The Philosophy of Science is Not Worth Caressing the Chin – It Makes Us Better Scientists | Marcus Munafo and George Davey Smith

… The worst part of philosophy is the philosophy of science; the only people, as far as I know, who read the works of philosophers of science are other philosophers of science.

This is the view of Arizona State University physicist Lawrence Krauss, author of the 2012 book A Universe from Nothing. He is certainly not the only physicist to criticize the philosophy of science. Richard Feynman, who shared the 1965 Nobel Prize in Physics for his work on quantum field theory, asserted that “the philosophy of science is as useful to scientists as ornithology is to birds.” It’s quite a position. In a recent commentary from Nature, we describe how a better understanding of one aspect of the philosophy of science, namely causal inference, can help us be better scientists.

What is causal inference? Quite simply, it’s about understanding whether X causes Y. For example, the majority of people who develop lung cancer are smokers, but does that mean that smoking causes lung cancer? In the 1950s, this was a legitimate research question: Could other differences between smokers and non-smokers explain the association, or could a third factor influence both? The causal effect of smoking on lung cancer is now clear, but it has taken years to establish and get the point across. This is because it is notoriously difficult to determine causation, and epidemiology has a knack for highlighting supposed links between, for example, behavior and health, which do not reflect cause and effect relationships. .

In science today, there is a huge interest in replicating the results, sparked by somewhat contested evidence that much of the published scientific results may be wrong, or at least misleading. If we took more care to repeat our studies, to verify that we get a broadly similar result each time, then our results would surely be more robust. Maybe, but a solid conclusion can always be wrong: X and Y can be correlated very reliably, but may not reflect a cause and effect relationship. This focus on replication stems from a widespread, but in our opinion incomplete, notion of falsification (supported by the philosopher of science Karl Popper) at the heart of the scientific enterprise. In fact, this is rarely how scientists work in practice.

Many scholars will refer to Popper if they are in a hurry to explain the basis of their inferences, but forgery is not everything. Another approach to questioning potentially causal associations is best explanation inference. Peter Lipton, the late Cambridge University science philosopher, described this as the search for the ‘most beautiful’, not just the ‘most likely’, an explanation characterized by ‘the scope, precision, mechanism. , unification and simplicity ”.

This process of arriving at the simplest and most probable explanation for an observation relies on the ability to approach the same question from different angles. Each approach will have its own biases and limitations, but if each gives the same answer, we can be more confident in the outcome. Known as triangulation, this is an approach that effectively complements traditional forgery.

In the current debate on reproducibility, too little is said about the need for triangulation. Scientists receive extensive training in a range of methods and skills, but very little in inference approaches. We can do better than that, emphasizing the need for triangulation and multidisciplinarity – approaching the same research question from multiple methodological perspectives that each have their own strengths and weaknesses. If we don’t, we can end up with solid results that are ultimately useless if our goal is to identify the causal risk factors that we can modify to improve health (as is the case with our own research. ). For example, having the yellow hand predicts a person’s risk of lung cancer, and this finding will replicate robustly in different studies. It might even help us predict who will get lung cancer. But unless we also use this information combined with other evidence to identify the underlying risk factor (cigarette smoking) and test it directly, this information alone will be of little help in understanding the causes of the disease. lung cancer.

What does triangulation look like in practice? We would like to see research teams from different scientific fields working on a single underlying question in a coordinated fashion. By approaching this question from different angles, they can determine whether their results agree or not. These teams work best when they understand the formal basis of the approach, and this comes directly from the philosophy of science. It might also help guard against the temptation to select outcomes from a range of approaches and present only those that work best. Research is already moving in this direction, with a renewed emphasis on “team science”. But a collaborative approach may not be enough on its own. We need to understand the intellectual foundations of causal inference to make real progress. Scientists cannot simply absorb this kind of framework by osmosis. We need formal training in some aspect of the philosophy of science to recognize its importance to our work.

Marcus Munafo is Professor of Biological Psychology at the University of Bristol. He studies the genetic and cognitive influences on addictive behaviors

George Davey Smith is Director of the MRC Integrative Epidemiology Unit at the University of Bristol, where his research focuses on the use of genetic information to improve causal inference in observational studies

Leave a Reply

Your email address will not be published.