The trouble with measurement part II

I'm moving onward from linear algebra, picking up some calculus (I'm rusty), and moving into quantum computing proper in the next couple months. With that said, here's a continuation of where I'm at on the philosophical journey on quantum interpretation.

If you read the last post, I started at Many Worlds, made my way to pilot wave theory, and had a short stopover at Copenhagen. More recently? I finished Something Deeply Hidden by Sean Carroll; he's a Many Worlds adherent.

I'm sold on the brilliance of Many Worlds, and the argument is compelling:

• There's not a plethora of wave functions. There's only a single, smoothly evolving wave function.
• The single wave function simplifies collapse: it's just the wave function evolving and branching.
• The branching yields the "many worlds." The world we live in is simply the branch that won out.

Carroll's argument is that this simplifies quantum mechanics to the absolute minimum it needs to work. He continues: every addition an alternative offers must justify itself. The kicker is that these additions from other schools of thought are usually just arguments to help their theories hang nicely, and they don't have clean proof.

Ah yeah, but my response is that this Many Worlds argument smuggles a big HUH? through the back door.

Back to the continuation

Part 1 ended on Copenhagen and the shifty split: the measuring device triggers the wave function, but nobody can say in physical terms what counts as a measuring device. That was the point where I stopped being able to sit comfortably in the orthodox view. Here's where I went next, and where I ended up.

BTW: I’m using Claude for help on this post because I pushed back on a lot of theories in chat and had it take notes.

GRW: if the cut matters, make it physical

The first interpretation I ran into that takes the shifty split seriously is GRW (Ghirardi, Rimini, Weber). Collapse is a real physical process, and it just happens. Every particle has a tiny chance of spontaneously localizing, on the order of once every hundred million years. For a single particle you'd never notice. But a detector, a camera, or a cat is something like 10²³ particles all entangled together, so the moment one of them localizes, the whole system gets dragged along, and a definite outcome appears in a tiny fraction of a second. You don’t even need an “observer” as far as I’m aware.

I liked this more than I expected to. It keeps Copenhagen's operational discipline (definite outcomes, the Born rule) while dropping the agent-relativity. It also sticks its neck out. It's a different theory, not just a different story, so experiments can catch it, and they've been squeezing it hard, including the gravity-based variants from Penrose and Diósi.

Where it stopped working for me:

• Two new constants, tuned so small things stay quantum and big things stay classical. That's the same kind of explanatory debt I hated in Many Worlds, just in a different form.
• GRW is built on non-relativistic quantum mechanics, and a fully satisfying relativistic version has been a long struggle.
• Once I took decoherence seriously, collapse started to look less like an event needing an engine and more like an accounting necessity. Decoherence doesn't pick which outcome happens, and I'm not claiming it does. But I think it explains why interference disappears, and that was doing more of the work of "collapse" than I'd credited. GRW builds a physical mechanism for something I'd started to suspect was an accounting exercise.

It was a very short jaunt into this theory.

QBism: collapse as updating

QBism (Fuchs, Schack, Mermin, among others) went straight at my background (see: this site). The wave function isn't a thing in the world; it's an agent's degrees of belief about what they'll experience next, and collapse is just what happens when they update on a result. No mystery in the update, no physical event, and the spooky-action worry mostly evaporates because nothing physical jumps when I learn something. Coming from data work, Bayesian updating is my native language, so "collapse is bookkeeping" clicked immediately. Some of that survived into where I ended up.

What didn't survive: if probabilities are only ever personal beliefs, what are the beliefs about? When two people run the same experiment and their statistics converge, something is being tracked. QBism's answer, roughly, is that the world is made of agents' experiences. I had just spent a post pushing back on observer-dependence in Copenhagen, and QBism doesn't fix it. It makes it the foundation. I wanted the bookkeeping without giving up a real thing being booked.

Relational QM: everyone is an observer

Relational Quantum Mechanics (RQM), Carlo Rovelli's interpretation, cleans up something QBism leaves fuzzy. Anything can be an observer: an electron, a rock, a lab. A system's state is only defined relative to another system it has interacted with, and there's no single description of the world from outside all perspectives. Wigner's friend stops being a paradox, because the friend and Wigner are each right relative to themselves. Rovelli even argues this dissolves nonlocality, since distant results only get compared through a later local interaction.

