Counterargument 1. Regard the statement All ravens are black. This statement will be false when one finds a non-black, say white, raven. So the statement would be an acceptable scientific hypothesis, since falsification is possible in principle. But, as the falsificationist would hold, it would remain a hypothesis, and we should be aware of the fact that is only a hypothesis, until it had been checked for all ravens (Tintner (1968:12)). This falsificationist view however is problematic, since most of us will sense that there is truth in All ravens are black, for example by our definition of a raven.
Counterargument 2. In the extreme, all scientific knowledge would consist of instances of falsification. It has been falsified that the Earth is flat, that atoms cannot be broken, that ... But the principle itself, i.e. that ‘all scientific knowledge would consist of instances of falsification’, is a definition and is not open to falsification.
While falsification may be a successful research strategy in many cases, it does not seem to be a fully satisfactory way of organising science, at least from these two points of logic.
(ad 2) Take stochastics next. Let us regard the typical modelling situation:
| The model: Estimation: Observation X[+1] forecasts: Final observation: | y = X ß + y = X b + e yest[+1] = X[+1] b + Exp[e[+1] y[+1] |
The question now is whether this new observation can falsify the hypothesis of the empirical estimate. This question is not as simple as the naive falsificationist first had in mind. The principle of falsification is formulated as for deterministic reality, while many empirical models are stochastic. In stochastics, there may be deviations, and sometimes large ones. There are problems of measurement in y and X, the choice of the functional relationship, missing variables, and the choice of the stochastic specification itself.
One useful empirical answer is optimal control, with the example of a rocket launched to the moon, where there is continuous adjustment to observed error (‘falsification‘). This control only works well when there is a proper definition of the loss function. The issue of the loss function is a crucial one, but this is not falsificationism.
Logic and stochastics cause me to take the following position.
There is a difference between all1 (universal) and all2 (generally, usually, normally). The statement All ravens are black can be seen as:
1. a definition. It then holds universally. Empirical truth then is conditioned to the logical tautology of the definition that we have chosen. If we find a white bird that looks like a raven, it cannot be a raven. (But we think that this definition covers reality, for example since we have some ideas about genetics and evolution.)