Civilization is kept afloat by a massive, decentralized body of often unseen knowledge.
This dark mass is made up of innumerable pieces of know-how accumulated by people mostly stumbling around, observing each other and the way things work. It’s what’s lodged in the head of the East German handyman that knows whom to bribe (and how) to get West German spare parts. It’s the idiosyncratic thought patterns and norms picked up by the students of Gerty and Carl Cori, who won the Nobel Prize in 1947, six of which went on to win the prize in turn. It’s your two-year-old learning the local language.
Since this body of knowledge is hard to quantify, and often even hard to spot, we tend to not think about it as deeply as it deserves. Talking about learning, and how to improve it, we often limit our discussions to the legible subset of knowledge transmission channels – that is, schools and universities. But as important as those institutions are, the education system plays only a minor part in knowledge reproduction.
As Lester Thurow argued in Education and economic equality: “Most actual job skills are acquired informally through on-the-job training after a worker finds an entry job and a position on the associated promotional ladder.”
This informal process, whereby job skills get picked up at work, social skills in friend groups, and language at home, is where virtually all knowhow is passed on. Education is the visible white tip kept up by this submerged iceberg; it is a formal order kept afloat by an informal one. Education cannot function without decentralized pathways of knowledge transfer, yet it often ignores them, and competes with them for resources.
As James Scott has observed:
The more highly planned, regulated, and formal a social or economic order is, the more likely it is to be parasitic on informal processes that the formal scheme does not recognize and without which it could not continue to exist, informal processes that the formal order cannot alone create and maintain.
In this essay, I’m going to look at the body of the iceberg – the decentralized processes that create and spread knowledge. To distinguish it from the education system, I’m going to refer to it as the learning system. Can we leverage it to more efficiently spread useful knowledge?
Decentralized Knowledge Reproduction
Whatever knowledge is spreading through society without a top-down mandate, is spreading through the learning system.
Exactly where to draw the line between the education system and the learning system is a bit arbitrary. On one side we have people involuntarily placed in classrooms, on the other people are disappearing down rabbit holes on Wikipedia. But there are many shades in between. One way to demarcate the line is by making a distinction between learning situations that, in Ivan Illich’s terminology, are convivial and those that are manipulative:
[Manipulative] institutions tend to be highly complex and costly production processes in which much of the elaboration and expense is concerned with convincing consumers that they cannot live without the product of the treatment offered by the institution. [Convivial] institutions tend to be networks which facilitate client-initiated communication or cooperation.
The education system is an attempt to manipulate the spread of knowledge through mandatory attendance. Attendance can be enforced by law, or, more subtly, by manipulating living conditions so that it becomes very difficult to live a decent life if one chooses to be an autodidact – that is, one who learns only through the learning system.
The learning system, on the other hand, is convivial. People go looking for it: they search for YouTube lectures, book clubs, mentors, and dynamic workplaces. It fills a need in them. It is self-directed.
The learning system is often highly efficient. Over the past five years, for example, millions of people, almost exclusively outside of the education system, have gained a conceptual understanding of how cryptographically secured tokens can unlock new software designs; and tens of thousands have acquired the skills required to implement these designs. This is not a trivial feat of knowledge spread.
Yet, uncontrolled, decentralized knowledge transfer will not necessarily lead to the spread of adaptive knowledge. There are many examples from history when important knowledge has been forgotten – as when the Polar Inuit of northwest Greenland lost the ability to make kayaks – or when maladaptive norms and practices have spread. We repeatedly see people who refused to learn things that could have saved their lives.
In 1841, a British parliamentary commission reported that they had found citizens who had never heard of a city called London. Meeting a group of working-class boys in the street, a commission member had asked if they knew who the Queen of England was.
“Yes, sir,” said the boys, “her name is Prince Albert.”
Literacy, which was common among the upper classes and spreading in the middle class, had not yet penetrated most of the working class.
This lack of knowledge can be viewed as a learning system failure (which can be thought of as analogous to a market failure, where a decentralized market does not allocate goods and services in a Pareto efficient way). And mass education can, in some regards, be seen as a reaction to these failures of the learning system.
