Relational Cognitive Value Under Capability Asymmetry

Description As artificial intelligence systems become increasingly capable, debates about human relevance often assume that humans must preserve some domain of cognitive superiority in order to continue contributing value. This work challenges that assumption by distinguishing individual cognitive capability from marginal systemic contribution. It introduces the framework of Relational Cognitive Value (RCV) for studying whether, when, and through which mechanisms a demonstrably weaker cognitive agent can improve a system containing a substantially stronger agent. RCV defines contribution relationally, as the net change in system utility produced by adding an agent under specified tasks, objectives, architectures, resource constraints, and integration costs. The framework distinguishes four potentially co-occurring mechanisms of complementarity: Informational complementarity, when the weaker agent provides relevant information unavailable to the stronger agent; Exploratory complementarity, when an intervention produces a useful change in the system’s search trajectory; Normative complementarity, when the participating agent contributes to the definition or evolution of the optimization target; Meta-selective complementarity, when a weaker agent creates value by deciding which trajectories generated by stronger agents should receive additional computation. The central empirical object is the mechanism-specific relationship between Relational Cognitive Value and independently measured capability asymmetry: [ RCV_m(\Gamma), ] where "\Gamma" represents the capability gap and "m" identifies the complementarity mechanism. The paper proposes six preregistrable hypotheses, six corresponding experimental families, explicit cost accounting, and hypothesis-level and framework-level failure-to-support criteria. Particular attention is given to two questions: whether a weaker agent can retain positive marginal value solely by allocating computation among trajectories generated by stronger agents, and whether a participant can remain systemically relevant when interaction changes not only what the system knows, but also what the coupled system is trying to optimize. The framework does not claim that weaker agents are necessarily useful, that diversity is automatically beneficial, or that humans possess permanently exclusive cognitive capabilities. Its central claim is deliberately limited: Cognitive inferiority does not, by itself, imply marginal redundancy. The broader research question is which forms of relevant difference persist, saturate, or become harmful as capability asymmetry increases.