Tim Palmer. The Primacy of Doubt.

"With four adjustable parameters, I will 
make you an elephant. With five, I
can make it wiggle its trunk". 

                              N. Wiener

      This is a strange book written by a beautiful mind, a genius who is a little bit crank. The most objectionable part of the book is his disquisitions about the quantum gravity. And not because he does not understand it. He studied for his PhD under the same Dennis Sciama who was also Hawking's dissertation adviser and had to suffer Robert Penrose's questions at his defense  -- but it used to be a tradition that one presents her/his ideas first to the professional community, and only when they are accepted, peddles them to the laypeople.  Palmer's desire to use fractal geometry and p-adic numbers is commendable, though I, understandably, cannot understand most of it. And his description in a book is lucid, as always. If I gather his intention correctly, his program must include formulation of the Quantum Field Theory on the fractal embedded into the manifold, which is locally isomorphic to the Minkowsky space, at least at some scale. Then he must build the field operators using p-adic coordinates on that fractal. 

        This program, to the degree of my limited comprehension, was already proposed by V. S. Vladimirov in the USSR, like Palmer, a member of the National Academy, a contributor to the Soviet counterpart to the "Manhattan project", but also a brilliant teacher, and a vicious anti-Semite (Владимиров В. С., Волович И. В., Зелёнов Е. И. P-адический анализ и математическая физика М., 1994). I cannot judge the success of this program but, at the first glance, it should not allow the existence of black holes because the points with strong gravity, which are close in our physical space, can be very distant in the p-adic metric. Or, so I think. And, after 2018, their existence is an experimental fact. 

    Using that much space to criticize the book (better to say, expressing DOUBTS on its treatment) I must say that Palmer is an excellent storyteller. Because of his main field of mathematical meteorology, he is biased towards gigantic models with billions of parameters and he wants more. But, as my epigraph suggests, and everyone in the field of deep learning knows, there is such problem as overfitting. With sufficient slack in a model and sufficient computer power, one can fit anything to everything. The unparalleled brilliance of the General Theory of Relativity was precisely the fact that Einstein's construct was so tight, not allowing variations, for instance, in the velocity of gravitational waves, or in relativistic corrections to the orbit of Mercury. 

            Unlike quantum physics, where I am an enlightened amateur, I understand next to nothing in meteorology. So I cannot question Palmer's assertion that gigantic agent-based models with umpteenth parameters can be as useful in biostatistics for propagation of epidemics, traffic control and finance, if only we reach a sufficient computing power. 

His proposal of a "stochastic computer", which uses thermal noise in semiconductors to simulate random processes is quite interesting, and he claims that the specialists in that field already appropriated his ideas. I believe him.