Dr Stuart Armstrong, Senior Researcher, InhibOx
Abstract: Friendly AI is devoted to managing the transition to smarter-than-human AI’s in such a way that does not exterminate humanity – and, along the way, granting us unlimited lifespans, eternal bliss and empathy, universal meaning, and all the usual suspects.
The current approaches are interesting and productive, but suffer from the drawback of being either probably impossible or insanely dangerous. Here I suggest an incremental approach that reduces the hard problem of friendliness to the somewhat easier problem of trust, while providing a framework on which many other precautions can be easily added. Most of all, it gives a way of solving one of the eternal problems of philosophy: what sort of world do we actually want to see, in the end?