Salmon on rational prediction

The philosopher Wesley Salmon wrote a paper called Rational Prediction criticising Karl Popper’s critical rationalism in 1981. The thrust of Salmon’s objection is the following:

According to Popper, negative instances provide rational grounds for rejecting generalisations. If, however, we make observations and perform tests, but no negative instance is found, all we can say deductively is that the generalisation in question has not been refuted. In particular, positive instances do not provide confirmation or inductive support for any such unrefuted generalisation. At this stage, I claim, we have no basis for rational prediction. Taken in themselves, our observation reports refer to past events, and consequently they have no predictive content. They say nothing about future events. If, however, we take a general statement as a premise, and conjoin to it some appropriate observation statements about past or present events, we may be able to deduce a conclusion which says something about future occurrences and which, thereby, has predictive content. Popper himself gives this account (Schilpp [1974],p. 1030) of the logic of prediction.

The problem of rational prediction concerns the status of the general premise in such an argument. One may claim, as Popper does, that we ought not to use a generalisation which has actually been refuted as a premise in a predictive argument of this sort, for we are justified in regarding it as false. We ought not to employ premises which are known to be false if we hope to deduce true predictions. The exclusion of refuted generalisations does not, however, tell us what general premise should be employed. Typically there will be an infinite array of generalisations which are compatible with the available observational evidence, and which are therefore, as yet, unrefuted. If we were free to choose arbitrarily from among all the unrefuted alternatives, we could predict anything whatever. If there were no rational basis for choosing from among all of the unrefuted alternatives, then, as I think Popper would agree, there would be no such thing as rational prediction. We are not in this unfortunate situation, Popper contends, for we do have grounds for preferring one unrefuted generalisation to another:

My solution of the logical problem of induction was that we may have preferences for certain of the competing conjectures; that is, for those which are highly informative and which so far have stood up to eliminative criticism (Schilpp [1974], p. 1024).

Popper’s concept of corroboration is designed to measure the manner in which conjectures have stood up to severe criticism, including severe testing. This, I take it, is the crucial thesis – that there is a rational basis for preferring one unrefuted generalisation to another for use in a predictive argument. If that is correct, then Popper can legitimately claim to have solved the problem of rational prediction.

Later in the paper Salmon writes:

I must confess to the feeling that we have been ‘given the run-around’. We begin by asking how science can possibly do without induction. We are told that the aim of science is to arrive at the best explanatory theories we can find. When we ask how to tell whether one theory is better than another, we are told that it depends upon their comparative ability to stand up to severe testing and critical discussion. When we ask whether this mode of evaluation does not contain some inductive aspect, we are assured that the evaluation is made wholly in terms of their comparative success up to now; but since this evaluation is made entirely in terms of past performance, it escapes inductive contamination because it lacks predictive import. When we then ask how to select theories for purposes of rational prediction, we are told that we should prefer the theory which is ‘best tested’ and which ‘in the light of our critical discussion, appears to be the best so far’, even though we have been explicitly assured that testing and critical discussion have no predictive import. Popper tells us, ‘I do not know of anything more “rational” than a well-conducted critical discussion.’ I fail to see how it could be rational to judge theories for purposes of prediction in terms of a criterion which is emphatically claimed to be lacking in predictive import.

There are a lot of problems and misconceptions packed into this paper. Popper criticised inductivism by pointing out that it is impossible. Induction involves starting with observations, using them to get a theory and then doing more observations that somehow confer higher probability or confirmation or something like that on the theory. How is one supposed to choose what to observe? Inductivists have no answer. How is one supposed to get from observations to a theory that is not implied by those observations? Inductivists have no answer. How do observations confer more probability or confirmation or whatever on a theory when the theory is true or false independent on how many of your observations agree with it? Inductivists have no answer.

Salmon is asking for a way of justifying rational predictions. This is impossible because justification is impossible. A justification is a procedure of some kind that supposedly makes a theory more likely to be true or good or useful or something like that. No such procedure exists. Your theory is either right or wrong. You can categorise it as right or wrong or say you don’t know if it’s right or wrong. In the case where you don’t know whether an idea is right or wrong your ignorance doesn’t provide any way to attach any kind of numerical value like a probability to your ignorance. An idea that some idea is probable or improbable is a vague, intuitive judgement that there are arguments that would persuade you to accept or reject that idea if you took the time to state those arguments. People should actually state arguments and assess them rather than try to muddle through with vague feelings. For explanations of how to do this see Elliot Temple’s writing on Yes or No Philosophy.

Another problem with Salmon’s insistence that we must somehow use induction on observations to get theories to use for rational predictions is that observations are guesses about what happened in a particular region. For example, there are devices called radiosondes – sets of instruments that scientists attach to a balloon and send up into the sky to record atmospheric conditions (temperature, pressure, altitude), cosmic rays and stuff like that. The scientist who uses observations from the radiosonde doesn’t float with the balloon up into the atmosphere to monitor the instruments and check they are working correctly. Rather, the scientist has to look at the readings and try to explain them by guessing about what’s happening to the instruments. The same would be true even if the instruments were in front of the scientist on a bench in a lab. Scientists can and do make mistakes in interpreting the results of measurements. I could write a long series of blog posts on examples of this from physics alone. Treating observations as anything other than fallible guesses is wrong and extremely dangerous. For more explanation of the point that measurements are guesses about what is happening in a particular region and involve guessing about the explanation for experimental results see “The Beginning of Infinity” by David Deutsch, Chapter 2 and also Chapter 5 of “The Logic of Scientific Discovery” by Popper (LScD). Salmon could have known about this by reading Popper and taking his writings seriously. For example, at the end of Section 25 of LScD, Popper writes:

This doctrine founders in my opinion on the problems of induction and of universals. For we can utter no scientific statement that does not go far beyond what can be known with certainty ‘on the basis of immediate experience’. (This fact may be referred to as the ‘transcendence inherent in any description’.) Every description uses universal names (or symbols, or ideas); every statement has the character of a theory, of a hypothesis. The statement, ‘Here is a glass of water’ cannot be verified by any observational experience. The reason is that the universals which appear in it cannot be correlated with any specific sense-experience. (An ‘immediate experience’ is only once ‘immediately given’; it is unique.) By the word ‘glass’, for example, we denote physical bodies which exhibit a certain law-like behaviour, and the same holds for the word ‘water’. Universals cannot be reduced to classes of experiences; they cannot be ‘constituted’.

I have one more point to make. Salmon is in part looking for a way to make practical predictions:

Third, we sometimes find ourselves in situations in which some practical action is required, and the choice of an optimal decision depends upon predicting future occurrences. Although wagering is by no means the only such type of practical decision-making, it is a clear and comprehensible example. We all agree, I take it, that scientific theories often provide sound bases for practical prediction.

In general solving practical problems involves a lot more than making predictions. The laws of physics don’t imply one particular solution to a problem. If you have a guess about a solution the laws of physics might tell you something about whether it will work or not. For example, a solution to a problem that involves breaking the conservation of energy won’t work in reality. Other general theories can play a role in eliminating solutions like the laws of biology or economics, but they don’t dictate a particular solution to a problem either. In addition, in general when you solve a practical problem you may have missed something and you will in general want to monitor whether your solution is working. There is no shortcut to having a working solution to a real problem: you have to guess and check your guesses, not wish for some way of guaranteeing correctness or probable correctness.