I generally subscribe to the signalling model of higher education: “education” is mostly about signalling how smart and diligent you are, rather than actually learning useful skills. Under this model, higher education doesn’t produce new human capital, it just makes it more obvious who has it already—it’s essentially marketing.
That said, it still seems like actually learning things does provide at least some value for at least some people—and personally, I think I’ve gained a lot more value than most out of learning things. A lot of this comes from being strategic: choosing what to learn, and how much time to spend on it, in order to maximize value. In order to do that, we need some idea of the mechanism by which learning things can provide value in the first place.
With that in mind, here are five different models for how things we formally learn (i.e. in school/university) can provide value.
Skills
A CS student learns a few programming languages, basic data structures, some aspects of low-level computer architecture and systems design… skills which are directly useful for real-world software development. Obviously not everything one learns in a CS degree is useful in real-world development, but there is a lot of directly-useful knowledge and skills in there.
General model: formal learning provides practical value by teaching how to use a useful tool or perform a useful task. Some examples:
Generic communication skills like legal/scientific/business writing or public speaking
Useful ways to frame problems, like thinking about risk and discounted expected value of potential business projects
When and how to use a physical tool, like a pipette or atomic force microscope
Mathematical/algorithmic tools, like techniques for solving ODEs/PDEs, statistical tests, or data structures
When and how to apply mathematical/algorithmic tools to real-world problems, e.g. turning a physical system into equations in physics or engineering
Most vocational education
This sort of thing is how I think most people imagine learning will provide value. We learn to use some technique which we will directly apply to valuable real-world problems. Keep it in mind as a baseline against which to compare the models below.
Gears
A doctor studies the physiology of the kidney—they gain a gears-level model of kidney function. Hopefully, this will later allow them to recognize/predict a wide variety of kidney failure-modes and their implications, by understanding how the effects of one particular failure will propagate throughout the system. In particular, they can hopefully handle novel problems, problems which are not like any problem they’ve seen before.
General model: formal learning provides practical value by teaching us the internal gears of a system. That, in turn, allows us to make predictions about how the system will behave in a wide variety of novel circumstances—e.g. when some component breaks, or when the environment changes drastically.
Consider suffix trees. Suffix trees (and their variants) provide a fast data structure for things like searching in text, comparing text, etc. Most programmers, most of the time, do not need to know anything at all about suffix trees. Even if you do suddenly need to know about suffix trees, you can google it and read up. But if you don’t even know when to google for it… then you’re liable to have a very rough time.
General model: formal learning provides value by removing unknown unknowns. Examples:
Intro statistics courses flag a number of common errors people often don’t realize they’re making
Even if we don’t remember how to use a particular tool from some class, knowing that it exists at all is usually enough to find it later if we need it—e.g. specialized algorithms or sensor types or experimental techniques
Along similar lines, knowing that some group of people works on a certain type of problem is often enough to find them later if we need to—e.g. experts specializing in a certain class of diseases or marketing to a particular segment or funding certain types of projects or …
In engineering fields, it’s useful just to know that we probably aren’t missing any key considerations of a design. Nobody wants to build the next Tacoma Narrows—the bridge which resonated with wind eddies, a factor the designers never even thought to consider.
One big upshot of this value-model: removing unknown unknowns requires relatively little investment of time and effort. A single pass through a book or set of lectures is often enough to recall an idea when it comes up later. (Of course, more practice may still help us recognize the idea in a wider variety of situations.)
Interface
Another model: formal learning provides value by teaching students to interface to a system. Some examples:
Learning some of a field’s jargon is very helpful for working with people in that field, even if you’re not in that field yourself. For instance, a technical product designer benefits from some exposure to programming and algorithms in order to interface with software engineers.
Law school largely teaches how to interact with the legal system; science fields largely teach how to interpret and evaluate the models used in each field.
Programming directly involves learning specialized languages. It is via these languages that we interface with the work of other programmers.
Learning human languages obviously allows us to talk to a wider variety of people around the world.
