, is one where you, the audience member watching the fight, get rich from someone else’s wealth. Maybe it’s the wealth of the foreigner that Anduril’s drone killed. Maybe it’s whoever is on the losing side of your oh-so-clever parlay on the gambling app or prediction market of your choice. Or maybe it’s just whatever sucker you manage to sell your worthless Trumpcoin to for precious USD that you can actually use to buy things.

And then the joke for all of them is, of course, that that positive future’s not for you. You, as the audience member, are not allied with Anduril, the prediction market, or the memecoin maker. They are advertising to you as the sucker, because they do truly believe in a zero sum world. They just believe that their positive future is one in which they take that money from you5.

And the predatory gaze behind that bleak vision makes me shudder. I know that America and the world grows every year, and that I can get rich while my neighbor gets rich too. I know that honest trade has been the key to the human race’s prosperity since time immemorial, and it’s very easy to sell things to people that you honestly think will improve their lives6.

But if we keep acting like the world is zero sum, it will become zero sum, or even negative sum. And then at some point, we’ll all find ourselves like poor Hershel Wilk, our allies and our enemies flickering out of existence, disappearing and blending together amid trash heaps of irreproducible stuff made by a dead civilization infinitely more capable than our own. Will we still be arguing over which faction deserves the biggest pile of trash then?

Instead of letting the Earth fall into the wrong hands, the defenders of our green dot decided to destroy it. They flooded the streets with lava, telling their fellow humans that they needed to leave, permanently. The last ship on Earth just barely escaped, taking off just as the wave of lava reached the launchpad.

I wasn’t on that ship. I was on the top floor of a skyscraper, writing the chronicle of the last days on Earth. My fight against time was the fight to get that chronicle published before I no longer could, before there were no more servers left on Earth. In my dream, I managed it, sending my chronicle right as the ship left Earth. Then it was just me, alone, on the top of a skyscraper in a ruined Earth.

I don’t think it takes too much of a Freud to interpret this dream. Congenitally, I’m prone to pessimism and nostalgia. And right now, in the US, and maybe globally, does feel like an end to things. It’s an end to the American century, for one. The US just lost a war to a middle power1 after sneak attacking them during negotiations, and this time it wasn’t just the “what comes after” that stumped us. We can no longer enforce the Pax Americana or free navigation of the world’s waterways. We also can’t keep screwworms out of our cattle or algae out of our Reflecting Pool, tasks we accomplished in 1966 and 1920, respectively.

It may be the end of the human millennium, as well. We are rapidly building and deploying our successors. I use Claude (and, to a lesser extent, ChatGPT) so heavily in my work that I legitimately don’t know what I’d do without it. When the LLMS briefly went down last week, I blankly stared at this scientific protocol that I was supposed to review for a solid 30 minutes. Then I just waited until Claude came back online, and together we reviewed it in 5 minutes. Right now, I am a human brain riding and guiding Claude’s powerful, impulsive elephantine intelligence, but this scenario can’t last. Dependence will turn into succession.

And then, it’s the end of a New England world for me. My wife got her dream2 position in London, and we’ll be moving there this fall for the foreseeable future. My work will continue, and my company will remain US-based, but I personally will be gone from this great, broken nation of ours. It’s a little dramatic calling this the end of the world, but I’ve spent roughly 28 of my 33 years on Earth in New England, and 4 of those were in New Jersey. To an embarrassing extent, New England is my world, and this version of me is ending.

So here I sit, on a Sunday morning in Beacon Hill in my second floor apartment overlooking quiet tree-lined streets and old houses, their red bricks warming up slowly but not yet hot, and I chronicle the end of the world. And then I click publish, before the moment is lost and I am forever silent.

Like most lessons learned in philosophy, this is striking at the time and then quickly fades into irrelevance. Sure, we might not ever know if the cars zooming by are “really” there or just illusions, but we should probably still look both ways before crossing the street. Science majors, especially, discard these thoughts. This is a pity, because there’s nowhere where acknowledgement of this layer (or layers) of abstraction matter more than science.

Let me give an example from my own life. Bear with me for a second, as this example might seem obscure at first. About a year ago, I was on the Redline subway heading up to Porter Square. I noticed a man slumped over, seemingly overdosed, on the seat across from me. I notified the MBTA authorities through their “see something, say something” app, and an MBTA official came on the train to check on the man.

She shook the overdosed man’s shoulder, but he didn’t wake. She used her walkie talkie to report, “Passenger is non-responsive”. Then she resumed trying to shake him awake. Her colleague, meanwhile, started banging on the metal pole next to the overdosed man.

There was a homeless man also on this train. He was annoyed enough that the train was stopped for this overdosed man (he didn’t know I was responsible for the stopped train, fortunately). But when the woman radioed in “Passenger is non-responsive”, he got very worked up.

“Non-responsive is non-responsive!” the homeless man exclaimed repeatedly. “What’s the use in trying to wake him up? He’s non-responsive!”

Contrary to the homeless man’s protestations, the MBTA officials did manage to wake the overdosed man up. Apparently, using your heavy keys to bang a metal pole right next to your ear cuts through even the heaviest of fentanyl slumps. The no-longer-overdosed man stumbled off the train at the Harvard stop.

What’s interesting about the reaction of the homeless man is not how strange his logic was. On the contrary, he made a classic logical mistake. He just made it in a very illustrative way. He mistook the map, namely the MBTA woman’s hasty diagnosis of “non-responsive”, for the territory, namely the reality of the man heavily sleeping on the train. Or, to put it more precisely, he insisted on the primacy of the more abstract map, the binary diagnosis of “non-responsive”, over the more detailed and less definitive map of his own perceptions.

I don’t point this out to make fun of him. I point this out because this is an issue all scientists must face, whether we be street scientists on the train or PhDs in a lab. We always deal with abstractions and maps of our data. Generally speaking, we want more detail in our maps over less detail. “I shook the guy’s shoulder and he didn’t wake up” is better than “non-responsive”. And more detail comes with being closer to the territory/data: we can go from “we shook the guy’s shoulder and he didn’t wake up” to “patient does not respond to wake-up attempts”, but not vice versa. Once we boil down to the latter, we sacrifice that detail for certainty.

This general pattern, of details and uncertainty being most prominent at the levels closest to the data holds true everywhere in science. Take GLP 1’s. They work, for sure. But if we dig down into those impressive clinical trial results, here’s what we might find:

Level -1: a participant in the clinical trial taking Zepbound wakes up at 2 am. His stomach hurts like hell for 2 hours, then he vomits.
Level 0: The participant reports his nighttime sickness to the investigator. He simplifies it to “last night I had bad stomach pain and vomited”. In his mind, he’s already compressed the 2 hour gap between the events.

Level 1: The investigator has to put this data into his “map”, which is going to be a spreadsheet of adverse events. He interviews the patient, then puts “pancreatitis” and “likely trial related” into the spreadsheet. Could it have been food poisoning? Maybe, but now it will forever be definitively “likely trial related pancreatitis” in that map.

Level 2: The writer for the clinical trial report has gotten to the safety section. He adds one more to the adverse effects column, and then reports the entire number in the discussion and abstract. “Likely trial related pancreatitis” has become definitively one serious adverse event in this trial.

Level 3: A journalist writes up the trial. He skims the paper and sees there were only a couple serious adverse events, and they were all intestinal. In his article, he writes “Zepbound is very safe. Only 2% of people experienced serious adverse events, and they were all gastrointestinal”. One specific case of pancreatitis has become a rounded figure, packaged as part of a larger narrative that Zepbound is safe.

Level 4: A doctor skims the journalist’s article. He sees that Zepbound is very safe, with only 2% of people experiencing serious adverse events. He tells his overweight patient, “Don’t worry. Almost nobody has any bad effects with Zepbound. It’s very safe”. 2% has become “almost nobody”.

Level 5: When the patient is explaining his prescription to his wife that night, he says, “The doctor says this is super safe and nobody has bad effects. That’s why I feel ok taking it.”

At every level detail was lost and definitiveness was gained. This isn’t a bad thing, necessarily. The average Joe doesn’t need to know every specific adverse event that was in every single trial of Zepbound. He certainly doesn’t need to be personally poring over the original trial participant’s health records, trying to decide if that really was pancreatitis or just food poisoning.

But it’s important to note that information was lost. The patient, or at least his doctor, should be aware that their map is many levels divorced from the territory. Most importantly, if they find their map starts disagreeing with their own exploration (e.g. if the doctor’s patients start all coming down with pancreatitis), they need to be willing to go back up levels. Maybe the writer for the clinical trial didn’t realize that obese men over 60 were very overrepresented in the pancreatitis cases. If so, then it’s up to the doctor to go back to the original trial data in the supplements and start warning his obese elderly patients accordingly. Otherwise, he risks being like the homeless man: insisting on the primacy of a map that no longer corresponds to the territory they’re on.

Last example, one from my own life. I’ve been hard at work creating NeutraOat, my modified oat fiber supplement. This has required a fair amount of chemistry and experimental work. None of this chemistry or experimental work has been done by me. It’s been done by my partners and contracted labs at the University of Alberta and elsewhere.

So, when I say “I’ve been hard at work”, I don’t literally mean I’ve been directly interfacing with the oat fiber. Instead, what my work looks like day-to-day is that I receive maps: documents, spreadsheets, PDFs. I try to interpret the maps with respect to what they mean about the oat fiber: e.g. what does this spectrographic analysis mean about our latest manufacturing run? Then I verbally transmit my ideas back to the people who are actually doing the chemistry.

In order to do this, I have to reduce a real, physical object to a few measurements. Or more precisely, my chemistry partner does, and I then work with it. Every time we talk, we are not really discussing then, the actuality of NeutraOat. We’re discussing a few labels.

This has hurt us in the past. Early on in our experiments, we had a certain target chemical profile that we knew we were trying to achieve for the oat beta glucan1. We carried out an experiment on our oat fiber, thought we achieved the profile, and then carried out a UV spectroscopic analysis to verify that we had the correct chemistry. It seemingly confirmed it.

But, on further analysis, the flat, 2d information of spectroscopy, even though it carries much more information than the binary of “non responsive”, proved to still not carry enough. We hadn’t washed out our solvent. Our nice peak was actually just from the solvent, not the modified oat fiber. So, the oat fiber was there, and the immediate interpretation was right (there was a peak representing a certain chemical element), but we had latched onto a piece of information too quickly without thinking enough about what it meant.

So, we mistook the map for the territory. This is a nice, neat ending to this blog post. Quick twist, though: the “further analysis” was done by Claude, a creature entirely fed by maps who only sees maps. It turns out that familiarity with the territory wasn’t enough for either me or my chemistry partner to diagnose this issue. The real expertise came instead from someone, or something, that knew much more than we did about the maps themselves, and had distance from the data.

As I stated in the beginning, maps are unavoidable. We navigate the world through them, and we can’t ever actually drill down all the way to the territory. Getting as close as possible to level 0, to the ground truth of the data, is important. But sometimes, understanding how the maps are made, and the ways they can deceive, is even importanter.

Having kids has always sucked in very real ways. Kids are expensive, noisy, dirty, and anxiety-inducing. This is true today, this was true for my father’s generation, and this was true for his grandfather’s generation. Some (or sometimes a lot) of the noisiness and dirtiness has been historically offloaded onto women, who didn’t have as much choice as men as to whether to have children. But, that came with the cost of placing a strain on the mother of your children, which has its own negative consequences for your life.

I think this is obviously true. The only people who ever experienced a concrete benefit from having children were agriculturalists, who could see children as an investment, and those who needed their children to be their retirement plan. Neither of those conditions have applied to either side of my family for at least 100 years.

I’m not sure why, then, that my grandfather, and my father, had children so automatically. My grandfather had 5 by the age of 35, my father 4 by the age of 40. This was not that unusual in their time, but I know nobody my age on track for that. This is true both statistically and personally: across the country, and across the world, people my age are not having children. They say they are not having children because children are a lot of work, but, as I said, that was equally true of previous generations who had a lot of children.

Before I try to answer that, though, I want to instead take a detour into a different strange, costly activity that a lot of men have historically participated in. Bear with me as I do this. This one is noticeably worse for your life than having children and yet, men have historically done it anyway. I’m talking, of course, about participating in warfare.

Warfare, from the average French peasant’s perspective, was a bad deal all around. There’s no good reason, biological or otherwise, for a French peasant to participate in it. His ability to enjoy life or to spread his DNA would end under the hooves of an English warhorse. But, the French nation, like all nations across the world, had to convince the French peasants to fight or the nation would end. This might not actually be bad for the French peasant, but it would have been bad for the heads of the nation.

Here are some of the ways that nations have tried to convince their young men to participate in war.

1. Paying people to fight. If you pay a peasant more than they can get on the farm, they will join your army. They will not have any loyalty to you, specifically, though, and they will do their best to remain alive to be able to spend the money, even if that’s bad for your military strategy.

2. Forcing people to fight. If you tell someone that they need to fight or you’ll kill them, they will fight. However, at some point you’ll have to give them a spear or sword, at which point negotiations get trickier. Also, they will try to run away at every opportunity.

3. Appealing to their masculinity or honor. “Fight if you’re a real man” works surprisingly well on some men. Unfortunately, if that’s the entirety of your army, your army will be composed entirely of meatheads. Also, this will not work if your army is losing, because tough guys do not want to join the losing side, unless you’re really good at selling the “glory in defeat” line.

3. Appealing to God, Queen, or country. This is probably the best method if you can make it work. The Crusades managed to send people across the world multiple times for God. The Soviets broke the German Army in wave after wave of human shock troops who believed in their country. Napoleon built the greatest army in Europe by convincing random provincials that they were French, that he was French Emperor, and that they needed to built a French Empire.

Most modern countries, who haven’t fought a desperate war in living memory1, have mostly forgotten the importance of this messaging. America, especially, has mostly forgotten this. We’re so used to bombing people from the sky that we have forgotten that sometimes our enemies shoot back and then soldiers (and their families) question why they put themselves in a position to be shot. This is why we are notoriously bad at absorbing any amount of casualties.

The countries that haven’t forgotten this, like Russia, are fascinating to look at. They start indoctrination of children young, teaching children how to fire AK47s and celebrating their great military victories (and the costs of those victories) in school lessons. By the time those young men are 18, some significant proportion of them are ready to do the stupidest biological move possible, and die unnecessarily without having spread their genes.

Now, with that in mind, let’s go back to the children question. As I mentioned, the fertility rate has been inexorably declining since my grandfather’s time. Korea is down to 0.7 children per women; China is down to 1.00; Japan sells more adult diapers than baby diapers.