That solves "who counts as an observer" neatly: everything does. But look at what it kept. The facts are still relative. I ended Part 1 wanting something that kept Copenhagen's discipline while dropping the agent-relativity, and RQM promotes relativity from a bug to a principle. The hard question for it is whether there's any fact about how the perspectives fit together, or only more perspectives. If there's no single objective story, I've moved the problem, not solved it.

Superdeterminism: the loophole I'm not moving into

Short section, because it deserves to be short. Bell's theorem assumes more than locality. It also assumes the experimenters' choice of what to measure is statistically independent of whatever hidden variables are doing the work. Superdeterminism denies that: settings and hidden variables are correlated through shared history, so a local hidden-variable account survives. It's the one loophole that's still logically open.

My problem is that the initial conditions of the universe would have to conspire so that every pair of detectors picks settings that hide the trick. As far as I can tell, nobody has a fully worked-out model that reproduces quantum mechanics this way. It also quietly undermines the assumption that makes any experiment interpretable at all. Maybe the loophole is real, but I'm not building a house in it.

Where I've landed: potentia

Heisenberg had a word for what the wave function describes: potentia, a real tendency toward outcomes, sitting between the idea of an event and the event itself. Popper later gave the same instinct a home in philosophy as propensity. My position is a version of that, with three commitments:

1. The propensity is real, and it's single-trial. It isn't a statement about frequencies over many runs, and it isn't my ignorance. A given electron in a given setup really has a chance of showing up here versus there. I'm not adding hidden variables underneath to settle which. Bohm's route is coherent, and I still have a soft spot for it, but I don't need to add particles to get the physics.
2. Collapse is bookkeeping. When an outcome becomes actual, we update the propensity we assign to the system. That update isn't a second kind of physical event needing its own mechanism. Decoherence accounts for interference vanishing, genuine chance accounts for which outcome, and I don't need GRW's engine for either.
3. The nonlocality is literal. For an entangled pair, the propensity belongs to the pair as a whole, and when an outcome at one end becomes actual, what's possible at the other end changes immediately. I can't have the flower promise, because Bell closed that door for local hidden variables. So I'm taking the correlations as they are. To be careful about the word: nothing usable travels faster than light and no message can be sent. But something about the world is genuinely non-separable, and I'm not going to pretend otherwise.

What this costs me

I don't have a story for what actualization is. Potentia says one outcome becomes actual; it doesn't say how. I'm not sure that's a debt I owe, but I should name it. And literal nonlocality is awkward in a relativistic world: for distant measurements, observers can disagree about which came first, so "changes immediately" can't be a frame-independent story. The predictions are fine and there's no signaling problem, but I don't have a clean picture. I'd rather live with a stated gap than fill it with a story I don't believe.

Why "simplest wins" isn't a tiebreaker

There's a pattern across every section here, the same one I flagged in Part 1. Each interpretation gets its appeal from something that seems to come free: Many Worlds' minimal postulates, Bohm's realism, Copenhagen's clarity, GRW's objectivity, QBism's tidy dissolution of collapse, RQM's end to the observer problem. Then you check the bookkeeping and something is being paid for: an unbounded ontology, a wave/particle asymmetry, a fuzzy cut, new constants, agent-centered reality, relative facts, or a conspiracy in the initial conditions.

That's why I don't accept "the simplest interpretation wins" as a tiebreaker. Simplicity counts postulates, but what matters is commitments, and those aren't the same currency. Simplicity arguments also have a rough history: [your historical example of Occam's Razor being misapplied]. When two interpretations make identical predictions, parsimony can't settle it, because there's no agreed unit of parsimony.

Where I am now...

What I'm not willing to explain away is genuine chance and genuine nonlocality. Others will make different trades, and that's fine as long as they show what they're paying. If you think I've mispriced one of these, tell me. I've changed my mind enough times in this series that I'd rather be corrected than comfortable.