First of all, your last sentence is not an accurate description of
what I am arguing for. I am not saying that "lack of positive
evidence" proves anything --- it doesn't. I am saying a failed
prediction constitutes positive evidence against a theory --- in fact,
it falsifies it. (When I wrote "lack of evidence" above, I was using
the sense Brian was using, namely "the predicted experimental outcomes
failed to eventuate" rather than the strict sense.)
Secondly, concerning Brian's example, the reason why the "failed
prediction" did not falsify Newton's theory was that the "prediction"
didn't really fail at all! The experimental results were wrong. When
the experiment was performed correctly, the results _did_ match the
theory. So Brian's example was not a true example of scenario 2).
Perhaps your point is that experiments can go wrong, and so we
shouldn't be too quick to dismiss a theory on the basis of one lot of
bad results --- we should investigate the possibility of experimental
error thoroughly first. I would agree with this. But this is a side
issue --- asking whether the evidence is valid or not. If there is a
problem with experimental technique then of course conclusions are
going to be questionable. But the question we are examining is the
underlying logic --- namely, that if a theory makes predictions, which
disagree with the results of non-flawed experimentation, then the
theory is false.
> The second reason is that you are ignoring all the evidence that was
> used to develop the theory in the first place. That evidence can be
> considered positive evidence in favor of the theory. To refute the
> theory you have to refute that evidence. A failed prediction does
> not refute this supporting evidence, so a failed prediction by
> itself does not falsify the theory. Only if the result directly
> contradicts the theory by refuting some of this supporting evidence
> (or directly contradicts known physical laws or other more strongly
> verified theories) can it then be considered positive evidence
> against the theory.
I have two serious concerns with your argument here.
1. It takes much evidence before one starts to have confidence in a
theory, but only a single piece of solid evidence to refute it. If my
theory is that all sheep are white, this theory may have come about by
the painstaking examination of many sheep --- and every extra white
sheep provides a bit more positive evidence for my theory. But all it
takes is for me to find one black sheep and my theory has been
falsified. I could go back to the drawing board and modify my theory
slightly to account for the new data, for example "at least 90% of all
sheep are white", but this will be a new theory.
2. It is considered a dangerous thing to use the same data for both
developing a theory and testing it. Ideally, having developed a
theory, one should devise means of testing it entirely independent of
data used in the development stage. Why? Well because it is often
possible to develop a theory which, due to the way it was developed,
predicts the data already available wonderfully --- but is completely
wrong. The falsity of the theory is only discovered by the fact that
it fails on new data. So indeed we should be "ignoring all the
evidence that was used to develop the theory in the first place" when
it comes to testing the theory.
Now of course, if a theory is falsified by new evidence, we may modify
the theory to take into account both this new evidence and the old
evidence. But the point is that this will be a new theory --- the
original one was false.
"They told me I was gullible ... and I believed them!"