you can learn everything: Nonlinear Function
Created: May 15, 2021
Modified: April 17, 2022

you can learn everything

This page is from my personal notes, and has not been specifically reviewed for public consumption. It might be incomplete, wrong, outdated, or stupid. Caveat lector.

Sometimes it's daunting how much knowledge there is in the world. For any given area, there are a thousand specialties and subspecialties, each with experts who've spent their whole life in the area and still feel like they know only a fraction of what there is to know. How can you or I hope to keep up?

One answer, of course, is that we can't. We all have limited time, energy, and motivation. We have to choose where to focus, what to specialize in. It's one of the tragedies of the human condition that life consists of a series of decisions each of which closes off paths and possibilities---ameliorated only by the fact that each of these choices also allows us to progress to deeper and more profound knowledge of the things we really care about. Without denying this, I want to propose a different and more optimistic model.

We might think of science as the practice of producing new knowledge. Much of what fills the brains of practicing scientists pertains to the process of producing knowledge rather than the final product. But the 'final product' represents the concentrated insight, the tested theories and validated concepts, the precious wisdom that scientists are trying so hard to discover.

Take for example a neuroscientist studying a brain disorder such as Alzheimer's disease. Their day to day work involves generating hypotheses and coming up with experiments to test these hypotheses. This may itself involve working with animal models, applying techniques such as brain imaging or electrical stimulation or genetic interventions that require deep practical knowledge and technical skill. Data must be collected and analyzed, and any results must be framed, presented, published, and sold to the community. These skills take many years to develop and practice (during a PhD, for example).

But critically, the end product of this work, if it is successful, is a set of new facts about the world. Learning these facts---that the cortex is a thin layer of tiled columns, say, or that neurons signal both electrically and chemically, that the blood-brain barrier exists, etc.---does not in itself require any of the wide range of knowledge-production skills that a practicing neuroscientist must have.

In any field, there is:

  1. Our current picture of how things work.
  2. Things we are unsure about---promising but unproven hypotheses. Scattered facts that have yet to be organized into theories. A literature full of old papers with intriguing ideas that didn't quite work at the time.
  3. Techniques and skills for doing research.

The second two of these are often the vast majority of the field. Most of (1) --- the really solid knowledge --- is taught in undergraduate courses.

A related take from Dwarkesh Patel:

David Deutsch points out in The Fabric of Reality that contra conventional wisdom, it actually is possible for a single person to understand most things - not in the sense of memorizing the names of ant subspecies or the GDP of different Asian countries, but in the sense of appreciating the main explanatory theories in each field.

One consequence of living in The Great Stagnation is that there is relatively little turnover in these fundamental ideas. Quantum mechanics, that nascent branch of physics which elicits the sense of woo woo from popular culture, is about a hundred years old. So is the theory of computation. The neo-Darwinian synthesis is over 50 year old.

So you don’t have to be scouring through the newest papers on Arxiv in order to know the most important things. A dozen or so textbooks even from a few decades ago contain about 80% of legible scientific knowledge.