Jon Williamson Causality and probability in the sciences

Causality and probability in the sciences Towards the end of the nineteenth century Karl Pearson noted that a prob-abilistic dependence between two variables does not necessarily imply that the two variables are causally connected (Pearson, 1897). This led Pearson, in the third (1911) edition of his book Pearson (1892), to argue that talk of cause and effect should be eradicated from science in favour of talk of prob-abilistic dependence (in the form of contingency tables). Around the same time, Bertrand Russell threw his intellectual weight behind this purge of causality (Russell, 1913). In the same vein, Ernst Mach (1905) argued that causality, understood as a way to explain phenomena, should be replaced by the concept of relation, which is a way to merely describe phenomena. These attacks had a profound influence on much of twentieth century science. Although scientists continued to reason causally—e.g., to find causes of phenomena, to devise experiments to measure interventions, to inform policy decisions—explicit mention of causality met with disapproval. Then came the 1980s. As explained below, causal methods developed in Artificial Intelligence (AI) in the 1980s helped to rehabilitate the concept of cause. While causality was no less controversial from a philosophical perspective, new formalisms for handling causality and probability together helped mathematise the notion of cause. Fig. 1 and Fig. 2 show the resulting transformation. A search of the Web of Science databases for papers whose titles include a word beginning with 'caus-' (e.g., 'causality', 'causation', 'causal') revealed a stark increase in the numbers of such papers after about 1990. This was so for the Science Citation Index Expanded (SCIE) database, which deals mainly with physical, biological and computational sciences, and for the Social Sciences Citation Index (SSCI) database. The growth in papers on causality in the Arts and Humanities Citation Index (AHCI), which covers philosophy, is rather more gradual. (Of course, the volume of all academic papers increased markedly in this period. In an effort to compensate for this general growth, for each of the three databases Fig. 2 portrays the yearly number of papers involving causal terms divided by 2 the yearly number of papers with an author whose name begins with the letter 'J', a rather arbitrary indicator of the general volume of papers in the database in question.) Figure 1. Total yearly numbers of papers involving causal terms. Given this rehabilitation of causal talk it is all the more important …