I think even the imperfect sensor data is enough to beat a human. My main argument for why self-driving cars will eventually be objectively safer than the best human drivers (no comment about whether that point has already done) is this:

A human can only look at one thing at a time. Compared to a computer, we see allow, think slow, react show, move slow. A computer can look in all directions all the time, and react to danger coming from any of those directions faster than a human driver would even if they were lucky enough to be looking in the right direction. Add to that the fact that they can take in much more sensor data that isn't available to the driver or take away from precious looking-at-the-road time for the driver to know, such as wind resistance, engine RPM, or what have you (I'm actually not a car guy so my examples aren't the best). Bottom line: the AI has a direct connection to more data, can take more of it in at once and make faster decisions based on all of it. It's inherently better. The "only" hurdles are making it actually interpret its sensors effectively (i.e. understand what cameras are seeing) and make good decisions based on this data. We can argue about how well either of those are in the current state of the technology, but IMO they're both good enough today to massively outperform a human in most scenarios.

All of this applies to an AI plane as well. So my money is on the AI.