Nucleation is the first step in forming a new phase or structure. For example, microtubules are hollow cylinders built from individual tubulin proteins, which stack almost like bricks. Once the base of the microtubule has come together, it’s easy to add more tubulin to the microtubule. But assembling the base—the process of nucleation—is slow without certain helper proteins. These catalyze the process of nucleation by binding and aligning the first few tubulin proteins.
What does learning have in common with nucleation? When we learn from written sources, like textbooks or a lecture, the main sensory input we experience is typically a continuous flow of words and images. All these words and phrases are like “information monomers.” Achieving a synthetic understanding of the material is akin to the growth of larger structures, the “microtubules.” Exposing ourselves to more and more of a teacher’s words or textbook pages does increase the “information monomer concentration” in our minds, and makes a process of spontaneous nucleation more likely.
At some point, synthesis just happens if we keep at it long enough, the same way that nucleation and the growth of microtubules spontaneously happens if we keep adding more tubulin to the solution. But for most people, I think the bottleneck to increasing the rate of forming a synthetic understanding is bottlenecked not by exposure to “information monomers,” the sheer amount of reading or listening we do, but by the process of “nucleation,” our ability to form at least a rudimentary synthetic understanding out of even a small quantity of content.
Might be about using the accumulated references to seek out and generate more training data (at the currently relevant level, not just anything). Passive perception of random tidbits might take much longer. I wonder if there are lower estimates on human sample efficiency, how long can it take to not learn something when getting regularly exposed to it, for a person of given intelligence, who isn’t exercising curiosity in that direction? My guess it’s plausibly a difference of multiple orders of magnitude, and a different ceiling.
This reminds me of a thing that gestalt psychologists call “re-centering”, which in your metaphor could be described as having a lot of free-floating information monomers that suddenly snap together (which in slow motion might be that first the base forms spontaneously, and then everything else snaps to the base). The difference is that according to them, this happens repeatedly… when you get more information, and it turns out that the original structure was not optimal, so at some moment it snaps again into a new shape. (A “paradigm shift”.)
Then there is a teaching style called constructivism (though this name was also abused to mean different things) which is concerned with getting the base structure built as soon as possible. And then getting the next steps right, too. Emphasis on having correct mental models, over memorizing facts. -- One of the controversies is whether this is even necessary to do. Some people insist that it is a waste of time; if you give people enough information monomers, at some moment it will snap to the right structure (and if it won’t, well, sucks to be you; it worked okay for them). Other people say that this attitude is why so many kids hate school, especially math (long inferential chains), but if you keep building the structure right all the time, it becomes much less frustrating experience.
So I like the metaphor, with the exception that it is not a simple cylinder, but a more complicated structure, that can get wrong (and get fixed) multiple times.
A “Nucleation” Learning Metaphor
Nucleation is the first step in forming a new phase or structure. For example, microtubules are hollow cylinders built from individual tubulin proteins, which stack almost like bricks. Once the base of the microtubule has come together, it’s easy to add more tubulin to the microtubule. But assembling the base—the process of nucleation—is slow without certain helper proteins. These catalyze the process of nucleation by binding and aligning the first few tubulin proteins.
What does learning have in common with nucleation? When we learn from written sources, like textbooks or a lecture, the main sensory input we experience is typically a continuous flow of words and images. All these words and phrases are like “information monomers.” Achieving a synthetic understanding of the material is akin to the growth of larger structures, the “microtubules.” Exposing ourselves to more and more of a teacher’s words or textbook pages does increase the “information monomer concentration” in our minds, and makes a process of spontaneous nucleation more likely.
At some point, synthesis just happens if we keep at it long enough, the same way that nucleation and the growth of microtubules spontaneously happens if we keep adding more tubulin to the solution. But for most people, I think the bottleneck to increasing the rate of forming a synthetic understanding is bottlenecked not by exposure to “information monomers,” the sheer amount of reading or listening we do, but by the process of “nucleation,” our ability to form at least a rudimentary synthetic understanding out of even a small quantity of content.
Might be about using the accumulated references to seek out and generate more training data (at the currently relevant level, not just anything). Passive perception of random tidbits might take much longer. I wonder if there are lower estimates on human sample efficiency, how long can it take to not learn something when getting regularly exposed to it, for a person of given intelligence, who isn’t exercising curiosity in that direction? My guess it’s plausibly a difference of multiple orders of magnitude, and a different ceiling.
This reminds me of a thing that gestalt psychologists call “re-centering”, which in your metaphor could be described as having a lot of free-floating information monomers that suddenly snap together (which in slow motion might be that first the base forms spontaneously, and then everything else snaps to the base). The difference is that according to them, this happens repeatedly… when you get more information, and it turns out that the original structure was not optimal, so at some moment it snaps again into a new shape. (A “paradigm shift”.)
Then there is a teaching style called constructivism (though this name was also abused to mean different things) which is concerned with getting the base structure built as soon as possible. And then getting the next steps right, too. Emphasis on having correct mental models, over memorizing facts. -- One of the controversies is whether this is even necessary to do. Some people insist that it is a waste of time; if you give people enough information monomers, at some moment it will snap to the right structure (and if it won’t, well, sucks to be you; it worked okay for them). Other people say that this attitude is why so many kids hate school, especially math (long inferential chains), but if you keep building the structure right all the time, it becomes much less frustrating experience.
So I like the metaphor, with the exception that it is not a simple cylinder, but a more complicated structure, that can get wrong (and get fixed) multiple times.