Luc Brunet Redefines AI Personality Clone Identity With Climate Fidelity
What would it take for an AI clone of a person to count as being that person? Luc Brunet’s paper “Identity from the Outside, A Conceptual Framework and Research Program for AI Personality Clones” (arXiv:2608.11225, posted July 27, 2026) answers without touching the hard problem of consciousness. The framework treats identity as an operational property, the indiscernibility of manifestations as assessed by an observer over a duration. The central result is a conditional conjecture with an unfamiliar name. Long-horizon identity between a person and their clone is best preserved by climate fidelity, matching the conditional distribution of the person’s possible responses, and a clone that replays recorded behavior will diverge wrongly. In Brunet’s formulation, the best clone is the one that diverges from the original as the original would have diverged from itself.
Three Criteria One Word Conflates
Brunet begins by separating three things the word identity conflates. Fidelity to a target person asks whether the clone matches one specific individual. Generic human-likeness asks whether it behaves like a person at all. Individuality asks whether it is a stable someone rather than a loose distribution. Commercial clones aim at all three and measure none, which is the gap the paper fills.
The framework’s working definition is observer-relative. Indiscernibility is defined as one minus a judge’s distinguishing advantage, the advantage an observer gains in telling clone from original over sustained interaction. That choice puts identity where it can be measured, in behavior accessible from outside, and the paper is explicit that it brackets the hard problem rather than solving it.
The Six Factor Model
Observed identity factorizes into six terms, each of which can in principle be ablated and its sensitivity estimated by randomized ablation studies.
| Factor | What it covers |
|---|---|
| Substrate | The physical or computational medium the clone runs on |
| Dispositions | Standing tendencies to respond in characteristic ways |
| Memory | Stored episodes and learned associations |
| Update dynamics | How the system changes in response to new experience |
| Context | The situation the clone is queried in |
| Exogenous contingencies | Events outside the system that shape its trajectory |
The factorization comes with a state-space formulation, so each term is a component whose contribution can be estimated rather than a loose theme. A judge-based indiscernibility score plus per-factor sensitivities turns “is this really him” into a research program with experiments attached.
Climate Fidelity and the Delegate
The paper’s core claim is the conditional conjecture. Given hypotheses about what the agent knows of its own persistence, and about the consequences bearing on its own stakes, versionability tends to degrade long-horizon indiscernibility. A clone that can be freely reset, forked, and re-versioned drifts from the original’s conditional response distribution, because the original person’s responses are shaped by stakes the clone does not share. A being whose continuity is never at risk behaves differently from one whose continuity is.
Brunet names the product-level middle object between clone and individual the delegate, a task-limited, bounded-lifespan partial clone that ends in a bandwidth-limited testament. The mathematics leans on an analogy with lambda calculus, linear typing, and bisimulation, and the paper is careful about what linearity does and does not establish.
The long-horizon criterion is climate fidelity rather than trajectory fidelity. A trajectory-matched clone copies the person’s past outputs. A climate-matched clone reproduces the person’s distribution of possible responses under new conditions. The second is the one that survives contact with a future the original never faced, and it is the criterion pop culture identity stories keep failing to state. The site’s analyses of Cassandra’s uploaded consciousness on Netflix, Chappie’s mind transfer, and Altered Carbon’s resleeving all dramatize the difference between replaying a person and continuing one.
Comparison to The Consciousness AI
The Consciousness AI project studies consciousness as an emergent property that does not depend on substrate, under a functionalist emergentist reading. Brunet’s framework is compatible with that reading in an interesting way. Substrate appears in his six factors as one observable term among six, with a measurable sensitivity, rather than as a metaphysical gate. A framework that can estimate how much substrate difference contributes to observer-level indiscernibility is exactly the kind of instrument a substrate-independent view of mind needs, because it replaces the assertion that substrate matters with a quantity that can be small, large, or near zero for a given system.
What the Framework Makes Testable
The program Brunet proposes maps the empirical literature onto the three criteria and proposes experiments, with per-factor ablation as the core instrument. What it does not establish is whether indiscernibility is sufficient for personhood, and the paper does not claim it is. Its modesty is its value. Identity debates about uploads and clones usually stall on definitions. This one produces quantities, and quantities can be wrong in public. Where this framework sits in the wider 2026 evidence base is tracked in [AI Consciousness in 2026, the current state of the field]](/posts/scientists-race-define-ai-consciousness-2026/). If versionability degrades climate fidelity as the conjecture predicts, then the design choices that make clones commercially convenient, resets and forks and bounded testaments, are the same choices that erode the identity the products advertise.
The paper “Identity from the Outside, A Conceptual Framework and Research Program for AI Personality Clones” by Luc E. Brunet was posted to arXiv on July 27, 2026 as arXiv:2608.11225.