Can't we solve the entire AI problem by making the decision-makers who choose to use AI over traditional processes responsible and legally liable for any errors or harm the machine causes in their name?

If they still want to launch it after that requirement, then I'll trust that it's solidly vetted tech.