If there are affinities with pragmatism or with later analytic work on normativity and practice, I’m happy to acknowledge them. But I’m not trying to force Wittgenstein into Hegelian inferentialism. I’m using later Wittgenstein to keep JTB anchored in how our practices actually operate, and to keep the discussion aimed at epistemic certainty, not Cartesian absolute certainty. — Sam26

Post-Sellarsianism is defined by where one locates normativity, which seems to be the same site you situate it, in public justificatory standing governed by mistake-conditions, defeaters, and correction. Like the post-Sellarsians, you treat JTB as a legitimate starting grammar, whereas Wittgenstein aims to dissolve this starting point. Wittgenstein uses hinges to stop explanation, not to underwrite it. Once hinges are recruited to keep JTB “anchored,” they have been absorbed into a normative architecture. That architecture is Sellarsian in spirit even if it is anti-foundational in tone.

If there are affinities with pragmatism or with later analytic work on normativity and practice, I’m happy to acknowledge them. But I’m not trying to force Wittgenstein into Hegelian inferentialism. I’m using later Wittgenstein to keep JTB anchored in how our practices actually operate, and to keep the discussion aimed at epistemic certainty, not Cartesian absolute certainty.
— Sam26

Post-Sellarsianism is defined by where one locates normativity, which seems to be the same site you situate it, in public justificatory standing governed by mistake-conditions, defeaters, and correction. Like the post-Sellarsians, you treat JTB as a legitimate starting grammar, whereas Wittgenstein aims to dissolve this starting point. Wittgenstein uses hinges to stop explanation, not to underwrite it. Once hinges are recruited to keep JTB “anchored,” they have been absorbed into a normative architecture. That architecture is Sellarsian in spirit even if it is anti-foundational in tone. — Joshs

That’s a fair challenge, and I’ll concede the affinity while rejecting the conclusion.

If “post-Sellarsian” just means locating normativity in assessable standing, mistake-conditions, defeaters, and correction, then yes, my account has overlap. But overlap isn’t identity. My use of Wittgenstein is meant to limit that normative architecture, not to extend it.

Here’s the key point: hinges aren’t part of justificatory standing in my framework. They aren’t reasons, warrants, or items that get graded as justified or unjustified. At the same time, I do think hinges are foundational, but not in the classical sense of “foundations” as justified premises from which knowledge is derived. Their foundational role is structural: they’re conditions of possibility for inquiry and justification. They stand fast as the background against which evidence, error, correction, and defeaters can have a role at all. So they’re foundational like the river-bed is foundational to the flow, not like axioms are foundational to a proof.

When I say hinges “anchor” JTB, I don’t mean they underwrite it with deeper reasons. I mean they prevent the J in JTB from being misconstrued as a Cartesian demand for foundations. Hinges mark where justificatory talk stops, they constrain the reach of “why?” questions, they don’t answer them.

Continuing with paper...
Post #6


4. Hinges and Limits: A Structural Parallel with Gödel

Our justificatory practices do not begin from nowhere. They move within a setting of certainties that stand fast for us and form the background against which our judgments make sense. These certainties do not arise from inquiry. They are the conditions from which inquiry proceeds. Wittgenstein’s image of a river with a relatively fixed bed helps bring this structure into view. Some propositions shift with experience, while others lie beneath that shifting surface, not as items known but as parts of our form of life. They are the points at which doubt gives out.

This helps explain why justification has limits that are not defects. Doubt always operates against a background that is not itself in doubt. To imagine otherwise is to lose the distinction between doubting and knowing. Hinges cannot be justified, and they do not need justification. Their role is not to supply evidence but to provide the stability against which the difference between evidence and error is intelligible. They do not enter the space of justification; they make that space possible.

A useful parallel can be drawn here with Gödel’s incompleteness results, understood in a strictly structural sense. Gödel showed that any sufficiently rich formal system operates against constraints it cannot, by its own rules, fully account for or secure. This is not a defect in the system but a consequence of its being rule-governed at all. The relevance of this result for epistemology is not mathematical and does not concern the formal incompleteness of human knowledge. Rather, it serves as an illustration of how practices governed by rules depend on background conditions that are not themselves established by the standards those practices deploy. Justificatory practices are similar in this respect. They cannot justify everything they rely on, not because of oversight or failure, but because justification itself requires a background that stands fast. Hinges mark these internal stopping-points. They are not axioms, and epistemology is not a formal calculus, but the comparison helps make clear how limits can arise from within a practice without undermining its authority or coherence.