The introduction of compulsory education, in Britain and elsewhere, was an attempt to correct for perceived inefficiencies in the learning system. Time that people had previously spent in the learning system was appropriated for education, where knowledge transmission could be centrally controlled. This gave curriculum designers a channel through which they could transmit knowledge deemed essential.
Or phrased another way: mandatory education was an intervention in the decentralized learning system.
How did the intervention affect the underlying system?
Interventions in Complex Systems
Interventions in complex systems have unforeseen consequences. Apply pesticide to stop budworm from killing off your spruce, and you will also kill off the budworm’s natural enemies, making future pest outbreaks more severe. Impose rent control to keep housing affordable and construction will slow, making it harder for people to move to cities with higher economic productivity – raising unemployment and economic inequality while lowering innovation and the rate of childbirth.
The problem with these types of interventions is that they limit the underlying system’s ability to self-organize. By shifting control from the system to a central authority tasked with managing it, the intervention reduces the system’s ability to adjust itself to changes in its environment. It is no longer in the system’s power to freely adapt.
The capacity to change in reaction to new circumstances is crucial for adaptability. Limiting a system’s ability to self-organize and adapt, induces costs that often are hard to see and connect to the intervention that stymied that ability. Rural deaths of despair seem unconnected from urban rent control. Pesticide seems like an unlikely cause of pest outbreaks.
To avoid these types of problems, you need another approach. You need interventions that leverage the system’s capacity to self-organize, rather than work against it. An intervention, if aimed at unblocking the system, can strengthen the ability of the system to shoulder its own burdens.
In forestry, this means cultivating ecosystems that in themselves are resilient enough to keep the budworm population from overwhelming the spruce.
In economics, it means some flavor of Keynesianism: instead of a Soviet-style command economy, you correct for market failures by leveraging the market’s own strengths. You break up monopolies and use free-market competition to hem in the market’s tendency to create winner-take-all effects. You use price-mechanisms, such as carbon rights, to force the market to internalize its externalities, that is: you incentivize the system to self-organize and overcome its limitations.
If we apply this insight to knowledge reproduction, what would it mean? The learning system is a self-organizing entity. It can change its structure to adapt to new circumstances: creating formal apprenticeships during the Renaissance, conjuring universities, building online communities. But when we have tried to promote learning during the modern era, we have often done so at the expense of this adaptability, and many of the structures developed by the learning system, such as apprenticeships, have withered.
Can we instead design interventions that strengthen the learning system, so that it can better overcome its own failures? I am not sure, but it strikes me as a promising question. I will sketch a few possibilities.
Enabling the Learning System
It must not start with the question, ‘What should someone learn?’ but with the question, ‘What kinds of things and people might learners want to be in contact with in order to learn?’ – Ivan Illich
If we want to enable the learning system and allow it to shoulder some of the burdens we’ve placed on education, we must first ask ourselves why the natural correction mechanisms are failing. Why can’t the learning system reproduce the knowledge needed to sustain civilization? It used to be our sole means of knowledge transmission; now it no longer seems sufficient. What is holding it back?
Two things come to mind, changes that might have rendered the learning system insufficient in the modern world. Firstly, the character of knowledge has changed. Whereas knowledge in premodern societies was easy to observe and immediately rewarding – like hunting or mending clothes – knowledge in the modern world is abstract, and the gratification of learning is delayed. This is the common criticism of self-directed learning. What is fun and intrinsically motivating is not necessarily what is useful; the incentives are off. So you misallocate your time.
Secondly, modernity has been a centrifuge. Rapid progress has separated everything – most importantly, home and work, children and adults – creating increasing specialization. Elderly in homes, working-age adults in offices, children sorted in age-graded classes. This segregation makes it hard for knowledge to pass from one group of people to another; children cannot, as in premodern societies, simply learn by working alongside adults; they can not get the widened perspective you get by living with the elderly.
How can these obstacles to learning be overcome?
