Category: Games
The great Ken Regan on AlphaZero
A must-read for anyone who has been following this issue, Ken considers how close to God AlphaZero actually came:
We must pause to reflect on how clarifying it is that this single heuristic suffices to master complex games—games that also represent a concrete face of asymptotic complexity insofar as their size n-by-ngeneralizations are polynomial-space hard…
It may be that we can heuristically solve some NP-type problems better by infusing an adversary—to make a PSPACE-type problem that hits back—and running AlphaZero.
That sort of thing. And don’t neglect the comments.
The Age of the Centaur is *Over* Skynet Goes Live
“Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm”
The game of chess is the most widely-studied domain in the history of artificial intelligence. The strongest programs are based on a combination of sophisticated search techniques, domain-specific adaptations, and handcrafted evaluation functions that have been refined by human experts over several decades. In contrast, the AlphaGo Zero program recently achieved superhuman performance in the game of Go, by tabula rasa reinforcement learning from games of self-play. In this paper, we generalise this approach into a single AlphaZero algorithm that can achieve, tabula rasa, superhuman performance in many challenging domains. Starting from random play, and given no domain knowledge except the game rules, AlphaZero achieved within 24 hours a superhuman level of play in the games of chess and shogi (Japanese chess) as well as Go, and convincingly defeated a world-champion program in each case.
In other words, the human now adds absolutely nothing to man-machine chess-playing teams. That’s in addition to the surprising power of this approach in solving problems.
Here is the link, via Trey Kollmer, who writes “Stockfish Dethroned.” Here is coverage from Wired. Via Justin Barclay, here is commentary from the chess world, including some of the (very impressive) games. And it seems to prefer 1.d4 and 1.c4, loves the Queen’s Gambit, rejected the French Defense, never liked the King’s Indian, grew disillusioned with the Ruy Lopez, and surprisingly never fell in love with the Sicilian Defense. By the way the program reinvented most of chess opening theory by playing against itself for less than a day. Having the white pieces matters more than we thought from previous computer vs. computer contests. Here is the best chess commentary I have seen, excerpt:
If Karpov had been a chess engine, he might have been called AlphaZero. There is a relentless positional boa constrictor approach that is simply unheard of. Modern chess engines are focused on activity, and have special safeguards to avoid blocked positions as they have no understanding of them and often find themselves in a dead end before they realize it. AlphaZero has no such prejudices or issues, and seems to thrive on snuffing out the opponent’s play. It is singularly impressive, and what is astonishing is how it is able to also find tactics that the engines seem blind to.
Did you know that the older Stockfish program considered 900 times more positions, but the greater “thinking depth” of the new innovation was decisive nonetheless. I will never forget how stunned I was to learn of this breakthrough.
Finally, I’ve long said that Google’s final fate will be to evolve into a hedge fund.
Kim Jong Un is rational in a very long-term game
That is the thesis of my latest Bloomberg column, note that Kim is only 33 and could be around for another fifty years or so he hopes. Peaceful exile probably is not an option! So how does one hold onto power and avoid those anti-aircraft guns? Here are some excerpts:
It is very difficult to predict the world a half-century out. Fifty years ago, China was just coming out of the Cultural Revolution, and Japan’s rise was not yet so evident. North Korea was possibly still richer than the South, which in 1960 was one of the poorest countries in the world. It’s unlikely anyone had a reasonable inkling of where things would stand today.
So if you are a dictator planning for long-term survival under a wide range of possible outcomes, what might you do? You don’t know who your enemies and your friends will be over those 50 years, so you will choose a porcupine-like strategy and appear prickly to everyone.
We Americans tend to think of Kim as an irritant to our plans, but his natural enemy in the long run is China. It is easier for North Korea to threaten Chinese cities with weapons, and its nuclear status stands in China’s way of becoming the dominant regional power in East Asia. Chinese public opinion has already turned against North Korea, and leaders wonder whether a more reliable, pro-Chinese option to Kim might be installed. Since assuming power, Kim has gone after the generals and family members with the strongest ties to China.
One way to interpret Kim’s spat with U.S. President Donald Trump is that he is signaling to the Chinese that they shouldn’t try to take him down because he is willing to countenance “crazy” retaliation. In this view, Beijing is a more likely target for one of his nukes than is Seattle.
