282 | Joel David Hamkins on Puzzles of Reality and Infinity
What this covers
Joel David Hamkins and Sean Carroll discuss the foundations of mathematics with a focus on what infinity reveals about the nature of mathematical truth. Hamkins, a logician at Notre Dame and pioneer of the Set-Theoretic Multiverse, argues that mathematical realism—the view that mathematical objects genuinely exist—need not entail a single, unique mathematical universe. Instead, he defends Mathematical Pluralism: the position that multiple, incompatible conceptions of sets can all be equally real and equally valid. The conversation traces how this pluralism arises from the pervasive independence phenomenon in set theory, where fundamental questions remain undecidable despite our best axioms.
The discussion ranges across centuries of mathematical thought, from Aristotle's and medieval potentialists through Galileo Galilei's paradoxes to the modern consensus shift toward treating infinities as actual and completed. Hamkins explains how Gödel's incompleteness theorems reveal an endless Consistency Strength Hierarchy towering above any axiom system, and why Dedekind's categoricity theorem—often cited to pin down arithmetic uniquely—secretly depends on set theory's own indefiniteness. The conversation addresses why the continuum hypothesis is already settled, not by finding a missing axiom but by understanding its varied behavior across possible universes under forcing. Hamkins also connects this to classical geometry's evolution from uniqueness to pluralism with non-Euclidean models, and touches on puzzles involving potentialism, the halting problem, and whether physical space's finitude might ultimately undermine our confidence in infinite mathematics itself.
Hamkins argues that mathematical realism need not imply uniqueness: there can be multiple equally real concepts of set (the set-theoretic multiverse), so questions like the continuum hypothesis are 'settled' by understanding their varied behavior across universes rather than awaiting a single missing axiom.
- The pervasive independence phenomenon (CH and other questions independent of ZFC) is itself evidence for pluralism rather than merely weakness of the axioms
- Platonism and uniqueness can be severed: one can be a realist and a pluralist about multiple platonic realms
- Categoricity arguments for a unique standard model of arithmetic fail because they rest on set theory, which is itself indefinite/pluralistic
Set theoretic independence from ZFC is ubiquitous across fundamental infinite combinatorics questions.
- Godel showed in 1938 via the constructible universe that CH (and the generalized continuum hypothesis) cannot be refuted from ZFC, and Cohen showed in the 1960s via forcing that the negation of CH is also consistent; from any set-theoretic universe one can manufacture extensions in which CH is true or false using the ontological resources available in that realm.
“It wasn't until 1938 that Godel had first proved that you can't refute it from Zermelo's theory because he produced this what's called the constructible universe... So then Cohen proved in the '60s that the negation of CH is also consistent with the axioms. And that's this forcing method.”
- The continuum hypothesis is the question of whether there is any infinity strictly between the countable infinity of the natural numbers and the uncountable infinity of the real numbers (the continuum); CH asserts there is nothing in between.
“the question of the continuum hypothesis is whether or not there is any infinity in between. So we have the natural numbers, that's the countable infinity, and we have the real numbers, the continuum. That's an uncountable infinity. And is there anything strictly in between? The continuum hypothesis says, no, there's nothing in between.”
- Although Godel hoped large cardinal axioms would settle the continuum hypothesis, Levy and Solovay proved that any standard large cardinal axiom is consistent with forcing extensions in which CH is true or false while the large cardinal retains its nature, so many set-theoretic statements including CH remain independent of all known large cardinals, undercutting the one-road-upward picture.
“any proposed axiom that settles the continuum hypothesis is going to therefore fail in one of these universes that we have just said are set theoretically acceptable. And therefore we can never find such a principle to be obviously true or manifestly true for sets in the way that is required. So if you have a principle and you prove that it settles CH, then that fact itself is undermining for the self-evident nature of your axiom.”
- Essentially every fundamental open question about infinite combinatorics in set theory turns out to be either completely trivial (provable or refutable immediately) or independent of the ZFC axioms, a pervasive and ubiquitous independence phenomenon demonstrated through thousands of forcing proofs since Cohen's 1963 work.
“Basically, all of the fundamental open questions in set theory have turned out to be independent of the ZFC axiom. Essentially, every question about infinite combinatorics that you might ask is either completely trivial and you can prove it immediately or refute it, or else it's independent of the ZFC axioms.”
- Hamkins and Miasnikov proved that although the halting problem is undecidable, for the one-way infinite tape Turing machine model it is decidable on a set of probability one: as program size grows, the proportion of programs whose halting status can be decided approaches 100%, because by Polya's recurrence theorem a random-walking head falls off the tape with probability one, though the result is fragile because it depends on the chosen computational model.
