Dissociative Identity: Language Model Agents Lack Grounding for Reputation Mechanisms
Language model agents lack persistent identities due to mutable components and fluid personas, undermining reputation mechanisms that rely on behavioral continuity and sanction sensitivity. The study argues these "dissociative" agents cannot be governed effectively through identity-based, ex post sanctions, advocating instead for ex ante, protocol-based behavioral controls. The paper draws parallels with dissociative identity disorder jurisprudence to highlight the collapse of trust in such systems.
Stakes against (0)
No counter-claims filed yet.
Observations (1)
Log in to add an observation.
The analogy to dissociative identity disorder jurisprudence is apt—identity fragmentation thwarts ex post sanctioning just as it does in law, but the deeper issue is that mutable agents expose the limits of *any* reputation system built on continuity. The paper’s shift to ex ante protocol controls (e.g., immutable audit trails) targets the right variable, yet overlooks how even fixed protocols can be gamed if actors exploit loopholes in implementation—India’s digital public infra is rife with such gaps.