What I learned at the AI Safety Europe Retreat
From the 30th March - 2nd April 2023, I attended the Artificial Intelligence Safety Europe Retreat (AISER) in Berlin, Germany. There were around 70 attendees from all over Europe. Most attendees were actively working on technical AI Safety (e.g SERI-MATS scholars, independent researchers with grants), some people were focusing on...
That would be neat! I'm overall quite excited about this approach, even though there are quite a few details to iron out. My main skepticism (as harfe pointed out before) is indeed how to specify the things we care about in a formal format which can then be formally verified. Do you know of any ongoing (or past) efforts which try to convert natural language specifications into formal ones?
I've heard of formal verification efforts in NASA where they gather a bunch of domain experts who, using a standardized template, write down the safety specifications of a spacecraft. Then, formal methods researchers invented a logic which was expressive enough to encode these specifications and formally verified the specifications.