People often need to apply the knowledge/tools of one field in another. Studying some physics helps biologists understand certain imaging tools; studying some electrical engineering helps programmers understand the lower abstraction levels of computer systems.
This value-model requires a fair bit of practice, but notice that it doesn’t necessarily require remembering all the details which come up in ones’ studies. A lawyer may need to write briefs or a programmer may need to write code to practice their skills, but they don’t need to memorize all the bits of information they had to look up while writing those briefs/programs. They do need to know where/how to find information, and how to correctly interpret/use that information, but they don’t necessarily need to store all that information in their heads.
Identifying Experts
King Louis XV of France died of smallpox the same year that an English dairy farmer successfully vaccinated his family. Louis had the resources of a literal king, and hired the supposedly-finest doctors in Europe, but at the end of the day he could not distinguish someone actually capable of curing the disease from astrologers and humoral experts. Even had he looked, he would not have picked that dairy farmer out of a crowd of people with clever-sounding ideas to protect from smallpox, and he wouldn’t have thought to run clinical trials.
Identifying people with more expertise on a topic than ourselves is Hard.
Model: formal learning provides value by giving us ways to distinguish actual experts from con-men and from people who just don’t know what they’re doing as well as they think they do. Some examples:
Learning some physiology and biology won’t teach you how to best handle every disease, but it will help you distinguish those who do understand a particular disease from those who market snake-oil.
Recognizing programmers who write readable/robust/maintainable code is Hard if you don’t know anything about programming
Basic statistics is a fairly general-purpose bullshit-detection tool
This is another model where we can potentially get a lot of mileage from a relatively small investment. We don’t need to fully understand the field ourselves, we just need to understand enough to recognize someone who understands more.
Of course, the catch is that we’ll learn to recognize people who are “experts” by the standards of the field studied, which may or may not be useful—the doctors of King Louis’ time would not have recognized the merit of vaccination any more than the King.
Models of Value of Learning
I generally subscribe to the signalling model of higher education: “education” is mostly about signalling how smart and diligent you are, rather than actually learning useful skills. Under this model, higher education doesn’t produce new human capital, it just makes it more obvious who has it already—it’s essentially marketing.
That said, it still seems like actually learning things does provide at least some value for at least some people—and personally, I think I’ve gained a lot more value than most out of learning things. A lot of this comes from being strategic: choosing what to learn, and how much time to spend on it, in order to maximize value. In order to do that, we need some idea of the mechanism by which learning things can provide value in the first place.
With that in mind, here are five different models for how things we formally learn (i.e. in school/university) can provide value.
Skills
A CS student learns a few programming languages, basic data structures, some aspects of low-level computer architecture and systems design… skills which are directly useful for real-world software development. Obviously not everything one learns in a CS degree is useful in real-world development, but there is a lot of directly-useful knowledge and skills in there.
General model: formal learning provides practical value by teaching how to use a useful tool or perform a useful task. Some examples:
Generic communication skills like legal/scientific/business writing or public speaking
Useful ways to frame problems, like thinking about risk and discounted expected value of potential business projects
When and how to use a physical tool, like a pipette or atomic force microscope
Mathematical/algorithmic tools, like techniques for solving ODEs/PDEs, statistical tests, or data structures
When and how to apply mathematical/algorithmic tools to real-world problems, e.g. turning a physical system into equations in physics or engineering
Most vocational education
This sort of thing is how I think most people imagine learning will provide value. We learn to use some technique which we will directly apply to valuable real-world problems. Keep it in mind as a baseline against which to compare the models below.
Gears
A doctor studies the physiology of the kidney—they gain a gears-level model of kidney function. Hopefully, this will later allow them to recognize/predict a wide variety of kidney failure-modes and their implications, by understanding how the effects of one particular failure will propagate throughout the system. In particular, they can hopefully handle novel problems, problems which are not like any problem they’ve seen before.
General model: formal learning provides practical value by teaching us the internal gears of a system. That, in turn, allows us to make predictions about how the system will behave in a wide variety of novel circumstances—e.g. when some component breaks, or when the environment changes drastically.
