What are the chief goals of axiomatic set theory, and to what extent have these goals been successfully
attained? In order to answer that question we must remember the circumstances which led
mathematicians, in the early years of the twentieth century, to search for an axiomatic basis to set
theory. The ideas of Cantor had already thoroughly permeated the fabric of modern mathematics, and
had become indispensable tools of the working mathematician. Algebra and analysis were formulated
within a framework of set theory, and some of the most elegant, powerful new results in these fields
were established by using the methods introduced by Cantor and his followers. Thus, when the
paradoxes were discovered and there arose doubts as to the basic validity of Cantor’s system, most
mathematicians were understandably reluctant to give it up; they trusted that some way would be
found to circumvent the contradictions and preserve, if not all, at least most of Cantor’s results.
Hilbert once wrote in this connection: “We will not be expelled from the paradise into which Cantor
has led us.”
With the discovery of newer paradoxes, and the failure of all the initial attempts to avoid them, it
became increasingly clear that it would not be possible to preserve intuitive set theory in its entirety.
Something—possibly quite a lot—would have to be relinquished. The best that one could hope for
was to retain as much of intuitive set theory as was needed to save the new results of modern
mathematics and provide an adequate framework for classical mathematics.
Briefly, then, axiomatic set theory was created to achieve a limited aim: it had to provide a firm
foundation for a system of set theory which—while it did not need to be as comprehensive as intuitive
set theory—must include all of Cantor’s basic results as well as the constructions (such as the number
systems, functions, and relations) needed for classical mathematics.
The systems of both Zermelo and von Neumann were successful in achieving this limited aim. But
the amount of intuitive set theory which they had to sacrifice was considerable. For example, in
Zermelo’s system, as we have already seen, the intuitive way of making sets—by naming a property
of objects and forming the set of all objects which have that property—does not take place at all. It is
replaced by the axiom of selection, in which properties are allowed only to determine subsets of
given sets. Furthermore, the only admissible “properties” are those which can be expressed entirely
in terms of the seven symbols ∈, ∨, ∧, ¬, ⇒, ∀, ∃ and variables x, y, z, … As a result, many of
the things that we normally think of as sets—for example, the “set of all apples,” the “set of all atoms
in the universe”—are not admissible as “sets” in axiomatic set theory. In fact, the only “sets” which
the axioms provide for are, first, the empty set Ø, and then constructions such as {Ø}, {Ø, {Ø}}, etc.,
which can be built up from the empty set. It is remarkable fact that all of mathematics can be based
upon such a meager concept of set.
While the various axiomatic systems of set theory saved mathematics from its immediate peril, they
failed to satisfy a great many people. In particular, many who were sensitive to the elegance and
universality of mathematics were quick to point out that the creations of Zermelo and von Neumann
must be regarded as provisional solutions—as expedients to solve a temporary problem; they will
have to be replaced, sooner or later, by a mathematical theory of broader scope, which treats the
concept of “set” in its full, intuitive generality.
This argument against axiomatic set theory—that it deals with an amputated version of our intuitive
conception of a set—has important philosophical ramifications; it is part of a far wider debate, on the
nature of mathematical “truth.” The debate centers around the following question: Are mathematical
concepts creations (that is, inventions) of the human mind, or do they exist independently of us in a
“platonic” realm of concepts, merely to be discovered by the mathematician? The latter opinion is
often referred to as “platonic realism” and is the dominant viewpoint of classical mathematics. We
illustrate these two opposing points of view by showing how they apply to a particular concept—the
notion of natural numbers. From the viewpoint of platonic realism, the concepts “one,” “two,”
“three,” and so on, exist in nature and existed before the first man began to count. If intelligent beings
exist elsewhere in the universe, then, no matter how different they are from us, they have no doubt
discovered the natural numbers and found them to have the same properties they have for us. On the
other hand, according to the opposing point of view, while three cows, three stones, or three trees
exist in nature, the natural number three is a creation of our minds; we have invented a procedure for
constructing the natural numbers (by starting from zero and adding 1 each time, thus producing
successively 1, 2, 3, etc.) and have in this manner fashioned a conceptual instrument of our own
making.
