> Most physical world thiigs have immediate feedback of incorrectness. We might not want it to enter such circumstances where there is such feedback - e.g. car

> Most physical world thiigs have immediate feedback of incorrectness. We might not want it to enter such circumstances where there is such feedback - e.g. car has run over a child. We generally aren’t interested in what the AI does in response to such feedback as such feedback should never ever occur.