This perspective clarifies the place of bedrock certainties. They are not hidden assumptions or unexamined beliefs. They are the inherited background against which justification takes its form. To recognize them is not to adopt dogmatism but to acknowledge the grammar of epistemic life. Doubt cannot extend everywhere, not because we refuse to question certain things, but because the possibility of questioning presupposes a foundation that is not itself the outcome of justification. The stopping-points are part of the practice.

This insight strengthens the refinement offered by JTB+U. Justification requires a background that is not itself produced by justificatory means, and understanding involves navigating the concepts that operate within that background. These limits do not undermine the classical model of knowledge. They show the conditions under which it can function. Gödel’s work reminds us that structured practices have horizons. Wittgenstein reminds us that epistemic life is no exception. Hinges mark the horizon of justification and teach us what must remain fixed for the river of belief and doubt to flow at all.

One way to describe what this paper asks the reader to do is to speak of aspect-seeing. Nothing new is added to our evidence by making these distinctions explicit. What changes is the way we see the epistemic landscape. We stop treating justification as if it were an inner glow that accompanies a belief, and we see it as a public standing within a practice, governed by routes, constrained by guardrails, and framed by what stands fast. The shift is not a new theory but a clearer view of the grammar that has been at work all along.

Continuing with paper...
Post #7

5. The Five Routes of Justification in Practice

Justification in our epistemic life does not take a single form. It moves along routes that reflect how our language-games operate, each showing a way in which a belief can acquire the standing required for knowledge. These routes are not methods competing for priority. They mark different moments at which justificatory support is given within a practice. Their unity lies in the grammar of justification rather than in any shared procedure.

Testimony is the route on which we rely most naturally. Much of what we count as knowledge depends on the word of others, taken within practices that give their claims standing. Testimony is not a lesser form of justification. It is part of the training through which we learn to distinguish the credible from the questionable and to recognize when a speaker’s role, authority, or position within a practice makes their statement a source of support. Testimony functions because our language-games already contain criteria for when a statement may be accepted.

Logical inference shows a different movement. Here, the standing of a belief comes from its relation to other propositions. Patterns of inference, whether deductive or inductive, reflect the conceptual links we inherit through training. They do not impose external rules on thought but express the grammar of our concepts. To follow an inference is to move correctly within a practice whose standards are already in place.

Sensory experience provides justification when what we perceive aligns with how things are described in our language-games. Perception does not stand outside our practices as an independent foundation. It is shaped by our capacity to apply concepts, to distinguish appearance from reality, and to place what is seen within the background that stands fast for us. Sensory experience supports belief because our training has already taught us how perceptual reports function within the practice.

Linguistic training is often overlooked, though it is involved in nearly every case of justification. To possess a concept is to be able to use it correctly, and much of what we call knowledge depends on this competence. When a belief rests on the proper application of a concept within a language-game, its standing reflects the training that guides our use. This route makes explicit why understanding is internal to justification rather than something added afterward.

Pure logic occupies a more limited role, though it remains important. It does not track experience or testimony, nor does it guide ordinary inference. Its task is to clarify the structural limits of what can coherently be said. Pure logic justifies by ruling out what cannot find a place within our language-games. It draws the boundary of sense rather than supplying support for particular beliefs.

These routes do not form a hierarchy. Each contributes to the standing a belief must have to count as knowledge, and each operates within the background of certainties that makes justification possible. JTB+U clarifies how these routes function together by showing that no route, taken on its own, guarantees justificatory standing. To see how that standing is disciplined in practice, we must also attend to the constraints that govern when support counts as support. That task is taken up in the next section.

. . . to argue that God isn’t merely one more claim inside the system, but part of what makes the system possible. My point is: that’s a coherent aspiration, but it’s also a very high bar. To succeed, the theologian would have to show that the denial of God undermines the grammar of justification itself, not just that God is a good explanation, or a satisfying metaphysical picture. And until that bar is met, “God exists” looks less like a hinge and more like a substantive claim that remains open to epistemic assessment, including defeaters, alternatives, and the usual standards of practice. — Sam26

