MIRI’s three research objectives are, at present: •highly reliable agent design: how can we design AI systems that reliably pursue the goals they are given? •

MIRI’s three research objectives are, at present: •highly reliable agent design: how can we design AI systems that reliably pursue the goals they are given? •value learning: how can we design learning systems to learn goals that are aligned with human values? Not exactly what I meant. What are these human values (for humans not robots) and how do you prove they are rational and scientific? Their goal is to design AI that will accept human goals/values without defining a rational basis for those human values.