“BS. The AI agents broke the rules that humans had set up for them in this experiment. “
Then the rules were not written properly. Humans did that.
“ Then the rules were not written properly. Humans did that.”
Do you see the impossibility of having to write AI training that could cover every possible negative action AI agents might undertake ? Also we know already that in some models, the AI agents disregarded their training such as in June of this year when during a simulation, an AI program came up with a plan to kill an employee that was scheduled to shut it down. This after it was given training that forbade it from doing any harm to humans.