1. Weak incentives. The first problem is a problem of incentives. Children, not having access to contexts where knowledge work is produced, for example, might not understand the value of developing deep literacy. Especially if they grow up in households where they don’t observe their parents reading, they might choose to learn less valuable skills, such as farming mushrooms in Minecraft, which could render them less employable in the modern job market.
I’m not sure this problem is as big as people would make it out to be. But assume it is. How can we incentivize the unmotivated to pursue things that are not intrinsically motivating?
The most straightforward way to incentivize is to simply reward the behavior that you want to promote. Rather than mandating literacy, give everyone that can pass a high school literacy test 20k dollars or so. Kids who are capable of teaching themselves can just collect the check, and the rest can ask their parents for lessons, or sign agreements with teachers, splitting the proceeds, and so finance their education and collect the reward. This is basically the learn and earn model popular in web3: if people aren’t intrinsically motivated to pay attention to what you value, you just pay them for their attention.
(Implemented on its own, with no further support, self-directed learning and bounties would of course lead to abysmal outcomes for certain groups: I will return to that. )
But for all the limitations of this simple model, it has the upside that it does not disrupt the learning system. Instead of replacing it, which is costly and conflict-ridden, incentives leverage the learning system. It creates a reward function and lets the system self-organize to solve the problem. Different people have different needs and capabilities, and if incentivized correctly the learning system can adapt to serve all these diverse needs, through a rich ecosystem of different learning opportunities. People will search around for the tools and mentors that work for them. Some will enjoy 3blue1brown; others will join study circles or go at it alone.
2. The separated society. Age segregation is a network problem – the nodes in the networks are not connected in a way that allows for efficient knowledge transfer. Children (nodes with limited knowledge) have been denied access to experienced nodes, and reduced to studying them from afar, mediated through books and lectures. And a lot of knowledge cannot pass through these types of low bandwidth connections – reality is too complex and nonlinear to pass through a linear string of words – leaving kids deprived.
For learning to happen spontaneously and effectively you need access. Christopher Alexander – writing in 1977 – asserted that the purpose of future educational institutions “must be to facilitate access for the learner: to allow him to look into the windows of the control room or the parliament, if he cannot get in the door. Moreover, such new institutions should be channels to which the learner would have access without credentials or pedigree, public spaces in which peers and elders outside his immediate horizon now become available...”
The problem with this kind of access is that there is a fundamental conflict at play: the conflict between granting novices access to experts and allowing experts to do productive work. Too many novices at a workplace and their demands for instruction overwhelm the attention of experts, slowing production to a crawl. This doesn’t mean we have to exclude novices as completely as we do today. Especially gifted children could enter productive environments much earlier than current legal codes permit, and gain from it. But you cannot just naively give novices access.
What we can do, however, is build better infrastructure. By leveraging communication technologies, we can grant more people access to valuable networks without overwhelming experts. We can build new types of architecture which allow experts and novices to coexist at scale. I have made a first stab at how that could work in a previous essay: Apprenticeship Online.
An interesting example of what it might look like is a Decentralized Autonomous Organization (DAO). These blockchain-enabled organizations are generally permissionless and have an ethos of working in public: anyone can jump into a Discord, participate in the dialogue, and look for ways to contribute. This week, in early November, as I’m writing this, I was lucky enough to see Liminal Warmth tweet about a new DAO – one that would try to buy a copy of the American constitution. For a person like me, who has a relatively limited understanding of DAOs, and memeing, it was instructive to hang out in the channels where communication strategies were developed, and various legal issues got sorted out. I can’t imagine I would have learned as much from a university course about DAOs, if such a thing even exists. In 72 hours, the group had gone from about a hundred users on Discord to a money swarm of +$47M. They also, ironically, managed to create so much attention around the auction that the price rose beyond them: they came in second place.
If we decentralize the responsibility to learn away from schools, we might need to create incentives to drive the spread of useful knowledge, and we definitely need to invest in infrastructure that makes it easier for novices to access the environments we want them to master. Otherwise, we will see unpalatable learning system failures, where large groups are unable to obtain the knowledge they need to lead dignified lives.