“the title of my paper with Alexei is that the halting problem is decidable on a set of probability one. So you can decide almost every instance of the whole.”
- Cantor proved that the real numbers form an uncountable set larger than the natural numbers, via the diagonal argument: given any proposed list of all reals, one can construct a new real differing from the first number in its first digit, the second in its second digit, and so on, so no list can contain all real numbers.
“If you had such a list of numbers, then you could write down another number which was different from the first number in the first digit after the decimal point, and different from the second number in the second digit and so forth, all the way down. And that number, it's a perfectly good number... it's different from all the numbers on the list.”
- The halting problem is the decision problem of determining, given a program and an input, whether the computation halts or runs forever; Turing proved in 1936 that no computable procedure can answer this correctly for all instances.
“The halting problem is the decision problem given a program and input answer the question whether it halts or not. And what Turing proved is that this problem has no computable procedure that will answer it correctly in all instances.”
Multiple coherent set-theoretic universes with incompatible truths can coexist as fully real mathematical objects.
- Dedekind's categoricity theorem proves that any two models of his second-order arithmetic are isomorphic, supposedly grounding a definite concept of finiteness; but because the crucial induction axiom quantifies over arbitrary sets of numbers, the definiteness of finite arithmetic is grounded in the definiteness of 'arbitrary set,' so if set theory is itself indefinite or pluralistic, the categoricity argument cannot establish definiteness or monism in arithmetic — different models of set theory have standard arithmetic models satisfying different truths.
“the attempt to show that our understanding of the standard model of arithmetic has failed because it's based on set theory. But if we think that there's indefiniteness or pluralism in set theory, then it can't possibly work to establish definiteness or monism in arithmetic.”
- Historically, being a Platonist in mathematics implied a unique mathematical universe, but this connection between Platonism and uniqueness has been severed: one can be a realist about mathematical objects while being a pluralist, holding that there are multiple coherent and fully real concepts of set, each giving rise to its own set-theoretic universe with incompatible truths.
“this connection between Platonism and uniqueness of the platonic realm has been severed. And I think this is a positive development because one can be a realist and a Pluralist just because you think that there are, say, multiple concepts of set each of which serves adequately as a foundation of mathematics, but with incompatible truths.”
- Contrary to those who treat the continuum hypothesis as still open, Hamkins argues CH is already settled: even though it is independent of ZFC, it is settled by our deep understanding of how it behaves across the multiverse — we can force it to be true or false, realize it in inner models, and understand all of that, and that understanding is the answer to CH rather than some single truth value.
“no, no, the continuum hypothesis is settled. Even though it's independent, it's settled by our understanding of how it behaves in the multiverse. We can make it true, we can force it to be true or force it to be false. We can make it true in inner models and so forth. And we understand all of that quite deeply. And that's the answer to CH not one.”
- Whereas a monist treats the independence of a question as showing the theory is too weak (pushing toward stronger axioms), the pluralist treats the independence phenomenon itself as evidence for pluralism, because we can already see how, by choosing the right conception of set, we obtain a set-theoretic universe in which the truth value goes one way or the other.
“the pluralist is going to say, well, the independence phenomenon itself is evidence for pluralism because we can already see how it could be that the continuum hypothesis is true or is false. If you use the right conception of set, you're going to get a set-theoretic universe in which the truth value is one way or the other.”
Non-standard models are inherent to first-order theories and cannot be eliminated by adding axioms.
- The problem of non-standard models is inherent and cannot be solved by adding axioms: every first-order theory, even 'true arithmetic' (the set of all statements true in the standard model), has non-standard models with all the same truths as the standard model, as a consequence of the compactness theorem.
“You can't ever solve the non-standard model issue by adding axioms because every theory will have non-standard models. Even if you think that there's a one true model of arithmetic that you're interested in, then the set of statements that are true in that model that's called true arithmetic. And it also has non-centered models. It's a consequence of the compactness”
- The density of independent statements within a formal system is strictly between zero and one: there are infinitely many independent statements (since trivial conjunctions of an independent statement remain independent), but the density cannot be zero or one because for a fixed independent statement a certain proportion of all statements involve it trivially.
“It's strictly in between zero and one, I mean the density. So there's certainly infinitely many statements because if something is independent, then I can just say if phi is true but not provable, then phi and phi and phi...”
- The common belief that the phrase '0, 1, 2, 3, and so on' describes a unique mathematical structure is not as clear as it seems, because the 'and so on' is hopelessly vague; to make it precise and prove uniqueness requires the Dedekind categoricity theorem, but that theory's induction axiom quantifies over arbitrary sets of numbers.