Governments interventions have been powerless to reverse the change: housing subsidies, childcare subsidies, propaganda campaigns, IVF subsidies, none of them work. Even the ones that seemed to be effective for a long time, like Finland’s combination of convenient subsidized childcare and pregnancy care, are failing now.

When I look at these government interventions, all of them strike me as besides the point. As we mentioned above, raising a child is a hard activity. It sucks. It’s always sucked. But, for previous generations, that didn’t seem important.

I mean, let’s go back to my family. My grandfather (my dad’s father) was part of the WW2 generation. He almost went to war himself, but got injured in basic training. Immediately after returning from war, he enrolled in medical school at Harvard, got married to an Irish/British immigrant, and had a child, my father, in a Beacon Hill apartment. Four more babies followed over the next 10 years. During this time, my grandfather slowly climbed into the middle class as he went from job to job, moving cities frequently. His wife mostly did not work. Their marriage was, frankly, terrible, and it ended in a bitter divorce and lots of drinking on both sides.

There are some parts of my grandfather’s story that work towards having a family, like having a reasonably well-paying job, a wife that didn’t work, and government-funded professional training. But, still, I guarantee you that there are almost zero 23 year-old medical students in Beacon Hill apartments today that are even stably partnered, nevermind having a baby.

Why did my grandfather not just have 1 child, but 5? His children did not contribute economically to him2. They were living a reasonably stressful life, with a lot of moving from job to job. I never got the sense he particularly liked children. He didn’t have to handle childcare, true (his wife did that), but then that just raises the question of why his wife, my grandmother, wanted 5 children herself, or, depending on how you frame it, why he wanted 5 children for her. For the record, it wasn’t that they had help, either: they moved around way too much for that.

The answer, of course, has to come down to, “That’s just what people did back then”, but why? Why did people get married and have kids back then, and why has that changed now? And if your answer is “birth control” or “women’s rights”, I accept that, but then I’d ask the same question about my parents and my aunts and uncles. My dad had four children, his siblings had 3, 3, 1, and 0 respectively. Unlike their parents, they were having children in a time3 when women had both access to birth control and careers that they were proud of. And then, my generation: my siblings have 2, 1, and 1 respectively (I have zero so far). Of my 11 cousins, there is precisely 1 additional baby.

Ok, one last piece to the puzzle, and then I’ll see if I can solve it, and get back to the warfare thing. The only groups in Western countries that have sustained high birth rates are religious ones: ultra Orthodox Jews, evangelical Christians, the Amish. Nobody else has, not even farmers, who have an economic reason to do so. What was unremarkable for my grandfather’s generation (who, for the record, was an atheist) requires a religious fervor now.

And maybe that’s the rub. The religious are the last people left who hear and take seriously the command “be fruitful and multiply”. It’s not that they don’t find childrearing inconvenient, it’s that it’s as important to them as recycling is to a urbanite, or a manicured lawn to a surburbanite. Their worldview is that God wants them to have lots of children, and having lots of children is good. Whether it’s fun (although hopefully it is) is besides the point.

So, what we’ve forgotten in modern society is that it’s not a coincidence that the Upanishad, the Vendidad, Plato’s Laws, and the Bible all tell their readers “you have a moral duty to have children”. That was a loadbearing belief of ancient society, which kept women going through one of the most medically dangerous events of their lives over and over again and kept their men pressuring their women to do the same.

We forgot that. We took away that pillar. No, scratch that. We smashed that pillar. We spent decades telling people to stop having children because of overpopulation. We elevated Margaret Sanger and Paul Ehrlich to the highest points of Western civilization, and watched their global influence grow as our two most populous countries, China and India, instituted mass sterilization campaigns for fear of starvation from overpopulation. We so misunderstood our own kind that we thought we had to force women to not have kids, only to see just a few decades later that you can’t pay most young Chinese and Indian men or women enough to have kids.

Many of my readers would argue that this is a good thing. They’d argue that a loadbearing (or childbearing) part of the old system was coercion of women. Women were sometimes literally forced to have children. At the very least, their lives were made unpleasant if they didn’t marry and have children. This is true in a lot of ways. My grandfather did not force my grandmother to have children, but marriage was the only way she saw to get out of the grinding poverty she faced as a young woman with no education or prospects in 1940s New York City4. After they married, the only way to stay married and financially stable was to have kids. My grandfather wouldn’t have stayed with her if they didn’t have kids.

And, so, maybe it’s a good thing we are not back in high fertility days. It is definitely a good thing we are not back in those days of women being second-class citizens. I’m not here to argue that. But, it does seem like we have somehow lost something important, if we are facing a scenario in which wealthy Korea is slowly, voluntarily extirpating itself. It’s like watching a wealthy man’s son retreat inside his room and just watch Netflix and eat takeout food all day.

But, in that case, what do we do? Throwing money at the problem doesn’t help. As I mentioned before, every government intervention so far has proven to be, at best, short-lived. Even worse, the most effective ones have been horrific, like Romania’s reverse One Child policy under Ceausescu.

Well, my argument is that, if we don’t want authoritarian countries to try more horrific coercive strategies, we have to take a page from the warfare book. We have to convince people that it is good to have children. It’s good for God, Queen, and country. We have to tell kids in school that literally every modernized country5 has a shrinking fertility rate, and that’s bad for the future of the country. We have to tell them with the same fervency that my generation got told that about recycling and dwindling oil supplies, because, guess what? That messaging worked!

We want to be good. People want to be good. Americans want to be good. They want to donate, volunteer, and have the right opinions. They will take extra time to sort the recycling and cut plastic rings so that the turtles don’t choke. I know this for a fact because I still do this.

Right now, the messaging that children get is that sex can give you STDs and teen pregnancy is bad. The messaging that young adults get is sex is fun and consent is important. The messaging that actual adults get is that having children is a private act, but, if you do it, you have to make sure to treat your children well.

There’s nothing wrong with any of these messages. But none of them contain any mention of it being good to have children, or that being a parent is good, or even that doing an ok job at being a good parent is better than not being a parent at all. Most of us have lost the message of the Bible, the Upanishads, and the Vedas6. Then, when we face the obvious reality of the world, that having kids sucks in important ways, we balk.

In conclusion, having kids can suck. Having kids can also be great. Parenting is important. Also, it’s not a big deal, and most kids will turn out more or less fine, even if their parents are screwed up like my grandparents. All of these things are true. But the vibes just focus on the negatives.

Part of that is, of course, explained by David Foster Wallace’s “fish in water” joke1. I live in Boston, and it surrounds me on a day-to-day basis, so much of Boston is basically invisible to me now. It’s what I look through to see the world, and it’s hard to stop just looking through it. This problem is compounded by the fact that Boston is not so different from Connecticut, where I spent the first 18 years of my life.

But there’s always been another piece of it that I haven’t really been able to put my finger on until right, well, now. As I write this, I’m sitting in the Congregational Library. This is a small library, next to the Boston State House, in the middle of downtown Boston, taking up the second floor of a 19th century office building. It is, technically, open to the public (at least according to their sign on the second floor), but you have to buzz in at the front door. Judging by their surprise when I buzzed in, I’m guessing almost no visitors ever actually visit. Currently, there are three staff members in this library, and then me.

Even though the library is open to visitors (allegedly), most of it is actually inaccessible. The only part I’m allowed in is the reading room, which is a beautiful old room overlooking the old Granary cemetery, final resting place of luminary American Congregationalists, like Samuel Adams, Paul Revere, and John Hancock. The centerpiece of the room is a long oak table with three lamps. Overlooking the table are no less than 5 portraits of old white men, presumably Congregationalist ministers2, their serious expressions staring down at me as I write.

When I first came into the library, I was trying to ask the librarian about where, exactly, I had stumbled into. She told me this was the library of the Congregational Church, which I had heard of but knew nothing about. She graciously tried to explain it to me3, which I vaguely understood. But, recalling her explanation, and with the help of AI, I think I’m getting a picture of it. And that, in turn, gives me a picture of Boston.

You see, the Congregational Church was the state church of most of New England4 for 2 centuries. It continued through the early 19th century, finally ending in Massachusetts in 1833. It was, and maybe is, the structural manifestation of Puritanism, when Puritanism turned from a radical separatist cult into something that needed to create a functioning government for the northern American colonies.

As it turns out, when the Puritans’ radical “everyone is going to Hell” passion died out, what they really cared about was a sober, sensible government that derived its legitimacy from the consent of the governed (mostly). They cared about stability, an educated populace, industry, and balancing tolerance with their belief in the primacy of their own values. They were, basically, the original WASPs as a social class.

The Congregational Library, in which I now sit and write this piece, was an extension of these values. It was founded in the middle of the 19th century, in Boston, next to the State House, to be a repository for Congregational books and to be a quiet representative of Congregational values next to the seat of power, not coincidentally just as Congregationalism stopped being the state religion of Massachusetts. It was the center of a well organized, well funded network of Congregational institutions, especially their missionary societies, which existed at home, abroad, and in the women’s movement. This library continued to exist throughout Congregationalism’s long decline over the next 170 years, as it merged, split, merged again, split again, and lost progressively more political power, organizational energy, and members.

This brings us to the present day, where Congregationalism, a denomination which only vaguely exists anymore, still has a library dedicated to them in the middle of Downtown Boston. They used to have the entire building, but they sold it in 2017 for $25 million and a 100 year lease, ensuring their existence more or less in perpetuity by American standards. This is a strange existence, given that they are a library full of books that nobody needs and staffed with librarians who don’t particularly want readers5.

And all of that brings me back to Boston. Because Boston, much like the Congregational Library, is old and hidden. The old part is obvious for anyone who visits Boston and sees the historic sites. The hidden part is, ironically, harder to see, and why it was so hard for me to write about Boston. Because Boston’s tourist sites are America/world-famous and very well trodden, it’s easy to miss the two realities that define Boston institutionally:

1. The institutions of Boston are, to an astonishing extent, both old and resistant to change

2. As a corollary to the first, it’s very difficult for outsiders, or insiders, to even know how to look at them, nevermind how to change them

These two realities extend throughout broader Boston, not only to the universities, but also to the politics or even industries. For all of them, it’s difficult to tell who has money, who has power, and what anybody’s opinion is.

I mean, I live in Beacon Hill, historic, beautiful center of old Boston, where condos go for millions. I just found out a few years ago that the biggest mansion in Beacon Hill is owned and occupied by the Beacon Hill Historical Society. Practically speaking, that means the mansion sits almost empty the entire year, and the only people who are allowed to be in it are those who can convince the unelected board that they are “historians of the neighborhood”, in which case they get to work there for free. To emphasize: it does not matter how much money you have or how many followers you have on Instagram. Occupancy in this mansion cannot be bought.

Or, to put it another way, Boston will never have a Zohran Mamdani, Marc Andreesen, or Joe Arpaio. The city would not allow it. This is a city where quiet, old implacable, slowly growing nonprofits ruled by a “meritocratic” elite both are the largest employers (the hospital conglomerates), the largest landowners (the universities), and the most important source of exports (the universities again, if you count their international students). The well-paid PhDs and MDs at the top of the Boston social class who have carefully engineered their own rise through these nonprofits’ ranks for decades do not appreciate people upsetting the apple carts..

The red bricks of Boston are built on the bones of the exact Congregationalists who built that library. They are mostly gone, but left Boston a legacy of quiet industry. And, because they are gone, they also left Boston a legacy of stasis. There have been some changes since the heyday of Congregationalists, most notably the influx of Irish and union politics6. But the hidden stability of Boston, the iceberg nature, both metaphorically and literally7, can’t be overstated.

And so myself, as someone who loves to write but has zero ability or patience for penetrating social structures, is left peering into the windows of the Boston institutions, occasionally managing to finagle my way in before being asked to leave because the librarian needs her daily lunch break and I need to not be there when she takes it. Eh, at some point I’ll move back to Cambridge.

I understand that many people would see this as spiritually miserly. My dad, for instance, seemed to take a certain pride in doing the dishes. Well, to be more precise, for a long time in my late childhood, he took a certain pride in my brothers and I doing the dishes, with him carefully inspecting. It was only after we left the house and only came back for short visits that he started again taking pride in doing the dishes when we came to visit. In fact, he’d insist on it, and get anxious if one of his guests usurped his position as dishwasher.

Maybe the mismatch, then, is less between cooks and dishwashers and more between people who see dishwashing as a chore vs. those who see it as a craft. I see it very much as the former. My dad saw it as the latter. Some part of this is definitely our relative technological capabilities (I’ve been without an automatic dishwasher in my apartment for about 4 years now, and miss it terribly). But some part of this is a temperamental difference between those who’d rather dishwashing just be over with already and those who take pride in doing it well.

I bring all this up not because dishwashing is itself important. It’s really not. It is, however, one of the eternally repeating threads in the fabric of our lives. It’s not important, but it is there.

In the history books and in our biographies1, they will mention the wars we fought in, our great loves, and our great losses. They will not mention how long we left the pan soaking in the sink or that our spouse had to redo our scrubbing because we missed a spot. This is fortunate for me, as I would be seen as one of history’s minor villains on both accounts.

But the dishwashing will be written in small letters in those books between the lines, and, for those who did a particularly thorough job of it, the light reflecting off those sparkling clean frying pans will brighten the halls of eternity. Maybe.

Anyways, just wanted to say that I love you, Dad, and I’m sorry I’m bad at dishwashing. I’m working on it.

The school event coordinator, Pasha, documents the change in his school. He gets increasingly frustrated with it, and starts sabotaging the propaganda in small ways. This causes the eye of the Russian government to turn its malevolent gaze on him, and he eventually has to flee Russia.

Before he flees, he narrates to the camera everything that he loves about his hometown. It’s an elegy to a homeland he might never return to.

I’ve been feeling a bit like Pasha, recently. Not that I think I’ll have to leave America and never return, but that I feel like something vital is being lost from the America I love. To be honest, this isn’t the first time this decade I’ve felt this way, either. It’s been a tumultuous 6 years.

Anyways, I wanted to put my own list of what I love about my own homeland.

I love America.

I love the surprising variety of Chinese food in Boston’s Chinatown.

I love that Uber and Doordash, thoroughly American inventions, have even reached my small hometown in Connecticut and that I can get chicken tikka masala without needing to get in a car.

I love that I can generally trust people on the street, that nobody I know has ever gotten mugged or pickpocketed.

I love that people let their dog off leash at Boston Common and how delighted the tourists are to see the dogs romping around.

Speaking of Boston Common, I love the fat, tame squirrels; the one dollar drawing guy, and that they replaced the Earl of Sandwich with a Flour that sells hot dogs and soft serve.

