While the Planck data is in excellent agreement with the predictions of single-field inflation, our incomplete knowledge of high-energy particle physics naturally leads us to consider the possibility of multiple scalar fields contributing to inflation.  In these scenarios the pre-inflationary dynamics can be chaotic, increasing the sensitivity of the onset of inflation to the initial data even in the homogeneous limit.  In this talk we will investigate the initial conditions problem for multifield inflation and examine how our ``a priori'' belief in the fields' background initial state affects the fraction of initial conditions that eventually inflate.  We will also discuss how one's choice of allowable initial conditions changes both the probability distribution for observables and the posterior probability for a model once confronted with data.