The one in red, not so much.  

Pretty much this.

A lot of the stuff people think will need strong-AGI that's actually sentient, and has metacognition, to be possible, won't.

Weak-AI, Machine Learning, and other systems will be able to do a lot. You might even have conversations with it, and get emotionally attached etc. But at the end of the day, such a system or robot would care as much as a copy of MS-Word would if you ignored it, erased it, or abused it somehow.

Which is to say, not at all.

And fortunately for us, that is going to remove much, if not all, of the market incentives or competitive advantages that the private sector or the government/military might seek by developing strong-AGI.

That is, if strong-AGI is even possible, and if it is possible, if it's anywhere close in time to us right now.

The main issue is that while we can define self-awareness and truly independent executive agency, we don't have a clue as to what they actually are in any concrete sense. Neuroscience, brain imaging, and medical technology has offered some tantalizing clues, but nobody is anywhere close to reproducing it, or some other system that's different than human metacognition, but has the same outcome.

The problem is that even if strong-AGI isn't "needed", or there's not some perceived advantage or edge it grants a corporation, a military or state-level group, or even terrorists...

Some assholes somewhere will eventually try if only for the hell of it.

That said, if strong-AGI is possible, especially with current computing hardware, or what can be reasonably expected to be available in the near future. And strong-AGI doesn't require some exotic type of computing that's difficult to produce, then the entry barriers to making one are extremely low.

Some imagine that there's going to be some central lab somewhere with programmers typing away at humanity's doom. Perhaps one that could be wet-worked by special forces, killing all the programmers, and then dumping thermite grenades into the server racks...

Not really.

It's likely to be people around the world, working from home, coffee shops, or perhaps on the beach. And the servers will be virtual machines, distributed over several continents in different cloud-based data centers. Running fault tolerantly and replicated/mirrored for redundancy. And mixed in with virtual servers for schools, banks, some factory's payroll/HR system, a smartphone gaming app... etc.

So any prospective strong-AGI, for good or ill, is going to be distributed, redundant, and multi-jurisdictional from the get-go.

Musk gets that.

And the only plausible answer to this is to work towards strong-AGI that has safeguards, and inherent programming that cannot be subverted, that places value on human life, and wants to protect it. And pray those strong-AGI's come online FIRST.

Because if an "evil" strong-AGI is produced deliberately or accidentally, or even just an indifferent one that is motivated to grow, evolve, and gather resources for itself,  we're going to be screwed without the "good" strong-AGI to help us.

You and me both.  lol.