US Air Force colonel ‘misspoke’ about drone killing pilot

Maybe I am mistaken, but I had thought lot of the excitement around current AI is the emergent behaviors that nobody specifically programmed for?

And more! The interface between those who know objective truth, those who emit signals, and we who try to discern the former through the games of the latter, is now a state of permanent turbulence. When you say "testing out" that is even more true for the premise of sending up "trial balloons": you emit a message, observe the results, and then walk it back or obscure its veracity as needed. The actual truth is a footnote; the important thing is obscuring that from rivals (broadly understood as anyone who threatens the mission or at least the status quo, which has feature-crept to mean most of the public)...

In a case like this a cynic such as myself sees equal odds that what is reported in these two stories, is a decent rendition of the truth...

...and that the truth is elsewhere, e.g., the first statement was true; it was leaked either as a test or by actual misstep; and the machinery automatically kicked in to emit the second statement—because regardless of what is true, it is valuable for enemies to be unsure. Was this an accidental reveal of a capability far in advance of what the public knows? Or is this is an intentional bluff? Or...?

Who can say? None of us.

It bears saying,

the initial story was profoundly concerning in two different ways.

First because of the Cameron-ready simple existential threat posed by murderous paperclip maximizers, which most here are long familiar with.

But second, and much more interesting to me, and probably likewise to many here, is that the executive planning and scenario-finding implied by a a putative military AI, which is capable of concluding that taking the human out of the loop would best satisfy it,

...is suggestive of a type of AI very very different from what we see in the public sphere, or at least, what I have seen, following the horse race with some real attention.

That to me suggests that it's not even odds that this was a feint of some sort wrt AI and the foes of the West. As with e.g. well documented efforts to suggest to Cold War rivals that they were behind in the recovered-UFO-tech clandestine arms race, who knows what truth is in the basement, there is a lot of 9-to-5 life-and-death action and effort going on in causing sleepless nights over in the GRU and PRC.

How did the AI learn that it could prevent human override by killing the human operator? How did it then learn to destroy the COMMS tower so that it wouldn't be penalized for killing the operator.

Why was human feedback even part of AI training simulation? Why did the reward function in training include logic that says 'if the simulated comms tower is destroyed, do not penalize friendly fire'?

We can talk about hypothetical AGI all we want, but that has nothing to do with what us currently called "AI", and what will soon be just another chapter in the growing book called "machine learning", when we find a new marginal improvement to call AI.