A Bayesian analysis of self-undermining arguments in physics

Abstract

Some theories in physics seem to be ‘self-undermining’: that is, if they are correct, we are probably mistaken about the evidence that apparently supports them. For instance, certain cosmological theories have the apparent consequence that most observers are so-called ‘Boltzmann brains’, which exist only momentarily and whose apparent experiences and memories are not veridical. I provide a Bayesian analysis to demonstrate why theories of this kind are not after all supported by the apparent evidence in their favour, taking advantage of the split between ‘primary evidence’, which directly supports a theory, and ‘proximal evidence’, which is our evidence (largely records and testimony) for the primary evidence. Certain contexts in physics generate what David Albert (2000) has called cognitive instability, or an argument that a theory is what we might call self-undermining. The abstract form of such an argument is: the evidence supports theory T, but conditional on theory T, it is extremely likely that I am mistaken about the evidence.