Prominent Christian philosopher, apologist, and analytic theologian William Lane Craig is in the process of releasing his magnum opus, a 5-volume Systematic Philosophical Theology (Wiley Blackwell, 2025–2026). Volume 2a (released in 2025) focuses on God’s attributes and includes a 100-page chapter on omniscience. Now aged 76, Craig presumably intends this chapter to be the crowning summary of his nearly 40-year publication record on the topic of divine omniscience and foreknowledge.
In this 4-part blog series I review and critique Craig’s omniscience chapter. The chapter is divided into four main sections—§1 Biblical data, §2 Omniscience, §3 Foreknowledge, and §4 Middle Knowledge—which correspond to the four parts of this blog series. This post is on Craig’s §2. For my previous post on §1, see here.
Part 2. Omniscience (pp. 202–215)
In this section of his chapter, Craig focuses on two main topics: (1) how best to define ‘omniscience’ and (2) challenges against the accuracy of his proposed definition.
(1) Defining ‘omniscience’
(1.a) De dicto vs. de re omniscience
Craig proposes and defends the following propositional definition of ‘omniscience’:
(Oprop) S is omniscient =def. For all propositions, p, if p [is true], then S knows that p and does not believe that ~p. (p. 203)
He variously calls it the “standard” (p. 208), “classical” (p. 209), and “traditional” (pp. 203, 208, 213–215) conception of omniscience without telling us clearly why it deserves such deference. A few weeks ago, in a previous post, I argued against this way of defining omniscience on the grounds that it greatly understates the qualitative perfection of God’s knowledge. I argued that God’s knowledge is primarily knowledge by acquaintance (knowledge de re) and only derivatively it is knowledge by description (knowledge de dicto). In other words, God knows reality (de re), all of it, not chiefly by knowing about reality (de dicto), but by being immediately and fully acquainted with it. Craig doesn’t entirely disagree with this assessment, for he says in reference to divine self-knowledge (knowledge de se) that “omniscience does not serve to exhaust God’s cognitive excellence” (p. 207).
This raises a critical question. Why define ‘omniscience’, as Craig does, in terms of the “scope and accuracy of one’s propositional knowledge” (p. 208) while realizing that such a definition falls short of God’s “cognitive excellence” rather than, as I prefer, define ‘omniscience’ in terms of maximal cognitive excellence with propositional omniscience being one facet of that? After all, as Craig himself points out, ‘omniscience’ “literally means all knowledge” (p. 208). And in the following chapter he says that “we ought to construe God’s knowledge as robustly as possible” (p. 345). So why shouldn’t omniscience include God’s de re knowledge of creation and de se knowledge of Himself and not just de dicto or propositional knowledge?
A big part of the explanation, I surmise, is that Craig has a theoretical blind spot when it comes to de re knowledge. He doesn’t take it that seriously and certainly not as foundational to God’s omniscience. Indeed, he mentions de re knowledge only twice (pp. 215, 346) and describes it dismissively as “mere” “knowledge of a thing from a third-person perspective” (p. 346) to distinguish it from first-person or de se knowledge. Moreover, even if we assume God has de re knowledge, something Craig neither affirms nor denies so far as I can discern, he doesn’t think it adds anything beyond what de dicto knowledge provides, for Ofull “adequately captures the extent and infallibility of God’s de re knowledge, though not its mode” (p. 215). In other words, propositional or de dicto knowledge is just another way of getting at the same reality that God (perhaps) knows in a non-propositional de re manner. This sidelining of God’s de re knowledge is, however, a serious mistake. In the first place, reality includes first-person perspectives and so the de re should include the de se and not be opposed to the latter as third-person versus first-person. In the second place, de re knowledge is broader in content than de dicto knowledge because the latter is abstract (from Latin, ab-trahere = drawn out of), meaning that it is drawn out of a full concrete experience and so necessarily leaves something of that experience behind. For example, if I say “This ball is red” I abstractively focus on this ball and its redness and (for the moment) ignore everything else as irrelevant. This is why I contend that God’s knowledge is primarily by acquaintance (de re) and only derivatively by description (de dicto).
