The Inconsistency of Arithmetic (2011)

> He taught himself to read by age 2.

Well, I taught myself to read by the age of 3...

> He was taking Calculus at age 7.

...OK, clearly something went awry with me between the ages of 3 and 7.

The IQ figure of 230 can't be calculated on the usual scale with a mean of 100 and a standard deviation of 15, because if so it would mean that a world with our current population only had a 1 in 57 million chance of containing someone as smart as him. How could we possibly know enough about population statistics to know such a thing? Especially given that we do in fact have him in our world.

Unless he was born on February 29th (good luck getting that drink at a bar), reading after only hitting two birthdays is the stuff of legend.

That said, having been told so by a genius, I feel like I'm now unshackled, ready to pursue a life of mathematics as a non-genius.

It’s very odd to put Erdos amongst Gauss, Euler and Tao. He was very proficient, but nowhere near as talented.

What's the source for some of those claims?

Calculus at seven I can believe, but teaching himself reading at 2 sounds like something that goes against what we know about biological and social development of the child.

And official IQs of 230 and 300 sounds like someone doesn't understand what IQ is or the meaningfulness of measuring/quantifying such in a standardised way.

Obviously the person may be exceptional, and I do not mean to take anything away from Terence in his work that i'm clearly unqualified to comment on (it wouldn't surprise me if a bit of digging shows him unconnected to such claims), but we shouldn't just accept such things as given. Extraordinary claims require extraordinary evidence...

  whereas the cognitive layers required for reading take time to develop in the child's brain, and you first have to wait for the verbal/aural language systems to develop the actual language structure first, then tack on some symbolic representation for letters, then understand composition for words, then attach words to their meanings, etc. You can skip some of those bits to perform party tricks (I.e rote symbolic recognition and repetition of aural sounds on cue), but this does not reading make. Reading we believe, from a cognitive science point of view, we think is a kind of a kind of kludge on top of later developed brain mechanics, whereas it's aural/verbal language that appears as an almost innate developmental ability around that age. (I say innate, but you still have to be exposed to other humans interacting and speaking for a young mind to learn the language).

I wonder if only geniuses can write "it doesn't take a genius" posts, because making truly hard things easier actually does take a genius.

Without even reading a single word of it, the non-irony is in your characterization. He taught himself. Ultimately we all taught ourselves and our peers and mentees. Intelligence is mostly rooted in curiosity and attention, not some innate capacity. People who are perceived to be at the lower end of that capacity often inform those who aren’t, and vice versa.

Classy retractions are rare, or at least uncommon. We should all be more like them.

So, Nelson had claimed to have proven Peano arithmetic inconsistent.

Peano arithmetic is a set of axioms that describe basic properties of the whole numbers (nonnegative integers). It's a pretty simple set of statements. The axioms are described in terms of 0, S (the successor function, S(n) = the next number after n), plus, and times.

The axioms are:

1. If Sn = Sm, then n=m

2. For all n, Sn is not 0

3. For all n, n+0=n

4. For all n and m, n+Sm = S(n+m)

5. For all n, n*0=0

6. For all n and m, n*Sm = n*m + n

7. [Induction] If P(n,m_1,...,m_k) is a predicate, and m_1,...m_k are whole numbers, such that:

A. P(0,m_1,...m_k) holds, and

B. For all n, P(n,m_1,...,m_k) implies P(Sn,m_1,...,m_k)

Then for all n, P(n,m_1,...,m_k) holds.

...OK, that last one is maybe a bit complicated. Technically, that one is actually what's called an axiom schema, that generates infinitely many axioms, one for each possible predicate of one or more variables (note you can have k=0). The theory is about whole numbers, not about predicates; the theory can't actually talk directly about predicates. Anyway, that's a bit of technical detail you don't really need to know for these purposes, so let's just move on. If you didn't understand that part, it's OK.

As you probably know, if we have a mathematical theory, and that theory is supposed to describe something that is supposed to actually exist (such as the whole numbers), that theory had better be consistent.

