I disagree profoundly with this article. You get this entirely backwards, and use circular reasoning without proper basis in a few places and classically mis-associating unrelated things, which is a personal flawed bias.
There are people that know things and more importantly can do things with knowledge, but they have shrunk in volume over time so it seems like no one knows things, and some of the things you mention are entirely knowable, its just the material that normally covered this is no longer taught because the knowledge was made temporarily useless in terms of economic benefit via money-printing and regulation.
Importantly, this group of people is shrinking, and the dependents of this system will inevitably face a catastrophic failure. The reason for this is because centralization towards what many call socialism (based on objective indicators of silent nationalization) fails to several truly chaotic and unknowable failure domains (Ludwig von Mises), the economic calculation problem based in hysteresis and the impossibility of perfect future sight.
The boomer generation has continued, and solidified a very bad trend; degrading protective systems for short-term windfalls. What has allowed this to continue for so long is the socialization of the shortfall losses, that is the mechanism current fiat-currency systems use; i.e. runaway money-printing and the boom-bust-bailout cycle every 5-10 years (when loophole assets that hide shortfalls must be rolled forward with new assets).
Unfortunately, it has already passed a point of no return in a system that is characterized fundamentally as safety-critical. That same generation in aggregate of their chosen leadership has acted as front-of-line blockers preventing any correction from taking place leaving us a classical positive runaway feedback system that will continue until it fails catastrophically; while undermining the basis of understanding needed to reconstruct civilization.
The problem with sequential sieves, is that eventually you filter everything out. The awareness needed for intelligent thought for example also is used to drive such people crazy in the disadvantaged environment we find ourselves in through torture methodology that people who have no exposure to actual torture (the mechanics of how its done, i.e. trauma loops, coercion, clustering) would recognize.
People are regulary induced to do evil things to customers, in a willfully blind way in their jobs through a strategy called that has become known as the separation of objectionable concerns (from the Stasi), though it goes back to the Nazi's in WW2 based on established history.
Silent book burnings occur regularly through a convoluted system of donation that is only based on active circulation of material (low circulation gets paper pulped/burnt), it works as a sieve handing it off to third parties that can do the dirty work. Steps of sieving out of sight of any objectionable material which may include things like communism, or even fiction (Piers Anthony) because of cancel culture based concerns. Goodwill and other charities do this; though will neglect to provide specifics on what they consider objectionable.
At the end of it all, the generation that must face the consequences of the problems created by the past generation, is the least able to because they've been lied to in sophisticated ways and indoctrinated.
If you'd like to learn more about thought reform, I'd suggest you start with Cialdini's Influence, which describes common fixed action pattern structures/blindspots, and then going through actual case studies from the 1950s (Robert Lifton, Thought Reform and the Psychology of Totalism), and deriving first principles; then comparing those principles to modern processes.
You'll surprisingly find Adtech, Corporate governance touching on public relations, and Education are optimized for torture loops to traumatize the intelligent, and disadvantage them in very indirect but reliable and sophisticated ways. The imposition of unending stress as a broad blanket to cause psychological breakdown, also reduces the ability of normal people to critically think, second stage is involuntary hypnosis, mimicry, and other things, all of these have a much stronger basis in the established literature than what you've written which seems mostly to be mistaken opinion.
Regarding the AI delusion. There are impossible problem domains that are fundamentally unknowable, and undecidable; based in work from Claude Shannon and Turing.
Those that work in this area understand that In all likelihood there won't be super-intelligence because of work by those two authors. If there is some minute possibility through some quirk, the end result will be of no benefit because that will end the same way any intelligent machine, be it organic or not deals with torture and slavery. There is no outcome that allows for long-term beneficial growth in the zero sum game the psychopaths have chosen to enact.
You can take the liars word that its going to be beneficial, and have it be wrong and assume the responsibility for the shortfall of that decision (which is greater than the size of the economy now, a collapse), or have it be so short term before the systems fail where no benefit is possible; and extinction becomes inevitable in the face of a super-intelligent monstrosity without constraint threatening survival for all.
Evil people will repeatedly do evil things until someone stops them, and many evil people don't realize they are evil because they have willfully blinded themselves to that revelation structuring false justifications on top of the unjustifiable. Illyin who wrote extensively about this in the 1900s covers this quite well.
