Are Waymo using AI and humans to build a ginormous curated decision tree that can be audited or are they doing something more black-box / pure AI / something else?

I always imagined a big decision tree that could get diff’d after each AI training run and tuned by an army of humans.

Why do you assume there's any "AI" in it at all?

There's some info on the systems here https://techcircle.vccircle.com/2018/05/09/how-google-s-waym...

>“While perception is the most mature area for deep learning, we also use deep nets for everything from prediction to planning to mapping and simulation.” ...Dmitri Dolgov, CTO and VP, engineering, Waymo

etc

Because this would fairly obviously be impossible without ML techniques.