More radically, think of Kim as auditioning to the U.S., Japan, South Korea and India as a potential buffer against Chinese expansion. If he played his hand more passively and calmly, hardly anyone would think that such a small country had this capacity. By picking a fight with the U.S., he is showing the ability to deter just about anyone.
There is much more at the link, and of course I consider the “are these people really all so rational?” critique. You will note by the way that this inverts the usual argument that a longer time horizon means more cooperation. In this case a longer time horizon means more signaling and a more rambunctious form of signaling, precisely because the time horizon is long.
Would North Korea consider a first strike?
Vipin Narang says yes:
The strategy turns on Kim’s main calculation that the United States will say it’s not worth losing a major American city to get rid of him.
Of course he could not knock out a major American or allied target, but he could use them somewhere. And the use would boost his, uh…credibility. In fact Charles Murray is worried.
I think of the model this way. If Kim is irrational, we have obvious reason to worry, and of course a first strike could not be ruled out. Remember Pearl Harbor? (Or is that “Remember Pearl Harbor!”)
Alternatively, say all involved parties are fully rational in the selfish sense. Fully rational agents make purely forward-looking calculations. So if Kim used a nuke to kill a sparrow in North Korea, we would not attack because fear of losing an American city would far outweigh desire to retaliate for the loss of the sparrow.
How about one sparrow in the DMZ? In Japan? In the Arctic? In a Malaysian airport? Or maybe one sparrow, three sled dogs, and thirty Inuit?
At what point do we give it a go, and risk a poorly aimed North Korean ICBM being shot off into the sky?
What if Kim uses “only” a biological or chemical weapon, designed for minimum but noticeable impact, on a nearby country? You should think of Kim’s strategy space as a continuous variable, with some noise added of course.
Is the space of “boosts his credibility and domestic stature, but without too much upping the risk of massive American retaliation” really the empty set?
Maybe. Maybe not. I give it about one percent, which in expected value terms is still a real worry.
Strategic tennis grunts (from our email)
Hi guys:
I can pass along that there’s another angle to the grunts (having played a lot of tennis). The sound of the ball hitting the racket provides useful information, particularly for a mishit or a powerful shot — because you have to move up or back quickly to cope. For years, top tennis players have used grunts and shrieks to conceal this sound from their opponents (e.g. I always thought Sharapova, and Seles years ago, were prime offenders). There’s no need for such noises as a function of effort, or events like NBA games would sound much different. But the tennis authorities haven’t done anything about it.
In table tennis, where I have a very long involvement, the spin on the ball is tremendous in high-level play — so much so that a concealed dead ball (with no spin) is a very effective tactic because the opponent will err by responding to the spin that isn’t there. Years back, a totally dead racket covering was developed for this purpose; even worse, it tends to continue the spin so that the originator effectively gets the reverse back of what he put on the ball. A top US player with whom I grew up developed a style where he used only one side of the racket for both forehand and backhand, while frequently flipping between the spinny and dead sides of his racket that were colored the same. Players could hear the difference, however, as the dead side made a little thud when struck. His innovation was to stomp his foot on the floor each time he struck the ball (going beyond the norm of the time of just stomping on the serve). A subsequent regulatory change required rackets to have one red and one black side, to facilitate keeping track of which rubber covering is being used for a given shot.
Best,
Carl [Danner]
And here is my previous link to a new study of tennis grunts. And Carl sends along this video for table tennis. Sports aside, what other social practices fit this “misdirection” model?
The orca wars escalate, with special reference to the economics of fisheries
The orcas will wait all day for a fisher to accumulate a catch of halibut, and then deftly rob them blind. They will relentlessly stalk individual fishing boats, sometimes forcing them back into port.
Most chilling of all, this is new: After decades of relatively peaceful coexistence with cod and halibut fishers off the coast of Alaska, the region’s orcas appear to be turning on them in greater numbers.
“We’ve been chased out of the Bering Sea,” said Paul Clampitt, Washington State-based co-owner of the F/V Augustine.
Like many boats, the Augustine has tried electronic noisemakers to ward off the animals, but the orcas simply got used to them.
“It became a dinner bell,” said Clampitt.
John McHenry, owner of the F/V Seymour, described orca pods near Alaska’s Aleutian Islands as being like a “motorcycle gang.”
“You’d see two of them show up, and that’s the end of the trip. Pretty soon all 40 of them would be around you,” he said.
