Because when the autonomous systems are trained on how to handle adverse conditions, its the sum of the knowledge of all machines that build how to handle these conditions. Whereas each human has to learn it individually. And given how bad people drive even with rainstorms, I'd put the money on the computers.
And frankly, we can also discuss if "driving in rain/sleet on frozen mountain pass" is even a good idea. Because computers can also provide data like "Estimated chance of wreck is X%", and decisions can be made using that.
Tl;Dr. One machine learns and can share with all nearly instantaneously. 1 human learns and that's it.