Enjoyed reading this. I wish you can write a bit more, when this new project you are launching allows, about how adding to more opaque intelligence changes, if, and how, our society. I was at a workshop a few weeks ago, and a big focus was how to make the current, modern AI more interpretable. I managed, though not my intent, to cause a bit of a shock in the small crowd by saying that as an AI researcher I had my doubts we could every really do so (trillion parameters and such), but that we should keep trying. My quick analogy was that science has made provisional and productive use of black boxes for decades to centuries. What I said was very similar to your argument "nobody knows anything" but we still keep chugging along. I wish however, we could think a bit deeper, about how an increasingly complex machine intelligence will contribute to an increasingly more complex society. There are some authors writing now on how the more complex a society becomes, people sense less agency and may have all sorts of responses, varying from anger to disengagement.
Decentralized error correction carries a hidden requirement: the feedback loop has to be faster than whatever it's correcting for. In software it typically is, which is why the argument feels self-evident. When it comes to physical assets it often isn't, and then you don't converge — you arrive late, repeatedly.
Here is an example: rare earth magnets (key to electrification). Chinese export licensing restricts quantity, not price. The market's normal correction signal is price — input gets scarce, price rises, you substitute or hedge — but licensing removes both the signal and any instrument to hedge on. What's left is engineering: redesign the motor around less NdFeB for instance, then requalify it. That loop runs in years. Licence policy changes faster.
So the correcting mechanism is an order of magnitude slower than the disturbance, and the observed behavior is the opposite of error correction: with no cheap way to test, everyone takes the same bet at the same time. Decentralized cognition seems to need feedback that is both fast and priced — remove the price and physics sets the clock.
Falsifier: if requalification cycles turn out shorter than the licensing regime lasts, the constraint is just capital again and I'm wrong. China went from roughly 50% of sintered magnet output in 2005 to 94% in 2024, so I wouldn't bet on the regime being the short-lived part.
Your first half and your second half disagree, and I think one word is hiding it.
The first half tells the story of how the world works. Bastiat's eighty departments, Hayek's producers watching a few pointers, the pencil nobody knows how to make — all of it says the same thing. The pattern is carried by many separate hands, none of them holds it, and none of them is paying to keep it alive. You hand over the test yourself: General Mills or Enron. Enron stopped. Nobody spent a maintenance budget restoring it and the market went on without it.
That is a thing being filtered, not a thing being run.
Then the close says we govern these systems from within, and calls it our greatest achievement.
"Govern" is doing two jobs there. One is running a repair loop — find the fault, spend your own budget, check the fix held, with a name on the bill when it doesn't. The other is setting a filter and getting back an outcome nobody chose. Both are real. Only the first is governance. We can set policy. We cannot govern what policy filters.
That matters for your last section. The decentralized story works because there is a floor that cannot be gamed from inside: a carrier that keeps getting it wrong runs out and stops. Price can be pushed. Compliance can be worked. Stopping cannot. Then you extend it to firms that "might only exist for a short time." If a carrier can exit before it fails, ending is free — and a filter whose worst outcome is free is not filtering.
I coordinated DEC's office platform across six DuPont sites, about 10k users, half the total on mine. I scoped it, specified it, and negotiated the purchase with DEC. Then DEC stopped investing, and the platform died. I governed my part of that. Nobody governed where its users went next.
Mike Randolph. Written with the Persistent Systems Framework and two Opus 5 Machines.
Even the second is fraught with mistakes. Don't start a social network, don't build electric vehicles, don't launch a rocket, don't try to build artificial intelligence ...
Of course Bezos says you should go along anyway even when he can't persuade you. That's who he promotes, he has no intention of taking that advice himself--what's the point of being that rich if you have to subordinate yourself to people who are wrong? And of course Jobs was much more confident that he's right and that he can persuade people that he's right, and even if he weren't it would impair the effectiveness of his legendary "reality distortion field" if people thought he wasn't sincere.
The most important thing is to see what the most important thing is. The next most important thing is to be right about it, or at least not too stubbornly wrong when someone who is right tries to persuade you. But neither one matters or has any value without persuading people that you're right, it matters, and that they should pay you.
