An assumption to prove a different theorem
Ernst Zermelo’s claim, made in 1904, was narrower than the axiom’s later fame suggests: he needed a way to guarantee that, given any collection of non-empty sets, even an infinite collection, it is possible to select exactly one element from each of them simultaneously, forming a single new set out of those selections. He introduced this as an assumption, now called the axiom of choice, specifically in order to prove a separate result, the well-ordering theorem, which states that every set, no matter how large or unusual, can be arranged into an order with a definite first element and a defined next element at every step. Without some version of this choice assumption, there was no way to guarantee such an ordering could always be constructed for an arbitrary infinite set, since no explicit rule might exist for picking elements from some of the sets involved.
Existence without a recipe
Zermelo built toward this over several years, publishing early work on transfinite cardinal arithmetic in 1902 before his 1904 proof of the well-ordering theorem, a result significant enough to earn him a professorship at Göttingen the following year. The proof drew immediate scepticism because it was non-constructive, meaning it established that the required choice function must exist without providing any procedure for actually specifying it, and Zermelo refined his argument in 1908 using a concept from Richard Dedekind to win somewhat wider acceptance. That same period, between 1905 and 1908, Zermelo also worked out a full axiomatic foundation for set theory itself, publishing it in 1908 even though he could not prove the system was free of internal contradiction. Abraham Fraenkel and Thoralf Skolem independently extended this system in 1922, adding further axioms to produce what is now called Zermelo-Fraenkel set theory, the standard foundation used in mathematics today.
One ball becomes two
What has held up, despite the discomfort the axiom originally caused, is its sheer usefulness across nearly every area of modern mathematics. Results now considered basic and indispensable depend directly on it, including the fact that every vector space has a basis, that an infinite product of compact topological spaces is itself compact, a result known as Tychonoff’s theorem, and that every ring with a multiplicative identity contains a maximal ideal. This breadth of application is largely why the mathematical community settled into broad acceptance of the axiom despite its non-constructive character: rejecting it would mean giving up theorems that much of algebra, topology and analysis now takes for granted, a cost most mathematicians have judged not worth paying merely to avoid an assumption that cannot be constructively justified.
A foundational system built around it
The clearest demonstration of how strange the axiom’s consequences can be is the Banach-Tarski paradox, which shows that, using the axiom of choice, a solid three-dimensional ball can be broken into a finite number of pieces and reassembled, using only rotations and translations with no stretching, into two solid balls, each the same size as the original. This does not describe anything achievable with an actual physical object, since the pieces involved are not measurable in the ordinary geometric sense the axiom’s critics found deeply counterintuitive, and it remains the clearest illustration of what is philosophically uncomfortable about the axiom: it does not merely make mathematics more convenient, it licenses genuinely bizarre constructions that would be impossible without assuming a choice function can always be made, whether or not anyone could ever specify what that choice actually is.
Neither provable nor disprovable
The question of whether the axiom is even necessary, in a strict logical sense, was answered decades later in two stages. Kurt Gödel showed in 1938 that assuming the axiom of choice cannot introduce a contradiction into set theory if none was already there, by constructing a model of the axioms, called the constructible universe, in which the axiom of choice holds. Paul Cohen completed the picture in 1963, using a technique called forcing to construct a different model in which the axiom’s negation holds instead, proving that the axiom of choice is logically independent of the more basic Zermelo-Fraenkel axioms: it can neither be proved from them nor ruled out by them. That leaves mathematicians free to adopt or reject it as a matter of choice rather than logical necessity, and the practical usefulness described above is largely why the overwhelming majority choose to adopt it.
Kept anyway, for what it makes possible
Yes, and the appeal here is in a genuinely unusual kind of mathematical drama: an assumption introduced quietly to prove one specific, fairly technical theorem turned out to carry consequences, the Banach-Tarski paradox chief among them, strange enough to unsettle mathematicians for decades, and important enough that giving it up would cost the field results it now considers foundational. It is also a clean illustration of a genuinely modern idea in mathematics, that some questions about the field’s own foundations can be shown definitively unanswerable from more basic principles, requiring a choice, made openly and for practical reasons, rather than a proof. Anyone who assumes mathematics is built entirely on statements everyone must accept will find that assumption usefully complicated here.