We probably also need general economic support for the young. The young and the unskilled tend to be financially vulnerable and will therefore underspend on their learning compared to the societal optimum. Knowledge has positive externalities; it should therefore, at least partially, be financed by the collective. This is the basic idea behind state-sponsored education, and remains true even if the resources are decentralized into the learning system.
People who come from homes without good access to skilled role models, and those that have learning deficits, will need extra support. Decentralizing the responsibility to learn would otherwise risk hurting these groups. Decentralization leads to increased variance. We want to have a high-pass filter, that allows us to keep the upside of that variance – the gifted children that can grow faster when left to their own devices, the kids with obsessive passions that can specialize early – while limiting the downside.
Regrowth from the Edges In
To sum up:
There exists something we can call the decentralized learning system.
Education is an intervention to correct for its (perceived or real) failures.
Heavy-handed interventions tend to undermine underlying systems, so they should be used as a last resort.
There might be some way to instead unleash the learning system, similar to how economics has unleashed the market.
That might be cool.
Looking at the problem of knowledge reproduction from a systems perspective allows us to find new solutions. The possibilities sketched here (granting access, incentivizing, and giving economic support) are only the most naively obvious, probably not the global optima. The design space is vast.
Instead of the lock-step of compulsory schooling in a fixed place, work in piecemeal ways to decentralize the process of learning and enrich it through contact with many places and people all over the city: workshops, teachers at home or walking through the city, professionals willing to take on the young as helpers, older children teaching younger children, museums, youth groups traveling, scholarly seminars, industrial workshops, old people, and so on. Conceive of all these situations as forming the backbone of the learning process; survey all these situations, describe them, and publish them as the city’s “curriculum”; then let students, children, their families and neighborhoods weave together for themselves the situations that comprise their “school” paying as they go with standard vouchers, raised by community tax. Build new educational facilities in a way which extends and enriches this network.
Exploring the full range of possibilities can probably most effectively be done outside of existing institutions, from the bottom up. It will need to be a messy, organic process.
Piece by piece, we can learn how to compensate for the shortcomings of decentralized and self-directed learning without relying too heavily on centralized control. This will allow for a richer, more dynamic ecosystem of learning services.
Acknowledgments
This essay has benefitted from several rounds of feedback, primarily by Gunnar Zarncke and Justis Mills. Don’t blame the outcome on them, though. It was a lot worse when they started.
The Learning System
Link post
Civilization is kept afloat by a massive, decentralized body of often unseen knowledge.
This dark mass is made up of innumerable pieces of know-how accumulated by people mostly stumbling around, observing each other and the way things work. It’s what’s lodged in the head of the East German handyman that knows whom to bribe (and how) to get West German spare parts. It’s the idiosyncratic thought patterns and norms picked up by the students of Gerty and Carl Cori, who won the Nobel Prize in 1947, six of which went on to win the prize in turn. It’s your two-year-old learning the local language.
Since this body of knowledge is hard to quantify, and often even hard to spot, we tend to not think about it as deeply as it deserves. Talking about learning, and how to improve it, we often limit our discussions to the legible subset of knowledge transmission channels – that is, schools and universities. But as important as those institutions are, the education system plays only a minor part in knowledge reproduction.
As Lester Thurow argued in Education and economic equality: “Most actual job skills are acquired informally through on-the-job training after a worker finds an entry job and a position on the associated promotional ladder.”
This informal process, whereby job skills get picked up at work, social skills in friend groups, and language at home, is where virtually all knowhow is passed on. Education is the visible white tip kept up by this submerged iceberg; it is a formal order kept afloat by an informal one. Education cannot function without decentralized pathways of knowledge transfer, yet it often ignores them, and competes with them for resources.
As James Scott has observed:
In this essay, I’m going to look at the body of the iceberg – the decentralized processes that create and spread knowledge. To distinguish it from the education system, I’m going to refer to it as the learning system. Can we leverage it to more efficiently spread useful knowledge?