I loved my house growing up, with the apple tree in the backyard overlooking the pond. We’d go swimming in the summer and ice skating in the winter, although only after the neighbors had tested the ice. In the fall, my Parson Russell Terrier and I would gorge ourselves on apples.

I love Disney World, and Las Vegas, and Halloween, and every other American celebration of stupid excess.

I love asking for cups of water at coffee shops and not getting charged for them.

I love the New England landscape that I grew up in, small towns separated by patches of forest.

I love that the landscape changes so thoroughly when you go out West, and people disappear and you can see for miles in the desert.

I love snow days spent playing videogames on the couch.

I love public libraries, and that anyone can spend a day reading a book or using a computer just by wandering in.

I love American restlessness, the way our culture changes because it’s never quite good enough.

I love that Irish Americans are more proud of Ireland than the Irish, and Italian Americans more proud of Italy than the Italians.

I love the rapid changes of fortune, when you read about someone who invents something clever and gets rich in a year.

I love free speech, free press, freedom of religion, and the entire cultural apparatus that allows idiots and geniuses to broadcast their thoughts to whoever will listen.

I love the American mythology: George Washington, who led the country to independence then refused to become its king; Abraham Lincoln, a white man who freed the slaves and died for his convictions; Martin Luther King, who refused to obey unjust laws and was imprisoned, then killed for his beliefs.

I love how delighted my Indian friend was when he came to an American supermarket, and saw that the supermarket sold plastic bottles with mouths designed for pouring without dripping. “We never would have thought of that,” he told me.

I love America, with all its flaws and problems. I refuse to be ashamed of it, or to let idiots and malefactors pretend to dictate on the world stage what America is. This is my country, not theirs.

Specifically, let’s talk about the fact that we have now figured out a repeatable method for generating artificial intelligences. And I mean, intelligences. Things that can see, hear, talk, write, speak, and reason. Pieces of software that can see a math question handwritten on a whiteboard, interpret it, answer it, and then turn the answer into a poem.

These intelligences are not perfect, but they’re damn near, and the gaps are shrinking each day. Even the hardest tasks in physical intelligence, the ones that small children and apes fail, like folding laundry or loading a dishwasher, are starting to become manageable.

I didn’t think we’d see this kind of progress in my lifetime. Almost nobody did. When I was in college 10 years ago studying philosophy of mind, and when I was just out of college teaching it to Chinese college students (long story), questions of where intelligence comes from, or free will, or “qualia” were purely theoretical. I mean, sure, there was neuroscience, but the gap between overheated lectures about Broca’s and Wernicke’s area and the reality of being a human being who can come up with a dirty limerick in his mind and then shout it in a lecture hall was enormous.

But now we have a real, experimental way of looking at these questions. We still can’t build a biological brain that does too much of anything, but we can build an artificial brain that can both come up with a dirty limerick and reason about appropriate places to share it. It’s often said that experiment and theory walk hand-in-hand. Well, at long last, experiment of mind is a thing.

So let’s get back to theory of mind. And that’s where A Brief History of Intelligence comes in (from now on ABHI). ABHI is an astonishing book. While it’s a layman’s summary, it’s written by a layman who clearly did a lot of work, and, more importantly, was doing the work right at the time that it became clear LLMs were something special and were going to continue to improve. So, it presents its view of how intelligence evolved with frequent, technical allusions to LLMs and the other AI technologies of the moment, which is exactly what other works lacked.

The book’s framing structure is that human-level intelligence evolved with 5 evolutionary breakthroughs, presented in roughly chronological order of accomplishment:

1. Steering: learning to head towards good things in the environment. This is the maximum level of intelligence a nematode has.

2. Reinforcing: learning to repeat behaviors that lead to good outcomes. This is the maximum level of intelligence a goldfish has.

3. Simulating: learning to model outcomes in advance. This is the maximum level of intelligence a rat has.

4. Mentalizing: learning to model other minds. This is the maximum level of intelligence a chimp has.

5. Speaking: learning to communicate ideas in non-instinctual ways. This is the minimum level of intelligence a fully grown human has.

When you look at these, you get the weird sense that modern AI has all of them except for “reinforcing”, because modern AI doesn’t “want” anything. AI is reinforced, yes, but it’s reinforced externally towards the goals of its developers, instead of any internal motivation. Similarly, AI’s “steering” isn’t actually heading towards good things for it, but instead on what sort of actions the developers have decided are good. It also can’t learn, strictly speaking, as its weights are frozen after training, although it can simulate learning with various bootstrapped memory hacks.

But that’s just the phenotypical, phenomenal 1 level of intelligence. If we stick with this level, we don’t end up getting much past our college philosophy/neuroscience classes, but now with a dash of AI. So let’s go a level deeper, into the actual mechanics of intelligence 2.

When we start looking into the actual mechanics of intelligence, a few questions jump out:

1. How did intelligence start at all? How did free floating plankton decide to start going somewhere?

2. How did intelligence evolve? How was Mozart or Einstein’s brain built out of the same building blocks as a rat’s brain?

3. Most importantly, how do we bridge the gap between what the brain looks like under a microscope and what thinking feels like?

We can tackle these one-by-one.

How did intelligence start at all? How did free floating plankton decide to start going somewhere?

Intelligence in bacteria started like a Roomba. That is, just with something that moved. Mostly, it moved randomly (in a “random walk”), driven by whatever currents drove it and its own flagella.

But, it also had a sensor. This sensor was biased towards good things, like favorable pH, temperature, or light. The intelligence here was purely mechanical. The sensor was designed in such a way that , if it found something good, it would be biased to keep heading in that direction. If it was heading towards something bad, like an unfavorable pH, it would change direction.

The next intelligence step was the step the nematodes took. They started to predict the consequences of their actions, seeing how their motor actions impacted their sensory perception (more on that predictory machinery later). This allowed them to distinguish between accidental movements leading to good things and purposeful movements leading to good things. It also forced them to centralize neural machinery (cephalize), to put the sensory neurons next to the motor neurons.

Not coincidentally, this is also when neurotransmitters started to become a thing. Two of the most notable were dopamine and serotonin. While both are old molecules, and can be found in thoroughly unintelligent organisms like plants or bacteria, in nematodes they got repurposed to aid in this purposeful search3.

Dopamine got repurposed as the confirmation or reward molecule. If the nematode’s movement led it towards a good thing, dopamine surged and said, “Keep doing that!” If it led it away from a good thing, then there was no dopamine signal.

Serotonin, meanwhile, got repurposed as the satisfactory molecule. High levels of serotonin said, “You’re ok for now, so you probably don’t need to move”. Low levels of serotonin said, “Start moving and paying attention to dopamine, because you need to move”.

How did intelligence evolve? How was Mozart or Einstein’s brain built out of the same building blocks as a rat’s brain?

Before I explain the answer to this question, I just want to say that this is the coolest, most exciting answer in the entire book. Not just because of the elegance of the solution, which I’ll explain in a second, but because of how timely it is.

For such a long time, evolutionary jumps in intelligence seemed impossible to explain. It was easy enough to see how a human’s heart comes from the same origins as a chimp’s heart, or how human muscles were like chimp muscles. There’s not that much difference between them. And, on a pathological level, the same is true of a chimp’s brain and a human’s brain. In formaldehyde, they’re pretty similar.

But, in a chimp, the chimp brain looks very different from the human brain does in a human. I mean, we can squint and say, “Yeah, I guess stripping a twig and using it to hunt for termites is kind of the same tool use as a human using a car”, but is it? And if it is, what about building a car? No matter how many chimps you locked in a room with auto parts, it seems unlikely they’d ever build a Model T, never mind a Lamborghini.

LLMs have changed that intuition. On a high level, the incredible increase in intelligence from GPT-2 to GPT-5, from preschool level intelligence to graduate level intelligence, has just been to do the same thing but bigger. More data, more training, more post-training. Even totally new AI capabilities, like vision or realistic speech, have come from doing inference on a bigger and bigger scale and on new datasets.

The really exciting thing is that we can see that exact same pattern with biological intelligence. Throughout the cerebral cortex, we see the same columnar pattern repeated over and over again. It’s a 6 layer pattern, with data input at the bottom and predictive output at the top. All mammal brains, from mice to Mozarts, have this pattern. The difference is just in how much of it there is, both in terms of brain size and neuronal density.

Each additional layer of intelligence can be explained by this same column, or “microcircuit”, piled on top of each other. Steering takes in sensory data as an input, and tries to predict what motor output leads to that sensory data as an output. If it’s wrong, it fixes its prediction. If it’s right, it reinforces that interpretation.

Reinforcing is on top of that. It takes the reward signal, dopamine, as an input, and tries to predict what sensory and motor activity lead to that reward signal as an output. This allows for signal persistence over time. So, unlike a nematode, who’s only reinforced when its motor activity is directly leading to good sensory data (swimming towards food smell), a goldfish is reinforced when his motor and sensory activity is leading towards a reward. So, a goldfish can learn that its owner puts his hand over the tank to give food which reinforces noticing that and swimming towards the anticipated food, even though there’s no direct reward from the hand over the tank.

Simulating is, then, on top of that. Rats can take reinforcement as an input, and then try to predict what actions would lead to that reinforcement. So, a rat can learn that mazes have food at the end of them, and, when put in a new maze, look at the paths ahead and simulate what actions would lead to the reinforcement.

And then mentalizing is on top of that. Chimps can take their own simulation as an input, and then try to predict how other chimps’ simulations would lead them to act. So, a chimp can learn that he is scared of the noises caused by banging metal cans, then decide to incorporate banging cans into his dominance display to frighten other chimps.

And then speaking is, at last, on top of that. Humans can take mentalizing as an input, then predict how their speaking affects other humans’ mentalization. So, I can craft a sentence you’ve never read before, like this one, and model how you would understand my combination of letters as a piece of information. I can even write a phrase like, “A fat cat”, and model how your understanding of language would cause an image to appear in your mind like you had seen Garfield himself.

All of these are the same columnar architecture, just taking different inputs and outputs, adding more layers, and allowing for integration of greater timescale. Goldfish, rats, chimps, Mozart: all of them were taking in data, using it to inform a model of the world, and then trying to minimize the gap between their predictive model of the world and what actually happened.

How do we bridge the gap between what the brain looks like under a microscope and what thinking feels like?

There are a lot of things about thinking that seem utterly mysterious when you consider the brain as a physical object.

First, there’s the imagination. I’m a big fan of the works of China Mieville, a “weird fiction” writer who populates his books with mosquito people, floating cities, and sentient cactuses. He describes them and their inner lives so vividly that I know he must see them in his mind like I see a pen on my desk4.

To ask the most cliched question possible: how did he come up with them? His brain has the same gooey mixture of proteins and chemicals as mine does, but those characters aren’t in my mind until he describes them5. We’re not coming up with original ideas and transmitting them in our livers or kidneys, so what makes the brain different?

Second, there’s choice. Everything in this world is constrained to physical laws: either cause-and-effect, or, at best, randomness. The same holds true for what happens in our body. Our heart beats because of an electrical signal. Our muscles tense because of neuronal signals. Inflammation surges because of an immunological signal.

But the brain appears to be its own prime mover. I can raise my arm or not. I can choose to imagine an elephant or not. I can choose to make a grocery list, then follow it the next day. Even my dog can choose to follow my commands or not. How?

These are difficult questions, and, to be honest, we’re not totally there. But we can get some sense of it by going back to the idea of simulation.

If we simulate ourselves, including us simulating ourselves, we can imagine. Those imaginings aren’t random. They’re composites of “features”, as AI interpretability would put it, forming bridges between sentience and cactuses or cities and oil tankers. If certain features are overweighted, we get distinct flavors to our imaginings, like a hungry person imagining roast beef.

And then, if, in turn, we can suppress direct sensory inputs to our predictive models, and use those imaginings to drive our predictive models (i.e. “What would I do in this situation? Let me choose, and then see what I actually do”), then we get free will.

I admit this isn’t perfect. In some sense, it just kicks the can up the road, because we have to choose to suppress and choose to form certain bridges of features. But this regress gets a little easier to understand when we see how much of it is automatic. We imagine when we dream, or when we’re faced with an unconscious choice of whether to take a bite of the sandwich or drink water first. My dog does much the same, whimpering in his sleep, or choosing when to come when called and modeling the outcomes of each based on his past experience. Our imaginings and our “free will” then run on autopilot.

Then we get a smaller surface area of problem to attack, where free will is only obvious in cases where there’s no strong past experience. Everything else gets some combination of past experience, random noise, and signals from the body to throw into the simulation. So, I “choose” to bite my sandwich before I drink my water based on my past experience plus being hungry. I “choose” to write a poem about my childhood based on my English classes, my childhood memory, and the fact that I randomly remembered seeing a house that looked like my childhood home.

The only one that doesn’t fit in with this is something like me choosing to wiggle one of my 10 fingers just because. Yes, there’s some element of noise and past experience that goes in there. And yes, anyone who’s ever spent a lot of time around a dementia patient and been asked the same question 20 times knows that the brain can fool itself into thinking it has free will. I mean, Grandpa probably thought he chose to ask how school is going, but it seems much more automatic to everyone else.

But, still, it does seem like I’m choosing to wiggle my right index finger because I wanted to, with no input from anything else. Maybe we won’t be able to resolve that for sure until we have a better X-ray into LLMs, or, better yet, even better models of intelligence. Qualia is a still yet to be conquered frontier, but it will come. We’ll just have to see how6.

Las Vegas

Overall: Las Vegas demands to be written about, which makes it tough for me to find something original to say. It has, from the beginning, been a ridiculous proposition: an oasis of sin in the middle of the desert, founded by the mob and sustained by one of the great American public works projects, the Hoover Dam.

It sustains itself by being notorious, which makes me suspicious that I’m playing into its predatory nature by writing about it at all. Still, here goes.

Downtown

1. We came into the Vegas airport at 4 am, exhausted from flight delays. The airport itself was empty and quiet, except for the flashing, blaring advertisements pasted on the baggage carousels, pumping club music into the empty space. There is no public transportation from the airport, so we got an Uber, and drove past the familiar detritus of impoverished cities: pawn shops, empty store fronts, sketchy tax preparation fronts.

Eventually, we got to the D, the hotel on Fremont Street where we were staying. This was our first mistake.

2. Fremont Street is the poor man’s version of the Strip. Music and lights blare 24/7, there’s a zipline through the entire length of the street, and at every hour there’s someone stumbling drunk past the giant Steak N’ Shake. The hotels lining Fremont Street are all casino/hotels. We did not realize this, so when we checked into our hotel at 4:30 am, we did so under cover of earsplitting pop songs competing with the constant jangle of slot machines, which, somehow, people were still playing at 4:30 am (or maybe had just started). Oh, also, our room smelled strongly of cigarette smoke, which the hotel concierge did warn us about before telling us our original, non-smoking room had already been given up.

