The Magic of Weak Links
As a scientist yourself, what should we make of the fact that despite having basically every known fact about the world memorized, these models haven’t, as far as I know, made a single new discovery? Even a moderately intelligent person who has so much stuff memorized would make all kinds of new connections (connect fact x, and y, and the logical implication is new discovery z).
That is a question that has bothered Dwarkesh Patel for a long time, and if you think about it, it should be bothering you too. Why do we not live in a world where LLM's are spitting out "new discoveries by logical implication, having connected facts x and y"?
And as I have been moaning about for the past two days, what stops me from being a little bit better at this with every passing month or year?
Friction is Clear, But What Are Weak Links?
I explained in yesterday's blog post about what friction is, and how friction might help me concretize ideas, and construct a series of them in my head. OK, but what are weak links?
To understand what they are, and why they matter, we must go back to a paper written in the field of sociology, about fifty years ago. The name of the paper is "The Strength of Weak Ties" and the author is Mark S. Granovetter.
Here's my way of explaining the main point of this paper to you - you want to hang out with your best buds, but are most likely to learn new things when you don't.
Or, if you prefer NotebookLM, here is its explanation:
Weak ties are more likely to transmit novel information because they connect individuals to diverse social circles. Strong ties, conversely, tend to provide redundant information from within one's immediate clique. "Weak ties, then, are presumably important because they provide people with access to information and resources beyond those transmited by strong ties."
When you meet your best buddies from college, you are going to talk about the crazy things you did in college, and that is as it should be. But the opportunity cost of doing so is that you tend to not talk much about much else.
When, however, you meet folks who were not in college with you, you are much more likely to NOT talk about the crazy things you did in college, and that too, thankfully, is as it should be. You are much more likely to talk about other things, and in the process, both you and those other folk might end up learning a fair bit. Try reading the excerpt again, particularly the part in inverted quotes. Now does it make sense?
TMKK?
I mean, ok great, I'll try and hang out with other folks too, you might be saying. But what's that got to do with what you've been going on about for three straight days now?
One of the tricks to learning better, if you ask me, is to spend time making connections across different things you have read, but by forcing yourself to explain or understand ideas in other fields by using what you know of your own area of expertise. Or the other way around: learn better about your own field by adopting ideas already in use in other fields.
If you're looking for an example, have you heard about the gravity model of international trade? It says that trade between two countries is likely if they are closer to each other, geographically speaking. It also says that trade between two countries is likely if they're both large economies. Can you figure out why it is called the "gravity" model of international trade?
(Recall from your school days that the gravitational force between two bodies is directly proportional to the product of their masses, and inversely proportional to the square of the distance between them)
The more you make these connections across two different subjects, the better you get at appreciating both subjects. The more you do this with more subjects, the better you get at knowing a little bit about more subjects, and about connections across all of them.
And finally, and most importantly, the more you do this with "weak" connections across different subjects, the better your "access to information and resources beyond those transmitted by strong ties"
An Example
What would be really cool would be a thingummy that knew I had written this blog, that knew I had started (but still haven't finished) a book called This Entangled Life by Merlin Sheldrake, and knew that I am a fan of the book Information Rules by Hal Varian and Carl Shapiro.
And if that thingummy could come up with a small quiz based on these three things, and then recommend what I could read next based on any one or all three of these things...
...and could therefore increase my knowledge and way to think about information networks, knowledge networks, biological networks...
...and do all of this by making the act of consuming information united by weak links frictionless...
...but make the act of concretizing and constructing across those weak links have just that little bit of friction...
...and if it could do this intuitively, iteratively and incrementally, day by day...
...well then, life would be very good indeed.
And that would be a very nice mix of friction and weak links indeed.
No?