A report this week in the Alaska Dispatch News outlined instances of aggressive orcas harassing boats relentlessly — even refusing to leave after a desperate skipper cut the engine and drifted silently for 18 hours.
These are not Coasean orcas, or are they? And sperm whales are now in on the act:
Fishing lines are also being pillaged by sperm whales, the large square-headed whale best known as the white whale in Moby Dick.
“Since 1997, reports of depredation have increased dramatically,” noted a report by the Southeast Alaska Sperm Whale Avoidance Project.
A remarkable 2006 video by the Avoidance Project captured one of the 50,000 kg whales delicately shaking fish loose from a line. After a particularly heavy assault by sperm whales, fishers are known to pull up lines in which up to 90 per cent of the catch has disappeared or been mangled.
Here is the full story, with video, and further points of interest. For the pointer I thank the excellent Mark Thorson.
Why was the Qatar cut-off so extreme and sudden?
I am considering hypotheses here, to see how game theory might apply, so don’t think of this is an actual description of the situation.
As an economist, what struck me was the quick and extreme cut-off of Qatar by the Saudis and six other supporting parties. In the simplest versions of principal-agent theory, we think of most incentives as being applied continuously and varied in small doses: was Qatar’s behavior the day before the Qatar embargo/boycott really so different than the day of and after? So why did it happen this way? I can think of a few possibilities:
1. The boycott is like suddenly firing misbehaving workers. For morale reasons, you don’t want to keep them around on lesser terms, because they will be destructive. This hypothesis implies that the cut-off of Qatar is a permanent one.
2. Demonstrations of power require large, discrete events. If the Saudis had simply tweaked the incentives facing Qatar, the Qatar citizenry might not have distinguished the effects of that tweak from random noise. This hypothesis suggests that once the Saudis have made their point, and received Qatari concessions, the cut-off will be lifted or at least modified.
Note that along this game path, Qatar may not wish to “fold” immediately, as that could make them an ongoing puppet of the Saudis, all too easily manipulated. And indeed Qatar still has significant open markets for its natural gas.
3. Donald Trump’s meeting with the Saudis gave them an unexpected green light, either explicitly or implicitly, and thus the sudden receipt of this new information motivated their sudden switch in behavior.
During my recent trip to the Middle East I stated that there can no longer be funding of Radical Ideology. Leaders pointed to Qatar – look!
— Donald J. Trump (@realDonaldTrump) June 6, 2017
#3 still may be consistent with either #1 or #2.
4. The Saudis actually are playing a game with Iran, not so much with Qatar. What appears to be a big, sudden snap to the Qataris is actually just a smallish, mid-sized tweak in the incentives being applied to Iran. Qatar, because it is so small, feels a high degree of collateral damage.
5. The punishment space is multi-dimensional. Once “duration of punishment” is viewed as a variable, even a big punishment applied for a short period of time can be viewed as a marginal tweak. In this sense there is no paradox.
6. The Saudis view the Qataris as the ones who made a “discrete break” from the previous equilibrium, by paying a $1 billion ransom to Iranian and al Qaeda-linked forces, to induce the release of some kidnapped royal family members. Discrete breaks are inefficient, but perhaps you have to respond to one discrete break with another, precisely because they are inefficient.
7. Ian Bremmer mentioned on Twitter that 90% of the Qatari food supply is imported, 40% of it from Saudi, and now that is at risk. There are some countries for which a partial degree of agricultural subsidies and protectionism may make sense, for national security reasons. In any case, the degree of allowed smuggling reintroduces the notion of a smoother punishment space.
In a rational actor model (ha), this cut-off would be lifted in about a week from now.
Those new service sector jobs average is over spelling bee tutor edition
In the world of competitive spellers, Sylvie Lamontagne is known as a juggernaut. She placed fourth in last year’s Scripps National Spelling Bee, and ninth in 2015. Last summer, she traveled to California and won the Spelling Bee of China’s North America Spelling Champion Challenge, a contest for kids in the United States and China.
Now that the 14-year-old from Denver is no longer eligible to compete in this week’s National Spelling Bee at the Gaylord National Resort and Convention Center in Maryland — which is televised on ESPN and often turns kids like Sylvie into momentary celebrities — she’s focusing on a new vocation: spelling bee coach.
Sylvie’s rate? $200 an hour.