Decentralized Knowledge Reproduction
Whatever knowledge is spreading through society without a top-down mandate, is spreading through the learning system.
Exactly where to draw the line between the education system and the learning system is a bit arbitrary. On one side we have people involuntarily placed in classrooms, on the other people are disappearing down rabbit holes on Wikipedia. But there are many shades in between. One way to demarcate the line is by making a distinction between learning situations that, in Ivan Illich’s terminology, are convivial and those that are manipulative:
The education system is an attempt to manipulate the spread of knowledge through mandatory attendance. Attendance can be enforced by law, or, more subtly, by manipulating living conditions so that it becomes very difficult to live a decent life if one chooses to be an autodidact – that is, one who learns only through the learning system.
The learning system, on the other hand, is convivial. People go looking for it: they search for YouTube lectures, book clubs, mentors, and dynamic workplaces. It fills a need in them. It is self-directed.
The learning system is often highly efficient. Over the past five years, for example, millions of people, almost exclusively outside of the education system, have gained a conceptual understanding of how cryptographically secured tokens can unlock new software designs; and tens of thousands have acquired the skills required to implement these designs. This is not a trivial feat of knowledge spread.
Yet, uncontrolled, decentralized knowledge transfer will not necessarily lead to the spread of adaptive knowledge. There are many examples from history when important knowledge has been forgotten – as when the Polar Inuit of northwest Greenland lost the ability to make kayaks – or when maladaptive norms and practices have spread. We repeatedly see people who refused to learn things that could have saved their lives.
In 1841, a British parliamentary commission reported that they had found citizens who had never heard of a city called London. Meeting a group of working-class boys in the street, a commission member had asked if they knew who the Queen of England was.
“Yes, sir,” said the boys, “her name is Prince Albert.”
Literacy, which was common among the upper classes and spreading in the middle class, had not yet penetrated most of the working class.
This lack of knowledge can be viewed as a learning system failure (which can be thought of as analogous to a market failure, where a decentralized market does not allocate goods and services in a Pareto efficient way). And mass education can, in some regards, be seen as a reaction to these failures of the learning system.
The introduction of compulsory education, in Britain and elsewhere, was an attempt to correct for perceived inefficiencies in the learning system. Time that people had previously spent in the learning system was appropriated for education, where knowledge transmission could be centrally controlled. This gave curriculum designers a channel through which they could transmit knowledge deemed essential.
Or phrased another way: mandatory education was an intervention in the decentralized learning system.
How did the intervention affect the underlying system?
Interventions in Complex Systems
Interventions in complex systems have unforeseen consequences. Apply pesticide to stop budworm from killing off your spruce, and you will also kill off the budworm’s natural enemies, making future pest outbreaks more severe. Impose rent control to keep housing affordable and construction will slow, making it harder for people to move to cities with higher economic productivity – raising unemployment and economic inequality while lowering innovation and the rate of childbirth.
The problem with these types of interventions is that they limit the underlying system’s ability to self-organize. By shifting control from the system to a central authority tasked with managing it, the intervention reduces the system’s ability to adjust itself to changes in its environment. It is no longer in the system’s power to freely adapt.
The capacity to change in reaction to new circumstances is crucial for adaptability. Limiting a system’s ability to self-organize and adapt, induces costs that often are hard to see and connect to the intervention that stymied that ability. Rural deaths of despair seem unconnected from urban rent control. Pesticide seems like an unlikely cause of pest outbreaks.
To avoid these types of problems, you need another approach. You need interventions that leverage the system’s capacity to self-organize, rather than work against it. An intervention, if aimed at unblocking the system, can strengthen the ability of the system to shoulder its own burdens.
In forestry, this means cultivating ecosystems that in themselves are resilient enough to keep the budworm population from overwhelming the spruce.
In economics, it means some flavor of Keynesianism: instead of a Soviet-style command economy, you correct for market failures by leveraging the market’s own strengths. You break up monopolies and use free-market competition to hem in the market’s tendency to create winner-take-all effects. You use price-mechanisms, such as carbon rights, to force the market to internalize its externalities, that is: you incentivize the system to self-organize and overcome its limitations.
