New Framework for AI Personality Clones Explores Identity

This paper introduces a new conceptual framework for understanding and evaluating AI "personality clones," moving beyond the philosophical "hard problem of consciousness" to focus on observable identity. The authors define identity based on the indiscernibility of an AI's manifestations as assessed by an external observer over time. They differentiate three key criteria for evaluating such clones: fidelity to the target person, generic human-likeness, and the AI's unique individuality. The framework proposes a six-term factorization of observed identity, encompassing substrate, dispositions, memory, update dynamics, context, and exogenous contingencies, using a state-space formulation. It suggests that indiscernibility can be quantified by a judge's ability to distinguish the clone from the original. A central conjecture is that "versionability" (the ability to create multiple versions) tends to degrade long-term indiscernibility. The paper also introduces the concept of a "delegate" – a task-limited, bounded-lifespan partial clone – as a third category alongside product-clones and individuals. It maps existing empirical literature onto its criteria and outlines an experimental program. Ultimately, it argues that the most appropriate long-horizon criterion for a personality clone is "climate fidelity," meaning the clone matches the conditional distribution of the original person's possible responses, rather than strict trajectory fidelity.