Hiring coaches isn’t new. But bee aficionados say a recent surge in competition, and a tightening of rules meant to limit co-champions, has spawned a demand for younger coaches such as Sylvie: high-schoolers or college kids, months or just a few years into their bee retirement, who can pass along fresh intelligence on words to memorize and how to decode bizarre words based on their language of origin.
That is from Ian Shapira at WaPo.
My Conversation with Garry Kasparov
Yes, the Garry Kasparov, here is the link to the podcast and transcript. We talked about AI, his new book Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins, why he has become more optimistic, how education will have to adjust to smart software, Russian history and Putin, his favorites in Russian and American literature, Tarkovsky, his favorite city to play chess in, his match against Deep Blue, Ken Rogoff, who are the three most likely challengers to Magnus Carlsen (ranked in order!) and who might win. Here is one excerpt:
GK: The biggest problem, and I’ve been talking about for quite a while, that we’re still teaching very specific knowledge in the schools. Instead of teaching what, we have to teach how because this knowledge may be redundant 10 years from now. We are preparing kids for the world that will change dramatically. By the way, we already know it will look different. So what’s the point of trying to teach kids at age 10, 11, 12 without recognizing the fact that when they finish college, when they will become adults looking for jobs, the job market will be totally different?
And:
COWEN: …If we look back on centuries of Russian history, do you think there’s something in Russian geography or demographics or geopolitics — what has it been that has led to such unfree outcomes fairly systematically?
Where do you find the roots of tyranny in the history of Russia? Is it a mix of the size of the country, its openness to invasion, its vulnerability, something about being next to a dynamic Europe, on the other side, China? What is it?
KASPAROV: It’s a long, if not endless, theoretical debate based on our interpretation of certain historical events. I’m not convinced with these arguments about some nations being predetermined in their development and alien to the concept of democracy and the rule of law.
The reason I’m quite comfortable with this denial . . . We can move from theory to practice. While we can talk about history and certain influence of historical events to modernity, we can look at the places like Korean Peninsula. The same nation, not even cousins but brothers and sisters, divided in 1950, so that’s, by historical standards, yesterday.
And:
Let’s look at Russia and Ukraine, and let’s look, not at the whole Ukraine, but just at eastern Ukraine. Eastern Ukraine is populated mostly by ethnic Russians. In the former Soviet Union, the borders between republics were very nominal. People could move around, it was not a big deal. Even after the collapse of the Soviet Union, the official state border between Russia and Ukraine was respected, but people still could move around. They didn’t need special visas.
When we look at ethnic Russians born and raised in Kursk and Belgorod on the Russian side and across the border, say in Kharkiv and Dnipropetrovsk on the Ukrainian side, there were people that could be hardly separated anything. They read the same newspaper, Pravda, watched the same television, spoke the very same language, not even accents. But somehow, in 2014, after Putin’s annexation of Crimea and invasion of eastern Ukraine, we saw a huge difference. Most of ethnic Russians in Ukraine signed for the Ukrainian army, fighting against Putin’s invasion, against the same Russians that came from the other side.
It could be a long debate, but I would say that one of the main reasons is that Ukraine experienced in 1994 a gradual transition of power from one president to another after sitting president Leonid Kravchuk lost elections and walked away. Ukrainians somehow got an idea that power is not sacred, and government can come and go, and they can remove it by voting.
And even despite the fact that Ukraine never experienced higher living standards than Russia, people realized that keeping this freedom, keeping this ability to influence their bureaucrats and government through the peaceful process of voting and, if necessary, striking, far more effective than Russia’s “stability” where the same leader could be in charge of the country with his corrupt clique for a long, long time.
On computer chess, I most enjoyed this part of the exchange:
KASPAROV: But I want to finish this because what we discovered in this process . . . I wouldn’t overweight our listeners with all these details. I don’t want just to throw on them the mass information.
COWEN: It’s amazing what people will enjoy, though. You’d be surprised.
Self-recommending! We cover many other topics as well, again you can read or listen here.
And I strongly advise that you buy and read Garry’s wonderful new book Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins.
Las Vegas average is over no arbitrage condition
Now operators have started scrutinizing complimentary drinks, introducing new technology at bars that track how much someone has gambled—and rewards them accordingly with alcohol. It’s a shift from decades of more-informal interplay between bartenders and gamblers.
