Hi, all! The Machine Intelligence Research Institute (MIRI) is answering questions here tomorrow, October 12 at 10am PDT. You can post questions below in the interim.
MIRI is a Berkeley-based research nonprofit that does basic research on key technical questions related to smarter-than-human artificial intelligence systems. Our research is largely aimed at developing a deeper and more formal understanding of such systems and their safety requirements, so that the research community is better-positioned to design systems that can be aligned with our interests. See here for more background.
Through the end of October, we're running our 2016 fundraiser — our most ambitious funding drive to date. Part of the goal of this AMA is to address questions about our future plans and funding gap, but we're also hoping to get very general questions about AI risk, very specialized questions about our technical work, and everything in between. Some of the biggest news at MIRI since Nate's AMA here last year:
- We developed a new framework for thinking about deductively limited reasoning, logical induction.
- Half of our research team started work on a new machine learning research agenda, distinct from our agent foundations agenda.
- We received a review and a $500k grant from the Open Philanthropy Project.
Likely participants in the AMA include:
- Nate Soares, Executive Director and primary author of the AF research agenda
- Malo Bourgon, Chief Operating Officer
- Rob Bensinger, Research Communications Manager
- Jessica Taylor, Research Fellow and primary author of the ML research agenda
- Tsvi Benson-Tilsen, Research Associate
Nate, Jessica, and Tsvi are also three of the co-authors of the "Logical Induction" paper.
EDIT (10:04am PDT): We're here! Answers on the way!
EDIT (10:55pm PDT): Thanks for all the great questions! That's all for now, though we'll post a few more answers tomorrow to things we didn't get to. If you'd like to support our AI safety work, our fundraiser will be continuing through the end of October.
As Tsvi mentioned, and as Luke has talked about before, we’re not really researching “provable AI”. (I’m not even quite sure what that term would mean.) We are trying to push towards AI systems where the way they reason is principled and understandable. We suspect that that will involve having a good understanding ourselves of how the system performs its reasoning, and when we study different types of reasoning systems we sometimes build models of systems that are trying to prove things as part of how they reason; but that’s very different from trying to make an AI that is “provably X” for some value of X. I personally doubt AGI teams be able to literally prove anything substantial about how well the system will work in practice, though I expect that they will be able to get some decent statistical guarantees.
There are some big difficulties related to the problem of choosing the right objective to optimize, but currently, that’s not where my biggest concerns are. I’m much more concerned with scenarios where AI scientists figure out how to build misaligned AGI systems well before they figure out how to build aligned AGI systems, as that would be a dangerous regime. My top priority is making it the case that the first AGI designs humanity develops are the kinds of system it’s technologically possible to align with operator intentions in practice. (I’ll write more on this subject later.)
Thanks! Could link there you will write about this subject later?