It's a really dangerous precedent that we say "a rogue computer did X". Especially when the police does so.
We don't say "a rogue knife stabbed someone", or "a rogue scalpel botched a surgery". There's a human behind this, who set everything up. They might have lost control at some point, but when someone loses control while driving, we don't say that "a rogue car crashed into some kids" and call it a day. When a surgeon "loses control of their scalpel" we question whether they're qualified to perform surgeries.
The person responsible who lost control is still responsible for losing control, and it's dangerous precedent that we pretend otherwise for these cases.
> We don't say "a rogue knife stabbed someone", or "a rogue scalpel botched a surgery".
Well, yeah. It's almost as if LLMs are different from knifes.
> The person responsible who lost control is still responsible for losing control.
That's valid, and yet one can advocate for AI companies to be culpable for actions of their agents, while still admitting that rogue AI agents are a thing that exists.
It is different perspectives on the same thing I guess. On a global statistical level it is risk = impact x likelihood . On a local level your controls need to be appropriate for the risk. If accepting those risks globally goes without a consequence and companies accept a huge risk appetite, it is time for law enforcement and policy making to correct this in the interest of third parties impacted.
We don't say "a rogue knife stabbed someone", or "a rogue scalpel botched a surgery". There's a human behind this, who set everything up. They might have lost control at some point, but when someone loses control while driving, we don't say that "a rogue car crashed into some kids" and call it a day. When a surgeon "loses control of their scalpel" we question whether they're qualified to perform surgeries.
The person responsible who lost control is still responsible for losing control, and it's dangerous precedent that we pretend otherwise for these cases.
Well, yeah. It's almost as if LLMs are different from knifes.
> The person responsible who lost control is still responsible for losing control.
That's valid, and yet one can advocate for AI companies to be culpable for actions of their agents, while still admitting that rogue AI agents are a thing that exists.