CAN OBSERVATION REPORTS BE REVOKED?

There is a supposition behind the title question: when should we ignore evidence in favour of a hypothesis? The supposition is, sometimes we may do so, sometimes we may not. Some philosophers have explicitly rejected this supposition, others have implicitly rejected it.

Francis Bacon said the hallmarks of a dogmatist were first, that he is unable to see facts that conflict with his theory, and second, that when others force him to notice them either he denies them or belittles them with the aid of a ‘frivolous distinction’.

At roughly the same time Galileo expressed admiration of Copernicus for sticking to his hypothesis in the face of observed facts concerning the brightness of Mars. It is difficult to make much of this fact because, unlike Bacon, Galileo did not explicitly offer a clear-cut methodology within which to fit observation. Whereas Bacon argued that it is better to avoid the formation of hypotheses, since they act as blinkers, Galileo never argued that it is better to avoid observations. Galileo sincerely hoped that the facts would fit his theory perfectly but he felt that a temporary discrepancy did not matter much.

Bacon’s point seems common-sense and normal, because in theory tradition goes along with him. But in practice tradition went along with Galileo, not with Bacon. Failure to see this may easily make us misjudge the history of the Copernican movement. Bacon charged Copernicus with trying to replace one dogma for another. Dogmas, we remember, were for Bacon marked by ‘frivolous distinction’ - the excuses dogmatists make for awkward facts. In the case of astronomy the excuses were epicycles and eccentrics. Bacon was willing to go further. He did not know, he said, whether Copernicus’s admission that one planet goes round the earth rather than the sun, was not a case of a dogmatist’s inability to accept facts which refute his theory.

Now Copernicus was definitely not scandalized by Ptolemy’s epicycles: he did not oppose the tradition of making temporary allowances for deviation of facts from theory. He deplored the deviations being allowed to remain so long, with not enough effort being made to remove them or to try a new hypothesis. A new hypothesis, he seems to have suggested, could tolerate blemishes during its early days, during its trial period. Both Giordano Bruno and Galileo seem to have said the same, though they were not articulate enough on these matters for me to let things rest here without further exegeses.

The official victory of Baconianism in the scientific societies in the generation following Galileo is perhaps the chief reason for our difficulty in interpreting Galileo. But it would be oversimple to think that Bacon’s view was endorsed without qualification.

Robert Boyle offered a crucial qualification. Whereas Bacon counselled avoiding the formation of hypotheses for as long as possible, Boyle was both a Copernican and a Cartesian. He explicitly allowed the formation of hypotheses-at least as candidates for tests and thus as stimuli to experi­ment and observation. His counsel was rather that wherever a hypothesis and a fact clash, the hypothesis must be given up.

The rule that reports of facts should always make us reject the hypoth­eses they conflict with, I shall call ‘Boyle’s rule’. We may use the label ‘Locke’s rule’ to refer to the rule, ‘accept as final only observation reports’ - though in historical fact it was also put forth by Boyle as a justification of‘Boyle’s rule’. And indeed Locke’s rule does justify Boyle’s rule. Locke’s rule is itself justifiable by what we may call ‘Bacon’s rule’ - which tells us not to accept any statement that is not final, plus the rider that only observation reports and what is properly based on them can ever be final. (Descartes accepted Bacon’s rule, but not the rider.)

Bacon’s rule is one which many philosophers and many scientists endorse. It is still very popular. I myself often heard it in childhood, in adolescene, and in my university days - usually from science teachers. It is plainly nonsensical. Anyone who thinks he endorses it is forced seriously to question such allegations (which he has not checked with his own eyes) as that he was bom to a human female. One does not believe this because one saw cases of human birth, or heard about human births from reliable witnesses. We simply accepted the facts of motherhood before we could understand them, and we never doubt them though we sooner or later learn to doubt other things we heard at that period, con­cerning Santa Claus and the like. If someone says, or even hints that the story of his birth is on a par with stories of storks bringing babies, he should be rushed to a psychiatrist.

Back to our title question. Unless we reject Bacon’s rule, we do not have to take hypotheses seriously, particularly after they have been discovered to clash with observation reports. And unless we reject Locke’s rule too, we still cannot take hypotheses seriously because Locke’s rule entails Boyle’s rule; and Boyle’s rule answers our question, ‘when should we reject observation reports in favour of a hypothesis?’, with the straightforward reply: ‘never!’

Yet Locke’s rule was a fixture of the scientific tradition until fairly recently.

The exact reversal of Kant’s view appears at the turn of the century with Pierre Duhem, the first scientist openly to reject Locke’s rule. He agreed with Locke that sense-data are final, but declared their scientific articulation, scientific observation reports, subject to modifica­tion together with modification of the very theoretical framework within which they are cast.

