#Meta's Diplomacy #AI agent shows both the promises and pitfalls of the ability of #AI agents to mental model humans. We know that lying is a natural extension of the ability to model humans, and studied how humans handle machines telling them (white) lies back at 2019 @AIESConf
#Meta's Diplomacy #AI agent shows both the promises and pitfalls of the ability of #AI agents to mental model humans. We know that lying is a natural extension of the ability to model humans, and studied how humans handle machines telling them (white) lies back at 2019 @AIESConf
@tdietterich @AIESConf Intentional deception involves mental modeling, whether explicit or implicit..
@tdietterich @AIESConf I am more with Bratman (..as long as we are doing proof by eminent citations ;-). Srsly though, I am not sure what line you are drawing. Me having a wrong belief about you--that you didn't cause me to have--is not considered deception on your part afaik..
@tdietterich @AIESConf The one I read of Bratman is the monograph "Intentions, Plans and Practical Reason" (but it has been a while..) Back to deception, we do say things like "that problem was deceptively simple" etc., but there we are attributing it to our own misjudgment IMHO..
@tdietterich @rao2z @AIESConf Another wrinkle in this discussion is that it looks like there is an explicit objective that messages convey the current (partial) plan: x.com/ml_perception/…
@tdietterich @rao2z @AIESConf Another wrinkle in this discussion is that it looks like there is an explicit objective that messages convey the current (partial) plan: x.com/ml_perception/…