Firstly, because if it's not consistent, it can't possibly be true (the technical term is "sound"). Reality is consistent, so if something is inconsistent, it's incorrect.

Secondly, because of the principle of explosion. If you know P and not P for some statement P, you can conclude any statement Q. This principle of logic may be counterintuitive, but it's pretty essential. Some people have made a version of logic that don't include it ("minimal logic"), and it's basically unusable.

Suppose you know both P and not P, and you want to prove a statement Q. Well, you know P, so you certainly know P or Q. But you also know not P; that eliminates P as a possibility, leaving Q. So in order to get rid of the principle of explosion, you'd have to get rid of the principle that if you know A or B, and also know not A, then you can eliminate A as a possibility to conclude B. That's a pretty big tool to go without!

(Well, or you'd have to get rid of the possibility that if you know A, you can conclude A or B, but that's even worse.)

So, if a set of foundational mathematical axioms is found to be inconsistent, it could be something of a disaster for mathematics. People would need to come up with new, weaker axioms that somehow didn't lead to this contradiction. This would be a difficult thing to do -- which principles do you keep, and which do you jettison? (And which do you replace with weaker versions? And what new weaker ones do you invent to take the place of ones that had to be tossed?) That's not an easy question!

So let's say that the ZFC axioms were found to be inconsistent. The ZFC axioms are a set of axioms (and axiom schemas) that describe set theory, and basically all of "ordinary mathematics" can be founded on them. (I'm not going to list them all here; they're not that complicated, but they're rather more complicated than Peano arithmetic.)

If the ZFC axioms were found to be inconsistent, well, it could be a project of years or decades to come up with a new, hopefully consistent, foundation of mathematics. It would be quite the problem. But... it wouldn't destroy mathematics. After all, ZFC itself is what people came up with after the earlier attempts to axiomatize set theory were found to lead to contradictions, so this has in a sense happened once before.

So why would it be such a big deal if the Peano axioms in particular were found to be inconsistent? Why would do people say that would destroy mathematics?

There are a few reasons for this.

Firstly, the Peano axioms describe the whole numbers, rather than set theory. Set theory is kind of out there -- who can really say what should or should not be true of vastly infinite sets? Yeah, the ZFC axioms all seem like they should be true, but so did the axiom of unrestricted comprehension, and that turned out to be no good. The Peano axioms, by contrast, describe the whole numbers, and are pretty basic statements about them. They had better be true, or else we are very wrong about the whole numbers!

But, that's not the only reason. After all, if that were it, it would still likely be possible to recover from the problem by passing to a weaker system of axioms. And there are various ones that have been proposed! The Peano axioms are mostly pretty simple, but that last one, induction, has a lot of hidden complexity to it. People have suggested ways you could weaken the induction axiom by limiting what sorts of predicates it applies to. And that would certainly be contentious, but if Peano arithmetic were found to be inconsistent, we'd have to. Thing is, that wouldn't solve the real problem.

The problem is that weakening the induction axiom is only really a possibility if you want to describe the whole numbers in isolation. In reality, we don't consider the whole numbers in isolation, described by the Peano axioms; but rather as part of the broader picture of mathematics, described by ZFC. We don't study just whole numbers, but sets of whole numbers, whole numbers interacting with real numbers and complex numbers and p-adic numbers, whole numbers interacting with graphs and groups and partitions, whole numbers interacting with vector spaces and manifolds and all the rest of mathematics.

So, in reality, if Peano arithmetic were found to be inconsistent, the real task wouldn't be to weaken the Peano axioms so they'd be consistent; it would be to weaken the axioms of ZFC, in such a way that that would weaken the Peano axioms to remove the contradiction.

But that's just about an impossible task. It'd be nearly impossible to weaken the axiom schema of induction, while also leaving a theory that can interact with the rest of mathematics. Because you see, if we want the whole numbers to be able to interact with the rest of mathematics, then we have to be able to talk about sets of whole numbers.

