The awarding of the 2024 Nobel Prize in Physics to a physicist (John Hopfield) and a psychologist (Geoffrey Hinton) for their contribution to the development of machine learning algorithms based on so-called “neural” networks came as a surprise to many. many physicists. The original purpose of the prizes created by Alfred Nobel at the end of the 19th centurye century being to reward an invention or a discovery useful to society, we understand that the Nobel committee did not want to miss what some hastened to describe as a “technological revolution”. But how can this prize be justified in the context of the Nobels? After all, the Turing Prize – often called the “Nobel in Computing” – had already saluted this computing breakthrough in 2018 by honoring Geoffrey Hinton and his two collaborators, Yann Le Cun and Yoshua Bengio!
Whatever one thinks, the choice of the committee of the Swedish Academy of Sciences raises the questions of what, ultimately, physics is and whether the work of the winners really falls within this discipline. The history of science teaches us that the boundaries between disciplines are shifting, or even disappearing. They were pretty much clear by the time Nobel wrote his will in 1896 and identified chemistry, physics and physiology or medicine. But only ten years later, the connection between chemistry and physics was already less so with the work of Ernest Rutherford and Niels Bohr on the structure of the atom. This is how, to his great chagrin, Rutherford was awarded the Nobel in chemistry and not in physics as he had hoped, while Bohr received the Nobel in physics in 1922. Since the 1960s, the border between chemistry and biology in turn weakened with the development of molecular biology.
One thing remains, however: the object of natural sciences remains non-living (physicochemical) and biological matter. But the surprise of the 2024 Physics Nobel is there: we do not see what objet physics is studied, because the work of the winners focuses on computer algorithms. This is where, in my opinion, the Nobel committee uses sophistry by asserting that because the winners used “tools” from statistical physics, what they are doing is physics.
When will there be a physics Nobel on the modeling of social phenomena?
In fact, it is essentially a use analog of the famous Ising model, dating from 1924, which analyzes the magnetism of a set of atoms having a spin of + 1 or – 1. A statistical physicist welcomed the committee’s choice and even said that neural networks have since long considered to be part of statistical physics! Now, there is a curious confusion here between methods and objects. The characteristic of mathematical methods, here those of statistical physics, is to be transposable, because they are in a way only syntaxes applicable to objects of very diverse nature which have nothing in common on the ontological level. Thus, we use the Ising model in “sociophysics” to model phenomena social like voting: voting for or against replaces the spin of the atom and we adjust the parameters of the model to reproduce empirical data. These same methods are also widely used in “econophysics”, a field whose subject here is the economic behavior of agents and not of atoms. And all this work appears in statistical physics journals!
In short, claiming that the winners do physics under the pretext that the mathematical methods used are the same is a sophism which implies that sociology and economics are now part of physics, because they are based on the same mathematics… When will there be a physics Nobel for the modeling of social phenomena?
