Kenji Urada

You know who the first guy was who was killed by a robot?

Some Japanese guy in a factory?

It was either the Japanese guy or a German guy in another factory, and the reason he got killed is because there was a robot arm that just went into where he was and crushed him to death, you know.

So even if that robot contained in it a vision system which uses a neural network and isn't... that you can't ultimately prove very much about, you could simply encase that neural network-based system in a scaffold that it can't break out of, and...

some traditional code saying outside of this that the robot arm may never have an x coordinate which is larger than 1.3 meters because that's where the human is right and then you can prove stuff about that that code and it doesn't matter now what the neural network does the robot is never going to be able to kill the human you could have similar things for self-driving cars and countless other very helpful technology where you end up with the assurance you want Because the thing which is acting directly on the world, unfiltered, is not a neural network.