Observation and Reliable Detection

Abstract

Empiricists worry about the division between the observable and the unobservable because they want epistemic security. Direct perceptual knowledge of the observable might be defeasible, they reason, but it does not have the shaky status of phlogiston, the superego, noxious effluvia, and all the rest of those entities that turned out to be unobservable because they did not exist. I am an empirical realist, so I worry about the security of my data too. Hence our first step ought to be to take the security issue as primary and the observable versus unobservable distinction as merely one attempt to resolve that issue. That is, it is not the intrinsic status of something as ‘observable’ or ‘unobservable’ that should concern us, but how good are the detectors we use at producing secure knowledge. In that respect, the traditional human sensory modalities are neither particularly reliable nor all that versatile. We all know that eyewitness reports of who was present at a crime scene are unreliable, and that evidence in the form of ‘unobservables’, such as DNA profiles and blood types, is far more secure than the unaided observations of humans. I know that on a dark Warsaw street, my eyes are not reliable indicators of the street names on a Polish map, and although I can confidently tell the aural difference between a piano and a harpsichord, I cannot reliably discern a trumpet from a cornet. The superiority of instruments over human senses is, of course, widely recognized. Photo-finishes in horse races, electronic timing in Olympic swimming and sprints, touch sensors in fencing, automatic pilots, light meters for camera exposures, and so on are all reflections of this superiority. Readers of a certain age may remember the use of a ‘personal equation’ as a corrective in some scientific experiments such as scintillation counts before the era of automatic detectors. As a last example, many of the optical illusions that fool the human visual apparatus do not affect computerized image analyzers.²

This is a much revised version of an informal talk given at the LMPS96 conference. The talk was illustrated with colour images that supplemented the oral argumentation in essential ways. Unfortunately, a combination of cost and copyright constraints prevented me from including those images in this paper. That is a pity, for part of the background to this paper, although it is not explicitly dwelt upon here, is that graphical output from experiments and simulations is now an essential component of much scientific activity.