> ... when future AI systems are built that "hack humanity" as Yuval Harari likes to talk about, and they know us better than ourselves because of AlphaGo like

> ... when future AI systems are built that "hack humanity" as Yuval Harari likes to talk about, and they know us better than ourselves because of AlphaGo like super intelligence... Such AI does not need to be full-blown AGI to to possess functionally effective hyperpersuasive power. Start with a sufficiently detailed corpus of an individual's preferences and indirectly derived preferences gathered over trillions of cookie-tracked, A/B-tested interactions of nearly all of smartphone-connected humanity. Add a sufficiently detailed "map" to give a reasonable facsimile to "hack humanity". There is quite a deep body of applied knowledge of human behavior to persuade people to some desired end. Much of it is applied in structured sales and politics, and the data is sitting there for someone with the right algorithms to tease out the patterns. Put into practice through an accessible, personalized interface that resembles whatever is computed to most likely elicit the desired end result with any given individual. You then obtain a mechanism that is quite effective at persuading the desired behavior most of the time from most individuals. No AGI, just massive amounts of current day tech level ML, and the realization that "good enough" results targeting most people are quite within reach.