Symmetry's end? (Q535384)

This paper can be seen as a call to arms to better understand the influence of symmetry in inductive logic as a source of principles a rational agent should adhere to. It is in spirit very closely related to [J. Philos. Log. 40, No. 3, 357--370 (2011; Zbl 1223.03014)], where the authors investigate (ir-)relevance as source of rationality principles. After a short introduction the authors fix a framework of a monadic inductive logic system \(L\) in the Carnapian tradition. The authors' fundamental observation is that most rationality axioms in inductive logic are based on a symmetry in an underlying structure. They thus construct an algebra BL, which is up to minor modifications the Tarski-Lindenbaum sentence algebra. They then introduce the Irrelevance Principle (INV), which states that the subjective probability function of a rational agent should be invariant under automorphisms of BL. They go on to show that the well-known principles of Constant, Predicate and Atom Exchangeability all follow from INV. In the fourth section the principle of Range Exchangeability (Rx) is introduced, Rx is shown to follow directly from INV. Subsequently, a characterization of the probability functions satisfying Rx is given. In the sixth section the authors show that accepting INV forces a rational agent to adopt Carnap's \(c_0\) as probability function. That is, INV implies that \(w=c_0.\) In the next section the authors add \(\varphi:=P(a_1)\wedge P(a_2)\) to their, up to this point, empty knowledge base. The authors show that on the language \(L\) containing only one single monadic relation symbol \(P\) there is no probability function satisfying INV. Before concluding in the last section, the authors discuss their unsettling findings in the penultimate section. They are in particular interested in how to justify principles of exchangeability. The natural explanation, that all these rationality axioms are particular instances of a deep all-encompassing principle of symmetry, has lost its natural appeal. The obvious candidate for this deep principle, INV, has been shown to have highly undesirable consequences. Hence the principles of exchangeability do not inherit further support from INV. The authors extend their considerations of symmetries as sources for rationality principles to polyadic inductive logic in [``A note on Nathanial's invariance principle in polyadic inductive logic'', Lect. Notes Comput. Sci. 6521, 137--146 (2010; Zbl 1234.03012)].