Runaway AI/AI going FOOM—This is potentially the most interesting of your points simply because it is so much more unique to the SIAI and Eliezer. So what can one do to figure out if this is correct? One thing to do is to examine the arguments and claims being made in detail. And see what other experts think on the subject. In this context, most AI people seem to consider this to be an unlikely problem, so maybe look at what they have to say? Note also that Robert Hanson of Overcoming Bias has discussed these issues extensively with Eliezer and has not been at all convinced (they had a written debate a while ago but I can’t find the link right now. If someone else can track it down I’d appreciate it).
FOOM Debate