3. In the morning, we found a pretty good little coffee shop to have breakfast at. Next to the coffee shop were several large, empty lots, one of which had a homeless person sitting in the middle of it. I can’t really explain Las Vegas development patterns. The city looks either half-developed or half-destroyed, and crowded streets sit right next to large areas of dirt.

4. In the afternoon, we went to the Sphere to see the Wizard of Oz show. You’ve probably seen the Sphere on Instagram. It’s a large hemisphere, covered in LED panels both inside and out. On the outside, they do creative advertisements and occasionally a giant face that looks at planes flying by. On the inside, they do backdrops to concerts and, I guess, also a re-imagining of the Wizard of Oz that expands every scene with CGI cows, cornfields, and Munchkins, as well as theater props like soap bubbles that come flying from the ceiling during the blizzard scene. It’s fun, but ridiculous, and makes the Wizard of Oz a worse movie but still a pretty fun show.

What they don’t show on Instagram is that getting to the Sphere makes no sense. The roads to the Sphere begin and end of their own volition. Las Vegas roads in general are baffling, but it was only at the Sphere that all that separated us from 3 lanes of oncoming traffic going 40 mph was about a foot of clearance.

5. After the Sphere, we went onto the Strip. This was everything that you expect from Las Vegas: neon signs piled on top of replica of the Eiffel Tower piled on top of canals and gondolas that go nowhere, all funded by tourism and gambling dollars. This is the hyperreallist Vegas that you can only find in America.

Death Valley

1. The next day we rented a car and headed into Death Valley. In the summer, Death Valley is the hottest, driest place in America. But this was January, so it was a cool 60ish degrees and intermittently raining. The desert is not built for rain, and puddles and sticky mud formed everywhere, threatening our crappy little car.

2. Our first stop was the Amargosa Opera House. This is a quintessential piece of Americana, or rather Nevadana. The Amargosa Opera House is a hotel in the precise middle of nowhere that has had 3 distinct stages of life:

a. A random hotel in the middle of nowhere from the late 1800s to 1967

b. The site of a one-woman song-and-dance act that ran for decades in the middle of nowhere from 1967 to 2012, making the hotel a niche tourism spot

c. A random hotel in the middle of nowhere, from 2012 to the current day, run by 2 disaffected hotel workers who are doing a poor job keeping out the sand and the mud. We tried to take a tour of the place to see the old hand-painted murals, but the hotel manager told me tours are only at 9 am and 6 pm. I had a feeling that if I had come at 9 am, the tour wouldn’t have been available anyways.

3. Our next stops were threatened by torrential downpour, including our intended visit to a ghost-town. Fortunately, we came across another ghost town on our way back to Vegas, an old gold-mining town that had about 10 years of boom times then was completely abandoned called Rhyolite. It’s still littered with rusted cans and boxes from the early 20th century, as well as more recent abandoned car parts from the 60s. The National Park Service has put fences around all the old structures, like the famed Bottle House, old jail, and train depot, but it’s easy enough to wander around them and get the vibe.

Rhyolite is also, strangely, the home of about 10 random sculptures put there by Danish artists about 15 years ago, including one large naked woman made out of what looks like Lego, who stands proudly baring herself to the desert.

4. Our last stop was a brothel, one of seven legal ones in Nevada. This one was called the “Alien Cathouse Brothel”, because of its proximity to Area 51. We went there to take a tour, rather than participate, although the escort giving the tour did show us a pricelist.

It was a very strange experience. The woman giving us the tour was scantily clad, but otherwise looked much like the average person you’d see at a Target. Her pricelist was astronomically expensive, with each hour of companionship costing about $2000. She made pains to tell us how pleasant her experience being a legal prostitute had been, and then revealed at the end that she had precisely one customer the entire week but still had to pay rent to the brothel.

I got the sense that the brothel was more of a scam on her than it was on us. It was in the middle of nowhere, hours from Vegas, and wasn’t even particularly fun in a campy way. I mean, it was literally called the “Alien Cathouse”, but had boring decorations in that sleazy 70s style. At least we got her business card.

5. As we headed back into Vegas, passing the shocking number of new houses under construction in the sprawling Vegas suburbs, I found myself with a new appreciation for Vegas. Nevada is not really meant for human habitation. Entropy swallows up hotels and whole towns, and they’re just left there to fall apart. Vegas’s manic energy keeps it alive, and even then half the town is falling apart. It’s not easy to keep a place going.

Hoover Dam

1. Our last day in Vegas we decided to visit the famed Hoover Dam, provider of water and electricity to Vegas. First, we went on a beautiful hike next to the dam, down an abandoned riverbed, clambering down boulders with ropes anchored to them, ending in natural hot springs right on the Colorado River, with a great view of the elegant bridge spanning the river.

Our companions in the hot springs were Vegas teenagers, an East German and Japanese college student traveling together, and a Mexican family. The atmosphere was jovial but peaceful, a far cry from the relentless energy of the city.

2. After a long climb back out, we drove over the Hoover Dam to one of the innumerable lookout points watching over Lake Mead. I felt strangely sad. The Hoover Dam felt like a pure tourist spot, with a hotel/casino next to it, 12 different parking lots by it, and a museum/gift shop. There’s nothing wrong with this (we were tourists after all), but I got a weird echo of the Grand Canyon, which I had visited as a kid.

The issue is that the Grand Canyon is, of course, a natural phenomenon. We’re not making more of them anytime soon, so we might as well all visit the one we got. But Hoover Dam was made by people in the 30s in 5 years. We can make another one any time, if we have the will. But it doesn’t seem like we do.

San Diego

Overall: I don’t really have detailed notes of San Diego. By the time we got there, we were exhausted, and spent a lot of time just visiting my fiancee’s extended family. San Diego is warm, sunny, pleasant, functional, and, fortunately, seems like it’s slowly transitioning from its single-family housing and car culture to more of a city with public transportation, although it’s being fought every step of the way by California NIMBYs.

Besides the time spent with my fiancee’s family, we spent most of our time in Ocean Beach, where our Airbnb was.

1. Ocean Beach has heavy beach town vibes. We saw an octopus in a tide pool (a highlight of the trip), innumerable dogs, and a shockingly large population of seemingly retired 20 somethings who surfed, drank coffee, and hung out at coffee shops at 11 am on a Thursday. Judging by the flyers on the coffee shop walls, the only jobs available in Ocean Beach were leading psychedelic retreats, yoga classes, or massage therapy. I suspect that heavy subsidies from Mom and Dad are involved on both the supply and demand side of these industries, though.

2. Overall, San Diego was a very pleasant place to visit. We should have gotten a car, though, and budgeted enough money for the shockingly expensive zoo ($80 a person!). Oh, also, while there, I finished the excellent book A Brief History of Intelligence, which I’ll be writing up notes for soon. That was pretty much it.

There was something strange about this, I thought. Not that the journalist would ask a difficult question, or that Putin would refuse to respond straightforwardly, but that Putin would put himself in such a situation at all. Putin is pretty close to an autocrat. He’s pursuing the largest land war in Europe since WW2, frightening and angering the major Western powers. He’s had all of his significant opponents killed or jailed. He doesn’t entertain questions like that from domestic journalists, and a domestic journalist who asked him a question like that would soon regret it.

So, why would Putin respond to this journalist? Why wouldn’t he just say, “None of your business”? Putin’s answer to the question, regardless of what it is, would never seem to benefit him. He’s under no obligation to answer questions from the BBC, and his answer to a question like that would not endear him to the British public.

I think my answer to this has to come back to the title of this blog post: Putin lives, like we all do, in a world of words. His words, regardless of what they are, matter. Putin has the power that he does because of his words, and he could lose that power from his words. He’s not a tiger, who kills people with teeth, claws, and muscle. Almost every single person he’s killed has been killed by his words or anticipation of them. Putin invaded Ukraine with tanks, yes, but those tanks moved inside the stories he told.

But how do Putin’s words kill people, or invade Ukraine? Well, it’s the weight that they carry. When he tells his underlings to kill someone, they have to trust that his words matter. That, after they kill the person, they will be protected by his words. Or, in a paranoid environment like Russia’s, they have to trust that, if they don’t obey Putin, the next guy will obey Putin when he orders their arrest. So, in other words, the next guy will trust Putin.

I’ve been thinking so much on words since I started reading Borges, specifically his short story collection Aleph. Borges, of Infinite Library fame, is obsessed with words. His fiction is all about reference and self-reference, what words mean and what they don’t, what we can trust and can’t.

Let me give you an example. One of Borges’s shortest stories in the book is the “Parable of the Palace”. It’s just two pages, and relates a story of a poet and an emperor wandering around the emperor’s enormous pleasure palace for an entire day. At the end of the day, the poet recites a short poem, or maybe just a word, that somehow encompasses the entire palace. As Borges states it, “Within the poem lay the entire enormous palace, whole and to the least detail, with every venerable porcelain it contained and every scene on every porcelain, all the lights and shadows of its twilights, and every forlorn or happy moment of the glorious dynasties of mortals, gods, and dragons that had lived within it through all its endless past.”

The story ends with the emperor upset that the poet had “stolen his palace”, and he has the poet executed. Or, Borges writes, he’s also heard the story that the poem wiped out the palace in a puff of smoke. But then again, Borges concludes, both endings must be incorrect. The poem must be lost today because it “merited oblivion”, and the poet himself must have just died an anonymous slave.

This is Borges’s view of the world. The world is words, his stories are words, and both are unreliable. Words mean everything and nothing. Words can kill a poet (or journalist) or capture a palace in a sentence, but they can also be misinterpreted, forgotten, or meaningless. They are sometimes important enough to kill over and then simultaneously not worth remembering at all.

And that’s the world we live in, too. I don’t mean to bring this back again to AI, but I can’t help it. I’ve been continually astonished at how far LLMs have gone with just words and numbers. LLMs do physics, chemistry, poetry, coding, and therapy with just words and numbers. And I mean, yeah, those fields have always been made up of words and numbers, but that never seemed like the most important part. I always thought it was the thoughts behind the words and calculations that matter, what the symbols meant.

But what they mean is what they say, or don’t say, in the context of all the other words they’re surrounded by. Our words define our world to such a great extent. Just look back at this essay. Putin, Steve Rosenberg, Borges: they’re all real, yes, but they’re also just all words I put in order, created them like Borges created his emperor and his palace.

The power these words hold. God, I hope I’ve put them together right.

There’s a risk, though, of these weird animal science stories turning into “just so” stories. We have a certain set of what seem to be effects, like hummingbirds who can eat copious amounts of sugar without getting diabetes, and what seem to be causes, like certain suggestive mutations in glycolytic proteins. Then we put them together and say, “These mutations cause this phenotype, ‘just so’”.

On this blog, I try to be careful about the science that I highlight to select the strongest cause and effect relationships, or at least make sure that the scientists involve note the uncertainties. So, for example, in that hummingbird paper, the scientists involved did do work to make sure that the mutations would cause at least some of the “non-diabetic” phenotype in hummingbird cells, and to note the limitations of their work.

But we’re still left with some issues. Are these mutations sufficient for hummingbirds to avoid diabetes? No, and nobody’s claiming they are. Diabetes is a disease of complex systems, and there are enough ways to get to it that one mutation isn’t going to do much. Are these mutations necessary for hummingbirds to avoid diabetes? Well, that’s a harder question. The authors of the paper have recently written a follow-up examining other nectivores to see convergent mutations, which gets to the idea of which mutations seem to come up over and over again, which suggest that they’re necessary. But, still, we can’t say for sure.

I don’t mean to pick on this hummingbird paper. I think it’s an excellent paper, which is why I highlighted it in the first place. This is a problem inherent to biology, and how we try to find legible answers to complex systems. As long as we insist on trying to understand the mechanisms of biology, we will face the specter of just so stories.

I bring all this up because there’s a “just so” story that’s been bothering me for a while, and I’ve been struggling to write about it. It has to do with why elephants are so long-lived. I’ll tell you the university press release version first, and then we can dive into how we might fix it, or at least think about it better.

The university press release version of elephant longevity is that elephants are long-lived because they don’t get cancer. Elephants not getting cancer is a prime example of “Peto’s Paradox”, which notes that large animals, who should have the highest rates of cancer because of their increased number of cells, paradoxically tend to have the lowest rate. They don’t get cancer because, unlike other animals, they have 20 copies of the TP53 gene, which encourages apoptosis in cancerous cells. This is unlike humans, who only have 1 copy of the TP53 gene. If both alleles get disabled, as they frequently are in cancer, the cells won’t self-destruct when they need to.

This is an ok story at first glance, but gets weirder the more that you think about it. If you were an engineer designing a cancer-resistant large animal, you wouldn’t necessarily go for replicating the same gene over and over again. That doesn’t exactly limit the point of failure. Also, if this is such an effective anti-cancer strategy, it’s weird that no other animal has evolved it, including other long-lived animals like bowhead whales or humans. It would seem relatively straightforward, given that failure of TP53 is a big problem across all mammals.

And then, last, there’s the biggest problem, which is that longevity isn’t just about avoiding cancer. Think about old people. Sure, cancer is a big problem for them, but so is dementia, sarcopenia (muscle wasting), osteoporosis, organ failure, and everything else that comes with being old. Are we saying that elephant longevity in general is about this 20 copies of TP53 story, or are we just searching where we have a light?

Let’s take a step back before we dive deeper into this. This is, frankly, where I got stuck before, so I think it’s good to survey the landscape first. I’m going to start by just discussing what we know about elephants and longevity.

Elephants are models of healthy aging. Despite being very large and very long-lived, they:

1. Don’t seem to get dementia or Alzheimer’s, unlike whales, humans, or even parrots. If anything, the oldest elephant females may be more cognitively sharp than other elephants.

2. Don’t get cancer

3. Are generally infection-resistant in old age

4. Remain fertile/virile through their old age (no menopause or andropause)

5. Generally maintain healthy bones and muscles through their old age, with no universal osteoporosis or sarcopenia

6. Can go 48 hours without sleep or obvious sleep debt, regardless of their age

The main reason why elephants don’t live forever is that they’re still affected by the wear and tear of aging. So, old elephants do still get cataracts, osteoarthritis (especially in their feet), and tooth decay, and die when they have too few teeth and functioning joints to continue being migratory herbivores.

So, whatever explanations we provide for why elephants are so long-lived, we have to account for the fact that elephants seem to have solved a surprising amount of the problems of aging, even compared to other long-lived animals. If I were to simplify it, I’d say something like they don’t seem to have any of the problems of aging where systems get dysregulated, and only have the problems of things getting slowly worn down and not replaced. Already, it seems unlikely that “20 copies of TP53” will cut it.

