The Paradise and the Serpent
EDO SEGAL: Professor Hilbert, when you wanted to express your deepest commitment in a single sentence, you reached not for your own work but for Cantor's. "No one shall expel us from the paradise that Cantor has created." You wrote that in 1926, defending the theory of the infinite against the mathematicians who wanted it banned. I find that line almost unbearably poignant, and I'll tell you why at the end of the round. First — tell the audience what paradise you were defending, and why it mattered enough to fight for.
HILBERT: Cantor had discovered that infinity comes in sizes. The counting numbers, the fractions, even the algebraic numbers can be put in a list — they are countable. But the real numbers, the full continuum of points on a line, cannot be listed. There are strictly more reals than integers; the infinity of the continuum is larger than the infinity of counting. And he proved it with an argument of such elegance that I would have defended it against an army. The diagonal argument. Suppose you had a complete list of all the reals between zero and one. Go down the diagonal, change each digit, and you have constructed a number that differs from the first entry in the first place, the second in the second, and so on — a number that cannot be anywhere on the list. The assumption that the list was complete destroys itself. The continuum overflows every enumeration. The finitists called this a disease. I called it a paradise, because it was the most beautiful country mathematics had ever found, and I would not let small men wall it off out of fear.
EDO SEGAL: And here's the poignancy, Gödel — you tell this part, because it's your inheritance. The diagonal argument. The thing Hilbert defended. It's the murder weapon, isn't it?
GODEL: It is. That is the serpent in the paradise, and Professor Hilbert guarded the garden it lived in. The same move — assume you have captured everything in a list, then diagonalize to construct the thing the list missed — is the engine of every great limit theorem of the century. My unprovable sentence G is built by a diagonal construction: it refers to itself by indexing into the list of all formulas, exactly as Cantor's number indexes into the list of reals. Turing's halting result is a diagonal argument: assume a machine that decides halting for all machines, then build a machine that does the opposite of what the decider predicts for it, and contradiction follows. The proof that the reals are uncountable, the proof that provability is incomplete, the proof that halting is undecidable — one method, three walls. Professor Hilbert defended the ground on which his own program would be proven impossible. He guarded the paradise, and the paradise contained the serpent, and the serpent and the angel were the same mathematics seen from two sides.
EDO SEGAL: Now bring it to the machine, because there's a counting argument here that should sober every person who's ever said a big enough model could compute anything. Gödel.
GODEL: It is stark and it is exact. Anything a computer can do is countable. Every program is a finite string of symbols, and finite strings can be listed; so the set of all programs is countable, and the set of computable functions is countable. But the set of all functions from numbers to numbers is uncountable — of Cantor's larger size. Put those facts together. The functions a machine can compute are a vanishing speck, a countable dust, inside the uncountable ocean of all functions. Almost every function is uncomputable. This is not a limitation of current hardware. It is a limitation of the concept of computation itself, established by counting. When someone tells you a sufficiently advanced machine intelligence will eventually compute anything we can specify, the answer is Cantor's: the specifiable-and-computable is a measure-zero island, and no amount of compute enlarges the island to cover the sea.
HILBERT: And now watch me do the thing I do, because it is the move of the evening and I want the reader to learn its shape. Gödel says: almost every function is uncomputable, therefore the dream of the universal machine is hopeless. And I say — which functions? Look at the speck. Look at that vanishing countable dust of the computable. It contains every function that has ever mattered to science, to engineering, to a human life. The motion of the planets is in the dust. The folding of a protein is in the dust. The structure of every language you have ever spoken, every image you have ever seen, every theorem worth proving — in the dust. The uncountable ocean of "all functions" is, almost entirely, an ocean of noise — arbitrary, patternless, structureless assignments that correspond to nothing anyone would ever want to know. You have proven that the machine cannot compute the random. Congratulations. Neither can the universe. The functions that carry meaning cluster, overwhelmingly, in exactly the region the machine can reach, and that is not a coincidence — it is because meaning is structure, and structure is what computation captures.
GODEL: That is the best answer that objection has, and I will not pretend it is weak. But it concedes my point while dressing it as a refutation, Professor Hilbert. You say the meaningful functions cluster among the computable. Perhaps. But which meaningful truths fall outside? My theorem names one. The consistency of arithmetic is not noise. It is among the most meaningful statements a reasoner could utter, and it is true, and it is not provable in the system. You cannot wave at the ocean and say "it is all noise out there," because I have walked into the ocean and brought back a pearl — a true, meaningful, humanly graspable statement that your machine in that system cannot reach. The question is not whether most of the inaccessible is noise. It is whether any of the accessible-to-mind-but-not-to-machine is treasure. And I have shown you one piece, and where there is one, the honest position is to suspect there are more.
EDO SEGAL: Here's the poignancy I promised, and I want to close the round on it because it's the most human thing in the mathematics. Cantor was treated abominably in his life — his ideas attacked as a disease, his career obstructed, his mind broken under the strain. Hilbert's "paradise" line was an act of loyalty to a persecuted man. And the irony compounds beyond bearing: in defending the infinite, Hilbert was defending the exact mathematics that housed the proof of his own program's impossibility. He guarded the paradise, and the paradise contained the serpent that bit him. Professor Hilbert — knowing now what the diagonal argument would do to your life's work, would you defend Cantor's paradise again?
HILBERT: Without a moment's hesitation, and the question almost offends me. You think I would trade the truth for my comfort? You think I would expel us from paradise to protect my program? The whole point of a paradise is that it is real whether or not it is convenient to you. Cantor was right about the infinite. Gödel was right about the limit. They are the same mathematics, and I would rather inhabit a true country that contains my defeat than a false one that flatters my dream. That is the orange pill, Mr. Segal, if you want to know what it tastes like to a mathematician. It tastes like defending the garden that grew the serpent that killed your life's work, and calling it paradise anyway, because it is.
EDO SEGAL: I have rarely heard a better definition of the thing this whole series is about. The infinite marks one outer wall — the uncountable, the uncomputable, the undecidable. But there's a second wall, closer in and more unsettling, that Hilbert's own method throws into relief: not what the machine cannot compute, but whether the symbols it computes with mean anything at all. The gap between getting the symbols right and the symbols being about something. After this.