Right. The necessity of God for the very structure of justification is -- if true -- deeply hidden. So much so, that it requires huge systematic effort to bring it to light. As you say, this is a coherent project, and many philosophers and theologians believe it can succeed. This does require abandoning what you're calling the substantive claim, which many are reluctant to do. I would say it also requires at least some explanation of why the hinge-certainty here is so much more difficult to establish than that of, say, "The world exists," and of why a person can evidently do all the practice-based work of justification while vigorously denying the God-hinge, even after it's been carefully argued for. (This is harder than it looks, because the idea of a hinge is that it represents a place you stop, and advocates of the God-hinge simply refuse to stop with "The world exists" etc. Must they be wrong? What is the persuasion for "You should stop at X"? I've often wondered this about Wittgensteinian phil. in general.)

the desire to make God hinge-certain is structurally parallel to the hinge idea, — Sam26

This was really my point, or observation. In both cases, what's wanted is an acknowledgment that justification or rationality depends upon a deep-background source that is not itself an item for justification. Theologians probably have better success showing God as ontologically, rather than conceptually (grammatically), necessary. Which raises an odd question: Could you consistently maintain that God is necessary for there to be Being at all, but not necessary for there to be rationality and justification?

There's some repetition in the paper I can cut out. Also, I think I need to clarify a couple of ideas.

As for why a “God-hinge” is harder to establish than “there is a world”: my answer is that the world-hinge is not something we arrive at by argument (i.e., we don't arrive at a hinge epistemically), it’s displayed in the most basic operations of inquiry. It’s not a thesis inside the practice, it’s part of what makes the practice of checking, correcting, and learning possible at all. That’s why it isn’t typically up for debate in the same way, and why someone who “denies the world” is usually no longer playing the game of justification in any recognizable sense.

By contrast, “God exists” is, for most people, a substantive metaphysical claim that remains active in the space of assessment. People can vigorously deny it while still fully participating in the practices of evidence, error, correction, and defeater sensitivity. That fact is exactly what makes the “God as hinge of rationality” project such a high bar. If the denial doesn’t collapse the grammar of justification, then the claim hasn’t shown itself to be a hinge in the Wittgensteinian sense.

On “must they be wrong” and “what’s the persuasion for ‘you should stop at X’”: I don’t think the hinge idea is a recommendation about where one ought to stop, as if Wittgenstein is issuing a rule. It’s a description of where our justificatory practices actually do stop, where reasons run out and the background stands fast. You can refuse to stop, but at some point the demands cease to be ordinary justificatory demands and become a different kind of philosophical ambition, for example a metaphysical demand for an ultimate ground. That ambition can be coherent, but it’s no longer the same as ordinary epistemic judgements.

On your last question, yes, I think you can consistently hold that God is ontologically necessary without claiming God is grammatically necessary for rationality and justification. That’s basically the difference between a metaphysical thesis and a thesis about the conditions of intelligible justification. A person could say: “God is the ground of Being,” while also saying: “human justificatory practices, as practices, can operate without explicitly presupposing that claim.” In such a case, God would be ontologically basic on their view, but not a hinge of rationality in the Wittgensteinian sense.

And that’s why I keep separating two things, viz., the ontological project asks what must exist for anything to exist. The hinge project asks what must stand fast for our practices of doubt, evidence, error, and correction to be intelligible. Those projects can converge, but they don’t automatically converge, and the burden is on the theologian to show that they do.

Continuing with paper...
Post #8

6. Guardrails and the Discipline of Justification

The five routes describe the ordinary ways in which justification proceeds. They show how a belief can be supported within our language-games, through testimony, inference, sensory experience, linguistic training, and the boundary-setting role of pure logic. Yet a route is not, by itself, a guarantee that a belief has the standing required for knowledge. A belief can travel along one of these routes and still fail to count as knowledge because something in the justificatory situation does not have the right shape.
This is why it is helpful to make explicit a set of guardrails, not as additions to the classical model, but as clarifications of what our practices already require when we speak carefully. These guardrails articulate constraints that belong to justification as it functions within a practice. Their point is grammatical. They mark what it is for justificatory support to count as support within a language-game, rather than as something that merely looks supportive from a distance.

I call these guardrails No False Grounds, Practice Safety, and Defeater Screening.