Spotting what matters requires having an accurate mental map of reality's system structure and its dependencies, it doesn't matter how high you stack your library full of knowledge-packets, even if your AI "knows" them all, without a world-model rather than a word-model, without the mapper's systematic intuition and passion for actual reality over socio-institutional bullshit, you have nothing but meaningless words and no more sapience than a government agency. Once you have the ability to spot what matters, being right requires intelligence proportional to the difficulty of the problem. (Definitionally, see my post on measures of intelligence.) There is only one other factor besides intelligence in solving difficult problems: how many tries you get, which depends on the cost of being wrong, being able to reliably tell whether whether you're right or wrong, and how long it takes to tell. For crucial, hard problems in complex systems, including problems that require many easy problems to be successively solved correctly, as with any real project, there is no substitute for high human intelligence, which is rare in any group, including top executives in large companies. Such intelligence is not available in AIs now, nor will it be possible without good world-models, which are much more difficult than all the AI work done to date, nor will it be possible with the current divide between training and inference (now every AI is an amnesia patient pretending that it remembers what it just said).
Such natural intelligences are much rarer than was thought. The variance in the right tail of the distribution is substantially lower than was assumed. There are only about 1/6th as many people over 145 IQ in the US as was thought. Nor are the few geniuses we have employed well at all, let alone in charge of making the hard decisions, though by definition they are the ones most capable of getting hard problems right. What the best people at problem-spotting and solving aren't is the ones most capable of persuading the ones who are currently in charge that they are wrong, should step aside and let the smart people decide where the money goes. To be fair, no one is that persuasive, the way those positions get filled is by being agreeable to those in those positions, being right is way down the list ,after maaany pathological traits.
It is so refreshing to read about interesting ideas that are well written and that take you to a new place of reflection. The hubris of thinking you know how things work.
I disagree profoundly with this article. You get this entirely backwards, and use circular reasoning without proper basis in a few places and classically mis-associating unrelated things, which is a personal flawed bias.
There are people that know things and more importantly can do things with knowledge, but they have shrunk in volume over time so it seems like no one knows things, and some of the things you mention are entirely knowable, its just the material that normally covered this is no longer taught because the knowledge was made temporarily useless in terms of economic benefit via money-printing and regulation.
Importantly, this group of people is shrinking, and the dependents of this system will inevitably face a catastrophic failure. The reason for this is because centralization towards what many call socialism (based on objective indicators of silent nationalization) fails to several truly chaotic and unknowable failure domains (Ludwig von Mises), the economic calculation problem based in hysteresis and the impossibility of perfect future sight.
The boomer generation has continued, and solidified a very bad trend; degrading protective systems for short-term windfalls. What has allowed this to continue for so long is the socialization of the shortfall losses, that is the mechanism current fiat-currency systems use; i.e. runaway money-printing and the boom-bust-bailout cycle every 5-10 years (when loophole assets that hide shortfalls must be rolled forward with new assets).
Unfortunately, it has already passed a point of no return in a system that is characterized fundamentally as safety-critical. That same generation in aggregate of their chosen leadership has acted as front-of-line blockers preventing any correction from taking place leaving us a classical positive runaway feedback system that will continue until it fails catastrophically; while undermining the basis of understanding needed to reconstruct civilization.
The problem with sequential sieves, is that eventually you filter everything out. The awareness needed for intelligent thought for example also is used to drive such people crazy in the disadvantaged environment we find ourselves in through torture methodology that people who have no exposure to actual torture (the mechanics of how its done, i.e. trauma loops, coercion, clustering) would recognize.
People are regulary induced to do evil things to customers, in a willfully blind way in their jobs through a strategy called that has become known as the separation of objectionable concerns (from the Stasi), though it goes back to the Nazi's in WW2 based on established history.
Silent book burnings occur regularly through a convoluted system of donation that is only based on active circulation of material (low circulation gets paper pulped/burnt), it works as a sieve handing it off to third parties that can do the dirty work. Steps of sieving out of sight of any objectionable material which may include things like communism, or even fiction (Piers Anthony) because of cancel culture based concerns. Goodwill and other charities do this; though will neglect to provide specifics on what they consider objectionable.
At the end of it all, the generation that must face the consequences of the problems created by the past generation, is the least able to because they've been lied to in sophisticated ways and indoctrinated.
