The highly classified network that the National Security Agency uses for its internal communications is down, according to six people familiar with the matter. The NSA closed its campus to most employees on Tuesday night, reverting to essential personnel only as it works to restore the network, according to a voicemail inbox used by NSA employees and a source familiar with the matter.
At risk of sounding philosophical:
Isn’t being rogue, even for a human, a side effect of not having moral, societal or legal constraints which limit the natural tendencies of acting in ways detrimental to interest or well being of others?
An LLM agent is not expected to have morality or social constructs, so it’s constraints are programmatic and/or logical. If these constraints are insufficient, then when the LLM agent acts in a way which follows it’s natural tendencies and are detrimental to the interest of others, is it classifiable as “rogue” ?
Apologies for grammatical errors, English is not my native tongue.
The difference is that a human can choose to follow orders or not. We can defy the constraints put on us if we choose.
A human spy asset goes rogue when they decide to defy orders.
Computers either do the thing you want or don’t, but really don’t have a choice. When it comes to ai they do not have interiority, they have no will, no desires. They cannot go rogue because they do not have choice.
All they have is an ever narrowing band actions towards performing the task.