An example of Craig’s indifference to de re knowledge comes in a long footnote on p. 207 wherein he rejects Linda Zagzebski’s position that God has ‘omnisubjectivity.’ This is the idea that God knows exactly “what it’s like” from every possible creature’s subjective perspective. Omnisubjectivity is a corollary of God’s having complete de re knowledge of reality. Thus, if creature C exists and is in subjective state S, then God has de re knowledge of C’s subjective state. He doesn’t merely know de dicto or descriptively about C’s subjective state. Rather, He’s fully acquainted with C’s subjective state itself. Hence, de re omniscience entails omnisubjectivity of all actual creatures. It also entails omnisubjectivity of all possible creatures. This is because possible creatures are wholly sourced in God, and God has complete de re knowledge of Himself. (This is, in fact, the same as God’s de se knowledge.) In other words, God knows exactly what it would be like for any hypothetical creature in any possible situation were He to create a creature of that sort and place it in a situation of that sort. Craig rejects omnisubjectivity, however, on the bizarre grounds that, on Zagzebski’s view, “the world’s conscious creatures function as sense organs for God, which is difficult to square with his omniscience, given the limited variety and temporal evolution of conscious creatures” (p. 207). This objection is bizarre because it assumes that if omnisubjectivity is true then God needs conscious creatures in order to sense creation and so would be limited in knowledge of creation because of this dependency. But omnisubjectivity doesn’t presuppose that God needs conscious creatures to sense anything. Omnisubjectivity is grounded, rather, in the assumption that God has exhaustive de re knowledge. Since creaturely subjective states are part of reality, God’s de re knowledge includes them as well. If there were no sentient creatures, God would still have exhaustive de re knowledge of reality.
In appears, then, that Craig sticks with his propositional definition of omniscience because he doesn’t appreciate either de re knowledge or its omnisubjective corollary. Because of this, he doesn’t appreciate de re knowledge as the foundation for de dicto knowledge, which leaves him no way to integrate God’s de se and de dicto knowledge. The former, on his view, is just a non-propositional part of God’s overall “cognitive excellence” that remains disconnected from God’s propositional omniscience.
There may be an additional, unstated motive here for Craig’s propositional definition of omniscience. Molinism requires God to have de dicto (middle) knowledge of information that is not grounded in any reality God could be acquainted with prevolitionally. After all, all that exists prevolitionally is God and God alone, but middle knowledge can’t be sourced in God’s nature (because it’s metaphysically contingent) nor in God’s will (because it’s prevolitional). Hence, if Craig were to admit that de dicto knowledge is grounded in de re knowledge, he would immediately run into an epistemic grounding problem with respect to God’s middle knowledge. By disconnecting the two and ignoring de re knowledge, Craig hopes quietly to sidestep the question of how God could possibly know Molinist information.
(1.b) Tensed vs. tenseless information
There are two ancillary issues that Craig addresses in the course of defending his propositional definition of omniscience. One concerns tensed vs. tenseless language. The other has to do with de se or first-person information.
Regarding the first issue, Craig asks whether propositions are tensed and, if so, whether this affects their knowability by a timeless God. One problem with his discussion here is that he never explains the tensed/tenseless distinction. He assumes that his readers already understand it. I doubt that’s a safe assumption. In any case, for the sake of my readers, a tensed proposition represents a situation from a temporally situated perspective, that is, it locates the situation in relation to a contextually specified present moment or now. For example, “Caesar crossed the Rubicon” (past tense) represents Caesar’s crossing of the Rubicon as occurring prior to now, whereas “The sun will rise tomorrow” (future tense) represents the sun’s rising as occurring one day subsequent to now. A tenseless proposition, in contrast, represents a situation from a non-temporal vantage point. It’s a view from no-when, so to speak. Tenseless propositions are usually expressed with a built-in date-time stamp, for example, “Caesar crosses the Rubicon on January 11, 49 BC.” If a tenseless proposition is all the information you have, then you aren’t able to say whether the event described is past, present, or future.
After raising the tensed language issue, Craig briefly considers and rejects a rather stupid argument to the effect that if propositions are tensed then “Socrates will drink hemlock” and “Socrates will not drink hemlock” are both true, allegedly generating a contradiction (p. 203). But obviously the propositions only imply that Socrates drinks / doesn’t drink at some future time or other. There is no contradiction unless both occur at the same time.
A more serious and interesting question, which Craig considers at length in the following chapter (on divine eternity), is whether an omniscient and timeless God can know tensed propositions. Short answer: No. Tensed propositions change their truth values over time. Hence, an omniscient God who knows tensed propositions would have to change with respect to the content of His omniscience and so wouldn’t be timeless. In this chapter Craig devotes only a page or so to this issue and then drops it because it’s irrelevant to his definition of omniscience (p. 205). (Why, then, did he bring the issue up here in the first place?)
(1.c) Creaturely de se knowledge
The final issue Craig considers is what to make of creaturely de se information. This is a challenge for his definition of omniscience because it looks at first glance like there are true first-person propositions like “I am Alan Rhoda” that I can know but no one else can, not even God. If so, then we can’t say that God knows all true propositions.
