ABSTRACT
This chapter examines how complex introspections can be approximated by code, sometimes in an opportunistic way. It shows how a failure need not imply a retreat and that the solution could be in even more ambitious introspection. The details of the design will be presented in an order that moves from some preliminary remarks, to the functionality visible in the introspective model, and then to the technicalities inherent in any design. The random element allows novel actions to be tried out within the normal process of responding to the environment, in contrast to many learning systems that have a distinct “learning” and “doing” phase. The chapter provides a partial list of issues and parameters that need to be tweaked is provided in order to demonstrate the complexity. It demonstrates a possible concrete manifestation of Gadamerian artificial intelligence as recommended by Terry Winograd and Fernando Flores.
