turpin@ut-sally.UUCP (Russell Turpin) (08/27/87)
Being one of the earlier posters who appealed to it as a philosophic criterion for comparing scientific theories, I now feel obliged to put forth my defense of Ockham's Razor. (The spelling "Occam" is also correct.) The first order of business is describing precisely what (in this article) is an application of Ockham's razor. From reading the many postings on the subject, it seems there is no concensus on this. In presenting the definition below, no claim is pressed that this is closest to what William originally intended. It is simply the version that I wish to defend. o The term "theory" refers to a set of statements. The set, though probably infinite in cardinality, must be finitely describable to qualify as a usable scientific theory. But in discussing the relative merits of theories, it is the entire set that is important. This point is expanded below. o I assume some noncontroversial method for identifying which statements are testable by observation. These might include such statements as "grass is green" or "if A is performed, B will be observed". (This is a very controversial assumption, but one that is not the subject of this article.) o A theory X can now be divided into an "observable component" C and its non-observable component X - C. Note that X - C includes mixed statements. (If the integration path is closed, conservation of flux is observed.) Only statements that are completely testable are in C. In general, both C and its complement are infinite. A theory Y is stronger than (equal to) a theory X if every statement in X is also in Y (and vice versa). A theory Y with observational component D has greater (equal) explanatory power than a theory X with observational component C if every statement in C is also in D (and vice versa). The version of Ockham's Razor that this article defends can now be stated. If X and Y are two theories and Y is properly stronger than X but has the same explanatory power, then X is preferred. (If C and D are the observational components of X and Y respectively, then C = D but X [ Y.) The justification for this is not difficult. Scientific theories are not deductively derived. The only reason for putting any stock in them at all is their explanatory power. A scientific theory is assumed true because of the observable statements it predicts (contains). In particular, the only reason for believing any statement y in Y that is not in D is because it helps explain D. But in the case above this is not true for the statements in Y that are not in X. They can all be dropped with no weakening of the explanatory power of the theory. By talking about entire theories rather than their finite descriptions via sets of axioms, the difficulties that arise because of different equivalent axiomatizations are avoided. Sometimes two axiomatizations can be directly compared. One set of axioms may properly contain the other. But this is rare, and can almost always be ducked by the defenders of a theory by changing the axiom set. In science, it is an entire theory that is assumed true. Discussing a theory only in terms of its axioms is to focus on a description of the theory, usually one of many, rather than the theory itself. At this point, a potential criticism of the above view must be presented and dismissed. It might be argued that in light of the strong argument above for weak theories, the best theory is simply the concerned set of testable statements, C. But C (probably) does not have a finite axiomatization. This puts it out of the running as a usable theory. While in comparing two theories one should not worry about particular finite descriptions, it is undoubtedly true that the motivation behind posing theories in the first place is to find finite (understandable) descriptions of the infinite range of observable phenomena. A second rejoinder that might be made is that taking such a strong stance for weak theories completely disallows assuming the existence of entities that cannot be directly observed. One cannot assume that electrons exist, only that observations will be in accord with QED. The first reply to this complaint is that this is not so bad. The second reply is that definitions come for free. One can define an electron as the locus of a set of behavior. Adding a definition to a theory, for simplifying discussion, should not be viewed as changing its content. Similarly, mathematical content is not important in comparing physical theories. Russell
kube@cogsci.berkeley.edu.UUCP (08/31/87)
In article <8851@ut-sally.UUCP> turpin@ut-sally.UUCP (Russell Turpin) writes: > Being one of the earlier posters who appealed to it as a >philosophic criterion for comparing scientific theories, I now >feel obliged to put forth my defense of Ockham's Razor. Thanks for taking the time to make such a thoughtful response. I have only a couple of further remarks. > o The term "theory" refers to a set of statements. Sounds good to me. > o I assume some noncontroversial method for identifying which > statements are testable by observation. > (This is a very controversial assumption, > but one that is not the subject of this article.) I agree that the observational/nonobservational distinction turns out to be pretty bogus when you lean on it too hard but for present purposes it seems fine. > The justification for this is not difficult. Scientific >theories are not deductively derived. The only reason for putting >any stock in them at all is their explanatory power. A >scientific theory is assumed true because of the observable >statements it predicts (contains). In particular, the only reason >for believing any statement y in Y that is not in D is because it >helps explain D. But in the case above this is not true for the >statements in Y that are not in X. They can all be dropped with >no weakening of the explanatory power of the theory. This is an elegant way of putting your point, but I worry that it might be *too* elegant. In particular: You are saying that a theoretical statement y in a theory Y cannot help to explain the truth of an observation sentence if the observation sentence is a consequence of Y - y. But in the "intuitive" sense of `help explain', this doesn't seem right; y can in practice be used to explain observations even though y is `in principle' eliminable. The problem is that `explanation' seems to be a partly psychological concept. It's natural to cash it out in terms of the consequence relation as you have (I'm inclined to do it myself), but it's not clear how much is given up with this abstraction. (It might be no problem here; I think maybe you need only confirmation, not the full concept of explanation, to run your argument; and confirmation is probably more closely tied to consequence.) What the philosophy of science needs, and does not have, is a good theory of explanation. --Paul kube@berkeley.edu, ...!ucbvax!kube