Musk vs. Zuck AI debate

The confusion here is a matter of timing. The key take away here is that as the system becomes more intelligent it also becomes more capable of affecting more change and hence more dangerous.

Short term - Zuck is right: AIs are likely to be narrow domain and unlikely to cause extinction level events or massive loss of life without some stupidity on the part of human systems designers. (E.g. Putting it in charge of ICBMs)...

Long term - Musk is right: If we assume that a recursively self improving system is possible to build at some point (a decade, a century...), then we have to consider the potential risks (and benefits) of such a system as a limiting case. The key assumption is that this system will be able to affect change in the real world on its own without human intervention, including making improvements to itself.

Benefits and risks include the very outlandish possibilities explored by various transhumanist thinkers, basically whatever is not limited by the laws of physics. (immortality, post human / post singularity existence, uploading, etc). Downsides include extinction, eternal enslavement, destruction of the planet, solar system etc.