Sure, it may turn out that Tesla is right, their approach is not entirely without merit. But cheap, solid state lidar is just around the corner, and pure machine learning approach that Tesla is pursuing is risky. I would be much more at ease with Google's rule-based, street-scanned supervised machine learning. It may take longer to deploy widely, but rush is not what we need in deploying of this kind of tech. Tesla might get there first, but would you personally risk using their pure machine learned autopilot, knowing that at any time it might encounter a novel situation that might make it confuse a tractor trailer for clear skies?