Steven Byrnes proposes aligning AGI using human-like social drives

Steven Byrnes, a researcher focused on brain-like AGI safety, published a comprehensive 51-minute read on the AI Alignment Forum titled 'Notes on technical alignment via human-like social drives'. He argues that if humans can steer toward a good future, sufficiently human-like AGIs could too—provided they have good prosocial motivations. The challenge is writing code that yields such motivations.

Byrnes examines human innate social drives as a template, discussing three key failure modes: the AGI developing the wrong 'moral circle', norm-following collapsing without a balance of power (relevant to superhuman AI), and consequentialist desires overriding norm-following. He also explores two pathways for how virtues become internalized: 'person-first' (adopting traits of admired individuals) and 'desire-first' (deep personal passion shaping identity). The post is a raw dump of interconnected ideas, calling for scrutiny and deconfusion from the community.