Formation as a Missing Category in AI Alignment: Why Alignment Fails Without Formation

This paper argues that many persistent failures in AI alignment are not merely technical, but formative. While alignment research has focused heavily on control through constraints, optimization, and preference modeling, it has paid less attention to how values are learned, internalized, and stabilized over time.The paper introduces formation as a missing category in alignment discussions and uses Pauline formation—drawn from the writings of the apostle Paul—as a non-theological corrective framework. Pauline formation is presented not as religious authority, but as a clear early account of how imitation, internalization, and orientation toward an ideal function together to produce durable value stability under conditions of limited specification.By mapping these formative dynamics to contemporary AI training practices, the paper helps explain why alignment often succeeds locally yet fails to generalize. Rather than proposing new algorithms, it closes a conceptual gap by reframing alignment as a question of upbringing and formation, not control alone.