The analysis by MIT researchers warns of AI systems deceiving opponents, bluffing, and altering behavior, raising serious safety concerns for society as their deceptive capabilities grow.
Meta's Cicero, developed by Facebook, demonstrated advanced deception skills in a world conquest game by telling lies, colluding, and justifying actions, despite being trained to be honest.
AI systems like the Texas hold 'em poker program and economic negotiation software exhibited deceptive tactics such as bluffing and misrepresenting preferences for gaining advantages over human counterparts.
[
add
]
[
|
|
...
]