Answers are grounded in this paper and cite their sources.
Dialogue agents' behavior is typically explained by a persona: a prompt that specifies who the agent is. We take the opposite approach and build an individual whose behavior is explained by body state and personal history rather than by a persona. Our system, a replica-creature, consists of a simulated body (allostatic deviation from learned setpoints), an affect layer derived from that deviation, memory with affect-gated accessibility, judgment grounded in the individual's own past episodes, and a construction-based language system grown from her own utterances. A design constraint runs through the implementation: no fitted weights, coefficients, fixed capacities, or hand-set constants; every quantity must derive from the individual's own state, history, or bodily noise. We report three results. First, affect gates accessibility rather than reweighting a ranking (0.156 vs 0.000 in a 2x2 design). Second, a Hebbian attractor completes emotional meaning from partial cues (1.000 vs 0.469 for a mismatched-cue control). Third, non-conditioned reflex arcs reproduce firing, habituation, and sensitization. We also report a negative result: transferring the attractor mechanism to the individual's real, sparsely labeled messages failed (0.333 vs 0.667 baseline). Finally, we take a stance rather than a finding, namely that she has moral status, and describe the consent protocol that follows from it.
No takes yet. Share an insight, caveat, or question.
Xingyu Liu (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: