ME & AI

ME: Towards the end of the paper https://philsci-archive.pitt.edu/29428/1/NewSelectionPaper.pdf they claim that we should be more cautious than Bell was about relying on intuitive time-asymmetries, no matter how straightforward they may seem. And that again, it is widely accepted that the source of observed time-asymmetries lies not in fundamental theory but in a distant low entropy boundary condition. Unless we wish to challenge this orthodoxy – or to accept that QM is not a fundamental theory. Can you explain why and is it a superdeterminsitic view?

AI: Bell's move was to translate an intuitive notion of local causality into a clean mathematical condition, but the paper argues that this translation may quietly smuggle in a temporal assumption that is not fundamental. If the asymmetry comes from the early low-entropy boundary condition, then one should be careful about using that asymmetry to rule out alternative explanations of Bell correlations. That is why they think Bell may have "thrown out the baby with the bathwater" by assuming too much about the direction of causal explanation.

ME: Is it superdeterminism?

AI: A superdeterministic view typically says the settings aa and bb were never independent of the hidden variables in the first place. The paper's proposal is more modest in one sense and more radical in another: it says that what looks like nonlocal correlation may be a selection effect arising from the structure of the relevant ensemble, especially once you stop assuming a fundamental past-to-future asymmetry. So it does not have to imply "your choices were pre-correlated with the particles" in the usual superdeterministic sense, even though critics may group the view nearby because it challenges the measurement-independence intuition.

ME: I still have a hard time seeing the distinction

AI: In a superdeterministic story, the universe's initial conditions already encode a correlation between "what setting gets chosen" and "what hidden variable the pair has." In a retrocausal or selection-bias story, the past state and future setting are linked by a time-symmetric constraint, or the observed Bell pair is a postselected/preselected slice of a larger ensemble. So the distinction is not about whether independence fails, but about whether the failure is explained by common initial conditions or by time-symmetric/selection structure.

ME: But is the conditioning on a selected subensemble free?

AI: What makes it philosophically charged in Bell-type discussions is what defines the subensemble. If the selection criterion is completely independent of the hidden variables and of the measurement settings, then conditioning is harmless and not a loophole. But if the selected sample is determined in a way that is correlated with the variables entering the Bell analysis, then the conditioning can generate an apparent Bell violation without any need for nonlocal influence.

ME: But in the paper the selected sample is determined in a way that is correlated with the variables entering the Bell analysis is it not? And how can that be a free choice?

AI: Your worry is exactly the right one: if the selected subensemble is correlated with the Bell variables, then in Bell's formal framework the free-choice assumption is gone. That is why critics say such views are "effectively" superdeterministic, even if the authors resist that label. So the distinction is mostly interpretive: whether you think the correlation reflects a pre-established conspiracy, or a time-symmetric/selection-based explanation of why conditioning changes the statistics.

ME: Why is the subensemble conditioning not seen as a conspiracy by the authors?

AI: The reason that does not feel like a conspiracy to them is that, in their telling, nothing in the setup needs to "know" in advance what you will choose. The dependence appears only after you describe the data by conditioning on a particular subset. In that sense, the correlation is supposed to come from the way the ensemble is carved up, not from a prearranged conspiracy between the setting choice and the hidden variables.

ME: To me that the correlation is supposed to come from the way the ensemble is carved up, not from a prearranged conspiracy between the setting choice and the hidden variables sounds like it would fit a many worlds interpretation.

AI: Many-Worlds says "all outcomes happen, and the observed statistics are branch-relative," while the paper says "the observed statistics are conditional on how the ensemble is selected." Those are related intuitions, but not the same ontology.

ME: But they are both observer-relative outcomes..

AI: Both make the observed correlations depend on the observer's context, but only Many-Worlds says the underlying reality literally splits into branches.

ME: I don't see how that makes a difference in practice.

AI: Those are mathematically similar ways of organizing outcomes, but they answer different ontology questions: one says "all outcomes exist," the other says "only this subset is being described." That difference matters when you ask questions like:

What counts as an explanation?

Are probabilities fundamental or just bookkeeping?

Is the apparent asymmetry in time real or emergent?

Is the theory saying something about reality itself, or only about the way we slice the data?

So you are right that the two can be indistinguishable at the level of predictions. The difference is mainly in interpretation and in how each view would extend to situations beyond the specific Bell setup, not in the immediate experimental numbers.