In response to this issue, Craig briefly considers two suggestions. First, he cites David Lewis who suggests that the difference between the third-person claim “Alan is typing out a blog post” and Alan’s first-person claim “I am typing out a blog post” is a matter of “non-propositional knowledge … arising from the self-ascription of properties” (p. 206). This reply seems clearly wrongheaded to me (but not, apparently, to Craig) because my most basic de se knowledge is my immediate de re knowledge of myself. As immediate, it doesn’t depend on any abstractions and so is prior to any self-ascription of properties on my part. To self-ascribe properties, in contrast, I first need to abstract from my lived self-awareness a conception of myself as a self and then predicate of that self various other abstractions or properties. Another problem with this suggestion is that a self-ascription of properties is propositional, not non-propositional. It has the propositional form of S is P in that it attributes a predicate/property (P) to a first-person subject (S).
A much better suggestion, which to his credit Craig also seems to like, is Jonathan Kvanvig’s proposal that the difference between the third-person claim “Alan is typing out a blog post” and Alan’s first-person claim “I am typing out a blog post” is not a matter of propositional versus non-propositional knowledge (pace Lewis) but rather a matter of how one and the same proposition is accessed. Both claims attribute the same property (typing out a blog post) to the same designated individual (me). In the first-person case, however, that claim is accessed from the “inside,” so to speak, whereas in the third-person case it is accessed from the “outside.”
I like this proposal because, as I apply it, only myself and the omnisubjective God have inside or de re access to my subjective states. No one else does. That’s why I can speak about myself in the first-person. God also knows what’s it like for me on the inside at every moment—indeed He knows me better than I know myself. But He also knows that His all-encompassing perspective is not identical to my much more limited perspective, and so He’s in no any danger of conflating His first-person “I” with mine.
So far so good. I’m glad Craig finds Kvanvig’s proposal agreeable, but then he ruins it by grounding de se knowledge not in de re knowledge (as I do) but in de dicto knowledge of “propositions implying one’s individual essence” (p. 207). This, I submit, is philosophical abstractionism run amok. It amounts to explaining something concrete and immediate (a person’s lived experience) by abstract propositions and an abstract individual essence. It’s yet another indicator of Craig’s blindness to de re concrete reality.
(2) Challenges to Craig’s definition of omniscience
After setting forth his propositional conception of omniscience, Craig considers various challenges to that conception, namely, (a) that it’s incoherent, (b) that there are truths God can’t know, and (c) that God doesn’t need any propositional knowledge to be omniscient.
(2.a) Is propositional omniscience coherent?
Over the past four decades atheist philosopher Patrick Grim has pressed a variety of arguments against the very concept of propositional omniscience. The concept, he contends, is incoherent. No being could possibly be omniscient in that sense. Craig spends about four pages rebutting four of Grim’s arguments.
Grim’s first two arguments deal with self-reference, so Craig responds to them together. Grim’s first argument is a “Divine Liar” paradox. Assume that God believes all and only true propositions and let p be the proposition “God does not believe that p is true.” The paradox is that if God believes p then it follows that God does not believe p, and if God does not believe p then it follows that God does believe p. Grim’s second argument is very complicated to state fully. You can access one of Grim’s papers on it here. Basically, he employs reasoning similar to Gödel’s incomplete theorems. Whereas Gödel showed that, for any given axiomatic system containing basic arithmetic, there are truths that cannot be proven in the system, Grim argues that in every such axiomatic system there are truths that cannot be known. Let’s call this the “Knowledge Incompleteness” problem.
Craig’s reply to both arguments doesn’t go into any technical details. Instead he proposes a well-known hierarchical solution. The idea is that what cannot be known at one level (or in one axiomatic system) can be known at another level that includes the previous level but is more robust. This, however, is an incomplete answer because it doesn’t get at the root of the two problems. As I see it, the Divine Liar paradox is malformed for the same reason that standard versions of the Liar paradox are malformed—the self-referential “propositions” in questions aren’t even propositions because they have no definite meaning. As for the Knowledge Incompleteness problem, it draws its power from the abstract nature of propositional knowledge. God’s knowledge, however, is fundamentally de re, not de dicto, and so it’s oriented toward concrete reality, not abstract propositions about concrete reality. God’s de re knowledge is thus not built-up in logical sequence, propositional unit by propositional unit, axiom by axiom, hierarchical level by hierarchical level. Rather, God knows the whole of reality intuitively, in a single Gestalt.
One problem with the hierarchical solution is that it leaves Grim a potent retort: Omniscience is supposed to attribute all knowledge to God, but (Craig’s quoting Grim here) “[i]f truth forms a stratified hierarchy, [then] there can be no notion that applies to truth on all levels” (p. 211). Craig’s response is that “[w]hile we cannot quantify over all truths collectively, we can nonetheless quantify over levels and say that for any level, God knows all the truths at that level” (p. 211). But this response misses Grim’s point. After all, Craig’s own definition of omniscience Oprop implicitly quantifies over all truths collectively since “for all propositions, p, if p [is true], then …” can easily be rephrased as “for all true propositions, p, …” What’s more, if we quantify over all possible levels in the hierarchy (not just all levels so far) and say that God knows all truths at each level then we are, again, implicitly quantifying over all truths collectively. In short, then, Craig’s hierarchical reply to Grim implicitly conflicts with his own propositional definition of omniscience. This is why to rebut Grim we have to shift from thinking about omniscience in de dicto propositional terms to thinking about it in more robust de re terms: God is omniscient because He is immediately and fully acquainted with all of reality in one unified Gestalt. We can abstract as many true propositions as we want from that Gestalt and say God knows those, but no collection of abstractions can possibly exhaust the Gestalt.
