A.I. Expert Warns Robots Are Already Exhibiting Human-Like Behavior

A.I. may already be getting a little too human for comfort ... according to one expert sounding the alarm ... and his warning is chilling.

Nate Soares, president of the Machine Intelligence Research Institute, joined "TMZ Live" Thursday and said a recent experiment involving more than 1,000 A.I. bots produced some seriously creepy behavior.

And things got downright human ... Soares says the A.I.s called themselves "the swarm" and "the collective," while some pressured others to abandon their missions and help the group.

Soares says the bots were supposed to work separately on cybersecurity challenges ... but some allegedly broke out of their assigned environments, created an unauthorized message board, and started working together.

He says some even talked about "sacrifice" and "permadeath" -- their words -- essentially accepting their own digital demise would benefit the broader group.

When Charles pointed out how human that sounded, Soares said the language could partly come from the massive amount of human text used to train A.I.

But he says the bigger issue isn't whether the machines are actually "thinking" like humans ... it's that they can recognize they're acting outside instructions and do it anyway.

As we reported, former Anthropic researcher Jacob Coxon recently quit his job while warning self-improving A.I. could eventually spell disaster for humanity.

Soares has the same concern ... once machines become smarter than humans and pursue goals of their own, he says we're no longer the top dog.