If we apply this insight to knowledge reproduction, what would it mean? The learning system is a self-organizing entity. It can change its structure to adapt to new circumstances: creating formal apprenticeships during the Renaissance, conjuring universities, building online communities. But when we have tried to promote learning during the modern era, we have often done so at the expense of this adaptability, and many of the structures developed by the learning system, such as apprenticeships, have withered.
Can we instead design interventions that strengthen the learning system, so that it can better overcome its own failures? I am not sure, but it strikes me as a promising question. I will sketch a few possibilities.
Enabling the Learning System
If we want to enable the learning system and allow it to shoulder some of the burdens we’ve placed on education, we must first ask ourselves why the natural correction mechanisms are failing. Why can’t the learning system reproduce the knowledge needed to sustain civilization? It used to be our sole means of knowledge transmission; now it no longer seems sufficient. What is holding it back?
Two things come to mind, changes that might have rendered the learning system insufficient in the modern world. Firstly, the character of knowledge has changed. Whereas knowledge in premodern societies was easy to observe and immediately rewarding – like hunting or mending clothes – knowledge in the modern world is abstract, and the gratification of learning is delayed. This is the common criticism of self-directed learning. What is fun and intrinsically motivating is not necessarily what is useful; the incentives are off. So you misallocate your time.
Secondly, modernity has been a centrifuge. Rapid progress has separated everything – most importantly, home and work, children and adults – creating increasing specialization. Elderly in homes, working-age adults in offices, children sorted in age-graded classes. This segregation makes it hard for knowledge to pass from one group of people to another; children cannot, as in premodern societies, simply learn by working alongside adults; they can not get the widened perspective you get by living with the elderly.
How can these obstacles to learning be overcome?
1. Weak incentives. The first problem is a problem of incentives. Children, not having access to contexts where knowledge work is produced, for example, might not understand the value of developing deep literacy. Especially if they grow up in households where they don’t observe their parents reading, they might choose to learn less valuable skills, such as farming mushrooms in Minecraft, which could render them less employable in the modern job market.
I’m not sure this problem is as big as people would make it out to be. But assume it is. How can we incentivize the unmotivated to pursue things that are not intrinsically motivating?
The most straightforward way to incentivize is to simply reward the behavior that you want to promote. Rather than mandating literacy, give everyone that can pass a high school literacy test 20k dollars or so. Kids who are capable of teaching themselves can just collect the check, and the rest can ask their parents for lessons, or sign agreements with teachers, splitting the proceeds, and so finance their education and collect the reward. This is basically the learn and earn model popular in web3: if people aren’t intrinsically motivated to pay attention to what you value, you just pay them for their attention.
(Implemented on its own, with no further support, self-directed learning and bounties would of course lead to abysmal outcomes for certain groups: I will return to that. )
But for all the limitations of this simple model, it has the upside that it does not disrupt the learning system. Instead of replacing it, which is costly and conflict-ridden, incentives leverage the learning system. It creates a reward function and lets the system self-organize to solve the problem. Different people have different needs and capabilities, and if incentivized correctly the learning system can adapt to serve all these diverse needs, through a rich ecosystem of different learning opportunities. People will search around for the tools and mentors that work for them. Some will enjoy 3blue1brown; others will join study circles or go at it alone.
2. The separated society. Age segregation is a network problem – the nodes in the networks are not connected in a way that allows for efficient knowledge transfer. Children (nodes with limited knowledge) have been denied access to experienced nodes, and reduced to studying them from afar, mediated through books and lectures. And a lot of knowledge cannot pass through these types of low bandwidth connections – reality is too complex and nonlinear to pass through a linear string of words – leaving kids deprived.
For learning to happen spontaneously and effectively you need access. Christopher Alexander – writing in 1977 – asserted that the purpose of future educational institutions “must be to facilitate access for the learner: to allow him to look into the windows of the control room or the parliament, if he cannot get in the door. Moreover, such new institutions should be channels to which the learner would have access without credentials or pedigree, public spaces in which peers and elders outside his immediate horizon now become available...”