And if we can talk about sets of whole numbers, then we can talk about the following variant of induction: Let S be a set of whole numbers, and suppose that 0 is in S, and, for each whole number n, n being in S implies n+1 is in S. Then all whole numbers are in S.

Now this may sound like the same thing I said above -- they're both just induction, right? But this one is about sets, not predicates, because now we're in a setting where we can talk about sets. And this one is more general -- because, if we have a predicate about whole numbers, we can form the set of whole numbers satisfying it. Whereas notionally one could have a set not described by any predicate.

So, if you have this statement (set induction), you get all of predicate induction. So you'd somehow have to weaken the axioms of mathematics such that:

1. You can still talk about sets of natural numbers, but

2. You don't get predicate induction.

And how on earth would one do that?? Stopping set induction sounds pretty much like a dead-end; like if you did that, how would you get any form of induction at all? (And if you don't have any form of induction, then you don't really have the whole numbers; it's kind of their defining feature.) I mean you could make special assumptions about whole numbers, but if we're trying to make more general set-theoretic axioms, where whole numbers aren't fundamental, those don't really belong.

So the remaining option then is to stop the link from set induction to predicate induction, by limiting what sorts of sets you can make, so not every predicate on whole numbers can be used to form a set of whole numbers.

But (while I think that's considered less impossible), that's not really a viable option either! Because that would mean that somehow you'd have to introduce, into your set theoretic axioms, restrictions on what predicates can be used to form sets; and while there are sensible ways one could formulate restrictions in the limited context of whole numbers, there's not really any good way to formulate such restrictions that would make sense in the broader context of mathematics as a whole.

So, basically, we'd be stuck. Coming up with new axioms if ZFC were proved inconsistent would be difficult but doable. Coming up with new axioms if Peano arithemtic were proved inconsistent seems basically impossible. So, finding a contradiction derivable from the Peano axioms could indeed destroy mathematics.

Of course, since people generally expect that the Peano axioms are true, nobody really expects that to happen. But Edward Nelson -- an adherent of the fringe mathematical school known as ultrafinitism -- disagreed, believing induction was probably not true and not consistent. And here he claimed to have found an actual inconsistency. But, as has been mentioned, Terry Tao found a hole in his argument. And so mathematics was not destroyed. At least not that day.

What is true is that since we have computer-based theorem provers / proof checkers, "working it out" becomes a matter of getting your challenge past one of those, and the orthodoxy will have a hard time arguing against that.

But OTOH you see many people in other disciplines challenging the orthodoxy (in Physics: string theorists, MOND proponents, etc.) and while their views aren't always taken fully serious, their houses aren't being torched either.

It sucks to hear that something you've worked hard on is fundamentally flawed, especially on such public display

Exactly what I say at parent-teacher conferences.

"Finding truth" isn't quite what happened here, though. Finding a contradiction/other invalidation of someone else's proof, doesn't advance the knowledge frontier. It means that there was a flaw in process of proving the original proof, but it doesn't disprove the idea behind the original proof. You'd need your own separate proof to do that—probably a proof by contradiction, of one of the main lemmas of the proof.

The useful skill, here, is one shared by logicians and compilers: the ability to quickly find flaws that invalidate a proof, without needing to fully understand what the proof is trying to prove.

Said skill, if it could be more widely-learned, would be very helpful in iterating toward sound proofs (or impossibility proofs for such.)

But when it's only a skill in the hands of another party external to the one writing the proof, that iterative aspect isn't really there.

Easy, there. It's just evocative language. I could have said "spot the core flaw" to the same effect.

Well, classical Euclidean geometry relies on the parallel postulate, although it can't itself be proven. Abstract mathematicians have put together alternate non-euclidean geometries that are consistent but don't rely on the parallel postulate - kind of along the same lines as Nelson here "rejecting" induction just to see what would happen. I've always wondered what might happen if you "allowed", in the same sense, a square root to be a real rather than imaginary number. Maybe nothing, but I always wondered why nobody ever tried.