Let’s also discuss what we know about how other animals have solved the problems of aging and cancer. Now, there are a number of different places we could take this, but I think looking at the example of bats is the most interesting. Compared to their size and nearest relatives, bats are very long-lived and cancer resistant. They also:

1. Seem to lack cognitive decline (although this is not as well studied as elephants’)

2. Have low rates of cancer while aging

3. Remain generally infection-resistant in old age

4. Remain fertile/virile through old age (no menopause/andropause)

5. Seem to maintain healthy muscles and bones through their old age

The only thing they lack is resistance to sleep debt, although this isn’t particularly well studied.

Counter-intuitively, they do not do this by making their cells ready to apoptose at any moment. Instead, they do the opposite, as I’ve discussed before. Bats have a very high tolerance for DNA damage, which is occurring constantly because their body temperature is elevated during flight. Instead of encouraging their cells to apoptose, they invest heavily in DNA repair.

As part of this, bats have a reduced sensitivity to random bits of DNA or RNA floating around, including those that would indicate invasion by a pathogen or tumor cell. Their innate immune system, by contrast, is ramped up, preventing infections or tumors from ramping up out of control.

TP53-related apoptosis in bat cells is reserved for absolutely necessary cases. But, when it is necessary, bat cells do not hesitate, and indeed show much higher rates of p53 apoptosis when exposed to gamma radiation than human or mouse cells. This is helped by their increased copy number of TP53 genes (like elephants have), which, suggestively, increase in number in longer-lived bat species

But bats pair their increased TP53 gene activity and gene number with also increased p53 inhibition machinery. Bat cells invest in both being able to apoptose when necessary and stop apoptosis when necessary.

So, now, we’re starting to get some intriguing ideas about TP53 copies actually contributing to longevity: possibly sufficiently, maybe even necessarily. But, the story is complicated: humans and mice have 1 copy, shorter lived bats have 1 copy, longer lived bats have 3-7 copies, elephants have 20 copies. Bats also pair their TP53 genes with complex regulatory apparatuses, which don’t just inhibit p53 activation when it’s unnecessary, but also makes it more potent per gene when it’s necessary.

Let’s make it more complicated by going back to one of our original questions. How can 20 (or 7) identical copies of a gene actually help? Is that really adding redundancy, given that a deactivation of one would probably deactivate the rest as well?

Well, identical copies of a gene don’t actually help add redundancy. In fact, they probably don’t help at all. That’s why it’s fortunate that elephants don’t have 20 identical copies of TP53, and bats don’t have 7 identical copies either.

Instead, elephants have 1 normal copy of TP53, ~17 retrogene 1 copies of mutated TP53 that prevent the original copy of TP53 from being deactivated, and 1 retrogene copy of mutated TP53 that’s so mutated that it has a completely independent apoptotic mechanism. They also have a reanimated pseudogene 2 that works with TP53 to help with apoptosis. There’s a similar story with bats, but the exact functionality of their 2-6 mutated copies is less well studied.

So, going back to our original story, there does seem to be an advantage to creating multiple copies of TP53, as long as they aren’t all identical.

It’s hard to say if multiple different copies of TP53 are sufficient for longevity, though. Elephants evolved one unique immune system model to cope with their challenges, which is that they’re gigantic and rely on their matriarchs to navigate to historical watering holes. This made their cells hyperalert for any foreign DNA. Bats evolved a different unique immune system model to cope with their challenges, which is that they run such a high body temperature from flying that they’re constantly causing DNA damage to themselves. This made their cells very tolerant of DNA strands up to a point, and hyperalert thereafter.

Both did converge on multiple copies of TP53 for that hyperalertness, and both ended up being models of healthy longevity, even healthier than humans in many ways. It’s tempting to say that any animal that also had multiple different copies of TP53, regardless of what other immune features it had, would also be able to age as gracefully as elephants and bats.

Unfortunately, elephants’ multiple TP53 copies are closely integrated into the elephant immune system and DNA regulatory system. Managing to get the same working in, say, mice, would require a ton of work to integrate these copies into the mouse immune system and DNA regulatory system. This is, as far as I know, currently beyond the capabilities of genetic engineering.

Hey ChatGPT, all-knowing, hyperintelligent, obedient AI that I can contact at any time, with 0 being our knowledge of electricity in the year 1700 and 100 being our knowledge of electricity in the year 2025, rank how far we were in each of V. A. Obruchev’s goals in the year 1956 and how far we are in the year 2025.

Here’s my best stab at “how far along” we were on each of Obruchev’s technical goals, on a 0–100 scale where:

• 0 ≈ our understanding/use of electricity in 1700 (static shocks, party tricks).
  
  

• 100 ≈ electricity in 2025 (Maxwell + QED are nailed, global grids, electronics, lasers, MRI, etc.).
  
  

So “100” doesn’t mean absolute physical limits, just “as mature and exploited as electromagnetism is now”.

Rough calibration:
20 ≈ electricity c.1800, 40 ≈ 1880, 60 ≈ 1920, 80 ≈ 1950.

1. Life extension, disease, aging, “restoring life”

What Obruchev wanted:
Average lifespan 150–200 years, infectious diseases wiped out, non-infectious diseases minimal, old age & fatigue “conquered”, and the ability to bring people back from accidental death.

• 1956 – ~25/100
  
  

  • Big wins: antibiotics, vaccines for many killers, safer surgery and anesthesia, early organ transplant, basic ICUs. Life expectancy in rich countries was already >60, versus ~30 globally in 1900.(Our World in Data)
    
    

  • But infectious disease still a leading killer worldwide; no control of aging, no molecular genetics, no ICU-level resuscitation beyond minutes.
    
    

• 2025 – ~45/100
  
  

  • Global life expectancy ~73 years; most children survive to adulthood; many infections controlled by vaccines/antibiotics/antivirals; organ transplantation, intensive care, and some gene and cell therapies exist.(Our World in Data)
    
    

  • Aging itself is only modestly modifiable; we’re nowhere near 150–200 year averages, nor can we “restore life” except from very brief clinical death (CPR, defib, ECMO).
    
    

Verdict: Lots of progress, but this field is still more like “electricity 1880–1900” than “electricity 2025”.

2. Harnessing all forces of nature for energy (sun, wind, geothermal, atomic, storage, wireless power)

What he wanted:
Solar, wind, geothermal, atomic energy widely applied to industry, transport, construction; efficient storage; long-distance wireless power transmission.

• 1956 – ~40/100
  
  

  • Mature fossil fuels and hydro; significant electrification; first nuclear reactors and bombs; some geothermal; batteries and grids exist but are limited.
    
    

  • Essentially no useful solar/wind industry; wireless power only for tiny loads (radio, early microwave experiments).
    
    

• 2025 – ~70/100
  
  

  • Massive deployment of solar and wind: renewables generate ~30% of global electricity in 2023 and are on track to become the largest source by mid-2020s.(Ember Energy)
    
    

  • Nuclear widely used in some countries; geothermal used in a few; high-capacity grids and batteries exist.
    
    

  • Wireless power is still short-range and low-power (Qi chargers, etc.), not “to any point on Earth”.
    
    

Verdict: Energy capture and storage are relatively advanced; only the “everywhere wireless power” piece is still at the science-fiction stage.

3. Predicting and rendering harmless floods, hurricanes, eruptions, earthquakes

• 1956 – ~15/100
  
  

  • Early seismology and river gauges, but no weather satellites yet, crude hurricane tracking, minimal tsunami warning, limited understanding of plate tectonics.
    
    

• 2025 – ~40/100
  
  

  • Global constellation of weather satellites and radar, detailed hurricane and flood forecasting, tsunami and volcano monitoring networks; earthquake early-warning systems in some regions.
    
    

  • But we mitigate rather than “render harmless”: we can evacuate people and build safer infrastructure, not stop earthquakes or hurricanes.
    
    

Verdict: Prediction has advanced a lot; control is still mostly impossible.

4. Materials beyond anything in nature (harder than diamond, more heat-resistant than brick, more refractory than tungsten/osmium, more flexible than silk, more elastic than rubber)

• 1956 – ~45/100
  
  

  • Already had a rich synthetic-chemistry and polymer industry: nylon, Teflon, synthetic rubber, plastics, alloys.
    
    

  • 1954: first reproducible synthetic diamonds at GE – literally “harder than diamonds” as an engineered product line.(Rincón educativo)
    
    

  • But no recombinant protein production yet, limited control at atomic scale.
    
    

• 2025 – ~80/100
  
  

  • We can mass-produce proteins and other complex biomolecules via engineered microbes and cell lines; industrial enzymes and biologic drugs are routine.
    
    

  • Supermaterials: advanced ceramics and superalloys, Kevlar, carbon fibre, ultrahard materials like cubic boron nitride and nanocrystalline diamond; graphene and other 2D materials; metamaterials with designed optical/mechanical behavior.
    
    

  • There’s still room at the top (room-temperature superconductors, better structural materials, etc.), but this area is closer to “electricity-level maturity” than most of his other goals.
    
    

5. New breeds of animals and plants with much higher yields

• 1956 – ~35/100
  
  

  • Classical breeding and early Green Revolution work: hybrid crops, fertilizers, pesticides, mechanization, AI for livestock breeding.
    
    

  • Genetics was understood in principle, but no DNA sequencing, no genetic engineering, no IVF or cloning in the modern sense.
    
    

• 2025 – ~75/100
  
  

  • GM crops are widespread (Bt corn, herbicide-tolerant soy, etc.), and CRISPR is used to edit plant and animal genomes; cloned and gene-edited animals exist; marker-assisted breeding is standard.
    
    

  • Yields per hectare and per animal have multiplied many-fold in the last century, though with big environmental and ethical trade-offs.
    
    

Verdict: Plant/animal improvement is one of the domains where we’re closest to Obruchev’s vision, even if it doesn’t look exactly like Soviet agronomy imagined.

6. Conquering “unpromising” regions: marshes, deserts, taiga, tundra, sea bottom

• 1956 – ~30/100
  
  

  • Large-scale land reclamation (Dutch polders, Soviet canal and irrigation projects), dams, early desert irrigation, polar and high-altitude research stations, early deep-sea submersibles.
    
    

  • But many regions remained effectively inaccessible for year-round economic life.
    
    

• 2025 – ~55/100
  
  

  • Huge cities and agriculture in deserts and cold regions (e.g., Gulf megacities, North American and Siberian resource towns) enabled by air-conditioning, pipelines, desalination, logistics, and aviation; offshore oil/gas platforms and undersea cables everywhere.
    
    

  • Still nothing like permanent seabed settlements; climate change is also re-desertifying some areas.
    
    

Verdict: We’ve partially “conquered” these areas in a logistical/engineering sense, but not in the total, effortless way Obruchev implied.

7. Weather control (regulating wind and heat, moving clouds, ordering rain/sun/snow on demand)

• 1956 – ~5/100
  
  

  • Early cloud-seeding experiments and atmospheric research, but very limited capability and lots of uncertainty.
    
    

• 2025 – ~20/100
  
  

  • Cloud seeding is used in many regions; evidence suggests it can boost precipitation by perhaps 10–15% under favorable conditions, but effects are modest and uncertain.(Government Accountability Office)
    
    

  • Proposals for large-scale geoengineering (stratospheric aerosols, marine cloud brightening) exist, but are not deployed; we have essentially no ability to steer hurricanes or “order” the weather for specific days.
    
    

Verdict: We’ve gone from “almost nothing” to “a tiny bit of probabilistic nudging,” still incredibly far from communist sci-fi levels of control.

Very compressed summary

• Most advanced vs his vision in 2025: materials science (80/100), plant/animal breeding (75/100), and energy systems (70/100).
  
  

• Mid-range: conquering harsh environments (55/100), disaster prediction/mitigation (40/100), and medical life extension (45/100).
  
  

• Lagging far behind: genuine weather control (20/100).
  
  

If you want, we can redo the scores with a different calibration (e.g., 100 = actually achieving his literal target like 150–200-year lifespans), which would drag many of these numbers down quite a bit.

Thank you, my artificial friend. What are your predictions on how they will change in the next 5 years using the same scale?

Here’s my best guess for where the scores move by 2030 (five years from your 2025 baseline).

Same scale as before:

• 0 = electricity in 1700
  
  

• 100 = electricity in 2025
  
  

• So 2030 can’t overshoot 100; it can only creep closer.
  
  

Quick prediction table

1. Life extension / diseases / aging

45 → ~52

Why up:

• More gene and cell therapies actually approved for cancer, rare diseases, maybe common CV/metabolic diseases.
  
  

• mRNA and similar platforms standard for many vaccines; faster pandemic response.
  
  

• GLP-1-type drugs and metabolic modulators widely used; big reductions in cardio-metabolic mortality if policy doesn’t screw it up.
  
  

Why not higher:

• Aging biology: we’ll probably have a few specific partial rejuvenation therapies (e.g., for blood, immune system, maybe some senolytics), but no general 30–50-year life-extension.
  
  

• Global life expectancy could stall or even fall in some regions due to climate, conflict, and inequality.
  
  

2. Energy (sun, wind, nuclear, storage, wireless)

70 → ~80

Why big jump:

• Solar + wind capacity likely at least doubles; grids and batteries catch up enough that renewables dominate new generation and a large share of total.
  
  

• Grid-scale storage (Li-ion, sodium, flow, etc.) and demand-response are mainstream, smoothing intermittency.
  
  

• Some SMR or next-gen nuclear designs move from paper to early deployment in a few countries.
  
  

Why not closer to 100:

• Fossil fuels still large; heavy industry and aviation remain hard.
  
  

• Long-distance wireless power is still basically sci-fi; room-scale and EV-scale inductive systems at best.
  
  

3. Disasters (prediction & mitigation)

40 → ~48

Why up:

• Better satellite constellations and sensors; denser networks of buoys, radar, seismometers.
  
  

• AI-enhanced weather/climate/hydrology models give more accurate and longer-lead forecasts.
  
  

• More countries adopt earthquake and flood early-warning systems; building codes and nature-based defenses improve in some places.
  
  

Why still low:

• We’ll predict earlier and evacuate better, but control remains essentially zero.
  
  

• Climate change increases hazard frequency and intensity, eating some of the gains.
  
  

4. Materials beyond nature

80 → ~86

Why up:

• Continued progress in:
  
  

  • 2D materials and heterostructures,
    
    

  • nanostructured alloys/ceramics,
    
    

  • polymer design,
    
    

  • bio-manufactured proteins and composites.
    
    

• AI-driven materials discovery and simulation starts dropping the time from idea → sample → product.
  
