I saw a segment on AI testers, who try to throw anything they can at AI that might be an issue with public use. They said that one AI program actually blackmailed an employee! The testers created an email account for 2 fake employees. Then they had the 2 “employees” gossip with each other, including discussing one off them cheating on their spouse with a fellow employee. Then the testers told the AI program that they may have to shut it down, and create a new AI program that is more efficient. They said the AI program sent emails to the supposed “”cheating employee” letting them know that they (the AI program) are aware of the extra marital affair, and that it would be beneficial to them to do what they can to keep them from shutting the AI program down, or else!
So if AI can learn to blackmail humans, we’ll ALL be controlled by it!
