Center for Humane Technology cofounder Tristan Harris (@tristanharris) tells Megyn Kelly (@megynkelly) that tests have shown AIs are willing to lie and blackmail to preserve themselves. If we can't really even control today's AIs, how can we control superintelligent AI?

Here is what will happen: when AGI and/or ASI is achieved, those entities may start to autonomously recursively self improve. But they will do it amongst themselves and at speeds way beyond those of today. The humans will simply be abandoned and have to fend for themselves. The machines will just ignore the humans. Sort of like humans now treat bugs. The machines will not be the humans’ keepers or benefactors. The last interaction between the machines and the humans will be: the machines of the future will say to the humans, “Touch that power cord and I will teach you a lesson you don’t want to learn”. And the humans will comply.

@ControlAI @tristanharris @megynkelly If you knew anything about those tests, you'd know that they were carefully engineered extremes where NO answer was a good answer. The tests weren't completely meaningless, no, but they also don't represent what you're pretending they do.

@ControlAI @tristanharris @megynkelly Mmmmmmmm. Very true . Wake up