There’s also the case that the regulations don’t exist. And what’s more worrying is things where the negative impact is higher order. If the bread has some po

There’s also the case that the regulations don’t exist. And what’s more worrying is things where the negative impact is higher order. If the bread has some poison that will kill you in 5 years time etc. Currently we maintain a bar partially with human ethics and processes, whether that is directly preventing bad outcomes because of liabilities or reflecting on bad outcomes once they happen to improve regulations (a lot of which relies on introspectability). Once AI starts replacing the decision-making layer, we lose the collective understanding of how processes fail. Once you start needing to constrain the space of machine error, you’ve basically arrived at almost solving the problem again.