Sports books have capitalized on big events, too. During March Madness, a five-person booth at the Harrah’s Las Vegas sports book cost $375 per person, which included five Miller Lite or Coors Light beers a person. In the past, seating at most sports books was free and first-come, first-served, even during big events. Placing a small bet or two could get you free drinks.
“The number-crunchers, the bean-counters have ruined Las Vegas,” said Brad Johnson, who lives in North Carolina and has come to Las Vegas almost every year since the early 1970s. “There’s no value to it; there’s no benefit.”
Casinos on the Strip now derive a smaller share of revenue from gambling. In 1996, more than half of annual casino revenue on the Strip came from gambling. Last year, the share was down to about a third, according to the University of Nevada-Las Vegas. More of the revenue comes from hotels, restaurants and bars.
That is from Chris Kirkham at the WSJ, via Annie Lowrey.
What should I ask Garry Kasparov?
I will be doing a Conversations with Tyler with him, no public event, podcast only. Today by the way is his birthday, so send along some good questions as a birthday present to him, and a non-birthday present to me!
Garry’s forthcoming book Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins is just superb, and the podcast will be released around the time of book publication in early May.
Does AI *really* help the defense in military situations?
Christina, an apparent MR reader, asked me whether it is really true that AI helps military defense more than military offense, as was previously argued by Eric Schmidt. I can think of a few parallel cases:
1. In chess, AI clearly has helped the defense. Top computer programs never play 32-move brilliant sacrifice victories against each other, a’ la Mikhail Tal. Most games are drawn, and a victory tends to be long and protracted. (Do note it is sometimes better to get the war over with and lose right away.)
2. In the NBA, analytics have helped offense more, for instance by showing that more attempted shots should be three-pointers. Analytics of course is not AI, but you can consider it a more primitive form of using information technology to improve decisions.
3. It is interesting to ponder the differences between chess and the NBA as potential analogies. In chess, the attack often “plays itself,” as the player with the initiative may be following fairly standard strategies of bringing the Queen and some lesser pieces in the neighborhood of the opposing King, or maybe just capturing material. Finding the correct defense is often a more complex matter, and the higher quality of the chess-playing programs thus boosts defense more than offense. Besides, under perfect information chess is almost certainly a draw, and the use of AI asymptotically approaches that outcome.
In professional basketball, the offense typically has more options and permutations, and given any offensive decisions, the defense often respond in fairly typical fashion, such as lunging at the player attempting a shot, or doubling Stephen Curry as he crosses the half-court line. In those cases where the defense has more options, however, analytics conceivably could help basketball defense more than offense. A (hypothetical) example of this would be using game tape and AI to see which kinds of tugs on the jersey best disrupt the shot or rhythm of the team’s leading scorer. That said, most of the action seems to be in honing the options for the offense.
4. Is warfare more like chess or more like the NBA?
I believe the USA has more options in most of its conflicts, and thus AI will help the United States, at least at first.
In the Second World War the Nazis had more options than their opponents. In the Civil War and American Revolution, however, the available offense was more static and predictable, and AI for those fighting forces might have helped the defense more. In the Iran-Iraq war I suspect the defense had more options too. Terror groups have more meaningful options than the forces defending against terror, and thus AI might help terror groups more than the defense, at least provided they had equal access to the data and to the technology (which is doubtful at this point, still as part of the exercise this is useful).
5. One important qualifier is that the chess and NBA examples already assume a game is on to be played. A war, in contrast, is started as a matter of volition on at least one side. If AI creates a new arms race of sorts, where one side at times opens up a decisive lead, that may provoke more decisions to engage and thus attack. The mere fact that AI increases the variance in the power gap between the two sides may increase the number of attacks and thus wars.
So there is more to this question than meets the eye at first, and I have only begun to engage with it.
Addendum: AI is also spreading in the legal world, will this help defendants or plaintiffs more?
Cyber plus nuclear equals Uh-Oh
Cyber weapons are different. If you are a state and you let potential enemies know about your arsenal of cyber attacks, you are giving them the opportunity to fix their information systems so that they can neutralize the threat. This means that it is very hard to use cyber weapons to make credible threats against other states. As soon as you have made a credibly-specific threat, you have likely given your target enough information to figure out the vulnerability that you want to exploit.
