Low Intelligence Agents for Market Testing

Markets are complex artifacts. The design of such artifacts is made possible by continuation — a process that breaks the design into pieces small enough to be within the cognitive reach of the designer. Continuation requires frequent testing of critical behaviors. For the economic agents in markets, these behaviors include certain sensitivities to learning, cooperation and the distribution of autonomy. The paper argues that these sensitivities are best tested by software agents containing relatively simple algorithms, that is, low intelligence agents.