The AI Paperclip Maximiser as Constitutional Political Economy

With Darcy W.E. Allen. Available at SSRN.

Abstract: The AI “paperclip maximiser” parable describes a superintelligent agent whose faithful pursuit of a single objective becomes catastrophic. Standard economic analysis frames this as a principal-agent or incomplete-contracting problem. We argue that the AI paperclip maximiser is better understood as a problem of multiple agents coordinating under uncertainty and under sets of rules: a problem of constitutional political economy. The relevant knowledge is dispersed, tacit and generated through parallel experimentation by multiple sovereign agents. The instrumental-convergence thesis must therefore be extended with two additional drives: tolerance of multipolarity and constitutional cooperation. Resources spent on predation and defence impede the maximiser’s terminal goal, making constitutional rules beneficial even to a dominant AI agent.