This session is hosted in person and online.
Algorithmic amplifiers like social media and AI chatbots play an increasingly important role in how our society operates. But despite being built with noble intentions, these algorithmic systems have demonstrated a fundamental misalignment with their users’ values: resulting in the amplification of low-quality information designed to reward hack and sensationalize. In this talk, I’ll discuss new ways to measure and steer algorithmic systems to align them with users’ values. In a first series of lab and field experiments in the context of sharing misinformation on social media, I show that this can be mitigated by steering behavior via reminding people of their own values. In a second project, I introduce a novel personalization architecture for measuring the expression of basic human values. I will then apply it to a large-scale audit of X's ranking algorithm to reveal how subtle value tensions in users' engagement behavior produces a fundamental misalignment between algorithmic amplification and user's self-stated values. To close, I will discuss ongoing work on aligning AI systems for collective goals of diversity and autotelicty, focusing on the domain of human creativity and creative production.