How does platonic realism affect the status of axiomatic set theory? From the point of view of
platonic realism, mathematical objects are given to us ready-made, with all their features and all their
properties. It follows that to say a mathematical theorem is true means it expresses a correct
statement about the relevant mathematical objects. (For example, the proposition 2 + 2 = 4 is not
merely a formal statement provable in arithmetic; it states an actual fact about numbers.) Now—if
we admit that mathematical objects are given to us with all their properties, it follows, in particular,
that the notion of set is a fixed, well-defined concept which we are not free to alter for our own
convenience. Thus the “sets” created by Zermelo and von Neumann do not exist, and theorems which
purport to describe these nonexistent objects are false! In conclusion, if we were to accept a strict
interpretation of platonic realism, we would be forced to reject the systems of Zermelo and von
Neumann as mathematically invalid.
Fortunately, the trend, for some time now, has been away from platonism and toward a more
flexible, more “agnostic” attitude toward mathematical “truth.” For one thing, developments in
mathematics have been conforming less and less to the pattern dictated by platonic philosophy. For
another, the cardinal requirement of platonism—that every mathematical object correspond to a
definite, distinct object of our intuition (just as “point” and “line” refer to well-defined objects of our
spatial intuition)—came to be an almost unbearable burden on the work of creative mathematicians
by the nineteenth century. They were dealing with a host of new concepts (such as complex numbers,
abstract laws of composition, and the general notion of function) which did not lend themselves to a
simple interpretation in concrete terms. The case of the complex numbers is a good illustration of
what was happening. Classical mathematics never felt at ease with the complex numbers, for it lacked
a suitable “interpretation” of them, and as a result there were nagging doubts as to whether such things
really “existed.” Real numbers may be interpreted as lengths or quantities, but the square root of a
negative real number—this did not seem to correspond to anything in the real world or in our intuition
of number. Yet the system of the complex numbers arises in a most natural way—as the smallest
number system which contains the real numbers and includes the roots of every algebraic equation
with real coefficients; whether or not the complex numbers have a physical or psychological
counterpart seems irrelevant.
The case of the complex numbers strikes a parallel with the problem of axiomatic set theory. For
the “sets” created by Zermelo and von Neumann arise quite naturally in a mathematical context. They
give us the simplest notion of set which is adequate for mathematics and yields a consistent axiomatic
theory. Whether or not we can interpret them intuitively may be relatively unimportant.
Be that as it may, many mathematicians in the early 1900’s were reluctant to make so sharp a break
with tradition as axiomatic set theory seemed to demand. Furthermore, they felt, on esthetic grounds,
that a mathematical theory of sets should describe all the things—and only those things—which our
intuition recognizes to be sets. Among them was Bertrand Russell; in his efforts to reinstate intuitive
set theory, Russell was led to the idea that we may consider sets to be ordered in a hierarchy of
“levels,” where, if A and are sets and A is an element of , then is “one level higher” than A.
For example, in plane geometry, a circle (regarded as a set of points) is one level below a family of
circles, which, in turn, is one level below a set of families of circles. This basic idea was built by
Russell into a theory called the theory of types, which can be described, in essence, as follows.
Every set has a natural number assigned to it, called its level. The simplest sets, those of level 0,
are called individuals—they do not have elements. A collection of individuals is a set of level 1; a
collection of sets of level 1 is a set of level 2; and so on. In the theory of types the expression a ∈ B
is only meaningful if, for some number n, a is a set of level n and B is a set of level n + 1. It follows
that the statement x ∈ x has no meaning in the theory of types, and as a result, Russell’s paradox
vanishes for the simple reason that it cannot even be formulated.
Russell’s theory of types is built upon a beautifully simple idea. Unfortunately, in order to make it
“work,” Russell was forced to add a host of new assumptions, until finally the resulting theory
became too cumbersome to work with and too complicated to be truly pleasing. For one thing,
corresponding t