I recall this high school history teacher who made every effort to inculcate us that historic research was made in a scientific way; I remained unconvinced. Political sciences and sociology are of slightly different quality because, even if they rest on history, a large part of their research is made in the present, with unverifiable experimentation except opinion surveys, ever changing as weather vanes do, or by so-called focus groups analysis which, chosen well, conveniently lead to precooked conclusions. It’s no science but these are nevertheless works of big interest as long as they are made with rigor and intellectual honesty.

In the scientific domain, there are broad areas which, similarly, cannot lead to refutable (falsifiable) theories. These are “…logies” that combine knowledge and theories from the basic sciences (physics, chemistry, biology, with mathematics as an essential link), with which specific explanations may be developed, but which in fine cannot submit themselves to validation or questioning.

Medical doctors understood this, knowing by experience that often it is the exception which becomes the rule; that is why they speak about the medical art and that is why, for them, “medical sciences” are only a part, essential but insufficient, of their discipline.

Ecology and climatology are such scientific agglomerates, we could call them supra-sciences. Economy also, but with much less secure theoretical bases. To understand such complex systems it is necessary to tie together subsystems in order to reveal connections and to find causal relationships; then significance, stability and reproducibility of such interactions must be verified to derive valid theories. Two obstacles make this an almost impossible task.

First, it lacks the ability to observe: we have only one laboratory, the earth, and only one ongoing experiment, our current history. Experimentation is thus impracticable.

Second, and bound to the first one: the lack of statistical relevance. Series of historical instrumental data are limited and have no sufficient accuracy and precision. The natural variabilities are also such that the signal-to-noise ratio is so tenuous that it becomes impossible to demonstrate any causality.

To palliate these defects, and thanks to the power of supercomputers, more or less complex models are then developed to try to reproduce portions of systems which, by the fitting of selected parameters, would behave in silico in the same way as the observable reality.

While model validation is relatively easy in engineering, it turns out almost impossible for systems having a large number of variables and which complexity and non-linearity make them barely intelligible. Even if a likely result is obtained, it would be false to infer from it that it is generally valid, because it is unverifiable outside of the range of the chosen parameters.

To develop a model is not a process of scientific or artistic research: it is made for a precise purpose. The designers of climate models try above all to incriminate the impact of the human activities, in particular greenhouse gas emissions; it is the mission assigned to them. If they had, simultaneously, to take other parameters of adjustment into account, the combinatorial complexity would exceed by several orders of magnitude the largest conceivable computing capacities. Besides, they would not obtain financial or technical support because they would not aim at the desired objective.

Scientific research does not know, must not know, what it is going to achieve; if it were the case we would already have found everything.

A model uses results from science, it cannot make science. Nevertheless, in economy, in ecology and especially in climatology we quickly fall in love with models which have the immense advantage to let believe in a technical objectivity. It is not enough to say that everything is connected and that it is necessary to think in an integral or global way so that, by the virtue of powerful computers, a virtual reality would model actual reality. But if the model output is in accordance with expectation then confirmation bias is guaranteed.

The alternative to models is the anecdote, widely used in militant ecology, but with very poor relevance beyond whistleblowing.

When we hang on to unverifiable thus irrefutable theories, to invalidated thus unusable models, we fall then in the ascientific domain of beliefs and oracles.

Merci de partager et de diffuser cet article !