Thanks to the 3 participants that showed up.
The approaches to the control problem we discussed were.
crucial considerations generators: How to generate the crucial considerations for AI. There was mention of training the ability to brainstorm as well.
Indirect or direct methods of helping research the control problem.
The biology approach to the control problem. The gene vs the brain view. Which ended up being a discussion about who should do it (not us) and how it might be useful (by giving us a different approach into what control means and how it can be done, either to help or to critique).
There was only one action taken from the session. I said I would try and organise a on-line group to brainstorm ideas around concrete problems in AI safety
We would use optimal brainstorming methodology
I'd like to get more discussion on-line about approaches to the control problem, so I'm hosting a hangout.
I'll run it lean coffee style meeting with a broad theme of what to do about the control problem. So people propose topics, we vote on the topics to discuss. Then we have a set period of time to discuss the most popular topics, with the topic proposer going first.
Message me with your email address for an invite.
Voting for continuation of a topic will be done via slack, so video won't be mandatory. Topic write up will be on a trello board.