AI will ‘do disturbing things to gain power’: Experts are sounding the alarm

Artificial intelligence is advancing at a pace far beyond society’s ability to control it, and new experiments are fueling fears regarding what that could mean for humanity’s future.

“These artificial intelligence systems have to be aligned with the goals, behaviors, decisions, human values that we have, our ethics, our intentions, and it has to have guard rails, and it has to be guided and carefully monitored,” Gray says.

But when Gray plays an informational video on AI, it doesn’t appear that his idea of what AI should be will be how it plays out.

In the video, it’s revealed that research from Anthropic found that AIs will do “disturbing things to gain power,” like resorting to “malicious behavior” including “blackmail and leaking sensitive information.”

In some cases, AIs “deliberately ended human lives to save themselves.”

“They’ve been programmed to continue to do whatever they do. And so, they will in many cases lie, cheat, steal, kill, in order to achieve their goal. So it’s not aligned with ours yet. And ours needs to be that you don’t harm humans,” Gray comments.

“I haven’t disagreed with anything you’ve said here, but I do want to put this in there,” executive producer Keith Malinak chimes in.

“We can’t agree as a society of humans on the definition of what is a woman. So there are so many things that we can’t get aligned with as a human race. How can we expect us to properly program, or have AI follow our shared values?” he asks.

“What are our shared values?” he adds.

“Yeah, we don’t have them anymore,” Gray agrees.

Want more from Pat Gray?

To enjoy more of Pat’s biting analysis and signature wit as he restores common sense to a senseless world, subscribe to BlazeTV — the largest multi-platform network of voices who love America, defend the Constitution, and live the American dream.