Science for the PostNormal Age

from uncertainty in knowledge and values; this is shown in the dominant teaching tradition in science, which created a universe of unquestionable facts, presented dogmatically for assimilation by uncritical students. Now scientific FUTURES September 1993 expertise has led us into policy dilemmas which it is incapable of resolving by itself. We have not merely lost control and even predictability; now we face radical uncertainty and even ignorance, as well as ethical uncertainties lying at the heart of scientific policy issues. For understanding the new tasks and methods of science, we can fruitfully invert Latour’s metaphor, and think of Nature as reinvading the lab. We see this in many ways; for example, our science-based technology, which for a while appeared to be a new man-made Nature dominant over the old, is now appreciated as critically dependent on the larger ecosystem in which it is embedded: and that it risks destruction of itself if that matrix becomes seriously perturbed or degraded. Similarly, the extension of modern technology to all humanity~ essential if equity between peoples is to be realized under the present system, would accelerate the self-destructive tendencies of the technological system itself. Thus Nature reasserts itself on all our scientific planning, for the technical and human perspectives alike. There have been other episodes in history when science has been transformed, when a particularly successful problem-solving activity has displaced older forms and become the paradigmatic example of science. These transformations have been identified with the names of such great scientists as Galileo, Darwin and Einstein. They have mainly affected theoretical science, because until quite recently technology and medicine were not generally influenced in the short term by the results of scientific research. The challenges to science were largely in the realm of ideas. Now, as the powers of science have given rise to threats to the very survival of hunlanity, the response will be in the social practice of science as much as in its intellectual structures. Centrality of uncertainty and quality Now that the policy issues of risk and the environment present the most urgent problems for science, uncertainty and quality are moving in from the periphery, one might say the shadows, of scientific methodology, to become the central, integrating concepts. Hitherto they have been kept at the margin of the understanding of science, for laypersons and scientists alike. A new role for scientists will involve the management of these crucial uncertainties; therein lies the task of quality assurance of the scientific information provided for policy decisions. These new policy issues have common features that distinguish them from traditional scientific problems. They are universal in their scale and long-term in their impact. Data on their effects, and even data for baselines of ‘undisturbed’ systems, are radically inadequate. The phenomena, being novel, complex and variable, are themselves not well understood. Science cannot always provide well founded theories based on experiments for explanation and prediction, but can frequently achieve at best only mathematical models and computer simulations, which are essentially untestable. On the basis of such uncertain inputs, decisions must be made, under conditions of some urgency. Therefore policy cannot proceed on the basis of factual predictions, but only on policy forecasts. Computer models are the most widely used method for producing statements about the future based on data of the past and present. For many, there is still a magical quality about computers, since they are believed to perform reasoning operations faultlessly and rapidly. But what comes out at the end of a program is not necessarily a scientific prediction; and it may not even be a particularly good FUTURES September 1993 Science for the post-normal age 743 policy forecast. The numerical data used for inputs may not derive from experimental or field-studies; the best numbers available, as in many studies of industrial risk, may simply be guesses collected from experts. Instead of theories which give some deeper representation of the natural processes in question, there may simply be standard software packages applied with the best fitting numerical parameters. And instead of experimental, field or historical evidence, as is normally assumed for scientific theories, there may be only the comparison of calculated outputs with those produced by other equally untestable computer models. Despite the enormous effort and resources that have gone into developing and applying such methods, there has been little concerted attempt to see whether they contribute significantly either to knowledge or to policy. In research related to policy for risk and the environment, which is so crucial for our well being, there has been little effort of quality assurance of the sort that the traditional experimental sciences take for granted in their ordinary practice. Whereas computers could in principle be used to enhance human skill and creativity by doing all the routine work swiftly and effortlessly, they have instead in many cases become substitutes for disciplined thought and scientific rigour.’ Even when there is empirical data for policy problems, it is not really amenable to treatment by traditional statistical techniques. As J. C. Bailar puts it: All the statistical algebra and all the statistical computations are of value only to the extent that they add to the process of inference. Often they do not aid in making sound inferences; indeed they may work the other way, and in my experience that is because the kinds of random variability we see in the big problems of the day tend to be small relative to other uncertainties. This is true, for example, for data on poverty or unemployment; international trade; agricultural production; and basic measures of human health and survival. Closer to home, random variability-the stuff of p-values and confidence limits, is simply swamped by other kinds of uncertainties in assessing the health risks of chemicals exposures, or tracking the movement of an environmental contaminant, or predicting the effects of human activities on global temperature or the ozone layer.s Thus, by traditional criteria of scientific method, the quality of research on these policy-related problems is dubious at best. The tasks of uncertainty management and quality assurance, managed in traditional science by individual skill and communal practice, are left in confusion in this new area. New methods must be developed for making our ignorance usable.” For this there must be a radical departure from the total reliance on techniques, to the exclusion of methodological, societal or ethical considerations, that has hitherto characterized traditional ‘normal’ science. An integrated approach to the problems of uncertainty, quality and values has been provided by the NUSAP system. In its terms, different kinds of uncertainty can be expressed, and used for an evaluation of quality of scientific information. We have to distinguish among the technical, methodological and epistemological levels of uncertainty; these correspond to inexactness, unreliability and ‘border with ignorance’, respective1y.l” Uncertainty is managed at the technical level when standard routines are adequate; these will usually be derived from statistics (which themselves are essentially symbolic manipulations) as supplemented by techniques and conventions developed for particular fields. The methodological level is involved when more complex aspects of the information, as values or reliability, are relevant. Then, personal judgments depending on higher-level skills are required; and the practice in question is a professional consultancy, a ‘learned art’ like FUTURES September 1993 744 Science for the post-normal age medicine or engineering. Finally, the epistemological level is involved when irremediable uncertainty is at the core of the problem, as when computer modellers recognize ‘completeness uncertainties’ which can vitiate the whole exercise, or more generally in post-normal science. In NUSAP these levels of uncertainty are conveyed by the categories of spread, assessment and pedigree, respectively. Quality assurance is as essential to science as it is to industry; and whereas in traditional research science it could be managed informally by a peer community, in the new policy issues of risk and the environment, quality of science must be addressed as a matter of urgency. The inadequacy of traditional peer review has been extensively analysed for the different areas of core science,‘l ‘mandated’ science,‘> and ‘regulatory’ science. I1 As we see, the evaluation of quality in this new context of science cannot be restricted to products of research; it must also include process and persons, and in the last resort purposes as well. This ‘p-fourth’ approach to quality assurance of science necessarily involves the participation of people other than the technically qualified researchers; indeed, all the stakeholders in an issue form an ‘extended peer community’ for an effective problem-solving strategy for global environmental risks. Problem-solving strategies To characterize an issue involving risk and the environment, in what we call ‘postnormal science’, we can think of it as one where facts are uncertain, values in dispute, stakes high and decisions urgent. In such a case, the term ‘problem’, with its connotations of an exercise where a defined methodology is likely to lead to a clear solution, is less appropriate. We would be misled if we retained the image of a process where true scientific facts simply determine the correct policy conclusions. However, the new challenges do not render traditional science irrelevant; the task is to choose the appropriate kinds of problem-solving strategies for each particular case. Figure 1 invol