Comment by Taylor Sorensen (taylor-sorensen) on The Intrinsic Interplay of Human Values and Artificial Intelligence: Navigating the Optimization Challenge · 2023-09-07T21:17:56.842Z · LW · GW

Fascinating post, Joe! We just published a research paper on modeling pluralistic human values, an I thought it might be relevant. Working with philosophers and cognitive scientists, we've tried to make a first attempt at concretely modeling pluralistic human values using language models. It is obviously imperfect, and assumes human values fixed in one point in time, but it is a computational attempt that, to our knowledge, no one has yet attempted.

Please let me know if you have any thoughts on our work and how it may relate to these thoughts, or if you'd like to discuss this sometime!
Paper: https://arxiv.org/abs/2309.00779
Demo: https://kaleido.allen.ai/