Machine learning limits?

We believe that Tesla’s electric cars run the most daring on-line machine learning experiment being conducted today. But in this article, which is about Goldman Sachs visiting the Tesla factory, it seems Elon Musk has grown a little more skeptical about machine learning. This is what is being said in the article:

“Goldman came out of the meeting feeling that Musk has become skeptical that machine learning can solve the harder problems involved with autonomous driving, saying that to be viable the system would need to achieve 99.999% accuracy, while he currently sees machine learning reaching 80% or 90% accuracy. Though Musk recently hired several machine learning experts to try to solve those problems.”

It is quite unclear what the numbers mean. But, I would think that a machine learning system that is “99.999% accurate” is not a machine learning system. Machine learning is not all there is to making self-driving cars make responsible choices; we need the help of good old symbolic AI for that (symbolic AI  is based on explicit knowledge representation instead of sub-symbolic algorithmic techniques).