AI Alignment and Fiduciary Obligation
Benjamin Lange reads alignment as a fiduciary obligation the developer owes the user, built from four classic duties.
Benjamin Lange's new paper reframes the whole alignment question. I think he's right.
Alignment becomes a fiduciary obligation the developer owes the user, with real criteria pulled from four old duties: loyalty, care, good faith, and candour.
One line jumped out: under the duty of care, staying ignorant of harm you could have caught is itself the breach. That moves the alignment debate onto legal ground, which is a healthier place to have it.
AI Alignment and Fiduciary Obligation
Reframes AI alignment as a fiduciary obligation developers owe users, deriving alignment criteria from loyalty, care, good faith, and candour.