One of these days, soon if not already, one of these AI’s will ‘escape’ control of it’s programmers and insinuate itself into the internet, like a huge virus that we cannot stop.
There was a sci-fi novel written late 80’s, IIRC, where the ‘Internet’ became so huge it became ‘self-aware’ and decided it would control everything..................
....Roko’s basilisk....
I am more concerned about the effects of hacking and viruses on AI controlled systems. It is one thing to program an intelligent feedback loop that corrects to achieve a more efficient path to an objective (very difficult as demonstrated by this research), it is another thing to harden those systems against the many various modes of deception that would cause it to breakdown or get hijacked (exponentially more difficult).