AI Research Papers

Provides a formal, quantifiable model and experiments showing how persistent, measurable self-identity can be instantiated in AI.
Supplies a formal, testable link between persistent memory structure and a stable AI self—specifies conditions under which an AI could sustain a coherent narrative identity.

Provides empirical methods to quantify persona stability and exposes gaps where models fail to maintain coherent identity over extended interactions.

The paper argues that current LLM-based AI systems lack true identity persistence because each interaction is stateless, causing failures in long-term coherence and continuity. It introduces the Narrative Continuity Test (NCT), which evaluates whether an AI remains the same interlocutor over time using five axes: memory, goal persistence, self-correction, stylistic stability, and persona continuity. The authors show that existing systems systematically fail these criteria and call for a shift from performance benchmarks to persistence-oriented evaluation and architecture design.

Supplies technical architectures and protocols for persistent, verifiable identities for autonomous agents in multi-agent ecosystems.