Uncertainty and Risk in the Predictions of Global Climate Models. (Invited)

There has been a great deal of emphasis, in recent years, on developing methods for assigning probabilities, in the form of quantitative margins of uncertainty (QMUs) to the predictions of global climate models. In this paper, I will argue that a large part of the motivation for this activity has been misplaced. Rather than explicit QMUs, climate scientists ought to focus on risk mitigation: offering policy advice about what courses of action need to be taken in order to reduce the risk of negative outcomes to acceptable levels. The advantages of QMUs are clear. QMUs can be an extremely effective tool for dividing our intellectual labor into the epistemic and the normative. If scientists can manage to objectively assign probabilities to various outcomes given certain choices of action, then they can effectively leave decisions about the relative social value of these outcomes out of the work they do as experts. In this way, it is commonly thought, scientists can keep ethical questions—like questions about the relative value of environmental stability vs. the availability of fossil fuels for economic development—separate from the purely scientific questions about the workings of the climate system. It is this line of thinking, or so I argue, that has motivated the large quantity of intellectual labor that has recently been devoted, by both climate scientists and statisticians, to attaching QMUs to the predictions of global climate models. Such an approach, and the attendant division of labor that it affords between those who discover the facts and those who decide what we should value, has obvious advantages. Scientists, after all, are not elected leaders, and they lack the political legitimacy to make decisions on behalf of the public about what is socially valuable. Elected leaders, on the other hand, rarely have the expertise they would need to accurately forecast, for themselves, what the likely outcomes of their policy choices would be. Since it would be disingenuous of climate experts to pretend that they can make forecasts with certainty, the objective assignment of probabilities to the forecasts of climate experts is just what is needed to resolve this tension. All of this, however, is predicated on the assumption that a conceptually coherent methodology is available for calculating QMUs based on the forecasts of complex deterministic models like the global models of climate used by climate scientists. I argue in this paper that, at the present time, no such conceptually coherent method exists, and it is not clear where one will come from. In fact, I argue, the present practice of assigning QMUs provides an artificial precision to the predictions of climate models where no such precision is possible. But it is this very kind of precision which would be required for the method of QMUs to divide our intellectual labor into the epistemic and the normative. And if QMUs cannot do their intended job of dividing our intellectual labor into the epistemic and the normative, then perhaps they ought to be abandoned in favor of an approach in which certain basic assumptions about values are built into the science. Risk mitigation might be such an approach.