(tl;dr : 'very very very few' is entirely relative )
I don't know if comparing frequencies is such a hot idea with regards to a paradigm shift.
If that's the comparison you want to make, the few AP accidents that have happened are infinitely more than came before Tesla.
Since there is no A/B comparison to make in order to judge severity, maybe we should treat every AP accident as important and research worthy until there is a standard to compare against -- unless of course you want to compare AP to non-AP crashes, but that metric is mostly only useful for marketing AP towards people, not improving the system itself.