Full Program »
Scientific Modeling and Limits to the Value-Free IdealAccording to the value-free ideal, the internal workings of science, including the evaluation of evidence, should be kept free from the influence of social values as much as possible. A standard challenge to the value-free ideal stems from the risk of error in inferring conclusions – that is, inductive risk (e.g. Douglas 2009). Arguments from inductive risk contend that it is appropriate for social values to influence at least some methodological choices in science, including the selection of standards of evidence; these standards should be more demanding in cases in which erring in accepting a hypothesis can be expected to have serious negative consequences.
Bayesians seem to have a way to resist these inductive risk arguments. A Bayesian can deny that scientists need to accept or reject hypotheses in the first place; the scientist’s job is merely to assign probabilities to hypotheses (see e.g. Jeffrey 1956). Such assignments should be arrived at via the application of Bayes’ Theorem, which does not require appeal to social values. For expanded inductive risk arguments encompassing other methodological choices, a similar strategy is open to the Bayesian: there may be different options available at a given methodological decision point, but the analysis will consider the probability of the hypothesis being true given the outcome of the chosen methodology, and will be able to do this irrespective of what the likelihoods of type I and type II error were for any such methodological choice. Call this the Bayesian Reply to Arguments from Inductive Risk (BRAIR).
BRAIR is subject to various counter-replies. Building on recent work focused on climate modeling (e.g. Winsberg 2012), we here develop a challenge to BRAIR that stems from facts about scientific modeling. First, we demonstrate some ways in which social values unproblematically influence choices in model development. The influences we identify stem not from inductive risk considerations but rather from the ways social values shape modeling purposes and priorities. We then show how this value influence creates trouble for BRAIR. In practice, scientists’ best attempts to implement the Bayesian approach often rely on models; these models are not straightforward embodiments of current background knowledge but rather deviate from that knowledge in ways that depend on modeling purposes and priorities, which in turn depend on social values. Moreover, correcting for differences between models and background knowledge, and thereby removing the model-based influence of social values on the probabilities produced, is sometimes infeasible. Finally, we consider the implications of our argument. We stress that it does not entail that value influence via modeling results in conclusions that are biased in problematic ways. Our argument does, however, call attention to some underappreciated limits to the extent to which the value-free ideal can be achieved in practice.
Douglas, H. (2009) Science, Policy, and the Value-Free Ideal. Pittsburgh University Press.
Jeffrey, R. (1956) “Valuation and acceptance of scientific hypotheses”, Philosophy of Science 22: 237-246.
Winsberg, E. (2012) “Values and Uncertainties in the Predictions of Global Climate Models”, Kennedy Institute of Ethics Journal 22(2): 111-137.
University of South Florida