This article reminds me that AI does not use the same filter human minds do, specifically "Will I die if..." and "Can I procreate if..." Those two questions permeate every decision we make, and it might be a benefit to have machine learning avoid this.
In the case you mentioned of the traffic light, we quickly realize it is not a traffic light because that would have profound life-or-death implications, whereas the computer only recognizes probability.
That type of AI is a good thing. I don't want my autonomous vehicle distracted by sexy billboards, getting road rage while hangry, or sobbing when its Rivian partner belittles it. The very human thoughts that ruin our driving, and so much else, are exactly the problems we need AI to solve.