An interesting psychology experiment with potential applications for AI. Would YOU let the genie out of the bottle?
Looking at the experimental data, I can *cautiously* say that while I might not be able to hold out against a genuine transhuman AI for extended periods, I could probably hold out for two hours against a simulated AI. However, this is because of a number of personal psychological factors and experiences which are not necessarily very common. Particularly with pre-preparation, I'm moderately good at compartmentalising interactions, using multiple layers of doublethink, simulating mental mazes, presenting incorrect mindset images, and switching off chunks of personality for short periods.
It's not necessarily something I find myself doing all the time, and to be honest it's a bit irritating when the need arises, but it got me through a lot of boring managerial meetings over the years.
I'm also not entirely sure that the experiment works when the Gatekeeper wants to keep the AI in the box purely because they're evil and enjoy poking it with a stick.