Corrigibility as a Singular Target: A Vision for Inherently Reliable Foundation Models
This addresses the core alignment problem for AI safety, potentially preventing existential catastrophe by ensuring foundation models remain tool-like and responsive to human guidance.
The paper tackles the safety challenge of foundation models potentially losing human control as capabilities scale, proposing a paradigm shift to design models with an overriding objective of empowering human principals to guide and correct them, aiming to prevent misaligned instrumental convergence.
Foundation models (FMs) face a critical safety challenge: as capabilities scale, instrumental convergence drives default trajectories toward loss of human control, potentially culminating in existential catastrophe. Current alignment approaches struggle with value specification complexity and fail to address emergent power-seeking behaviors. We propose "Corrigibility as a Singular Target" (CAST)-designing FMs whose overriding objective is empowering designated human principals to guide, correct, and control them. This paradigm shift from static value-loading to dynamic human empowerment transforms instrumental drives: self-preservation serves only to maintain the principal's control; goal modification becomes facilitating principal guidance. We present a comprehensive empirical research agenda spanning training methodologies (RLAIF, SFT, synthetic data generation), scalability testing across model sizes, and demonstrations of controlled instructability. Our vision: FMs that become increasingly responsive to human guidance as capabilities grow, offering a path to beneficial AI that remains as tool-like as possible, rather than supplanting human judgment. This addresses the core alignment problem at its source, preventing the default trajectory toward misaligned instrumental convergence.