This means that offensive cyber weapons are better for gathering intelligence or actually taking out military targets than for making threats. In this regard, they are the opposite of nuclear weapons, which are more useful as threats than as battlefield options. Nuclear weapons can create stability because they deter attacks. In effect, they create a stable system of beliefs where no state wants to seriously attack a nuclear power, for fear that this might lead to a conflict that would escalate all the way to nuclear war.
Nuclear weapons and cyber attacks don’t mix well.
Unfortunately, this means that the advantages of cyber operations become an important liability for nuclear deterrence when they are used for “left of launch” attacks on nuclear launch systems. By secretly penetrating another state’s launch system, you may undermine the stable system of beliefs that discourages an attack.
Consider what might happen in a tense standoff between two states that both have nuclear weapons, where one state has penetrated the other state’s launch system, so that it could stop a nuclear counter attack. The state that has penetrated the launch system knows that it has a military advantage. However, if it reveals the advantage to the target state, the target state will be able to patch its system, destroying the advantage. The target state does not know that it is at a disadvantage, and it cannot be told by the attacker.
The result is that the two states have very different perceptions of the situation. The attacker thinks that there is an imbalance of power, where it has the advantage. The target thinks that there is a balance of power, where both states have nuclear weapons and can deter each other. This means that the first state will be less likely to back down, and might escalate conflict, secure in the knowledge that it can neutralize the other state if necessary. However, the target state may too behave in provocative ways that raise the stakes, since it mistakenly believes that at a certain point the other state will have to back down, for fear of nuclear war. Thus, this creates a situation where each side may be more willing to escalate the tense situation, making it more likely that one state will decide to move toward war.
That is from Eric Gartzke and Jon R. Lindsay, there is more interesting material at the link.
Wesley So, talent extraordinaire
This 23-year-old Filipino-American should be starring in a Malcolm Gladwell column. A few years ago, he was an up-and-coming aggressive, tactical chess prodigy, presumably lacking in the strategic niceties of the game at its highest levels. I recall John Nye, my Filipino-American colleague (and chess player) coming into my office to discuss the astonishing fact that So had risen to number 9 in the world rankings. I suggested that a bit of regression to the mean was in order, and So would not be returning to the top ten anytime soon.
Since that time, So has won four top tournaments in a row, besting Magnus Carlsen, and has had a 56-game non-losing streak, against very high caliber players, and recently he was selected best chess player of the last year. Arguably he is the second best player in the world, and the one most likely to dethrone Carlsen from the world championship.
A turning point for So came in 2014 when he left university and moved to Minnetonka, Minnnesota to live with his adoptive parents, Lotis Key and Renato Kabigting, Key being a former Filipino movie star and now Vice President of the Minnesota Christian Writers Guild. She serves as So’s manager and insisted he not check on-line NBA scores when doing his chess training. Later, So turned away from the internet more fundamentally to focus on chess.
One Filipino international master remarked: “He cannot afford decent training given by well known GM-coaches and has to rely on his pure talent.” Last month he brought on Vladimir Tukmakov as a coach, but he’s had less formal training than any top player in recent memory.
So hopes to learn how to drive a car, and he enjoys playing in Las Vegas: “I like Las Vegas,” So said, laughing. “People are usually drunk. Makes them easier to beat. Just keep drinking.”
So’s style now has evolved to the point where…he doesn’t seem to have a style. He is renowned for his calm and he simply limits the number of mistakes. At the top, top levels, a player without a real style is a player who is hard to train for and hard to beat.
So is religious, and he is considered mild-mannered and humble. The story of Wesley So is not over. Yet Wesley So is an American, and an American hero, and he has received virtually no mainstream media attention.
What accounts for the cost disease?
The basic post is too long, but some of it is interesting and here is the best part:
The pattern of Cost Disease seems to be related to things that inextricably require the unsubstitutable labour and attention not just of human beings but of human beings somehow comparable to the buyer. (Americans, for the US focus of most of this discussion.) Education not only requires teachers who are part of the same cultural milieu as their students, but it requires the attention of the students themselves, and attention is inherently expensive. As the only thing that can be expensive in the final Strong Heaven, attention predictably gets more expensive in a culture that moves more and more toward general post-scarcity. Health care similarly requires local human involvement.
That is from Ansuz, via Matthew Fairbank. And here is Scott Alexander’s survey follow-up post.