6.1 No False Grounds

The first guardrail is straightforward. A belief cannot be justified when the support doing the work is false in a way that matters to the case. This is not a demand for certainty. It is a demand that the grounds on which the belief is placed in standing not be defective at the point where the practice treats the defect as disqualifying. When a belief is supported by a mistaken identification, a fabricated report, a misdescribed circumstance, or a misapplied concept, the belief may still be true, but its standing is not the standing of knowledge. It is not properly grounded within the practice.
This is one reason Gettier-style constructions do not function as counterexamples to JTB. They rely on the appearance of justificatory support while allowing false grounds to do essential work. Once justification is understood, as argued above, as a standing within an epistemic practice rather than the mere presence of supporting considerations, the pressure of these cases dissolves. What they reveal is not a defect in the classical model, but the need for greater care in describing what counts as genuine justification.

No False Grounds therefore clarifies the discipline internal to our epistemic life. It is not a philosophical invention. It is already present in the way we withdraw standing when a belief is traced back to something that is not so.

Continuing with paper...
Post #9

6.2 Practice Safety

The second guardrail concerns stability. In ordinary life we distinguish between a belief that is true and a belief that is true in a way that is secure enough to count as knowledge. Practice Safety names this distinction without turning it into a demand for infallibility. A belief is practice safe when, given the way it is supported, it would not easily have been wrong under the nearby variations that the practice itself treats as relevant.

This matters because a belief can be true by luck, and yet still look supported. In such a case the belief has a kind of success, but it does not have the standing of knowledge. Knowledge is not merely arriving at the truth. It is arriving there in a way that fits the discipline of a practice of justification. Practice Safety captures the sense in which our language-games require more than coincidence, not by imposing an abstract condition, but by reflecting the ordinary difference between what is dependable and what merely happens to work out in a single instance.

The point applies across all five routes. Testimony is practice safe when the source and the circumstances supply stable standing for the report, not a fortunate accident. Inference is practice safe when the transitions hold in the way the practice requires, not only in the one case where they happen to land correctly. Sensory experience is practice safe when the conditions are not the sort that regularly generate error signals. Linguistic training is practice safe when the relevant concepts are applied in the way the language-game calls for, not in a way that merely sounds right. Pure logic, in its boundary setting role, is practice safe when it draws limits of sense correctly, rather than excluding or permitting claims by a mistake in grammar.

Practice Safety therefore records a feature of our epistemic practices that many discussions of knowledge ignore. We do not treat every true belief with surface support as knowledge. We treat knowledge as something that stands within a stable pattern of justification, stable in the sense our practices recognize.

And that’s why I keep separating two things, viz., the ontological project asks what must exist for anything to exist. The hinge project asks what must stand fast for our practices of doubt, evidence, error, and correction to be intelligible. Those projects can converge, but they don’t automatically converge, and the burden is on the theologian to show that they do. — Sam26

Good, that works for me. And we can notice that the liminal area between the two projects would lie in how to understand the term "stand fast". It has to refer to something different than ontology, but exactly how would we characterize what it is that is standing fast? And in what space, so to speak, does this occur -- is it strictly a linguistic/grammatical inquiry?

On “must they be wrong” and “what’s the persuasion for ‘you should stop at X’”: I don’t think the hinge idea is a recommendation about where one ought to stop, as if Wittgenstein is issuing a rule. It’s a description of where our justificatory practices actually do stop, where reasons run out and the background stands fast. You can refuse to stop, but at some point the demands cease to be ordinary justificatory demands and become a different kind of philosophical ambition, for example a metaphysical demand for an ultimate ground. That ambition can be coherent, but it’s no longer the same as ordinary epistemic judgements. — Sam26

I'm not sure I agree with this. It implies that Witt is innocently noting a fact of the matter -- "Here is where our justificatory practices stop" -- rather than making a recommendation to avoid what he considers to be nonsense or, indeed, a kind of mental illness. If the question were really so simple, everyone would be a Wittgensteinian, and that would be that. Closer to the truth, I think, is what's implied in your phrase "at some point," and what Witt thought happened after that point.

The point at which an exploration of language becomes a metaphysical inquiry is far from clear, I would say. And while you generously call coherent "a metaphysical demand for an ultimate ground," I'm really not sure Witt would. Isn't his attitude more like "Fine, go ahead and refuse to stop at the point I've indicated, but what you think you're doing is no longer what you're in fact doing. So you really should comport your practice with what I'm showing you to be the limits of sensible discourse"? Two different references for "ought," I guess. It's as if someone said, "I'm going to win this 500-yd foot race by running 1,000 yards." Witt might reply, "You ought to stop at 500 yards, not because that's better in some way, but because past 500 yards you're no longer running the race." He's not saying, "The race ought to stop at 500 yards." That's the part he claims he's merely indicating, as a fact. But you should stop there, if you want to still be part of this practice.