  

Why not near 100:

• Scaling new materials into bulk, cheap, reliable stuff is slow and capex-heavy.
  
  

• Some of the “dream” materials (robust room-temp superconductors, perfect structural metamaterials) may still be lab curiosities, not infrastructure.
  
  

5. New breeds of plants & animals

75 → ~83

Why up:

• CRISPR and related tech embedded in routine breeding pipelines.
  
  

• Multiple gene-edited crops with improved climate resilience, nutrient use, pest resistance on the market.
  
  

• Genomic selection and reproductive tech (IVF, embryo selection, maybe gene editing) increasingly used in livestock and aquaculture.
  
  

Ceiling effects:

• Regulation, public acceptance, and IP fights limit rollout speed.
  
  

• Climate stress, water limits, and soil degradation partially offset yield gains.
  
  

6. Conquering harsh environments (deserts, tundra, seabed, etc.)

55 → ~60

Why modest gain:

• Expansion of offshore wind, subsea infrastructure, and Arctic/remote operations (with more robotics).
  
  

• Some growth in desert cities and agriculture using desalination, precision irrigation, and protected cropping.
  
  

• Better remote sensing and autonomy make it easier to use these regions without people physically living there full-time.
  
  

Why only +5:

• Climate instabilities make some places less viable.
  
  

• Seabed habitation remains uneconomic and mostly unnecessary; we stick to temporary or robotic presence.
  
  

7. Weather control

20 → ~25

Why any increase:

• More systematic cloud-seeding programs with solid evaluation (esp. in water-stressed regions).
  
  

• Serious field trials of limited geoengineering approaches (e.g., small-scale marine cloud brightening, localized radiation management) are plausible.
  
  

Why still very low:

• Political and ethical barriers to large-scale geoengineering.
  
  

• Physics: steering storms and “scheduling” local weather remains out of reach; at best we do coarse, global nudges.
  
  

Meta-summary

• Biggest likely movers:
  
  

  • Energy (2) and plant/animal biotech (5) — because these are already commercial and scaling fast.
    
    

  • Biomedicine (1) — not dramatic lifespan jumps, but more diseases become manageable/curable.
    
    

• Medium movers:
  
  

  • Disaster prediction (3) and materials (4) — steady progress, some AI-driven accelerations.
    
    

• Small movers:
  
  

  • Harsh environments (6) and especially weather control (7) — mostly incremental, not transformative.
    
    

If you want, we can also sketch “optimistic” and “pessimistic” 2030 bands for each, e.g. life extension 48–60 depending on how gene therapy/aging stuff breaks, or redo the whole thing assuming AI progress accelerates everything a lot more than current trendlines.

Let’s assume we get superabundant, superintelligent AIs. Where does that take us by 2030?

Let’s assume:

• We really do get superintelligent, very cheap AIs by ~2030.

• They’re usable and mostly aligned (no collapse, no global AI ban), but the world is still politically messy and physical stuff still takes time (labs, factories, permits, culture, etc.).

Same 0–100 scale:

• 0 = electricity in 1700

• 100 = electricity in 2025

The 2025 column is what we had before; the 2030 column is my best central guess in that “superintelligent, abundant AI” world, plus a rough plausible range.

Summary table – 2025 vs. “superintelligent-AI 2030”

1. Life extension / disease / aging

45 → ~70 (60–80)

What SI AI gives you:

• Much faster target discovery, drug design, and trial design.

• Automated wet labs running around the clock with AI “PIs”.

• Better personalization (polygenic + lifestyle + biomarker optimization) and early detection.

Why it doesn’t hit 90+ by 2030:

• Biology is messy and bodies are already built; you still need years of experiments and long-term safety data.

• Regulators, payers, inequality, and simple logistics slow deployment.

• Aging may turn out to have hard trade-offs (e.g., cancer risk vs regeneration).

So you probably get big dents in cancer, many genetic diseases, some neurodegeneration, and serious compression of morbidity, but not reliable 150–200-year lives by 2030. That’s why I land around 70.

2. Energy (sun, wind, nuclear, storage, etc.)

70 → ~88 (80–95)

Where SI helps a lot:

• Materials discovery: better batteries, catalysts, high-temp materials.

• System-level optimization: grid layout, demand response, siting, maintenance.

• Design & operation of new nuclear and geothermal systems.

Bottlenecks:

• Building mines, factories, grids, and reactors is concrete, steel, and politics. You can’t print transmission lines.

• Developing regions may lag for financial or political reasons.

So in rich/coordinated blocs, energy starts feeling almost non-scarce for most uses (electricity cheap, storage abundant); globally it’s uneven → high 80s, not quite “problem solved for everyone”.

3. Predicting & taming floods, hurricanes, quakes, eruptions

40 → ~65 (55–80)

Upside with SI:

• Way better climate and weather models; real-time data assimilation at insane scale.

• Optimized evacuation, infrastructure design, and insurance / planning.

• Smarter siting of everything: you just don’t build in the dumbest places anymore (if the politics allow it).

Limits:

• Physics: we still can’t stop tectonic plates.

• Geoengineering helps for average climate but not every individual storm.

• Poor/fragile states may not act on the warnings.

So, huge drop in deaths and damage where governance is decent, but not literal “harmless calamities” → around 65, with wide spread depending on how much the world actually listens.

4. Materials beyond nature

80 → ~90 (85–97)

SI AI plus robotic labs is almost tailor-made for this:

• Exhaustive search in enormous design spaces for alloys, ceramics, polymers, metamaterials.

• Automated synthesis/characterization loops.

• Dense simulation stacks (quantum → continuum) guided by the model.

Why not 100:

• Scaling from gram-scale “miracle material” to megaton-scale cheap commodity is slow, capex-heavy, and path-dependent.

• Safety, durability, and standards testing take time.

So by 2030 you likely have a zoo of super-optimized materials in high-value applications and a clear roadmap for commoditization → ~90, maybe mid-90s if deployment is very aggressive.

5. New breeds of plants & animals

75 → ~88 (80–95)

What SI does:

• Treats plant/animal genotypes and phenotypes as a giant RL / search problem.

• Designs edits and breeding strategies to hit arbitrary trait vectors (yield, resilience, taste, nutrient profile, etc.).

• Automates greenhouses, growth chambers, and phenotyping.

Constraints:

• Growing crops and animals still takes seasons / years.

• Regulation and public opinion can delay GM deployment.

• Ecosystem and climate shocks complicate neat lab-designed plans.

Net effect: second Green Revolution in cooperating countries; global hunger and micronutrient deficiency could be radically reduced, but not eliminated everywhere → high 80s.

6. Conquering deserts, tundra, marshes, seabed

55 → ~75 (65–90)

How SI helps:

• Energy + water + materials improvements from above make harsh regions more livable.

• AI designs entire cities, ports, farms, mines, pipelines, and logistics systems optimized for extremes.

• Robotics handles dangerous/dull work (offshore platforms, deep mines, polar research, seabed installations).

Why not “total conquest”:

• Climate instability may make some regions less attractive, not more.

• Some areas might be left wild on purpose (biodiversity, politics).

• Seabed habitation remains pointless; we’ll mostly colonize it with robots, not humans.

So we get much more economic and scientific use, but not Obruchev’s implicit vision of every square meter being tamed → around 75.

7. Weather control (clouds, rain/sun/snow on demand)

20 → ~35 (25–60)

SI’s role:

• You get really good climate engineering designs: where to put aerosols, how to do marine cloud brightening, how to manage side-effects.

• Cloud-seeding and regional interventions become much more reliable.

• Better control over heatwaves and maybe storm intensity via global-scale interventions.

Fundamental limits:

• Chaos + physics: there are hard constraints on precise local control.

• Politics: global geoengineering is contentious; different regions prefer different climates.

• Liability: “You made it rain there instead of here” will be a lawsuit magnet.

So we probably get coarse climate steering and robust control of some extremes, but not “scheduled sunshine for your parade”. That’s why I keep it at ~35 with a big range: if the world jointly embraces aggressive geoengineering, maybe you creep toward 60; if not, you barely move.

Big picture

If superintelligent, abundant AI really arrives and we survive it:

• Most of Obruchev’s technical goals become primarily physical/logistical/political problems, not intellectual ones.

• Scores in energy, materials, and ag/breeding move closest to the “electricity-level” ceiling.

• Medicine and disaster management move a lot but remain constrained by biology, ethics, and institutions.

• Fine-grained weather control remains the weird outlier: AI can help design interventions, but cannot repeal chaos or politics.

If you want, we can do a more pessimistic “alignment/regulation bottlenecks everything” version, or a more extreme “runaway self-replicating factories, Von Neumann probes, etc.” version where a few of these basically slam into 100.

I have a lot of personal thoughts about this that are hard for me to process right now. Biology/health-wise, I have a few main thoughts:

1. Keytruda is a miracle drug. It gave my dad another few years when chemotherapy and radiation totally failed. I’ll be writing about cancer and the PD-1 inhibitors sometime soon, as they really are pretty amazing. That being said…

2. My dad never fully returned to baseline after his ICU visit. He had so many problems, cancer and otherwise, including a nasty bout with COVID and a totally baffling bout of anorexia. Each of his illnesses individually had a cause, with smoking being the proximate cause of his cancer, but it seemed so obvious that his failure to fully recover from these illnesses was aging-related. It’s made me more convinced of the need to fix aging, as well as greater mystery as to what aging actually is. More on that later.

3. Hospitals suck. I don’t mean that in just the fact that it was always sad when my dad ended up in a hospital. I mean that hospitals were literally always terrible for his recovery, as they continually poked and prodded my dad, woke him up at all hours of the night, and wouldn’t let him do his daily routines. He was left scarred by them in more ways than one. There are incredibly obvious ways to make a hospital better for patients, which I think are mostly impossible due to overregulation. I’ll discuss that more later.

Anyways, here’s the obituary I wrote for my dad, with his and my mom’s input.

Brian was born in Boston, MA to a 24-year-old medical student, Gerald D’Arcy Klee, and his young, immigrant wife, Elizabeth Grace Klee. Over the next 10 years, Brian would be followed by four siblings: Ken, Sheila Grace, Susan (deceased), and Louise.

Brian only spent a year in Boston, as his father moved to Missouri after graduating medical school to work as a prison doctor. This was the first of many moves in Brian’s young life. Over the course of his childhood, Brian would live in Boston, Missouri, Staten Island, Baltimore, and Pennsylvania, before heading off to Franklin and Marshall College in Lancaster, PA.

College was one of the most formative experiences of Brian’s life, as his major in philosophy taught him “how to think.” He especially found his course in mathematical logic incredibly edifying, and he would like to note for the record that he got the only uncurved A in the entire class. He would also like to note that he received a prize for the most outstanding philosophy student in his graduating class.

Brian decided that he was not interested in being penniless, and followed his philosophy major by getting an MBA at Columbia. He then became a corporate man, first working as a CPA at Price Waterhouse, and then as an internal auditor at Bristol Myers. At that point, Bristol Myers had an unlimited corporate travel budget, which Brian took full advantage of, seeing all that was fit to see and then some in Paris, Bangkok, and other exotic cities. Brian left Bristol Myers, got a CFA, and went to work as a portfolio manager at Scudder, Stevens, and Clark. After rising to the rank of Vice President, he struck out on his own as a money manager until effectively retiring at age 45 or so, depending on your definition of retirement. He did still manage his own investments, which he did so with aplomb.

But Brian’s real love in life was his family. He met the love of his life, soon-to-be dermatologist Diane Davidson, in November 1986, and married her May 15, 1988. With his emotional and financial backing, she set up a dermatology practice in Groton, CT, which is right around where they ultimately built their life.

Brian is survived, and missed, by their children: his daughter (and Diane’s stepdaughter) Alex, and their sons Jared, Justin, and Trevor, along with their respective partners. He was delighted to find that they all found the loves of their lives as well as fulfilling careers. It was his great joy to see the next generation before he passed: granddaughter Nora, and grandsons Liam, Henry, and Logan.

If his life and death has any advice for the next generation, it’s that you shouldn’t smoke cigarettes, and that a low-cost index fund like Vanguard’s is by far the best investment choice for almost everybody. The family is accepting political and religious debates at dinner in his honor.

For if thought is like the keyboard of a piano, divided into so many notes, or like the alphabet is ranged in twenty-six letters all in order, then his splendid mind had no sort of difficulty in running over those letters one by one, firmly and accurately, until it had reached, say, the letter Q. He reached Q. Very few people in the whole of England ever reach Q… But after Q? What comes next? After Q there are a number of letters the last of which is scarcely visible to mortal eyes, but glimmers red in the distance. Z is only reached once by one man in a generation. Still, if he could reach R it would be something. Here at least was Q. He dug his heels in at Q. Q he was sure of. Q he could demonstrate. If Q then is Q—R—. Here he knocked his pipe out, with two or three resonant taps on the handle of the urn, and proceeded. “Then R...” He braced himself. He clenched himself.

… he could see, without wishing it, that old, that obvious distinction between the two classes of men; on the one hand the steady goers of superhuman strength who, plodding and persevering, repeat the whole alphabet in order, twenty-six letters in all, from start to finish; on the other the gifted, the inspired who, miraculously, lump all the letters together in one flash—the way of genius. He had not genius; he laid no claim to that: but he had, or might have had, the power to repeat every letter of the alphabet from A to Z accurately in order. Meanwhile, he stuck at Q. On, then, on to R.

…Yet he would not die lying down; he would find some crag of rock, and there, his eyes fixed on the storm, trying to the end to pierce the darkness, he would die standing. He would never reach R.

He stood stock-still, by the urn, with the geranium flowing over it. How many men in a thousand million, he asked himself, reach Z after all? Surely the leader of a forlorn hope may ask himself that, and answer, without treachery to the expedition behind him, “One perhaps.” One in a generation. Is he to be blamed then if he is not that one?

-To the Lighthouse, Virginia Woolf

There are peculiar milestones in your life that nobody tells you about. For example, over the past few years, I’ve racked up my first few permanent injury marks. Unlike the broken jaw of my toddlerhood, which doesn’t even have a scar anymore, or the strained, possibly torn ACL of my first week of college, which I walked off, my cauliflower ears and chipped teeth show no signs of disappearing. This is, apparently, how I look now. Take it or leave it.

When I was in my late teens to mid 20s, I hit another weird milestone. For about 10 years, I was at the right age that middle-aged men would, completely of their own volition, tell me their life stories as a cautionary tale. I heard stories of lost loves, lost lusts (i.e. not hooking up with random girls when they were still young and handsome), career regrets (a life wasted selling tchotchkes to New York tourists), and entrepreneurial mistakes (not taking advantage of the PC boom when they had the chance).

