>If you assume something will automatically happen, without a particular mechanism
He didn't mention it, but there are some pretty obvious mechanisms for this to be true:
1. Automated drivers don't get tired.
2. Automated drivers may "see" more of their environment than a human can. Human sensors aren't upgradeable.
3. Automated drivers don't get drunk.
4. Automated drivers don't have emotions.
5. Automated drivers aren't distracted by the infotainment system, the children in the back seat, or even some thought about what just happened at work.
6. With time, automated drivers can be improved, while human weaknesses stay constant.
There are of course drawbacks:
1. Humans have intuition about situations they've never been in. Most AI/machine learning still generalizes poorly.
2. The programming may be buggy.
3. The datasets may be inadequate.
So what will happen is definitely out there, but there are clear mechanisms for this to happen.