One of the most persistent commonsense accounts of science is that in which scientists are understood systematically to assemble observations and arrive at reliable generalizations based upon them. Sometimes, wrongly, this simple inductive empiricist view is laid at the door of Francis Bacon (1561-1626) and dubbed ”Baconian inductivism.” In fact, Bacon’s views were considerably more complex than this, but the hare that he set running – inductive inference as the heart of scientific method – has subsequently been pursued by all manner of hounds. The Scottish Enlightenment philosopher David Hume (1711-76) was preeminent among the early pursuers, and to this day ”Hume’s problem” continues to preoccupy philosophy of science. In the mid twentieth century, there was a period when the seemingly more powerful hypothetico deductive model of scientific inquiry appeared to have run inductivism and Hume’s problem to exhaustion. However, it rapidly became apparent that the issues surrounding inductive inference had a peculiar capacity to reemerge from the coverts of deductive certainty, not least where the nature of observation itself was questioned. Into the space thus created have hastened newer, more relativistic epistemologies and, in full cry, the sociology of science.
Although Bacon was by no means a naive inductivist, he did insist on the necessity of ridding the mind of certain kinds of preconceptions when examining the facts, so as to better discover the true workings of natural phenomena. In effect, then, the inferential process moved from neutral observation to generalization unencumbered by misleading beliefs likely to obstruct proper knowledge. In its period this was a bold formulation, and one crucial to the subsequent development of natural philosophy into modern science. But it immediately raised difficulties for those eager to underwrite the legitimacy of scientific method in inductive terms. For while deductive reasoning had a lengthy logical pedigree, inductive inference was to prove far more slippery.
It was David Hume who presented the central problem of inductivism in its most influential form. In essence, the argument is simple: that however many instances we may find of a specific phenomenon, this gives us no reason in logic to expect that observed pattern to continue in the future. In other words, we have no justification for making any reliable inference from past evidence. Nor, of course, can we lay claim to probabilistic justification in as much as at the heart of the problem is the very unpredictability of the future in relation to past experience. The future will hold surprises. And against those who suggest, more pragmatically, that our past successes with this kind of inductive inference should lead us to expect success in the future, Hume levels the charge of circularity: attempting to justify inductive inference by inductively inferring future success from past instances of inductive inference itself.
Unsurprisingly, then, the difficulties consequent upon accepting inductive inference as the distinguishing feature of scientific method gave way by the mid twentieth century to more deductively inclined models of science. Rather than seeing science as founded on generalizations from data, these approaches afforded greater emphasis to the relative autonomy of theory. So, for example, variants of the hypothetico-deductive model were little concerned with the grounds on which we actually arrived at our theories and generalizations. Their interest lay, rather, with deducing predictive hypotheses from theory which could then be subjected to (experimental) test. In strict versions – for example, Popperian falsificationism – the test could only falsify and not confirm (Hume’s problem again). But this, too, presented problems. Science clearly did not proceed on the basis of rigid, deductive tenets (let alone falsificationist ones), and at the heart of any process of testing lay “observation” – which apparently relied upon some form of inductive inference from experience to the observation statements describing that experience. Since this would re raise a variation of Hume’s problem, deductivism was forced to recognize the unavoidably theory laden and conventional character of observation. The traditional logical positivist reliance on the distinction between theory language and observation language simply would not do. The language of observation was no less theoretical than the language of theory.
None of this, of course, dissolves the problem of inductive inference, if problem it is, for even if induction is not the defining element in so called scientific method it remains an important feature of actual scientific practice. Scientists make inductive inferences, albeit within a context of inquiry which also involves deduction, intuition, competition, and even sheer bloody mindedness. Accordingly, philosophers of science have continued to examine induction with a view to somehow resolving or bypassing the Humean difficulties. Within the pragmatist tradition, for example, Rescher (1980) has sought to reconceive inductive inference as essentially a kind of cognitive method, while others, such as Howson (2000), who retain more formal concerns, have leaned toward Bayesian probability theory as providing grounds for resolving at least some of Hume’s problem(s). Such approaches are often illuminating about what kinds of presuppositions are involved in inductive practice, although their apparent goal of providing “justification” seems far less significant in a period which has come more fully to recognize the importance of sociological and psychological factors in scientific inquiry. Here, Collins’s (1985: 145) ”sociological resolution of the problem of induction” is interesting. By empirical examination of what he calls the ”experimenters’ regress,” he seeks to show that the nature of experiments as ”skillful practice” means that an attempted replication always leads to the necessity for yet further experimental tests to confirm the quality of each experiment in the chain. This regress can only be halted by contingent, collective decision. Observing, experimenting, and constituting ”facts,” then, are socially constructed achievements of human agents; they cannot, without loss, be rendered as logically justified processes under the grand rubric of Inductive Scientific Method.
- Black, M. (1967) Induction. In: Edwards, P. (Ed.), The Encyclopedia of Philosophy, Vol. 4. Macmillan, New York.
- Collins, H. M. (1985) Changing Order: Replication and Induction in Scientific Practice. Sage, London.
- Goodman, N. (1973) The New Riddle of Induction. In: Goodman, N., Fact, Fiction, and Forecast. Bobbs-Merrill, New York.
- Howson, C. (2000) Hume’s Problem: Induction and the Justification of Belief. Clarendon Press, Oxford.
- Hume, D. (1999) An Enquiry Concerning Human Understanding. Oxford University Press, Oxford.
- Latour, B. & Woolgar, S. (1986) Laboratory Life: The Construction ofScientific Facts. Princeton University Press, Princeton.
- Medawar, P. B. (1969) Induction and Intuition in Scientific Thought. Methuen, London.
- Rescher, N. (1980) Induction: An Essay on the Justification of Inductive Reasoning. Blackwell, Oxford.
Back to Sociology of Science