Simplicity is arguably the deepest and most persistent
mystery in the philosophy of science. When several theories fit the data,
Ockham's razor enjoins us to choose the simplest. But how could such policy
possibly help us find the truth, for Ockham's razor is a *fixed* bias toward
simplicity, that can no more indicate truth than a broken thermometer fixed
on a particular reading can indicate temperature? Standard responses either
beg the question by assuming that the world is probably simple or urge wishful
thinking by recommending belief that the truth theory has some nice property
associated with simplicity. That is why the realism debate is still a debate.
I will argue in a novel way that both the realist and the anti-realist are right: Ockham's
razor helps us find the truth (realism) -- but without indicating what the truth is (anti-realism).
In fact, Ockham's razor helps us find the truth by minimizing the number of retractions or
scientific revolutions prior to converging to the truth. Moreover, any deviation from Ockham's
razor does worse, so Ockham's razor is *necessary* for retraction-optimal convergent performance.
Proving this demands a general definition of simplicity, which I will state. Time permitting, I
will sketch how these ideas apply to uniformity of nature, grue, counting things, curve fitting
with noise, statistical testing, causal inference, inference of quantum conservation laws, and
model selection problems in statistics.