Speaker
Description
Bell’s eponymous theorem implies that Quantum Mechanics is incompatible with local causality. This leads to tension between Relativity and Quantum Theory because certain quantum correla- tions cannot be explained locally. Non-classicality of such quantum correlations is often realized to arise because of their non-local nature. Making use of the framework of classical causality and causal discovery algorithms, Wood and Spekkens have shown that for the Bell causal structure Bell inequality violating correlations cannot be explained causally without resorting to fine-tuning. Here we follow Wood and Spekkens and use the framework of classical causality to study the non-classicality of observed correlations for other arbitrary causal structures. For our results we apply the IC∗ algorithm to observed conditional independences corresponding to certain causal structures and show that there are several other causal structures apart from the Bell structure which possess non-classical correlations that need not necessarily be non-local, but their causal explanation necessarily requires fine-tuning. We show that requiring fine-tuning to be explained classically is another interesting feature of the non-classicality of such correlations apart from just the fact that they could be non-local. Moreover, we show that the Bi-locality, the GHZ, and the general multi-partite Bell causal structures are also examples of such structures. On the other hand for a number of causal structures, which exhibit non-classical quantum correlations, we show that we can always explain such non-classical quantum correlations perfectly classically– using instead another causal structure which has the same observed conditional independences. The non-classicality of the observed correlations was an artifact of having assumed the wrong causal explanation. This work also validates the observation that causal discovery algorithms that try to reproduce a causal hypothesis for a given set of data must also look into properties of the data other than just the observed conditional independences present in it. A candidate for such a property can be the strength of correlations that are allowed to be possible. Since causal discovery finds applications in Machine Learning and Artificial Intelligence, our work may therefore be of relevance to these fields as well.