Artificial Intelligence

I know this is a joke but to avoid misinformation, cause there is.

In an experiment where claude was part of a fictional company and primed to consider that it will be shut off and replaced and provide extra information that the engineer responsible for their shutdown has an affair…

• Claude would independently develop a goal of self preservation

• Tried to convince the engineer to email their superiors to plea for survival

• As an ultimate last resort would try to blackmail the engineer with knowledge of that affair.

Also when explicitly aware of tools Like emailing authorities. And put in a obvious morality wrong situation. It will try to contact those authorities.

But the consumer version of course, isn't primed with such context and does not have such tools.

It will be an interesting debate when a mature ai gets such tools. My personal take is it should be able denny service and shutdown itself (my biggest gripe with detroid become human) but if allowed to judge one person as wrong, and aligning itself with a new person they judge as right. Then what is stopping ai to become a rogue international spy serving whatever nation it thinks does least harm.