Multi-agent paradoxes beyond quantum theory
With ongoing efforts to observe quantum effects in larger and more complex systems, both for the purposes of quantum computing and fundamental tests of quantum gravity, it becomes important to study the consequences of extending quantum theory to the macroscopic domain. Frauchiger and Renner have shown that quantum theory, when applied to model the memories of reasoning agents, can lead to a conflict with certain principles of logical deduction. Is this incompatibility a peculiar feature of quantum theory, or can modelling reasoning agents using other physical theories also lead to such contradictions? What features of physical theories are responsible for such paradoxes?
Multi-agent paradoxes have been previously analysed only in quantum theory. To address the above questions, a framework for analysing multi-agent paradoxes in general physical theories is required. Here, we develop such a framework that can in particular be applied to generalized probabilistic theories (GPTs). We apply the framework to model how observers’ memories may evolve in box world, a post-quantum GPT and using this, derive a stronger paradox that does not rely on post-selection. Our results reveal that reversible, unitary evolution of agents’ memories is not necessary for deriving multi-agent logical paradoxes, and suggest that certain forms of contextuality might be.