Androids and agents: do we need a non-computational economics?

In this paper we probe the limits of the computational method in economics. This method involves modeling individual behavior and economic processes in terms of constrained optimization. In neoclassical economics human behavior is explained entirely computationally. Alternative paradigms include the evolutionary and the complexity-based approaches that model behavior and processes as non-optimizing or boundedly rational. But many of the models used in 'complex-evolutionary economics' are cellular automata or their equivalents. This means that neoclassical economics and complex-evolutionary economics are both committed to a computational vision of the economy. A highly complex computational economy can evolve and self-organize but it also displays computational universality that means that many problems are not decidable. The inherent limits of computability become evident. This paper proposes incorporating a particular (constructive) non-computability into our view of economic behavior and processes. The paper defines constructively non-computational behavior, discusses its origins in Roger Penrose's writings, and provides an application of this concept to the question of realistic counterfactuals in economic models.