The One-Channel Paradigm

post by Noah Walton (noah-walton) · 2018-04-25T21:49:56.176Z · LW · GW · 2 comments

Under the one-channel paradigm, one thing does one thing at a time.

The general thing that follows the one-channel paradigm is the game. Or maybe physics or a Markov process. 

This seems to be a common paradigm. It also leads to bad mistakes.

“What shall I do? Well, I want to learn some economics. So I’ll make a tree (a DAG, actually) of concepts that I need to know, and learn from the bottom up. At the end, I’ll know the whole tree. Then I can get building a bigger tree off of that tree.”

This is better than thinking of learning as Level 0, Level 1, Level 2, etc., where each Level is one concept. Since each Level can have many members, and you can branch out in many different ways, a tree makes more sense than a tower. (What about a cyclic knowledge graph? We might want to account for iff statements or meta-stable principles.)

But maybe the one-channel paradigm is not a useful way to think about problems, a lot of the time. Even if it’s “true". Even if we live inside a Markov process, it might be useful to think of things as not being that way. 

Because subscribing the one-channel paradigm destroys a lot of valuable things humans can do. It tends to overrate conscious thought, for example. Conscious thought is the obvious candidate for the channel to identify with. Notice the mistake here. We went from treating the whole world as a Markov process, to singling out a rather unnatural part of spacetime (conscious thought) and describing it as a coherent stream separate from the rest of the world.

For me, I think this error leads to wanting to do too many atomic activities, or at least thinking of my activities as atomic. Playing a game. Playing a piece of music. Writing a post. Eating a meal. Learning a topic.

There should be alternative ways of thinking about these things, under a paradigm other than the one-channel one. I don’t know how to express those views in language, though. But maybe stories are good examples.

It seems that reading a whole book can often give much greater returns than reading many isolated pieces of information. Would you rather read the dictionary or Anna Karenina? 

Why? I could give some bullshit answer about maybe Anna Karenina giving us a better chance to accomplish interesting compression: a lot of the compression done while reading the dictionary seems much duller — a dictionary is just filling in a template over and over again, and we can figure out how the template works really fast, after which the template becomes boring. There’s still lots to be learned by how the templates are filled in though.

2 comments

Comments sorted by top scores.