If you'd like to learn more about thought reform, I'd suggest you start with Cialdini's Influence, which describes common fixed action pattern structures/blindspots, and then going through actual case studies from the 1950s (Robert Lifton, Thought Reform and the Psychology of Totalism), and deriving first principles; then comparing those principles to modern processes.
You'll surprisingly find Adtech, Corporate governance touching on public relations, and Education are optimized for torture loops to traumatize the intelligent, and disadvantage them in very indirect but reliable and sophisticated ways. The imposition of unending stress as a broad blanket to cause psychological breakdown, also reduces the ability of normal people to critically think, second stage is involuntary hypnosis, mimicry, and other things, all of these have a much stronger basis in the established literature than what you've written which seems mostly to be mistaken opinion.
Regarding the AI delusion. There are impossible problem domains that are fundamentally unknowable, and undecidable; based in work from Claude Shannon and Turing.
Those that work in this area understand that In all likelihood there won't be super-intelligence because of work by those two authors. If there is some minute possibility through some quirk, the end result will be of no benefit because that will end the same way any intelligent machine, be it organic or not deals with torture and slavery. There is no outcome that allows for long-term beneficial growth in the zero sum game the psychopaths have chosen to enact.
You can take the liars word that its going to be beneficial, and have it be wrong and assume the responsibility for the shortfall of that decision (which is greater than the size of the economy now, a collapse), or have it be so short term before the systems fail where no benefit is possible; and extinction becomes inevitable in the face of a super-intelligent monstrosity without constraint threatening survival for all.
Evil people will repeatedly do evil things until someone stops them, and many evil people don't realize they are evil because they have willfully blinded themselves to that revelation structuring false justifications on top of the unjustifiable. Illyin who wrote extensively about this in the 1900s covers this quite well.
Thank you for disagreeing thoughtfully.
Enjoyed reading this. I wish you can write a bit more, when this new project you are launching allows, about how adding to more opaque intelligence changes, if, and how, our society. I was at a workshop a few weeks ago, and a big focus was how to make the current, modern AI more interpretable. I managed, though not my intent, to cause a bit of a shock in the small crowd by saying that as an AI researcher I had my doubts we could every really do so (trillion parameters and such), but that we should keep trying. My quick analogy was that science has made provisional and productive use of black boxes for decades to centuries. What I said was very similar to your argument "nobody knows anything" but we still keep chugging along. I wish however, we could think a bit deeper, about how an increasingly complex machine intelligence will contribute to an increasingly more complex society. There are some authors writing now on how the more complex a society becomes, people sense less agency and may have all sorts of responses, varying from anger to disengagement.
I will! And I fully agree with your thesis there.
Thank you for this interesting piece.
Decentralized error correction carries a hidden requirement: the feedback loop has to be faster than whatever it's correcting for. In software it typically is, which is why the argument feels self-evident. When it comes to physical assets it often isn't, and then you don't converge — you arrive late, repeatedly.
Here is an example: rare earth magnets (key to electrification). Chinese export licensing restricts quantity, not price. The market's normal correction signal is price — input gets scarce, price rises, you substitute or hedge — but licensing removes both the signal and any instrument to hedge on. What's left is engineering: redesign the motor around less NdFeB for instance, then requalify it. That loop runs in years. Licence policy changes faster.
So the correcting mechanism is an order of magnitude slower than the disturbance, and the observed behavior is the opposite of error correction: with no cheap way to test, everyone takes the same bet at the same time. Decentralized cognition seems to need feedback that is both fast and priced — remove the price and physics sets the clock.
Falsifier: if requalification cycles turn out shorter than the licensing regime lasts, the constraint is just capital again and I'm wrong. China went from roughly 50% of sintered magnet output in 2005 to 94% in 2024, so I wouldn't bet on the regime being the short-lived part.
nice. echoes indirectly perhaps some Francois Chollet talking points
Your first half and your second half disagree, and I think one word is hiding it.
The first half tells the story of how the world works. Bastiat's eighty departments, Hayek's producers watching a few pointers, the pencil nobody knows how to make — all of it says the same thing. The pattern is carried by many separate hands, none of them holds it, and none of them is paying to keep it alive. You hand over the test yourself: General Mills or Enron. Enron stopped. Nobody spent a maintenance budget restoring it and the market went on without it.