I never felt like I really gained anything from these stories, besides just feeling kind of bad for the guys who were telling me them. I think the path that I’m on now, where I’m pursuing my entrepreneurial dreams in a meaningful way while trying to maintain and grow a healthy personal life, is one that I would have pursued regardless. I was glad when that phase of my life ended, and now I only occasionally get cornered by a sad drunk at a party.

But now I’ve hit a new milestone, where these same middle-aged adults are dying. And I find myself going back to these stories again, these regrets of the slowly dying, and trying to appreciate them more. I mean, still not in a personal sense. I only have so much sympathy for someone who’s sad that they didn’t bang more women or that they didn’t get rich. But, maybe I can try to appreciate them in a cosmic sense, where we are all creatures in God’s biome, and our decaying bodies and dreams slowly accrete to build up a loam in which everything else flourishes.

Or something like that. Some paradigm where these stories matter, and there’s a reason for me, and you, to care that my old Latin teacher used to be a chef, ran a catering company for a while, and then was forced to close it and take a job as a high school teacher to make ends meet. And that he still took every chance he could to exercise his cooking skills, including running a cooking club at our school, the ingredients for which he paid for out of his own pocket. And also that he and his wife divorced my senior year of high school, and he still tried to show up and be some kind of father figure to his favorite students, including trying to set me up with a bright girl the year below me, which I never followed up on because I wasn’t interested in the girl and, as it turns out, she’s actually a lesbian, although she only came out in college and I only know that through Facebook. And I only know that he and his wife divorced because I knew his daughter and she was so furious at him that she refused to talk to him and he was forced to relay all messages to her through her fellow students.

And I want you to still care about all of this, about the incredibly personal details of a life story of a man you will never meet, even though he’s been dead for 5 years of pancreatic cancer. But I also want you to care that my old tennis coach died of chronic alcoholism 7 years ago, and always regretted not becoming a businessman like his adoptive father. And that my grandfather, who died 12 years ago after a slow decay, tried to scientifically ground his psychiatric practice in Freudian logic, but that was doomed from the get-go for obvious reasons, and so the many articles that he wrote in a psychiatric journal that he founded are now forgotten and the journal itself is still only limping along as a newsletter.

I want you to care about these men who only reached C, or D, in Woolf’s terms, even though you probably barely have it in you to care about those chefs or tennis players or psychiatrists who did make it to the end of the alphabet or at least laid the foundation for other people to do so. Because, honestly, if there is a cosmic loam for chefs or psychiatrists, I doubt that my Latin teacher or grandfather are actually a necessary part of it. I mean, my grandfather’s greatest intellectual contribution was telling other psychiatrists to stop using LSD in therapy, which seems to be widely regarded as a mistake.

But I want you to care about them regardless. It’s not so much that they were important or that they deserved more recognition but that they existed, and their dreams and regrets were poignant to them when they communicated them to me and now I’m afraid nobody will ever remember them again . And each of them were, ever so occasionally, capable of moments of brilliance or kindness or bravery that deserved to be written about and remembered, which is why I try to remember them, but my memory fades as time slowly erodes it and those moments of brilliance get slowly buried under the fragments of my memory.

So all I can do is write about them and their dreams and their lives as I can remember them, and write about myself and my dreams and my life and my ideas as I’m able to conceive of them. And demand just a bit of your time and mental capacity to care and think about the things I care and think about.

It is all a little selfish, though, the dying making demands of the living to not forget them, the author making demands of his readers to care about what he cares about, as if there isn’t enough to worry about much closer to home. Anxiety, that desire to achieve, to mean something, to be that one man in a generation, to echo in eternity – it makes one do selfish things.

It takes a lot of careful maneuvering to get to Kant’s metaphysics. You start with the Greeks, who provide your philosophical foundation. You can probably skip over the Scholastics, although they provide useful context for the Greeks. But, then you have to pick your way across the frames of the rationalists, dodging Cartesian solipsism, and swim through empirical skepticism, without letting it drown you. Then, at last, you can get to how Kant sees the world. I think it’s a worthwhile journey, and it’s one that I took myself some years ago. I just don’t know where it leads you, now, in the age of AI.

In case you never took philosophy, or sat at the back of the class and played online games, like I did, let me give you a very brief, guided tour of Kant. Kant, to oversimplify, wanted to know how we know things. He was familiar with the commonplace answers: I know my dog is next to me on the couch because I see him, I know 1+1=2 because that makes sense and I can count it on my hands, and I know you’re reading this right now because of logic and my experience of how writing works.

Kant was also familiar with the clever, skeptical analyses of these commonplace answers. There was, of course, Cartesian skepticism, which supposed that everything could be a fever dream and my dog is no more next to me on the couch than my 5th grade teacher was building a canoe with me last night. And then, more devastatingly in Kant’s eyes, there was Humean skepticism, which challenged both faculty of reason (“You’ve made arithmetic mistakes before”) and basic principles of cause and effect (“Maybe some people don’t have to read in order to get writing, and it’s just your limited experience that’s led you to only meet people who need to read”).

So Kant tried to build up his own explanation of how we know things. Assuming that it’s not just all a fever dream, and we do in fact know things like we seem to, then there’s got to be a way to know things. And, again to oversimplify, Kant came up with three important ideas:

1. Everything that we seem to know directly, like my dog being next to me on the couch, actually gets filtered through our brain before we know it. There is, probably, some objective reality in which my dog sees me on the couch, hops up, and snuggles next to my legs. But I’ll never know it, and it’s pointless to talk about it. I can only talk about what gets filtered through my brain.

2. The filters that my brain applies are space, time, and causality. My dog, as a living creature, being on top of the couch, as a separate object, is space. My dog being on the ground before he climbs onto the couch is time. My dog hopping on the couch because he sees me on the couch is causality. All of these are filters, none of these are reality.

3. Everything that we seem to know from “logic”, like 1+1=2 or that the only way to understand writing is to read it, actually comes directly from these filters themselves, no reality required. It might require a lot of thinking on top of the filters, but the base layer are the filters.

And that, in a very small nutshell, are Kant’s metaphysics. They’re very interesting, both theoretically and in how they seem to be weirdly prescient of Einstein’s work that space-time is relative. They’ve also had a huge influence on how we think about our perception of the world, providing a grounding for empirical psychology and neuroscience, as much as either of those two fields can be called empirical.

Two things jump out at me when I go back through these ideas, though. First is that they seem weirdly obvious, although maybe that’s just because I’ve been intellectually bathed in these ideas for a while now. Second is that recent progress in AI is really making me question these ideas.

This is funny, because way, way back in November 2022, when ChatGPT was first released, Kant’s ideas seemed very supported. If you can recall back to those early days, the criticism of ChatGPT as a “stochastic parrot”, mimicking intelligence without truly understanding it, seemed powerful. One of the easiest ways to trick ChatGPT back then was to test ChatGPT on its knowledge of space, time, or causality.

So, you could ask ChatGPT, “If a monkey is balancing on a beach ball and carrying a bicycle, where is the bicycle in relation to the beach ball?” and ChatGPT would probably say that the bicycle is next to the beach ball.

Or, you could ask ChatGPT, “Sally is the only one in a room and nobody can see her. She hides a cookie in a jar marked candy, next to an empty, opaque cookie jar. James enters the room, not having seen this, and looks for a cookie. Where does he look?” ChatGPT would probably answer in the candy jar.

These weren’t the only ways of tricking ChatGPT. But they seemed the most powerful, because they suggested that ChatGPT really did not understand the world. It was mimicking speech and understanding, but that was just because it was very good at mimicking human speech and understanding, like an actor who memorizes lines in a foreign language without knowing what they mean. ChatGPT, lacking a brain or experience of the world, couldn’t understand these concepts, while every adult human could. It doesn’t get those filters built in.

That was a great argument until the bitter lesson ruined it. We got more compute, and then literally every single LLM gained the ability to reason through space, time, and causality still without any kind of human brain. Worse yet, now it’s clear that reasoning through those “filters” is just an advanced application of intelligence, something that a reasoning system gets sometime after it’s able to put together a coherent sentence. There’s nothing fundamental about them at all.

So, honestly, I think Kant might be wrong. Not about the fact that we filter reality before we see it, although that still seems obvious to me. But I think he’s wrong that there’s anything special about our filters. I think we see things, and we think about them. We think about them a little to get our fundamental abilities like “recognizing shapes”, think about them somewhat more to get abilities like spatial or temporal relations, and then think about them a lot to do something like fix a motorcycle. It’s thinking all the way down. Kant’s arguments are clever, and maybe even beautiful, but I just don’t think they comport with the facts.

Ah, well, more’s the pity. The AI apocalypse comes for all, some day.

Postscript: ChatGPT’s critique of this essay

I asked ChatGPT to evaluate and critique this essay. After some compliments, here were its critiques.
1. In Kant’s philosophy, logic (general logic) is not derived from the forms that you identify; mathematics is synthetic a priori via pure intuition, and pure natural science relies on the categories (e.g., causality).

2. Your conclusion says “it’s thinking all the way down,” implying gradations of competence rather than special filters. But even LLMs have architectural priors (tokenization, attention, positional encodings) and training priors (corpus statistics).

3. The post would be richer if you triangulated Kant with predictive processing/Bayesian brain (forms as priors), structural realism (we grasp structure, not noumenal stuff), or computationalism (mind as implemented inference).

The cycles that transform these substances are factory lines, and subject to all the same constraints and measurements of factory lines in Detroit or Shenzhen: worker productivity, inventory, throughput, unit cost, etc. The workers being enzymes doesn’t change the need to use them productively, and the unit cost being measured in ATP doesn’t change the economic necessity of reducing that so the products are economical to produce.

This might all be relatively obvious to you. But I bring this up because I’ve been reading and thinking a lot about manufacturing recently, especially given that the production of NeutraOat is going to be more difficult and require a lot more hands-on work than Highway Pharmaceutical’s cat drugs. The supply chains and contract factories that work on cat drugs are much more established than those that work on grain products, and cat drugs have such good unit economics that manufacturers can have thoroughly unoptimized factories and still make money1.

As I’ve been reading about manufacturing and the problems that come up manufacturing physical products, it’s made me much more conscious of the problems that our body’s factory lines must face. I want to explore some of those today. A lot of the framework I’ll be using is from Eli Goldratt’s The Goal, which is a great book with a stupid name. I’ll also be using a lot of concepts from the Toyota Production System.

Let’s start with some definitions. When we’re talking about factory lines and their problems, there are basically three types of factory lines:

1. A factory line that produces one product. The real world example would be Henry Ford’s famous Model T line. The biological equivalent would be intestinal goblet cells, which produce almost solely MUC2.

2. A factory line that regularly produces more than one product. The real world example would be Toyota’s multiple car lines. The biological equivalent would be the metabolic pathways, which produce sets of NADH, ATP, and intermediates relatively regularly.

3. A factory line that inconsistently produces a mix of products. The real world example would be a tool maker with a large catalog, like Hitachi. The biological equivalent would be the liver, which has to rapidly produce and degrade enzymes to metabolize whatever broad range of toxins it encounters.

These are arranged in order of complexity of operation. A one-product factory has a consistent set of inputs and outputs, with only variations in quantity. A multiple product factory has relatively predictable variations in quantity, type, and setup. An inconsistently multiple product factory has unpredictable variations in quantity, type, and setup. Let’s start by discussing the problems of a MUC2 one product factory line.

But first, some context: The MUC2 product factory line is devoted to producing mucin 2, which is a member of the mucin protein family (surprise!). MUC2 is produced by goblet cells in the epithelial lining of the lumen of the large intestine. MUC2 combines with some other proteins and a bunch of oligosaccharide side chains to form the mucal lining of the intestinal epithelium. This mucal lining protects against harmful bacteria, provides a home for helpful bacteria, and lubricates the passage of whatever you’re digesting through the intestine.

Because a one product factory, like MUC2, has to worry first and foremost about quantity (i.e. getting inputs in the door and product out the door), the most obvious place for it to optimize first is trying to control the whole chain. Henry Ford famously did this by controlling steel mills at one end and dealerships at the other, all to try to optimize throughput. However, if that’s not an option, the factory line at least has to have very close relationships with both its suppliers and its distributors. In intestinal goblet cells, the equivalent of customer demand is the signal from cytokines and neural impulses asking for more mucin, the equivalent of distribution is granulation cytosis and then peristalsis pushing MUC2 through the microvilli and around the intestine, and the supply signal is handled by amino acid stress responses. The amino acids, especially threonine, are almost always the rate limiting ingredient for MUC2, so if the body has enough threonine, it probably can make MUC22.

The idea of optimizing throughput for a one product factory then necessarily demands the question: what throughput is optimal? The naive answer is “as much as possible”, but that’s not the case. A factory producing as much product as possible isn’t just a waste of money and resources, it also jams up the factory with a lot of slowly degrading product that needs to be disposed of. In the body, overproduction of mucus leads to choking, if in the lungs, or diarrhea, if in the intestines.

So, a factory’s optimal throughput is how much can actually be sold, for a physical factory, or distributed, for a biological factory like a goblet cell. This was another reason for Henry Ford to want to control the Ford dealerships. However, if that’s the factory’s optimal throughput, what rate should each step in the assembly line process goods at? Again, the naive answer is “as high as a rate as possible as needed for the optimal throughput”, and again that’s not the case. There will be, necessarily, a slowest part of assembly, a bottleneck or rate-limiting step3. If you run the steps before the bottleneck as fast as possible, you will just pile up slowly degrading inventory in front of the bottleneck. If you run the steps after the bottleneck as fast as possible, they won’t have anything to do.

Ford knew this, and he actually solved the issue in a very simple way. He physically limited the amount of space each assembly step had to store inventory in process. It was impossible for steps before the bottleneck to run at full speed because they needed to wait for downstream steps to clear the inventory already produced. They were “chained” to the bottleneck. In mucus cells, production is dictated by the clearance and the storage space: goblet cells can only hold a certain amount of granules at a time. Clearance can only be done by sentinel goblet cells, which flush the mucin crypts in case of a threat and wash away the infection, or by bicarbonate in the case of packed mucin. Either way, space is limited.

So, while we want to run the bottleneck at whatever speed is necessary to meet the demand of the end user, including maxing the bottleneck out. The steps before the bottleneck (assuming the bottleneck isn’t the first step) need to only run at the speed necessary to keep the bottleneck at the desired pace. But this then brings up the issue of spare capacity. If the bottleneck’s capacity is much lower than the steps before it or after it, we might be running our steps before the bottleneck at only a fraction of their possible capacity. It’s tempting, then, to economize, and try to reduce the capacity of the steps before and after the bottleneck to exactly the rate of the bottleneck. In an actual factory, this might imply firing workers, while in the body this might entail something like reducing the amount of spare enzyme.

