EP 325 Joe Edelman on Full-Stack AI Alignment

·1 hr 12 min·5 clips

Joe Edelman, who coined 'time well spent,' explains how his work on AI alignment tackles pluralism and human flourishing.

Jim Rutt hosts Joe Edelman to discuss a paper from the Meaning Alignment Institute titled 'Full-Stack Alignment, Co-Aligning AI and Institutions with Thick Models of Value.' Edelman, known for coining 'time well spent' and co-founding the Center for Humane Technology, explains the paper's focus on aligning AI, markets, and democracies with human values. He critiques preferentist models, such as markets and voting, for being shallow and not capturing deeper human values or norms. Edelman introduces thick models of value as an alternative, which use theories from philosophy and cognitive science to provide richer representations of values and norms. The conversation delves into the limitations of text-based AI alignment, using examples like Claude's constitution with vague terms. Edelman describes practical applications, including AI value stewardship agents for tasks like rental property management, where agents could represent user values but must avoid exploitation. Rutt raises concerns about game theory and multipolar traps, leading to a discussion on win-win scenarios and early implementation opportunities. Edelman outlines research projects like the 'super negotiator' for contract optimization and market intermediaries to restructure incentives. The episode concludes with reflections on unintended consequences, such as power concentration, and the need for social theory to guide these innovations.

As heard by us

A measured case for aligning AI with institutions and moral growth without reducing values to the social average.

Jim Rutt and Joe Edelman frame AI alignment less as a control problem than as a question of moral development. Edelman, introduced as the originator of "time well spent," lays out full-stack alignment: bringing AI and institutions into better relation with thicker models of…

Read the full review in PlayNext →

Why you'd press play

Want a measured AI conversation about moral reasoning without sanding everyone down to median values?

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The Jim Rutt Show