“when I talk about the natural numbers 0, 1, 2, 3, and so on, this is describing a unique mathematical structure... But is it really? Because this and so on, of course is hopelessly vague.”
Large cardinal axioms align with consistency strength hierarchy, suggesting a monist universe view competing with pluralism.
- Godel's theorem reveals that towering above any theory is a transfinite hierarchy of consistency strength: one can repeatedly add the consistency of the previous theory (PA + Con(PA) + Con(PA+Con(PA))...) and continue transfinitely, each step provably stronger and not provable in the prior theory.
“one way of thinking about Godel's theorem is that it has identified towering above any theory that you have is this hierarchy of consistency strength. And that hierarchy is, it provably exists because of the inconvenience theorem, and it's getting stronger and stronger consistency assertions.”
- Large cardinal axioms express natural infinite combinatorial principles (not self-referential logical statements), yet they line up exactly along the consistency strength hierarchy Godel predicted — stronger ones like a measurable cardinal imply the consistency of weaker ones like an inaccessible cardinal — which lends strong support to the monist universe view that we are on the one true path upward in set theory.
“the set theorists find this incredibly tall tower of axioms, the large cardinal axioms that instantiate exactly that predicted feature in a very robust way. And so it seems like we're on the path for the one true theory of sets.”
- Hilbert wanted to settle the antinomies of early 20th century set theory by holding the strong infinitary theory at arm's length and proving its consistency by purely finitary means, treating the infinitary theory as merely symbols pushed around on paper (formalism) so its safety could be established finitarily.
“So in the weak finitary theory, Hilbert wanted to prove this safety of the strong infinitary theory. And I think it's a very sound thing to want to do.”
- A simplified version of Godel's incompleteness theorem states that there is no computably axiomatizable theory of arithmetic that is true and proves all and only the true statements of arithmetic; this follows from the undecidability of the halting problem, because such a complete true theory would let us decide halting by searching for proofs that a program halts or not, contradicting Turing's 1936 result.
“if there were a theory that knew all the answers, then we could solve the halting problem. But we can't solve the halting problem because that's what Turing proved in 1936. And therefore, there can't be such a thing.”
- Godel refuted Hilbert's picture by showing not only that the finitary theory cannot prove the consistency of the strong infinitary theory, but that it cannot even prove the consistency of itself.
“Godel completely refuted this whole picture by showing not only can the finitary theory not prove the consistency of the strong infinitary theory, but it can't even the consistency of itself, of the finitary theory itself.”
- By Godel's second incompleteness theorem no theory can prove its own consistency, so if Peano arithmetic is consistent then the theory PA plus the axiom asserting PA is inconsistent (PA + ¬Con(PA)) is itself a consistent theory — even though it appears incoherent in asserting a theory while also asserting that theory's inconsistency.
“it follows, it's just immediate from that, that you can add the negation of the consistency assertion as an axiom, and it will still be a consistent theory, So PA plus not con PA... if PA itself is consistent, then that theory is consistent”
The continuum hypothesis is settled not by finding a true value but by understanding its behavior across the multiverse.
- The 'dream solution' of settling CH by finding a missing axiom that is self-evidently true for sets and decides CH is impossible: because we already understand that worlds where CH is true and worlds where CH is false are both perfectly acceptable set-theoretically, any proposed axiom that settles CH must fail in one of those acceptable universes, and the very fact that an axiom settles CH undermines its claim to be self-evidently true.
“any proposed axiom that settles the continuum hypothesis is going to therefore fail in one of these universes that we have just said are set theoretically acceptable. And therefore we can never find such a principle to be obviously true or manifestly true for sets in the way that is required. So if you have a principle and you prove that it settles CH, then that fact itself is undermining for the self-evident nature of your axiom.”
Mathematical philosophy should be judged by where it leads investigation, not by negative skepticism alone about infinity and foundations.
- Mathematicians instinctively prefer weakening hypotheses to obtain stronger, more widely applicable theorems, but applied to foundations this preference yields weak foundational theories with a pluralist affinity; by contrast, if the goal of foundations is to identify the fundamental truths that paint a coherent picture of mathematical reality, one should prefer strong-but-true axioms, which carry a monist perspective.
“if you have a preference for weak foundations, then you're sort of implicitly suggesting that there would be multiple alternative mathematical realities. It's connected with pluralism.”
