An omniscient God faces a self-referential contradiction
The argument is intended to challenge omniscience from within the concept of knowledge itself. It does not begin by denying that God exists, nor does it appeal to scientific explanations or religious authority. Instead, it constructs a proposition about God's own knowledge:
"God does not know this proposition."
Call this proposition P. The proposed dilemma is straightforward. If God knows P, then what P says is false: God does know it. But if God does not know P, then P is true, and there is at least one truth God fails to know. Either way, omniscience appears impossible.
This resembles the liar paradox, in which a sentence says, "This sentence is false." It also resembles well-known results concerning self-reference in logic, including the problems examined by Alfred Tarski and Saul Kripke. Serious philosophers of religion and philosophers of logic have therefore treated self-reference as more than an internet puzzle. The important question is whether this particular construction succeeds against a coherent account of omniscience.
The strongest form of the argument
The argument is strongest when its premises are stated carefully. Suppose that:
- There is a proposition P whose content is that God does not know P.
- Propositions are either true or false.
- Knowledge is factive: if someone knows a proposition, that proposition is true.
- An omniscient being knows every true proposition.
Now consider the two exhaustive possibilities. If God knows P, then P is true by the factivity of knowledge. But P says that God does not know P, so it is false. This is a contradiction.
If God does not know P, then what P says is true. But an omniscient being is supposed to know every true proposition. Thus God is not omniscient.
In a more formal idiom, the proposition would satisfy something like:
P is true if and only if God does not know P.
If omniscience is understood as knowing every truth, then, for every proposition q, God's knowing q tracks the truth of q. The self-referential proposition would therefore behave like a fixed point of the form P if and only if not P. Under classical logic, that has no consistent truth-value.
This is considerably stronger than the popular claim that "God cannot know what we freely choose." That separate argument depends on controversial assumptions about freedom and foreknowledge. The present argument targets the structure of omniscience itself. It also avoids the weaker mistake of saying that God's knowing something makes it true. Knowledge does not create the truth of its object. Rather, because knowledge is factive, knowing a proposition requires that the proposition already be true.
The central reply: the proposition does not exist
The most important reply is that the sentence may fail to express a genuine proposition. It may be a semantic construction with no consistent content, rather than a true or false statement that God ought to know.
This is the standard kind of response to the liar paradox. When a sentence says, "This sentence is false," the problem may not be that reality contains a mysterious object that is both true and false. The problem may be that unrestricted self-reference has produced an illegitimate semantic expression. A theory of truth must impose restrictions, distinguish object language from metalanguage, or use a more sophisticated account of truth-values.
The same response can be applied here. "God does not know this proposition" appears to presuppose that there is already a proposition whose content is fixed. But the reference of "this proposition" is being used to generate a claim about the very proposition whose content is being specified. If the resulting content is inconsistent, there may be no proposition there for God to know.
This reply is not an arbitrary escape. Contemporary philosophy of language and logic takes self-reference, semantic hierarchies, and truth predicates seriously. Tarski's work showed why a language containing its own unrestricted truth predicate faces severe difficulties. Kripke's later work explored ways of handling semantic paradoxes with partial truth-values and other restrictions. The lesson is not that every self-referential sentence is meaningless, but that self-reference cannot simply be treated as harmless in every context.
The reply also reflects an important distinction between sentences and propositions. An English sentence can be grammatically well formed without expressing a coherent proposition. Moreover, an omniscient being need not know every string of words that can be produced by a language. Omniscience is ordinarily defined as knowledge of every truth, not as possession of a belief corresponding to every possible piece of syntax, including malformed or semantically defective constructions.
Does diagonalization defeat this reply?
A proponent of the argument can respond that the problem is not dependent on an accidental English phrase. In sufficiently expressive formal systems, diagonal or self-referential constructions can be generated systematically. The diagonal lemma, familiar from mathematical logic, shows that under suitable conditions a sentence can be constructed that refers to its own relation to a predicate. If the relevant predicate is "is known by God," perhaps the construction yields a genuine sentence saying that it is not known by God.
This makes the challenge more serious. The defender of omniscience cannot dismiss every self-reference merely by pointing out that some everyday formulations are vague. If a precise formal language can define a proposition with the relevant self-referential content, the burden shifts to the defender to explain why that language does not describe a possible domain of propositions.
Even so, diagonalization does not automatically produce a contradiction. In Gödel's incompleteness theorems, a sentence that says, roughly, "This sentence is not provable in the system" can be true without being provable, provided the system has the relevant limitations. "Not provable" is not the same as "not known by an omniscient, factive knower." The argument succeeds only if the knowledge predicate is interpreted in a way that preserves the exact self-reference and the relevant assumptions about truth and completeness.
The defender can therefore distinguish between a formal sentence generated by diagonalization and a proposition with a coherent, determinate truth-value. A formal construction may show that unrestricted principles cannot all be combined. It does not by itself show which principle must be abandoned.
How much force does the argument have?
The argument establishes a genuine conditional result. If there is a proposition that says of itself that God does not know it, and if that proposition must be either true or false in the ordinary classical sense, then an omniscient God cannot exist under the stated definition of omniscience. At least one assumption must be restricted.
Possible restrictions include denying that the self-referential expression forms a proposition, adopting a hierarchy between levels of language, weakening the claim that every truth is knowable as a proposition, or developing a non-classical theory of semantic paradoxes. Each option has costs, and none should be presented as an effortless solution.
The argument does not, however, establish that every account of divine omniscience is incoherent. It does not show that atheism is true, that naturalism explains the problem better, or that no being could possess comprehensive knowledge. It shows that a particular combination of unrestricted self-reference, classical bivalence, factive knowledge, and an unrestricted definition of omniscience is unstable.
It is also not enough to reply that God could know the proposition to be false. If P is a genuine proposition, knowing that P is false is not the same as knowing P. The real issue is whether P is an intelligible proposition at all. That is where the debate must be conducted.
Academic philosophers generally recognise this distinction, even when they disagree about the outcome. Some popular presentations treat the dilemma as an immediate refutation of theism. That overstates the case. The serious version presents a substantial challenge to certain formulations of omniscience, while the serious reply appeals to the established difficulties surrounding semantic self-reference. The result is a philosophical pressure point, not a conclusive proof of atheism.