Aretē

The central argument of H.M. Collins' book, Changing Order, is his so-called “experimenters' regress”. He claims that experimentation can only be used as a test if we can break into the regress. He goes on to say that the only way to break into the regress is to find a criterion on which to judge the excellency of an experiment that makes no reference to the outcome of the experiment (84). But I will argue here that the very notion of the regress and the propositions that lead to it miss the point of using experimentation as justification for belief in hypotheses entirely.

Collins presents the experimenters' regress as the result of five propositions he makes about the nature of experimentation. The basic idea of these propositions is that an experimenter can't perform an experiment unless the correct outcome is known. The propositions emerge from the separation of ‘algorithmic’ and ‘enculturational’ learning, the main idea being that experimenters can only become good at performing a particular experiment by actually performing it; theoretical or algorithmic knowledge is not sufficient. The fifth proposition states that the proper operation of an experiment is defined by its ability to obtain the correct outcome. So, an experimenter can only become good at performing a particular experiment if the correct outcome of that experiment is known (73-74). The experimenters' regress takes over from here: it is the argument that we can't know the correct outcome of an experiment unless we can actually perform it. Collins illustrates the experimenters' regress with an example experiment for detecting gravity waves. He states, “What the correct outcome is depends upon whether there are gravity waves hitting the Earth in detectable fluxes. To find this out we must build a good gravity wave detector and have a look. But we won't know if we have built a good detector until we have tried it and obtained the correct outcome! But we don't know what the correct outcome is until . . . and so on ad infinitum” (84). This example illustrates the circular reasoning involved in determining the excellency of an experiment. It shows that the correct outcome of the experiment must be known in order to judge the experiment's excellency, but that the correct outcome can't be known without an excellent experiment. Collins recognizes the epistemological implications of this regress.

He writes, “Experimental work can only be used as a test if some way is found to break into the circle” (84). This is the problem Collins takes the regress to be for science. The idea is that no knowledge can come out of experimentation unless we find a way to avoid the regress. He goes on: “The experimenters' regress can only be avoided by finding some other means of defining the quality of an experiment; a criterion must be found which is independent of the output of the experiment itself” (84). Collins' claim here makes sense; if we can judge the excellency of an experiment without referring to its outcome, then we can test for the existence of gravity waves or whatever, and genuinely learn something. This solution amounts to rejecting Collins' fifth proposal—judging the excellency of an experiment without referring to its outcome amounts to finding other indicators of the proper working of the apparatus and experimenter than obtaining the correct outcome. Next Collins traces the practice of scientists to determine how they accomplish this.

Collins interviews a number of scientists wrangling with the gravity wave experiment. He traces the debate over which experiments are good ones and which are not. After analyzing the scientists' arguments, he finally concludes that “no set of ‘scientific’ criteria which can establish the validity of findings in this field [exists]” (88). In other words, there is no actual ‘science’ taking place in the debate over which experiments are good and which are not. It's entirely political. This finding leads Collins to his next two propositions. One, that “when the normal criterion—successful outcome—is not available, scientists disagree about which experiments are competently done,” and two, that “where there is disagreement about what counts as a competently performed experiment, the ensuing debate is coextensive with the debate about what the proper outcome of the experiment is. The closure of debate about the meaning of competence is the ‘discovery’ or ‘non-discovery’ of a new phenomenon” (89). To distill these propositions, we can say simply that in practice, scientists break into the regress by deciding what the proper outcome of an experiment is or which experiments they take to be ‘good’ by employing non-scientific reasoning. Later, however, Collins suggests that some scientists attempt to break into the regress in another manner: by calibration.

Calibration is what Collins calls a “test of a test”, in that it tests the excellence of an experiment without referring to the correct outcome (100). To illustrate, Collins continues with the gravity wave experiment. In order to calibrate the gravity wave detectors, scientists adopted a “well-understood surrogate phenomenon” and tested the apparatus' ability to detect that properly (101). There are a few problems calibration must overcome if it is to succeed in breaking into the regress. First, it must be shown that the surrogate phenomenon is similar to the unknown phenomenon in all the relevant ways. That is, if an apparatus is to be used to detect a certain phenomenon, the surrogate must behave in the same way, and so have the same effects on the apparatus as the phenomenon in question. By accepting a certain method of calibration, one also accepts constraints on the interpretation of the results of the experiment. Collins again refers to the gravity wave experiment: “In bowing to pressure to calibrate electrostatically, Weber set at least two assumptions beyond question. First he accepted that gravitational radiation would interact with the substance of his antenna in the same way as electrostatic forces… [Second he] put it beyond question… that the insertion of a localized pulse into one end of a bar antenna would have a similar effect to the insertion of energy into the bar as a whole from a source at a great distance” (103-104). In other words, by asserting that electrostatic calibration of the apparatus solves the experimenters' regress, the experimenter is limited to discovering phenomena that behave in relevantly similar ways to the electrostatic forces.

Because calibration solves the experimenters' regress in a very limited way, it places equally limited constraints on what we can learn from calibrated experiments. This is, of course, a step up from strictly political means of breaking into the regress, but still leaves much to be desired. But now I'll argue that this whole business of the experimenters' regress and Collins' propositions would never even have been considered if we held close the commonly cited reasons for accepting experimentation as justification for belief in hypotheses.

Collins wrote, “The closure of debate about the meaning of competence is the ‘discovery’ or ‘non-discovery’ of a new phenomenon” (89). This is where he gets closest to the reason we accept experimentation as useful in science. The idea is that through experimentation we learn something about phenomena; information which we take to support or undermine our hypotheses. But when Collins observes the construction of a complicated apparatus, he notes that trial and error are an enormous part of the scientist's practice. This leads him to his fifth proposition, that the success of an experiment is judged by its ability to obtain the correct outcome. The word ‘correct’ in that proposition leads him to the regress. He is right that the ‘correct’ outcome of an experiment is supposed to reflect the way the world is (and so his reasoning about using a device to detect gravity waves in order to know the ‘correct’ outcome is warranted), but the idea that deciding on the ‘correct’ outcome prior to judging an experiment is dubious. In order to actually learn about the way the world is, it doesn't make sense to posit that information, perform an experiment to test it, and then say that you “learned” it as a result. As soon as Collins takes the regress to be a problem, he excludes any hope of learning anything. The problem lies in the fifth proposition. The only way we can learn something useful regarding our hypotheses from an experiment is if the outcome is not known going into it. It might help to formulate my argument in a way that closely resembles Collins' regress, but remains true to the belief that experimentation is useful to science.

If we are to take experiments to be justification of our belief in a hypothesis, then we must be able to provide justification for our belief that the experiments are good indicators of the existence or non-existence of the phenomena the hypothesis predicts. If no justification can be provided other than the experiments actual indication of the independently known existential state of the phenomena, then the experiment immediately becomes useless as justification for our belief in the hypothesis (because the experiment was supposed to give us new information about the existence of the phenomena). So, if we are to hold that the experiment tells us something about our hypothesis, we must show that the experiment is a good indicator of the predicted phenomena without making reference to the phenomena themselves. Hence, if we accept Collins' fifth proposition, then experimentation cannot be justification for our belief in hypotheses.

Deep down, I think Collins knows this. His statement, “The experimenters' regress can only be avoided by finding some other means of defining the quality of an experiment; a criterion must be found which is independent of the output of the experiment itself,” embodies my argument (84). But this statement doesn't merely avoid the regress, it preempts the reasoning of the regress entirely.

last updated 2 years ago #