The problem with this kind of access is that there is a fundamental conflict at play: the conflict between granting novices access to experts and allowing experts to do productive work. Too many novices at a workplace and their demands for instruction overwhelm the attention of experts, slowing production to a crawl. This doesn’t mean we have to exclude novices as completely as we do today. Especially gifted children could enter productive environments much earlier than current legal codes permit, and gain from it. But you cannot just naively give novices access.
What we can do, however, is build better infrastructure. By leveraging communication technologies, we can grant more people access to valuable networks without overwhelming experts. We can build new types of architecture which allow experts and novices to coexist at scale. I have made a first stab at how that could work in a previous essay: Apprenticeship Online.
An interesting example of what it might look like is a Decentralized Autonomous Organization (DAO). These blockchain-enabled organizations are generally permissionless and have an ethos of working in public: anyone can jump into a Discord, participate in the dialogue, and look for ways to contribute. This week, in early November, as I’m writing this, I was lucky enough to see Liminal Warmth tweet about a new DAO – one that would try to buy a copy of the American constitution. For a person like me, who has a relatively limited understanding of DAOs, and memeing, it was instructive to hang out in the channels where communication strategies were developed, and various legal issues got sorted out. I can’t imagine I would have learned as much from a university course about DAOs, if such a thing even exists. In 72 hours, the group had gone from about a hundred users on Discord to a money swarm of +$47M. They also, ironically, managed to create so much attention around the auction that the price rose beyond them: they came in second place.
If we decentralize the responsibility to learn away from schools, we might need to create incentives to drive the spread of useful knowledge, and we definitely need to invest in infrastructure that makes it easier for novices to access the environments we want them to master. Otherwise, we will see unpalatable learning system failures, where large groups are unable to obtain the knowledge they need to lead dignified lives.
We probably also need general economic support for the young. The young and the unskilled tend to be financially vulnerable and will therefore underspend on their learning compared to the societal optimum. Knowledge has positive externalities; it should therefore, at least partially, be financed by the collective. This is the basic idea behind state-sponsored education, and remains true even if the resources are decentralized into the learning system.
People who come from homes without good access to skilled role models, and those that have learning deficits, will need extra support. Decentralizing the responsibility to learn would otherwise risk hurting these groups. Decentralization leads to increased variance. We want to have a high-pass filter, that allows us to keep the upside of that variance – the gifted children that can grow faster when left to their own devices, the kids with obsessive passions that can specialize early – while limiting the downside.
Regrowth from the Edges In
To sum up:
There exists something we can call the decentralized learning system.
Education is an intervention to correct for its (perceived or real) failures.
Heavy-handed interventions tend to undermine underlying systems, so they should be used as a last resort.
There might be some way to instead unleash the learning system, similar to how economics has unleashed the market.
That might be cool.
Looking at the problem of knowledge reproduction from a systems perspective allows us to find new solutions. The possibilities sketched here (granting access, incentivizing, and giving economic support) are only the most naively obvious, probably not the global optima. The design space is vast.
Here, for example, is another potential design sketched by Christopher Alexander:
Exploring the full range of possibilities can probably most effectively be done outside of existing institutions, from the bottom up. It will need to be a messy, organic process.
Initiatives such as Sudbury Schools, learning experiments in web3, homeschooling families exploring ways of integrating learning back into life, Agile Learning Centers, new types of apprenticeships models in the open source community – these, and many other, experiments can gradually help us regrow the learning system from the edges in towards the center.
Piece by piece, we can learn how to compensate for the shortcomings of decentralized and self-directed learning without relying too heavily on centralized control. This will allow for a richer, more dynamic ecosystem of learning services.
Acknowledgments
This essay has benefitted from several rounds of feedback, primarily by Gunnar Zarncke and Justis Mills. Don’t blame the outcome on them, though. It was a lot worse when they started.