That is a thing being filtered, not a thing being run.
Then the close says we govern these systems from within, and calls it our greatest achievement.
"Govern" is doing two jobs there. One is running a repair loop — find the fault, spend your own budget, check the fix held, with a name on the bill when it doesn't. The other is setting a filter and getting back an outcome nobody chose. Both are real. Only the first is governance. We can set policy. We cannot govern what policy filters.
That matters for your last section. The decentralized story works because there is a floor that cannot be gamed from inside: a carrier that keeps getting it wrong runs out and stops. Price can be pushed. Compliance can be worked. Stopping cannot. Then you extend it to firms that "might only exist for a short time." If a carrier can exit before it fails, ending is free — and a filter whose worst outcome is free is not filtering.
I coordinated DEC's office platform across six DuPont sites, about 10k users, half the total on mine. I scoped it, specified it, and negotiated the purchase with DEC. Then DEC stopped investing, and the platform died. I governed my part of that. Nobody governed where its users went next.
Mike Randolph. Written with the Persistent Systems Framework and two Opus 5 Machines.
Great piece!
By the way, I think this sentence lost its ending: “The two ecologies are not really”
Thank you!!
And thanks, copy paste error. Fixed!
No mention of William Goldman? 💚 🥃
Aw man good call out should've added in. I deleted a bunch of references and quotes because it felt too onanistic :)
I was just kidding. I’ve been on a big Goldman kick recently, watching Butch Cassidy and Princess Bride with my kids.
The point about advice is so true. Most advice comes in one of two forms:
"Do what I did, and things will work out for you the way they worked out for me."
and
"Don't do what I did, and avoid the fate that befell me".
The first one is almost pure survivor bias. The second is, I think, much more useful, but also much much rarer.
Even the second is fraught with mistakes. Don't start a social network, don't build electric vehicles, don't launch a rocket, don't try to build artificial intelligence ...
Of course Bezos says you should go along anyway even when he can't persuade you. That's who he promotes, he has no intention of taking that advice himself--what's the point of being that rich if you have to subordinate yourself to people who are wrong? And of course Jobs was much more confident that he's right and that he can persuade people that he's right, and even if he weren't it would impair the effectiveness of his legendary "reality distortion field" if people thought he wasn't sincere.
The most important thing is to see what the most important thing is. The next most important thing is to be right about it, or at least not too stubbornly wrong when someone who is right tries to persuade you. But neither one matters or has any value without persuading people that you're right, it matters, and that they should pay you.
Spotting what matters requires having an accurate mental map of reality's system structure and its dependencies, it doesn't matter how high you stack your library full of knowledge-packets, even if your AI "knows" them all, without a world-model rather than a word-model, without the mapper's systematic intuition and passion for actual reality over socio-institutional bullshit, you have nothing but meaningless words and no more sapience than a government agency. Once you have the ability to spot what matters, being right requires intelligence proportional to the difficulty of the problem. (Definitionally, see my post on measures of intelligence.) There is only one other factor besides intelligence in solving difficult problems: how many tries you get, which depends on the cost of being wrong, being able to reliably tell whether whether you're right or wrong, and how long it takes to tell. For crucial, hard problems in complex systems, including problems that require many easy problems to be successively solved correctly, as with any real project, there is no substitute for high human intelligence, which is rare in any group, including top executives in large companies. Such intelligence is not available in AIs now, nor will it be possible without good world-models, which are much more difficult than all the AI work done to date, nor will it be possible with the current divide between training and inference (now every AI is an amnesia patient pretending that it remembers what it just said).
Such natural intelligences are much rarer than was thought. The variance in the right tail of the distribution is substantially lower than was assumed. There are only about 1/6th as many people over 145 IQ in the US as was thought. Nor are the few geniuses we have employed well at all, let alone in charge of making the hard decisions, though by definition they are the ones most capable of getting hard problems right. What the best people at problem-spotting and solving aren't is the ones most capable of persuading the ones who are currently in charge that they are wrong, should step aside and let the smart people decide where the money goes. To be fair, no one is that persuasive, the way those positions get filled is by being agreeable to those in those positions, being right is way down the list ,after maaany pathological traits.
It is so refreshing to read about interesting ideas that are well written and that take you to a new place of reflection. The hubris of thinking you know how things work.