"Clearly, predictions can only be made probabilistically, but this is already true in standard physics. The fact that we can observe only one sample (our own universe) does not change the method qualitatively and still allows the refuting of models at given confidence levels. The key points here are the well known peculiarities of cosmology, even with only one universe: the observer is embedded within the system described; the initial conditions are critical; the experiment is “locally” irreproducible; the energies involved have not been experimentally probed on Earth; and the arrow of time must be conceptually reversed.

However, this statistical approach to testing the multiverse suffers from severe technical short cuts. First, while it seems natural to identify the prior probability with the fraction of volume occupied by a given vacuum, the result depends sensitively on the choice of a space-like hypersurface on which the distribution is to be evaluated. This is the so-called “measure problem” in the multiverse. Second, it is impossible to give any sensible estimate of fi. This would require an understanding of what life is – and even of what consciousness is – and that simply remains out of reach for the time being. Except in some favourable cases — for example when all the universes of the multiverse present a given characteristic that is incompatible with our universe — it is hard to refute explicitly a model in the multiverse. But difficult in practice does not mean intrinsically impossible. The multiverse remains within the realm of Popperian science. It is not qualitatively different from other proposals associated with usual ways of doing physics. Clearly, new mathematical tools and far more accurate predictions in the landscape (which is basically totally unknown) are needed for falsifiability to be more than an abstract principle in this context. Moreover, falsifiability is just one criterion among many possible ones and it should probably not be over-determined."