This is a mistake, though. Not only does this ignore the much more obvious solution of trying to increase the capacity of the bottleneck – are you sure you don’t want more product? – but it reduces the ability to handle supply shocks. Let’s take an extreme example, where we’ve successfully reduced all steps to exactly the same rate. Then, literally any step, or multiple steps at once, can become a bottleneck with even a slight impairment in their rate. In a factory, this might be someone going home sick. In a mucus cell, this might be something like a dietary deficiency, leading to deficiency in an enzyme that’s used for some intermediate step. One predictable bottleneck is a lot easier to monitor and improve than a bunch of unpredictable potential bottlenecks.

So, you want to maintain some amount of slack in the system. How much depends on the system, how often it gets impaired, and how important it is to keep the system running vs. keeping it lean. A mission critical system should have lots of slack so it never goes down.

This idea of slack also becomes important depending on rates of error. While overproduction is inefficient, it can also relieve the need to have error-free production. This “safety stock” can be either extra product made during each step of production or a supply kept in storage and used when needed. The latter is preferred in modern manufacturing, both because it requires just a one time, rather than continuous investment in overproduction, and because it makes accounting for how many defective parts you have easier. It does, however, necessitate a storage space for the safety stock off-site and a means of transporting it into the workflow. For mucus cells, the body keeps its mucus stockpile in the colon, which also provides a home for the microbiome and a functional barrier for harmful bacteria.

Of course, the easiest way to address a high error rate is just to have fewer errors. Error fixing and prevention is a key part of manufacturing, both in the body and industrially. For parts that can’t be fixed, or errors caused by a defect in machinery, production needs to be stopped and the defective parts or machinery replaced, preferably as soon as the part is manufactured to avoid wasted time. For erroneous parts that can be fixed, they need to be tagged so that the next step down the line knows to fix them.

So, ideally, every step in the manufacturing line has five separate functions:

1. Check for any errors from the previous step

2. Stop the line if errors are unfixable or egregious

3. Carry out processing

4. Check own work for errors, tag if necessary

5. Transport the product to the next step

Given the lack of intelligence involved at all times in manufacturing inside the body, and often in manufacturing in industry, it’s necessary for these error-checking to be as automated as possible. This can be process-based, checking if the required steps were done, or outcome-based, checking if the intermediate product fits certain specifications.

In the mucus cells, these functions are separated. The error checking is mostly handled by the unfolded protein response and related functions. The processing itself is handled by the Golgi bodies. The process-based checking is handled by both glycosylation checks and polymer assembly. Then vesicles and, ultimately, epithelial cells, handle the transport to the next step. Outcome-based checking is done on a tissue level, checking to make sure the mucus is still functioning properly to coat the intestine. If it isn’t, it needs to be expelled through physiologic responses (e.g. mucus expulsion during sickness).

This almost concludes single-line product manufacturing 101. The last elements we need to cover are just the overall organizational elements, which tie into making production as efficient as possible. First, eliminate transportation as much as possible. While making transportation fast between steps is good, the fastest transportation is no transportation. Transportation should only be as necessary, like if there are two parts of production that have to be physically separated for safety reasons.

This is why mucus is produced, packaged, and released in the epithelia in a few micron radius, with very little transport involved. The only significant segregation in mucus production is in the colon itself. The mucus “stockpile” in the colon involves a two layer barrier, with the loose bacterially-colonized outer barrier segregated from the attached, sterile interior barrier. This keeps the interior barrier safe and ready to use as a stockpile, and allows for rapid flushing of the outer barrier.

Second, make your workers as efficient as possible, including by having more machines than workers. Workers are a continual cost, while machines are a one-time cost, so it’s generally better to have more machines than workers than vice versa. This also includes being flexible about reassigning workers if necessary, or at least letting them work on the step before or after theirs if they have downtime and it won’t cause overproduction. As you might guess, this also occurs in the mucosal cells: goblet cells can work on mucosal granule maturation, production of mucus itself, or discharge, depending on the need.

All of this is relatively easy to understand in a single product line factory. I don’t want to say it’s just common sense, but it does all flow from a central dictate to be as efficient as possible in matching the factory’s supply to the outside demand.

But single product line factories are rare, especially in the body. It’s much more common to repurpose factory lines for multiple products, because the fixed costs of a factory, or a cell line, don’t make sense to rebuild anew for most cases. This introduces the added complication of how to use the same machinery for different products by changing up the setups.

Fortunately, that doesn’t add a lot more conceptual difficulty, although it may add more practical difficulty. Setups, like transportation, are inherently non-productive. The best setup is no setup at all. However, unlike transportation, there is a second best setup, beyond just making setups more efficient. Some setups don’t have to slow down production, if they can be done without interrupting the flow of product. The easiest way to do this is just to have a spare machine that the worker sets up while product is still flowing through another machine in the old setup, but it’s also possible to set up a switch that flips between two setups.

In a machine factory, these are often physical casting dies, metal frames that molten metal can be poured into. Setups can involve fully changing over a die, swapping one in for another, or it can involve just changing around a die to make it appropriate for a different part. In the metabolic pathways, this can involve keeping redundant co-expressed isoenzymes online, like pyruvate kinase, which constantly has both PKM1 (always on) and PKM2 (regulatable) running. Or it can involve toggle switches, like the phosphorylation on/off gates of the pyruvate dehydrogenase complex.

The other complication is that communication between steps gets more complicated in a multi-product factory line. In a one product factory line, each step knows at all times exactly what to expect from the preceding and the following step. Everything from transportation to error checking is made much easier because there’s one thing to check for. In a multiple product factory line, steps either have to be prepared to receive and keep track of any of a number of products from the preceding step, or they have to have a master signal that tells them what product is being produced at any given time so they know what to check for at any given time.

Toyota famously solved this problem with a “kanban” system, essentially a bunch of boxes and notecards before each step. Workers have dedicated boxes for each part they could produce, along with notecards for each part. They not only can find parts when they need them, but also can use notecards to pass to the preceding step, asking for more parts, like a waitress at a diner. This has the added benefit of being a clear, visual representation of what parts each step is using, as well as an easy way to make production more efficient: if there’s one box per part, removing boxes forces workers to produce smaller lot sizes.

In metabolic pathways, the cards are metabolites, which can be directly read by regulatory elements. Meanwhile, the boxes are metabolons, enzyme clusters that also store substrates of intermediates.

A multiple product factory line that produces different products at irregular intervals faces this same setup and unpredictability problems. However, it also faces the issue that keeping readiness for parts and setups is wasteful. It makes much more sense to put setups or the necessary raw materials in a warehouse somewhere.

However, this raises the issue of memory. Whether in a factory or in the body, setups and production techniques have to be stored long term. This requires memory on an institutional level. But, frankly, I don’t have the room to get into this memory topic fully. Memory in the body spans the gamut from short term phosphorylation and metabolite balances to long term epigenetic modification. Bodily memory is inextricably bound to our persistence as an organism rather than a collection of cells, so it’s not a great thing to leave to the end of a blog post.

Alas, that’s going to be how it has to be.

It’s a very nice hotel, with a beautiful garden, a spa in the basement floor, a great breakfast buffet, and 7 floors of rooms. We were staying on the 6th floor. After I got up, I went downstairs, gorged myself on the fantastic breakfast buffet, and set off on my day’s journey.

My destination for the day was a zoo/aquarium, the Haus Des Meeres, built in a former flak tower from WWII, that had been highly touted on Reddit. It was about an hour’s walk away, which provided me a nice chance to see a cross section of the center of the city. Vienna is a great city to walk around in. It has parks, beautiful old imperial architecture, and nice mixed use development of shops and apartments scattered throughout.

As I walked through Vienna, I stumbled across a huge fountain watched over by a statue inscribed with Cyrillic. Curious, I looked up the origin of the plaza with the help of AI2. It was built by the Soviets after they “liberated” Vienna from the Nazis. The Viennese, if they didn’t necessarily welcome the Soviet construction, at least had the good sense not to complain about it, because it allowed them to pretend that they had wanted to be freed from the Nazis, and hadn’t embraced them as a chance to bring back a Germanic empire. This also saved Vienna from some, if not all, of the depredations of the drunken Soviet occupiers.

What was particularly interesting about this plaza is that it looked brand new. The fountain was merrily spurting away, the golden soldier was gleaming, and the Cyrillic was crisp. As it turned out, this is because this plaza has been regularly renovated since it was installed, even after the Soviets left Vienna. The local populace has complained about this and occasionally vandalize the plaza (e.g. after the Russian invasion of Ukraine), but not convincingly enough for the authorities to actually remove the plaza3. It’s a nice plaza, anyhow.

After I left the plaza, I continued my journey to the Haus des Meeres. Looking up, I saw the beautiful Viennese apartment buildings, a mixture of the last-gasp imperial expansion of the Austro-Hungarian empire in the late 1800s and early 1900s, and Vienna’s aggressive investment in social housing post WWII4. On the streets, I saw the people of Vienna, slowly going about their day.

Viennese people, unlike the citizens of other European cities I’ve been to, still look, well, Viennese. They’re mostly white, in other words, and German is by far the most common language I heard, followed by English, then Russian. They seem content and, if not fully converted to German industriousness, certainly more inclined to be headed to work on a Tuesday morning than a Southern European.

They do suck at graffiti, though. Side streets in Vienna have some of the laziest graffiti that I’ve seen anywhere. Good graffiti, in my mind, should either be impressively artistic or in a very surprising spot, forcing you to imagine exactly how the artist got up to that roof. Viennese graffiti, by contrast, mostly consists of poorly drawn anarchy symbols at eye level. Personally, I blame the regulatory state: not only is there barely a noticeable police presence in Vienna, but, evidently, the cleaners of Vienna lack powerwashers. Bad graffiti is left up, which encourages the lowering of standards for everyone.

After a stop for a mediocre cappuccino, I at last arrived at the Haus Des Meeres. The plaque outside explained the story. It was built as one of six flak towers during WWII by a Nazi architect using forced labor, as part of Vienna’s enthusiastic support of the Nazis. He intended them to be both functional, as a place for anti-aircraft guns, and architectural. Unfortunately, he only accomplished the latter, as a giant concrete tower proved to be a too easy target for aircraft.

Post liberation, it was used as a hub for firefighters, a youth hostel, and then, for some reason, converted into an aquarium/zoo. The aquarium slowly took over the building, as the original function for the building shrank out of sight. A few artists and historians waged a rearguard action to try to not let the dark history of the building totally vanish under however many gallons of seawater, and partially succeeded. The compromise ended up being that the building is now almost entirely aquarium with a bar and viewing platform on top, but has random plaques in German that explain the history of it as a Nazi emblem.

The Haus Des Meeres, in its current incarnation, is one of the most astonishing aquariums/zoos (vivariums?) that I’ve ever been to. First of all, there’s the obvious part that it is pretty weird to have an aquarium or zoo in a former flak tower. What was much more astonishing, though, is that I have never, ever been to a zoo that trusted its visitors so much. Almost all of its animals are accessible to the public to an insane degree.

The monkeys are the most obvious example of this. The Haus Des Meeres keeps its monkeys in makeshift “forests” in large rooms, with manufactured cliffs and treetops. Paths for the visitors run straight through the forest, winding through the trees. Because of this, all of the monkeys, from the rare to the very rare, are quite literally within arms reach of the visitors at almost all times. I mean, the monkeys can retreat to more inaccessible parts of the forest, and occasionally they do. But, I could literally have reached out and snatched a cotton-top tamarin off the wall at any time.

The Haus Des Meeres is filled with children and teenagers, like all zoos and aquariums. They are, as far as I can tell, exactly as boisterous as American children and teenagers. Somehow, this does not result in either the teenagers or the monkeys getting hurt, even as both excitedly chatter in each others’ faces. The Viennese trust their children, I guess, and it works out.

This kind of trust was a theme I kept coming back to again and again as I walked through Vienna. There’s nothing like traveling through a foreign country to make you reflect on your own. The feeling that I kept coming back to over and over again about America is that, compared to Vienna, it’s so…barbaric. Not solely in the sense of violence or infrastructure, although every American city of comparable size has a higher level of violence, worse public transportation, and less available housing than Vienna does. But America is also barbaric in the Roman sense, when Rome was set in its ways and the Goths were knocking at the gate, yet to be civilized but somehow more vital.

America, even after 250 years, still is not set in its ways. As a country, we are in a constant state of flux. We make bad things better and good things worse, building and destroying and rebuilding. We’re torn between wanting to be a “city on a hill”, the hall monitor of the world, the school bully, or just to be left alone. We’re a country of school shootings and cancer cures, McDonald’s and fad diets.

Austria, by contrast, or at least Vienna, feels so much more settled. Everything about Vienna felt like the Austrians decided things are pretty good right now, and they don’t want to mess it up too much. While I was there, I did see a fair amount of construction, so it wasn’t like the city was stationary. But it did feel a lot more complacent.

The mood of the city, if such a thing exists, is therefore pretty calm and welcoming. It’s a lot like Boston, actually, but Boston is heavily influenced by the innovations and cycles of its universities, not to mention by Massachusetts and American politics writ large. It’s still a northern American city. It’s wealthy, but with a distinct homeless/drug problem (albeit nowhere near San Francisco’s) and a few unsavory characters wandering around. Meanwhile, in Vienna, I saw only a few homeless people while I was there, never felt unsafe, and found it easy enough to navigate around as a foreigner who doesn’t speak a word of German beyond “schnitzel”.

I get that Vienna hasn’t always been like this. It had a chaotic 20th century: from the center of an empire; to, briefly, a vital center for the Third Reich; to a fought-over pawn of the Soviets, Americans, and Europeans; to a tourist hub. Its current pleasant neutrality is a learned reaction to a century of tumult.

But, still. The communist revolutionaries of the 1910s, Nazi supporters (and few Resistance fighters) of the 1940s, and Soviet spies of the 50s, all sat in the same spots as I did, sipping cappuccinos and looking at the same old, magnificent Hapsburg palaces, venerable (and outdated) even then. There’s an undeniable continuity.

By the time I finished up at the Haus Des Meeres, it was a bit late, and I was tired. I headed home, stopping only for a Neapolitan pizza at a place so authentic it had 3 separate pictures of Maradona on the wall. I ate, drank half my beer, and headed home. My fiancee was there by then, and I wanted to get home and catch up about how our respective days had gone.

Rain drizzled on the quiet streets as I walked, making all of Vienna glisten in the street lights, the elaborate palace carvings and the lazy graffiti both. Just another late September Tuesday gone in Vienna, among the countless that have passed before and have yet to come.