Abstract

This paper explains the scientific rationale and research value of the SICS Human-State Proxy Benchmark Track. The central claim is that AI should not be evaluated only by what it produces. It should also be evaluated by what it leaves in the human being. The paper argues that conventional AI evaluation metrics such as accuracy, speed, benchmark performance, productivity gain, computational efficiency, and user engagement are necessary but incomplete. As AI systems become conversational, persuasive, adaptive, emotionally expressive, and embedded in daily life, a second evaluation axis becomes necessary: measurable human-state impact. The SICS Human-State Proxy Benchmark Track is proposed as a research-stage, non-diagnostic, non-therapeutic, non-clinical, non-surveillance, non-coercive benchmark support track for Human-State-Aware AI Interaction, Human-State Cost, multimodal proxy benchmarking, measurement-layer simplification, and future Sal-Meter A/B comparison. The paper introduces Human-State Cost as a non-diagnostic benchmark construct for comparing measurable proxy burdens left during or after interaction with an AI system. Human-State Cost is not a medical score, psychiatric score, clinical score, consciousness score, CAIS output, Sal-Meter output, certified benchmark, or human-ranking measure. The contribution claimed here is structural: organizing existing human-state proxy modalities into a synchronized, leakage-controlled, non-diagnostic benchmark layer for AI performance versus human-state impact evaluation, measurement-layer simplification, and future Sal-Meter I/G-channel A/B comparison. This paper does not redefine CAIS. It does not redefine Sal-Meter. It does not grant CAIS compliance. It does not validate Sal-Meter. It does not certify any benchmark, dataset, model, dashboard, laboratory, implementation, or technology. A companion boundary record is available as SICS Human-State Proxy Benchmark Track: Public Boundary and Program Charter v0.1.