David Hilbert

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI asks what it would mean to see these systems clearly. Hilbert is the right ancestor for that project because he is the one who most rigorously tried to turn reasoning into a mechanical procedure—and the negative results that demolished his program are now the most solid ground available for understanding what the machines we have built can and cannot do. The incompleteness theorems and the halting result are not speculations about the limits of AI. They are theorems, as solid as anything in mathematics, and they apply to any formal system of sufficient power, which every capable AI system is.

The most immediately practical limit concerns alignment and self-certification. Gödel's second theorem says that no sufficiently powerful consistent system can prove its own consistency: self-validation, from within, is exactly what cannot be had. Every dream of a self-certifying, provably-aligned artificial intelligence—a system that audits its own trustworthiness and guarantees its own safety from the inside—runs directly into this result. Trust must be anchored somewhere other than the system's own assessment of itself, because no system of sufficient power gets to be the final judge of its own soundness. Hilbert wanted mathematics to guarantee itself; Gödel proved none can. The engineers who hope AI will self-verify its alignment are making Hilbert's error.

His story also provides the cycle's clearest warning against the optimism that does not know its own limits. The conviction that a large enough model trained on enough data will eventually answer anything we can frame is the Hilbert program in the language of machine learning. The Entscheidungsproblem is its clearest ancestor, and the answer Turing produced is as definitive as any negative result in the history of thought. There is no general decision procedure; no algorithm that handles every case; no scaling path from “computes many things” to “computes everything,” because the set of computable functions is countably infinite and the set of all functions is not. The machines can get better and better at the instances that matter in practice, but they cannot become a general solver of an undecidable problem, because none exists to be approximated toward.

Origin

Born near Königsberg in 1862, Hilbert established himself at the University of Göttingen in 1895 and built it into the world center of mathematics. His early career reshaped algebraic number theory and invariant theory before his 1899 Grundlagen der Geometrie rebuilt Euclid from explicit axioms, demonstrating that the proofs held regardless of what the words “points, lines, and planes” referred to—the proof-theory lived in the structure of the axioms, not in the meaning of the terms. This was the formalist move, and everything followed from it.

In 1900, before the International Congress of Mathematicians in Paris, he presented twenty-three unsolved problems that set the agenda for a century of research. His second problem demanded a proof of the consistency of arithmetic. His tenth demanded an algorithm for solving Diophantine equations. Both were posed in confident expectation of positive answers; both turned out to be impossible in the forms he intended. The 1900 address also contained his direct rejoinder to the philosophical pessimist Emil du Bois-Reymond, who had argued that some questions lie permanently beyond science. Hilbert's answer was total: there is no ignorabimus in mathematics. We must know—we will know.

That confidence was tested and broken within his own lifetime. In 1930, the day before Hilbert delivered a radio address that closed with the words carved on his tombstone—Wir müssen wissen—wir werden wissen—a twenty-four-year-old named Kurt Gödel remarked at the same conference that he had proven any consistent system rich enough to do arithmetic must contain true statements it cannot prove. The refutation and the creed share a calendar, and the man who built the paradise had not yet heard of the theorem that proved it permeable.

Key Ideas

Formalism. The claim that mathematics is the rule-governed manipulation of symbols, and that whether something is a proof can be checked without understanding what any of it means. This is simultaneously the founding insight of formal logic, the founding intuition of every symbolic AI system, and the source of a gap—between syntax and semantics, between manipulating correctly and meaning something—that neither mathematics nor artificial intelligence has closed. A large language model inhabits this gap with the same structural condition: it manipulates tokens by learned arithmetic rules, with no symbol that is intrinsically about anything. The meaning is supplied by the reader, exactly as Hilbert placed it.

Hilbert’s program and its impossibility. The three-part demand that mathematics be complete (every truth provable), consistent (no contradiction derivable), and decidable (a mechanical procedure to settle any question). Gödel's first incompleteness theorem destroyed completeness, his second destroyed self-certified consistency, and Turing's halting result destroyed decidability. The negative answer to the Entscheidungsproblem required defining computation itself, and the definition—the Turing machine—became the blueprint for every computer. Hilbert's refuted dream is the load-bearing wall of the digital world.

The incomputability ceiling. The set of all computable functions is countably infinite; the set of all functions from numbers to numbers is uncountably larger. Almost everything—almost every function that exists—is uncomputable by any algorithm. This is not a temporary engineering limitation but a permanent structural fact, established by Cantor's diagonal argument before transistors existed. There is no scaling path to universal computation; more capable machines occupy a larger island of computability in an ocean whose shore they cannot reach.

We must know—the productive form of being wrong. Hilbert's optimism was wrong as a prediction and productive as a method. A sharp wrong conjecture is more useful than a vague right caution: it tells people exactly what to attack, and the attack, even when it overturns the conjecture, produces the real knowledge. The incompleteness theorems are the children of his second problem; Turing's theory of computation is the child of the Entscheidungsproblem. The field of computer science exists because Hilbert demanded answers that turned out not to exist, and proving their nonexistence was one of the most consequential intellectual achievements in history. His tombstone creed is right in the chastened form: we must keep trying to know, especially about the things we have proven we cannot fully know.