Évariste Galois

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI traces the river of intelligence from hydrogen atoms through biological evolution to artificial computation, and everywhere asks what structure lies beneath the surface of capability. Galois is the clearest historical instance of the answer: the structure is symmetry, and the symmetry is a group, and the group is prior to the numbers and the equations and the computations—which is exactly what the field of geometric deep learning discovered independently. He offers the cycle something rarer than a precedent: a mathematical proof that the right way to understand a hard problem is to find the group of transformations that governs it, and to let the architecture follow from the symmetry rather than imposing a structure and hoping it fits.

He also offers the cycle’s hardest gift: the discipline of impossibility. The dominant mood of the AI moment is that enough scale and enough compute will eventually reach anything. Galois spent his life proving the opposite: that certain walls are not engineering gaps but mathematical facts, established with the same rigor as any positive result. The No Free Lunch theorems, the undecidability results of computability theory, and the limits that Judea Pearl’s ladder of causation identifies are all Galois-shaped truths: not failures awaiting a next version, but permanent features of the logical landscape. A culture that has internalized Galois holds two convictions with equal rigor—that symmetry-aware design is extraordinarily powerful, and that the very same mathematics of structure can prove that no amount of power will reach certain destinations.

His life also carries a corrective to the AI field’s dominant mythology. The legend that Galois scrawled all of group theory in one desperate night before the duel is largely false. The mathematics had been developed over years of rejected manuscripts and ignored submissions. The bottleneck was not creation but reception—the failure of the institutions of his time to recognize work that already existed. The AI field is full of the same story in reverse: ideas proposed before their time, ignored, and rediscovered when the conditions finally caught up. The romance of the midnight genius conceals the institutional and material conditions that determine whether good ideas survive—and attending to those conditions is more useful than celebrating the heroic individual.

Origin

Born in Bourg-la-Reine in 1811, Galois encountered mathematics at fifteen and was seized by it with an intensity that consumed the rest of his life. He failed the entrance examination to the École Polytechnique twice; was expelled from the École Normale for a political letter; was imprisoned more than once for his ardent republicanism; submitted memoirs to the Academy of Sciences that were lost, rejected, or returned as too compressed to judge. The mathematicians who received his work—Cauchy, Fourier, Poisson—were not stupid; they were encountering a level of abstraction the field had not yet developed the vocabulary to read.

The thread that made him indispensable was his answer to a three-century-old question: which polynomial equations can be solved by a formula involving only the basic arithmetic operations and the extraction of roots? For equations up to degree four, such formulas exist. For degree five—the quintic—Ruffini had argued and Abel had proved that no general formula exists. What Galois added was incomparably deeper: a complete criterion for which equations of any degree are solvable and which are not, expressed entirely in terms of the symmetry group of the equation’s roots. The unsolvability of the quintic was not an isolated fact but a consequence of an internal structural property of a specific group. He converted a single impossibility into a general theory of structure, and the theory would take mathematics half a century to absorb and the rest of the century to extend into every branch of the discipline.

He was killed in a duel on 30 May 1832, dying the next day at twenty. His letter to Auguste Chevalier, written the night before, was not the creation of group theory—that work already existed in his rejected manuscripts—but a plea for the existing work to be recognized and published. Joseph Liouville read it in 1843, recognized what it was, and published it in 1846. The world was then, finally, ready to read it.

Judea Pearl

Key Ideas

The group and the symmetry. A group is the mathematics of symmetry: the collection of all structure-preserving transformations of an object, together with the rule that performing one and then another is itself such a transformation. Galois’s founding act was to attach a group to a polynomial equation by asking which permutations of its roots preserve all the algebraic relations between them. The geometric deep learning program has performed the same reorientation on neural network architecture: instead of asking what function to learn, ask what symmetries the problem has, identify the group, and build an architecture that respects it by construction.

The impossibility theorem. An equation is solvable by radicals if and only if its symmetry group can be decomposed, step by step, into simpler pieces of a specific well-behaved kind—a property called solvability. The general quintic’s group cannot be so decomposed; it contains an indivisible core that resists the decomposition. The impossibility of the formula is a structural fact about the symmetry, not a failure of ingenuity. This is the template for Galois-shaped impossibilities throughout mathematics and computer science: the halting problem, the No Free Lunch theorems, the limits that Pearl’s ladder of causation marks between rungs.

Abstraction as the primary act. Galois’s abstraction was understood abstraction: he could say what a group was, why it captured the structure of the equation, what each step of the reasoning meant. The abstractions that deep neural networks build are different in kind: real and powerful, but opaque—the network finds structure without grasping it. The machines perform Galois’s act of abstraction without Galois’s comprehension of what has been abstracted. Whether finding-without-understanding is the same kind of thing as finding-with-understanding, or a categorically different process, is among the deepest open questions of the present.

The myth of the midnight genius. The legend of the night before the duel compresses years of rejected labor into a single romantic moment and falsifies the actual shape of the achievement. The lesson is institutional, not individual: Galois had the ideas; the conditions failed him. A culture that celebrates the lone breakthrough while ignoring the reception infrastructure that determines whether breakthroughs survive is a culture that will keep producing Galoises and losing most of their work.