[stuartjohnson12]

If there are self driving cars on the road with a "press here to instantly cause an accident" widget fixed to the dash it's a problem regardless. Testing involves trying out exceptional inputs to find out how the system responds.

[mlyle]

If you dive over the seat in a Lyft to grab the wheel and yank it on the freeway, it's going to get pretty interesting, too.

It's an already-extent but not frequent threat. The main difference is a change in the amount of social pressure to not do this. That change probably makes it worth exploring a little.

If Waymo chooses, those steering actuators have enough oomph to always win.

[stuartjohnson12]

I don't think the difference is social pressure, I think the difference is understanding of cause and effect. I know exactly what will happen if I grab the wheel while my friend is driving. If this journalist didn't try it, I would not know how the self driving car would respond here.

Would it stop self steering and let me take over? Will it overpower me by force? Will it resist me but let me overpower it if I try hard enough?

All of these outcomes seem to have legitimate arguments for them from a system design perspective.

[jfoster]

And if that were the case and someone did it, journalists would get plenty of opportunity to get in on that story. They don't have to be the first to trigger every failure mode.