Aligning AI systems with human interests.

Nov 30, 2015 · I think that AI alignment research should focus on building scalably aligned versions of contemporary systems — i.e. ML systems that are just as competent as unaligned systems, but …

Apr 7, 2018 · The “alignment problem” is the problem of building powerful AI systems that are aligned with their operators. This is significantly narrower than some other definitions of the alignment …

Apr 26, 2021 · I’m now working full-time on the Alignment Research Center (ARC), a new non-profit focused on intent alignment research. I left OpenAI at the end of January and I’ve spent the last few …

Nov 19, 2016 · I’ve claimed that prosaic AGI is conceivable, that it is a very appealing target for research on AI alignment, and that this gives us more reason to be enthusiastic for the overall tractability of …

Feb 23, 2016 · ALBA: An explicit proposal for aligned AI In this post I propose an explicit procedure for aligning powerful learning systems with their users’ interests. The goal is to introduce minimal …

Feb 6, 2017 · In the first half of this post, I’ll discuss three research directions that I think are especially promising and relevant to AI alignment: Reliability and robustness.

I’m tentatively moving towards the term “AI alignment” as a substitute for what is called “AI control” here, though with a slightly different meaning that better captures my work.

Apr 30, 2021 · So I like to focus on a more precise special case of alignment: solve alignment when decisions are “low stakes.” I think this case effectively isolates the problem of “find a good objective” …

Dec 5, 2015 · My goal is to find scalable approaches to AI control that can be applied to existing AI systems. For now, I think that act-based approaches look significantly more promising than goal …