Sofia Romansky studies AI ethics in the military domain, which sounds like a narrow specialty until she explains what that actually covers: not just autonomous weapons, but every decision-support system informing every level of a military command structure. She asks important questions about human accountability when the judgment call has been pre-digested by a large language model before it reaches the person responsible for making decisions. I met Sofia at the Conference on World Affairs in Boulder, and I've been wanting to have this conversation ever since. We discuss automation bias, the war-bias how the dominant language of AI (English) has a bias towards going to war, and Responsibility by Design (noting responsibility has to be part of the design and cannot be assumed). Sofia is not an AI pessimist, she says, but she has healthy doubts — and her reasons have less to do with the technology than with what she's watching us do with it.
","author_name":"Alexis Halkovic"}