The Superhumans: Artificial Cognitive Amplification, Concentration of Agency, and Systemic Risk in the Era of Autonomous Agents

The dominant debate on the risks of artificial intelligence (AI) frequently centers on the possibility that highly capable, autonomous systems come to act in ways incompatible with human interests. This paper examines an earlier and potentially complementary transition: the amplification of individual human agency through increasingly capable AI. Our unit of analysis is neither the model nor the worker in isolation, but the composite system of human, agents, tools, memory, compute, and capital, treated as a single effective economic agent. We introduce the Functional Superhuman (FSH): a biologically ordinary individual whose effective cognitive and operational capacity is substantially amplified by artificial agents, computational resources, tools, memory systems, and capital. We argue that economically and politically relevant superhuman capability can emerge from distributed human-AI systems without requiring a single artificial system of vastly superhuman individual intelligence. Our contribution is twofold. First, we formalize the condition governing the emergence of large-scale FSH: the relationship between the useful cognition each additional agent contributes and the marginal cost of coordinating and verifying it —a version of the Coase-Williamson boundary of the firm applied to artificial cognition—. We derive the optimal number of agents n* and its comparative statics. Second, we articulate a falsifiable causal chain —from human amplification to the competitive delegation of autonomy (the Delegation Ratchet), formalized as a collective-action game— linking the concentration of agency to the long-run control problem. The paper derives seven testable predictions, six falsification conditions, an experimental design, and a panel-data strategy, and it proposes measuring human-AI amplification before the emergence of more speculative forms of superintelligence.