Abstract

The evidence of your own eyes has often been regarded as unproblematic. But we know that people make mistaken observations. This can be looked on as unimportant if there issome class of statements that can serve as evidence for others, or if every statement in our corpus of knowledge is allowed to be no more than probable. Neither of these alternatives is plausible when it comes to machine or robotic observation. Then we must take the possibility of error seriously, and we must be prepared to deal with error quantitatively. The problem of using internal evidence to arrive at error distributions is the main focus of the paper.