This is an unscientific essay written to spark debate at the AAAI Fall Symposium on applications of AI theory to real world autonomous mobile robots. Towards this end, the paper argues that it is not possible to devise a formal theory of autonomous mobile robots operating in unengineered environments which has strong predictive power. The argument is based on the observation that the behavior of implemented robotic systems tends to be chaotic in the mathematical sense, i.e. very small variations in initial conditions can lead to large variations in the subsequent behavior of the robot. Because unengineered environments contain variations which cannot be modelled the effects of these variations will always lead to unpredictable behavior. This argument is offered in the hope that it will be refuted by an actual theory which makes falsifiable predictions that are subsequently verified by controlled experiments. In the event that the argument turns out to be correct, the paper makes some recommendations on how to proceed.