News
Reminder: Guest Lecture today at 4 pm
Written on 25.04.2024 15:29 by Sarah Sterz
Hello everyone,
I just wanted to give you a quick reminder of today's guest lecture. Elijah Millgram is a very proficient philosopher and usually gives awesome talks that are very entertaining. He even took the time to write an alternative abstract for his talk for you (see below).
Title: Instrumentalism, Moral-Theory Overlays, and the Control Problem
Speaker: Elijah Millgram, professor of philosophy from the University of Utah
Date: today at 4pm at the
Place: German Research Center for Artificial Intelligence in building D3 2 (that's next to the Mensa), room Reuse HG 2.17.
Abstract (normal): How do we make sure that the much more capable AI agents of the future behave themselves and don't get out of line? The default response to worries about alignment and the control problem overlays constraints on AI decision making processes which are normally adapted from some familiar moral theory. Examining Stuart Russell's representative proposal makes it clear that recommendations of this kind will not work, and shows why it is a mistake to see the underlying issue as a question of control.
Abstract (fun): You know, evolution tried to build its theory of what we're supposed to be doing into the wetware, and look what *we* did with it: ultraprocessed junk food, rom-coms, and don't forget the bichon frisees and shih tzus. Why think our own attempts to build moral guidelines into future AI are going to fare any better? We should be taking a completely different approach, where our objective is not to *control* the AI agents we build, but to make healthy relationships with them possible, by ensuring that we can get a divorce when we want one.
The lecture is optional, but it will for sure be interesting! The event is public, so if you know anyone else who wants to attend, you can bring them along. The lecture is not recorded or streamed.
I hope to see many of you around!
Sarah