Mathematics · the language of reality
Mathematics is what's left when every reference to physical material has been removed from a description that still works.
Kernel
Beneath science is mathematics — the layer at which the relations the universe seems to obey are stated without reference to any particular material instantiation. A circle is not a wheel; it is the locus of points equidistant from a center. A wave is not water; it is a partial differential equation. The 1960 Wigner essay "The Unreasonable Effectiveness of Mathematics in the Natural Sciences" names the central mystery of this layer: mathematics, invented from inside human cognition, turns out to fit physical reality so accurately that the fit cannot be coincidence. Whether mathematics is discovered (Platonism) or invented (formalism) is the oldest unresolved question in the philosophy of science.
Geometry, calculus, and the second opening
Euclidean geometry (c. 300 BCE) is the first canonical mathematical structure: a small set of postulates from which an unbounded family of theorems is derived. Two thousand years later Newton and Leibniz (independently, c. 1665–1675) open calculus, the second canonical structure, in which continuous change becomes algebraically tractable. The pattern of mathematics is: invent a vocabulary; show that the vocabulary's grammar generates structures with properties that were not visible in the vocabulary; find those structures already present in physical reality. The pattern is so consistent that it is itself one of mathematics's deepest open questions.
Group theory, topology, and the geometry of physical law
Évariste Galois (1811–1832) discovered group theory while dying in a duel at twenty. Sophus Lie generalized the concept to continuous symmetries (1870s). Emmy Noether showed in 1918 that every continuous symmetry of a physical system corresponds to a conserved quantity. Henri Poincaré founded topology by asking what is preserved under continuous deformation. Hermann Weyl, Élie Cartan, Eugene Wigner, and the postwar generation built the gauge-theoretic language in which the Standard Model is written. Modern physics is, structurally, applied group theory and topology. The fact that nature's deepest currently-known structure is expressible in mathematical vocabulary invented decades or centuries before it was found in physics is the strongest empirical evidence for the Platonist position.
Probability and the geometry of belief
Probability theory begins as a 17th-century correspondence about gambling (Pascal–Fermat, 1654). Kolmogorov axiomatizes it in 1933. Bayes's 1763 essay, neglected for a century, eventually becomes the mathematical foundation of modern statistical inference, modern machine learning, modern rational-epistemology theorizing, and the AI safety community's vocabulary. Probability is the mathematics that lets quantitative reasoning operate over incomplete information. Without it, the layers above (computation, AI, neuroscience) do not exist in their current form.
Gödel, Turing, and the limits of mathematical knowledge
Kurt Gödel's 1931 incompleteness theorems showed that any sufficiently expressive consistent mathematical system contains true statements it cannot prove. Alan Turing's 1936 paper showed that the question "will this computation halt?" has no general algorithmic answer. The two results together establish that mathematical knowledge is genuinely bounded in a way physical knowledge was not previously thought to be. The strange implication is that mathematics itself contains a frontier that the scientific procedure cannot fully reach — a result about the layer rather than about any object in it.
Mathematics and AI
Modern AI is, mechanically, linear algebra at planetary scale. A transformer is a matrix of attention weights operating on a sequence of high-dimensional embeddings. A diffusion model is an iterative application of stochastic differential equations on image manifolds. Reinforcement learning is dynamic programming on Markov decision processes. Each AI capability has a mathematical structure that pre-existed the AI by decades or centuries. The pattern repeats: mathematics, invented in advance, becomes the operational vocabulary of the next layer. The 2020s AI revolution is, on this reading, less a discovery than a sufficient-compute deployment of mathematical structures that had been ready for a century.
Open questions on this layer
- — Is mathematics discovered or invented?
- — Why is mathematics unreasonably effective at describing nature?
- — Are mathematical objects more real than physical ones?