I imagine your ground for believing in a single form in each case
is the following. When it seems to you that a number of things are
large, there seems, I suppose, to be a certain single character which is
the same when you look at them all; hence you think that largeness is a
single thing . . . But now take largeness itself and the other things
which are large. Suppose you look at all these in the same way in your
mind's eye, will not yet another unity make its appearance-a largeness
by virtue of which they all appear large?... If so, a second form of
largeness will present itself, over and above largeness itself and the
things that share in it, and again, covering all these, yet another,
which will make all of them large. So each of your forms will no longer
be one, but an indefinite number.
Parmenides 131a1-b2
The notes below are adopted from S.M. Cohen's notes from http://www.aarweb.org/syllabus/syllabi/c/cohen/phil320/tmalect.htm. Some additions come from Cohen's article 'The Logic of the Third Man.'
Plato does not give the full argument of the so-called "Third Man Argument" with explicit steps. He gives the starting and ending points. We have to construct the steps in between. Doing so involves applying the principle of charity that a philosopher will not produce an inconsistent argument unless he or she is unaware of that inconsistency.
The result is an infinite regress: "So each of your forms will no longer be one, but an indefinite number." (132b2)
The result is an epistemological problem: if the Form F-ness is that by which you know that things are F, then in order to know that F-ness is F or to explain why F is F, you need to know another form, and another, and another. Which means you can never simply KNOW that things are F, or finish explaining it. This applies to belief as well, if you think that Forms are required for beliefs.
One might say that after the first few steps, one should notice that each successive Largeness n does not differ in any way relevant to largeness from Largeness n-1 and not worry about it, since the task was to explain largeness.
The result is also a problem for a theory of predication. In order to explain why x is F, Plato thinks, we need a Form F-ness. But we will never have the ultimate explaining factor of all F things, because explaining all the F things we can identify will require there to be another F thing, namely another Form F-ness. "Trying to explain predication in terms of the notion of participating in a (paradigmatic) Form leads to an infinite regress, and hence is no explanation at all." S.M. Cohen
"We need three premises, traditionally called "One Over Many", "Self-Predication" and "Non-Identity." I quote Cohen's description of them:
OM=One-Over-Many, SP=Self-Predication, NI=Non-Identity (Cohen
reformulates it to Non-Self-Participation).
Note that 1 above requires a single
Form F-ness, but the meaning of single
may be problematic: we think of it as a unified thing, but why should the
Form F-ness have any other qualities than simply F-ness? In other words,
why need it 'be' (at least in space and time)? Why need it be one (i.e. be
unified in the way that physical objects are usually held to be)? Perhaps
the Form Largeness is somehow what the parentheses around a set as well as
the label that says what it is a set of are, but not quite in the
dangerously spatial terms in which we usually think of parentheses and
labels. Plato clearly does not entertain these notions in the Parmenides, but they might, if they had a more subtle
defender than myself, get him out of the traps he has set for himself in
the Parmenides.
Note that 3 above simply says that you cannot explain the fact that F-nessn is itself F by using F-nessn. I.e. Forms do not explain the fact that they themselves are instances of what they are forms of. It is not clear to me why we should accept that principle, but it is clear to me that Plato does/must. At least he must do so in this text to generate the infinite regress he generates.
What happens if Plato gives up each of those principles?
Different people give different answers. Some people use the Parmenides as a way to date Plato's dialogues: those dialogues with a theory of Forms that defies the argument of the Parmenides must come before it. Some people say that Plato rejected the idea that Forms are models as well as recollection. He gave up self-predication. Perhaps he rejected or modified the One-Over-Many.
Some Definitions
These definitions are sufficient to produce the infinite regress of the Third Man Argument.
Note that there are no self-predication or non-identity (non-self-explanation) assumptions apparent there.
But self-predication is built in at D1 and D3. HOW SO?
Non-identity is also built into the argument.
Cohen thinks that this highlights a feature of the Third Man Argument. Namely, the One-Over-Many Axiom is the only one of the three principles that Plato explicitly formulates.
Cohen adds:
Will the "over" relationship be a relationship of one to one or of one to many? Cohen thinks it must be a many-many relationship! Many forms at any level n or higher are 'over' all the particulars as well as the forms at any level n-1.
Cohen concludes that to keep the uniqueness of forms, the One-Over-Many Principle will have to be rejected or modified.
Cohen's References