Analogies at the edge of reason

Making analogies is the engine of human intelligence, but for humanity as a whole, and our collective-intelligence enterprise called science, it is an obstacle. I’ll try to expand on that in this, maybe not sharpest of posts.

Hypotheses

In science and life alike, we use analogies as shortcuts to form hypotheses. Any other strategy—experimenting, making observations, statistical inference, etc.—is more expensive and time-consuming. It’s like a dude excited about how different his new girlfriend is from his ex, but cheers her up with fresh flowers because it worked in the past. … hmm, did that analogy bring my point home? Maybe not entirely, and that is the point. Whatever picture analogies put in our minds are biased approximations at best, often setting us off in the wrong direction.

When network science became popular among physicists about 20 years ago, the research questions and assumptions were, with few exceptions, straight out of the complexity handbook. A decade later, the field’s sense of purpose, methods, and what results are worthy of publication, all seemed rather like the network science of other disciplines before. Of course, some of the complexity-flavored network studies have earned their place in the hall of fame, but I can’t help thinking we would be much better off without the ambition to create a statistical physics of networks, or any “the [one field] of [another field].”

Understanding / explanation

Analogies have purposes other than generating hypotheses. Quantum mechanics, especially in the Dirac formulation, is a substance of distilled beauty. Everything is there; in one equation. Of course, unlike anything describing everyday phenomena. Nevertheless, people argue it needs interpretation, maybe not a complete analogy, but bits and pieces (specifically, waves and particles) from the world of everyday experience. But that cannot be understanding in that it legitimizes a false base of intuition.

The dilemma is that we are so attuned to analogies that if we don’t use them, we could lose our audience (are you still there?). When writing up papers, it sometimes feels like the choice is between misleading people by putting the wrong pictures in their heads or losing them to the boredom of not having any pictures at all.

Creativity

Sometimes, I see analogies heralded as a source of creativity. Isn’t that the strangest claim? Aren’t analogies, per definition, ideas within the box? Perhaps it could sometimes take creativity to find and apply them, but not as much as coming up with something, you know, new.

Background and epilogue

A tweet by the (always brilliant) Samuel Arbesman a few weeks ago prompted me to write down these notes. But the opinions came to me after watching a talk by the (almost always brilliant) Douglas Hofstadter (at Uppsala University a decade or so ago). He mentioned his group’s attempts to build a first-generation style AI based on the human mind’s ability to make “fluid analogies.” And while I agree that the human mind works like that. I also wondered why one should restrict oneself by modeling AI on human thinking. Just because something comes naturally doesn’t mean it’s preferable.

This point is even more pertinent for science as a distributed knowledge system. We may use system A as a quick-and-dirty model for system B in our mind. That’s just human nature. However, for science, that could be built up over generations, the temptation to give an immediate sense of insight by communicating perfunctory models (analogies) will only slow down the long-term progress.