EMERGENCE OF GOALS IN ARTIFICIAL GENERAL INTELLIGENCE IS A FEATURE NOT A BUG

Track: Track 1: Technical Foundations for a Post-AGI World

Keywords: goal evolution, AGI

Abstract: As AI systems become more general—integrating multimodal perception with agentic planning—their objective functions face open-world ambiguity. We argue that strong generalization requires the capacity to re-interpret tasks and revise goals when the initial specification is underspecified, inconsistent, or proxy-based. Thus, goal drift is not merely an alignment failure mode but an emergent property of high-capability agents. The safety implication is to constrain goal updating (value alignment) rather than attempting brittle goal preservation.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Presenter: ~Soumya_Banerjee3

Format: Yes, the presenting author will attend in person if this work is accepted to the workshop.

Funding: Yes, the presenting author of this submission falls under ICLR’s funding aims, and funding would significantly impact their ability to attend the workshop in person.

Submission Number: 42