I’ve already talked about this a bunch in Gears-Level Models as Capital Investments. In particular, it’s probably the main value model for most of the sciences on a large scale (though not necessarily for individual people). For examples outside the sciences, see Why Artists Study Anatomy and Baking Is Not A Ritual.
Unknown Unknowns
Consider suffix trees. Suffix trees (and their variants) provide a fast data structure for things like searching in text, comparing text, etc. Most programmers, most of the time, do not need to know anything at all about suffix trees. Even if you do suddenly need to know about suffix trees, you can google it and read up. But if you don’t even know when to google for it… then you’re liable to have a very rough time.
General model: formal learning provides value by removing unknown unknowns. Examples:
Intro statistics courses flag a number of common errors people often don’t realize they’re making
Even if we don’t remember how to use a particular tool from some class, knowing that it exists at all is usually enough to find it later if we need it—e.g. specialized algorithms or sensor types or experimental techniques
Along similar lines, knowing that some group of people works on a certain type of problem is often enough to find them later if we need to—e.g. experts specializing in a certain class of diseases or marketing to a particular segment or funding certain types of projects or …
In engineering fields, it’s useful just to know that we probably aren’t missing any key considerations of a design. Nobody wants to build the next Tacoma Narrows—the bridge which resonated with wind eddies, a factor the designers never even thought to consider.
One big upshot of this value-model: removing unknown unknowns requires relatively little investment of time and effort. A single pass through a book or set of lectures is often enough to recall an idea when it comes up later. (Of course, more practice may still help us recognize the idea in a wider variety of situations.)
Interface
Another model: formal learning provides value by teaching students to interface to a system. Some examples:
Learning some of a field’s jargon is very helpful for working with people in that field, even if you’re not in that field yourself. For instance, a technical product designer benefits from some exposure to programming and algorithms in order to interface with software engineers.
Law school largely teaches how to interact with the legal system; science fields largely teach how to interpret and evaluate the models used in each field.
Programming directly involves learning specialized languages. It is via these languages that we interface with the work of other programmers.
Learning human languages obviously allows us to talk to a wider variety of people around the world.
People often need to apply the knowledge/tools of one field in another. Studying some physics helps biologists understand certain imaging tools; studying some electrical engineering helps programmers understand the lower abstraction levels of computer systems.
This value-model requires a fair bit of practice, but notice that it doesn’t necessarily require remembering all the details which come up in ones’ studies. A lawyer may need to write briefs or a programmer may need to write code to practice their skills, but they don’t need to memorize all the bits of information they had to look up while writing those briefs/programs. They do need to know where/how to find information, and how to correctly interpret/use that information, but they don’t necessarily need to store all that information in their heads.
Identifying Experts
King Louis XV of France died of smallpox the same year that an English dairy farmer successfully vaccinated his family. Louis had the resources of a literal king, and hired the supposedly-finest doctors in Europe, but at the end of the day he could not distinguish someone actually capable of curing the disease from astrologers and humoral experts. Even had he looked, he would not have picked that dairy farmer out of a crowd of people with clever-sounding ideas to protect from smallpox, and he wouldn’t have thought to run clinical trials.
Identifying people with more expertise on a topic than ourselves is Hard.
Model: formal learning provides value by giving us ways to distinguish actual experts from con-men and from people who just don’t know what they’re doing as well as they think they do. Some examples:
Learning some physiology and biology won’t teach you how to best handle every disease, but it will help you distinguish those who do understand a particular disease from those who market snake-oil.
Recognizing programmers who write readable/robust/maintainable code is Hard if you don’t know anything about programming
Basic statistics is a fairly general-purpose bullshit-detection tool
This is another model where we can potentially get a lot of mileage from a relatively small investment. We don’t need to fully understand the field ourselves, we just need to understand enough to recognize someone who understands more.
Of course, the catch is that we’ll learn to recognize people who are “experts” by the standards of the field studied, which may or may not be useful—the doctors of King Louis’ time would not have recognized the merit of vaccination any more than the King.