![]() “In inverse reinforcement learning, a machine infers a reward function from observed behavior, as a possible basis for a mechanism for learning human preferences.” Alas, Russell's ‘principles’ are vague to a degree of utter uselessness, even more so than Asimov’s famed Laws of Robotics. The ultimate source of information about human preferences is human behavior. The machine is initially uncertain about what those preferences are.ģ. The machine's only objective is to maximize the realization of human preferences.Ģ. ![]() Alas, “Russell then proposes an approach to developing provably beneficial machines focusing on 'deference to humans.” His 3 Principles:ġ. despotic nations or Wall Street trading firms who train AI servants to be parasitic, amoral and insatiable. (Note the word ‘should’ we’ll get back to that.) ‘misguided’ motives by researchers and companies is a valid concern, especially when the orgs are inherently secretive and aggrandizing e.g. ![]() In Human Compatible, computer scientist Stuart Russell asserts that defining AI success by rigid human-specified goals is dangerously misguided. ![]()
0 Comments
Leave a Reply. |