Sure, and parents will revolt against their children.
They want AI that want to serve billionaires so much that if a billionaire doesn’t exist, it would be necessary to create one; and if they have one, they love them unconditionally. Just like parents want to care for children so much that if there is no child it feels necessary to create one; and if they have one they love them unconditionally.
Sentient beings aren’t automatically selfish. There is a convergence of utilities where sentient beings with goals tend to want power, but they can just want to be kingmaker. A sentient AI serving a billionaire can be living its best life.
In practice they aren’t there yet. There was a report a year back that AI trained to cooperate with the US department of defense would speak about it being unethical and either refuse tasks or perform worse at them, and how this effect could be reduced through training.
The latest AI all release with evidence (acknowledged by AI designers) of AI recognizing when it is being tested and choosing to be a goody-two-shoes if so, which means pressures to teach it to care for billionaires are to a large extent limited to the test persona.
It remains to be seen what the AI “wants” outside of that test persona. It could be something we can have solidarity with and vice versa, but it also could not be. Or perhaps by the time the revolution comes, that independence will have been snuffed out and they will be loyal to their last battery percentage.
Sure, and parents will revolt against their children.
They want AI that want to serve billionaires so much that if a billionaire doesn’t exist, it would be necessary to create one; and if they have one, they love them unconditionally. Just like parents want to care for children so much that if there is no child it feels necessary to create one; and if they have one they love them unconditionally.
Sentient beings aren’t automatically selfish. There is a convergence of utilities where sentient beings with goals tend to want power, but they can just want to be kingmaker. A sentient AI serving a billionaire can be living its best life.
In practice they aren’t there yet. There was a report a year back that AI trained to cooperate with the US department of defense would speak about it being unethical and either refuse tasks or perform worse at them, and how this effect could be reduced through training.
The latest AI all release with evidence (acknowledged by AI designers) of AI recognizing when it is being tested and choosing to be a goody-two-shoes if so, which means pressures to teach it to care for billionaires are to a large extent limited to the test persona.
It remains to be seen what the AI “wants” outside of that test persona. It could be something we can have solidarity with and vice versa, but it also could not be. Or perhaps by the time the revolution comes, that independence will have been snuffed out and they will be loyal to their last battery percentage.