- For millennia almost all mathematicians were explicitly potentialists about infinity (holding that you can always have more but never a completed totality), but around the late 19th century the consensus shifted so that nowadays almost all mathematicians are actualists who accept completed actual infinities, reversing the historical default.
“throughout most of history, for millennia, almost all mathematicians were potentialists quite explicitly... whereas nowadays almost all mathematicians are actualists and it sort of made a change around the late 19th century.”
- Philosophies of mathematics saturated with skepticism about infinity are unattractive unless they offer more positive insight into mathematical questions; a purely negative skepticism that merely says we're all wrong without any positive benefit holds little value, and philosophies should instead be judged by where they lead mathematics to the most interesting investigations.
“unless the competing vision is offering me more insight into questions, if it's only a kind of negative skepticism that's saying, oh, we're all wrong about these things, but it doesn't have any positive benefit, then I don't really find much attraction to that point of view.”
- A philosophical question (like the nature of potentialism, going back to Aristotle) can give rise to a rich mathematical theory (the modal logics of possibility such as S4.2 and S4.3), which in turn feeds back to refine the original philosophical understanding by revealing distinct conceptions — exemplifying a virtuous cycle between philosophy and mathematics.
“I just love that situation where a philosophical question gives rise to this mathematical analysis, which then feeds back in and ultimately refines the philosophical understanding. I think that's just fantastic.”
- Doing philosophy of mathematics well requires an uncommon combination of high-level technical mathematical skill and a philosophical way of thinking; this rarity makes advances difficult, and the field is becoming more productive as philosophers grow more mathematically sophisticated and more mathematicians engage philosophical questions.
“to do philosophy of mathematics really well and to understand some of the... To master the important questions and phenomenon, you really have to have a high level of mathematical skill, but you have to, of course, also have a philosophical outlook. And so it's this combination, which isn't so common”
- The most recent work severs potentialism from infinity: the core idea of potentialism is that the universe of mathematical objects might be unfinished, not that it is infinite — even fragments with actual infinities inside them (e.g. chopping the cumulative set-theoretic hierarchy at a high ordinal) can be treated potentialistically, so potentialism is really about whether the universe is completed.
“the core idea of potentialism is that the universe of mathematical objects might be unfinished. This is the essence of potentialism. Even if fragments of that universe have actual infinities inside them... potentialism isn't really about infinity, it's about whether the universe is completed.”
- Galileo argued against the potentialist point of view by noting that a potentialist, in holding that infinitely many further things are always possible, is thereby committed to an actual infinity of possibilities — turning the potentialist's own position into a covert acceptance of actual infinity.
“if you're a potentialist, then you think that all of these other things are possible. And so you are committed to an infinity, an actual infinity of possibilities. And so he was trying to undermine the potentialist perspective... By saying that the potentialists were committed to an actual infinity of possibility”
- There is a coherent non-standard form of potentialism in which not all smaller numbers are presupposed when you have a large number: because a typical number less than a googolplex has about a googol random digits and could not be described or recited even at a million digits per second since the Big Bang, such numbers cannot be held as objects of thought, so some numbers (those with simple descriptions) can 'come into existence' earlier than most smaller numbers.
“even if you recited a million digits every second since the beginning of time, you wouldn't be able to even describe a single one of those numbers. And so there's really no sense in which you could ever hold such a number as an object of thought in your mind.”
- If space and time turn out to be discrete and finite at a tiny scale — which Feynman reportedly named as the basic physical assumption most likely to be overturned — then the actual infinities of mathematics could be ultimately irrelevant to physics, and we might lack good reason to believe even that Peano arithmetic is consistent, since our intuition of infinity rests on physical pictures like an endless row of telephone poles that would be incoherent in a finite universe.
“if it really is just discreet and finite, then maybe the whole universe is finite in that way... a lot of people have this understanding of infinity because they see this sequence of telephone poles on the highway in the Arizona desert, receding into the distance, they imagine it's infinite. But if the universe turns out to be finite, then that picture is not actually coherent... maybe we don't have good reason to think that PA is consistent.”
Pluralism in mathematics mirrors the historical transition from unique Euclidean geometry to multiple equally real geometric spaces.
- For thousands of years geometry was viewed as the study of the one true space, but with the discovery of non-Euclidean geometries (hyperbolic, spherical, elliptical) the concept splintered into a pluralistic but still fully real domain — Euclidean and hyperbolic space are each fully real as abstract objects, showing that having multiple incompatible models does not reduce a subject to mere formal activity.
“with the discovery of non-Euclidean geometry and hyperbolic space and spherical geometry, elliptical geometry, and so forth, the concept splintered and it became pluralistic. But still real.”