Our question, then, only begins to signify at the beginning of the present century. It was not felt strongly, however. First, little obligation was felt that each and every known fact accorded with accepted theories. Freud said it first: it was permissible to overlook some facts for a while. The physicist P. A. M. Dirac stressed that in his work he favoured aesthetic considerations over the close agreement between theory and facts. This, and similar declarations, were certainly not overlooked by philosophers. They lead immediately to the problems put in the title of the present essay. To date, however, the question has seldom been asked - and I know of no other attempt to put it on the agenda so as to raise the question of whether we should reinstate Boyle’s rule or not. The issue may be of concern to scientists because Boyle’s rule is often applied by editors when rejecting scientific papers proposing hypotheses which conflict with accepted scientific observation reports.

There is no doubt, however, that the question has been raised before. Amongst scientists, Dirac’s move is often cited as a repudiation of Boyle’s rule. Moreover, Michael Polanyi has repeatedly said this in lectures and publications. Polanyi is philosophically opposed to any rule; he thinks that scientific method is ineffable and its practice is a matter of the intuitions of leading scientists.

Finding Polanyi’s theory authoritarian, my preference would be to propose, however tentatively, rules to replace Boyle’s rule, and then try to examine them. Thus far, there are three or four proposed rules: my own (1966), that of Lakatos (1970), and, by implication, that of Kuhn (1970 or perhaps even 1962). My own proposal, however, was within Popper’s philosophy of science. I now prefer Bartley’s view of the matter (1968) which puts a much better perspective on it, and makes it much less important than it seemed.

Before discussing the question, ‘when should an observation report be rejected in favour of a theory?’ we may tackle the more general question, ‘when should an observation report be rejected?’ Of course, any complete answer to the more general question contains an answer to the more specific question. Yet, in response to a general question one may, and often does, offer an incomplete list of cases and then one may start to discuss each of the specific items on the list in depth while totally ignoring the initial specific item or forgetting to put it on the list. And, when one declares the incomplete list to be complete - either absent­mindedly or for very different reasons - one thereby answers unwittingly the initial specific question.

Thus, the question, ‘when do we reject an observation report?’ may be answered by the statement, we reject it when we can replace it by a better and conflicting one. For example, we reject empirical assessments of atomic weights when better ones are available. Well and good. Does this mean that this is the only occasion on which we reject an observation report? It is hard to say, since the English language is flexible on this point and allows one to read the answer one way or another, depending on context. If strong context-dependence indicates that this is the only way, we can say that the author clearly implies that we may never reject an observation report in favour of a theory except when the theory is backed by a better observation report (and hence Dirac’s conduct was reprehensible).

The point just made will turn out as crucial at the end of Section V below. Meanwhile we may notice an interesting historical example of context-dependence - the case of Duhem. Duhem clearly said that observation reports are not final and are often in need of replacement. (One cannot say more about this without going into subtle details of Duhem’s doctrine. This I am reluctant to do, since the long and detailed study that the subtlety of Duhem requires is too cumbersome to publish.)

Duhem views the set of theoretical statements accepted by scientists at a given time as a system; the system is both theoretical and linguistic: it enables us to store factual information; it is a language in which factual information is stated in neat and compact ways. When too much factual information is stacked, the neatness of the system is overcharged, and the system becomes complex. A change of theory then occurs, which is also, of course, a change of the linguistic system (since the theoretical system is the same as the linguistic system). What Duhem claims is that the factual information is now translated from one language into the other: there is no loss of factual information. Only some old verbal formulation of certain information must indeed be rejected, as the result of the rejection of the theories whose terminology is reflected in them.

This is not to say that all the older formulations of observation reports must be rejected. Indeed, even observation reports which employ modified terms may be retained, even though now their meaning is somewhat altered. An old observation report may have to be rejected a priori (if it clashes with a new theoretical statement), or it may be rejected a posteriori (as the result of an observation report of a new experiment), or it may be accepted a priori (as a part of the new theory) or it may be accepted a posteriori, or it may be in doubt pending further experimenta­tion. There are historical examples for each of these cases! One thing can be said generally: once a new theory is accepted, and as long as it is, it forces us to reject some reports when taken literally, yet we can rescue these reports by translation. More cannot be said.

To Duhem, the answer to our question, ‘when should we ignore evidence in favour of a hypothesis?’ is, in a sense (translation) never, in a sense (transplanting with no alteration of words) always! The chief trouble with this answer is that it is utterly useless, say, for Dirac: it does not tell him whether to be concerned about the discrepancy or not. It may help a historian to look back but not a scientists to look forward.

This concludes the historical survey and brings me to the 1935 views of Karl R. Popper.

II.