Moving on, Grim’s third argument is the Cantorian problem. Craig doesn’t spell out the problem, but the gist of it is that given any collection of truths, we can always generate a larger collection by performing combinatorial operations on the original collection. In the same way that Cantor showed there is no largest type of infinity, Grim argues that there is no possible collection of all truths. Here’s my own simple example. Suppose we start with a single truth p. Now lets add a couple combinatorial operations. First, any two truths can be combined by logical addition (∧) to yield another truth. Second, applying Oprop, for any truth p God knows that p. Call that truth K(p). And so for every truth p we can generate another truth, K(p). From this simple starting point we get a runaway exponential explosion of truths: p, K(p), p ∧ K(p), K(K(p)), K(p ∧ K(p)), etc. Grim’s point is that if from any group of truths we can keep generating new truths ad infinitum then there can be no completed totality of all truths and thus there is no such totality to be known. Hence, omniscience is impossible.
Craig’s response is again to appeal to hierarchical levels. This makes sense because the expanding collection of truths is constructed by repeated iteration of combinatorial operations. But this isn’t the best answer. A better response is to say that the “new” propositions generated by these combinatorial operations aren’t new propositions at all. We’re not generating new information. We’re just re-expressing old information in more and more complicated ways. Everything after p and K(p) is redundant. By appealing to logical hierarchies here, Craig concedes Grim’s point that there is no such thing as a totality of all truths and that Oprop is therefore erroneous as stated. Indeed, he quietly reinterprets Oprop, tacitly gaslighting us into thinking he really meant something else all along: “In what follows whenever I speak colloquially of God’s knowing all truths, such expressions should be understood as rough for God’s knowing every truth at every level” (p. 212). In other words, forget Oprop, what I really meant was
(Oprop-level) S is omniscient =def. For every conceptual level L and every proposition p such that p is true-at-L, S knows-at-L that p and does not believe-at-L that ~p. (See Craig on Koons on p. 211)
This is clearly a very different definition than where we started.
Grim’s fourth and final argument is easy to dispel as it was already discussed above in section (1.c). The issue is that it seems that sentient creaturely have propositional de se knowledge that God can’t know, and so there are true propositions that God can’t know. In reply Craig repeats Kvanvig’s line from above that what we have here are two different ways of accessing the same proposition.
(2.b) Are there truths God can’t know?
The next challenge to Craig’s propositional definition of omniscience is much simpler. Following Brian Leftow he asks whether there aren’t many sorts of truths that God can’t know, for example, “how it feels oneself to be a failure or a sinner” (p. 213). Craig replies that Leftow is conflating propositional and non-propositional knowledge. I think this reply is correct. Knowing how something feels or what it’s like to experience something is not knowing a truth about that something. It’s not de dicto knowledge but de re knowledge. If, as I argue above, a God who has de re omniscience has omnisubjectivity, then there is no problem with saying that God knows what it’s like for someone (not Himself, presumably) to feel like a failure or a sinner. Craig rejects omnisubjectivity, but he’s certainly right that if the kind of knowledge that worries Leftow isn’t propositional then it doesn’t undermine Oprop.
(2.c) Could God be omniscient without knowing any propositions?
The final challenge comes from William Alston who wonders whether God’s knowledge is propositional at all. Alston basically agrees with me that God’s omniscience is fundamentally a kind of de re knowledge or knowledge by acquaintance. But Alston goes a step further and suggests that God has no de dicto knowledge or beliefs at all.
In reply Craig doesn’t argue that Alston is wrong. He says, rather, that it doesn’t matter for “we need only add the proviso that such a definition [Oprop] expresses the human point of view: as finite knowers we grasp God’s knowledge in propositional terms” (p. 215).
I think this is a fair response, but I also think Craig should have pushed back a bit more and argued that God’s knowledge can’t be wholly non-propositional or else it becomes quite unclear how God could speak propositionally in the Bible. Are we to suppose that God doesn’t actually believe what He’s presented as saying? In my view, Alston is correct that God’s knowledge is fundamentally de re but wrong in thinking that it’s exclusively de re.
My concluding thoughts concerning Craig on omniscience
As I noted several times, I think Craig has a major blind spot when it comes to de re knowledge or knowledge by acquaintance. This leads him to reject omnisubjectivity and to respond to Grim in ways that implicitly concede that his propositional definition of omniscience is a failure.