A hinge, or related group of hinges, is meant to be where epistemic justification stops. But any unclarity about what counts as justification is going to be a problem for Witt as much as for JTB. That's part of why I like your work on this topic. An explanation of justification must itself be justified, and that's why you're writing your paper. You don't claim to be able to simply show the place where "justification" now holds water as a concept; you argue for it carefully. I often wish Witt was as careful, brilliant aphorist though he was. But then I've only read parts of "On Certainty," so I may be talking out my ass.

First, on “stand fast.” I do think this has ontological bite. Hinges aren’t merely linguistic habits or conversational conveniences. They are bedrock commitments by which we take ourselves to be in contact with reality. They function as conditions of possibility for inquiry, but precisely because of that they carry ontological commitments: “there is a world,” “objects persist,” “other minds exist,” and so on are not neutral placeholders. They are ways of taking reality to be foundational (not foundational in the traditional sense).

Here’s the crucial point: hinges don’t function as claims inside our epistemic framework, they function as the framework. They aren’t supported by evidence in the ordinary way, and they aren’t normally overturned by ordinary counterevidence, because they are what make evidence, counterevidence, mistake, and correction intelligible in the first place. If a hinge were routinely up for the same kind of assessment as ordinary propositions, if it had the same kind of defeaters and the same criteria of confirmation as the claims we test within inquiry, then it wouldn’t be doing hinge work. It would just be another hypothesis among hypotheses. That’s why doubting everything at once doesn’t produce a deeper epistemology, it dissolves the background that makes doubt possible.

Second, on whether Witt is merely “innocently noting a fact.” I think this is where the hinge point is easy to miss. It can sound as if Witt is mainly giving a recommendation, “stop here or you’re doing nonsense.” But a hinge isn’t a proposition we choose to stop with because we’re tired of explaining. It’s what has to stand fast for the practice of giving and asking for reasons to have traction at all. So when Witt indicates a stopping point, the point isn’t “here is where I recommend you stop.” It’s “past this point your demands no longer operate as epistemic demands.” The “ought” is internal to the practice, not moral policing. Your race analogy captures it: past 500 yards you may still be running, but you’re no longer running that race. Likewise, you can refuse to stop, but then you’ve shifted from ordinary epistemic assessment into a different philosophical ambition, a demand for an ultimate ground.

That’s also why everyone isn’t automatically a Wittgensteinian. The disagreement isn’t whether practices have hinges, they do. The disagreement is where to locate them, how to describe them (I think there's a hierarchy of hinges), and whether someone’s philosophical demand has genuinely left the space of epistemic assessment or is still a legitimate request for further justification within it. My paper is trying to make those boundaries explicit without using hinge talk as a conversation-stopper.

Finally, I agree completely that unclarity about justification is a problem for Witt and for me. That’s why I’m writing the paper. Hinges are unavoidable, and they mark a legitimate limit: you can’t treat justifiication as an all-purpose demand that reaches all the way down. What you can do is make the standards of justificatory standing explicit, error, correction, defeaters, and the role of understanding in tracking mistake-conditions, so that “this is where we stop” isn’t a gesture, it’s a disciplined account of how our epistemic life actually works.

Hinges aren’t merely linguistic habits or conversational conveniences. They are bedrock commitments by which we take ourselves to be in contact with reality. They function as conditions of possibility for inquiry, but precisely because of that they carry ontological commitments: — Sam26

No, the hinges are neither habitual nor pragmatic/convenient. But . . . "By which we take ourselves to be in contact" -- that's the rub. What is the practice being described which can result in contact with reality, aka, that which ontology studies? This question isn't limited to Witt-related thought, of course, but nor do the Wittgensteinian moves render it unproblematic.

Another way of saying it: To carry ontological commitments -- which I agree that hinges do -- is not to be part of what ontology studies or describes. There remains the question of the status of our epistemic practices as they relate to what we're pleased to call "reality." That is an Ur-metaphysical question, so possibly out of bounds for the Witt line of thought?

More to say about the rest of your reply -- as always, you nail the issues beautifully -- but gotta run now.

This having just read the first post. (I feel I should always say that all my statements of the way I take things to work are meant as provisional and questioning (brainstorming, however arrogant in appearance), subject to being accepted or clarified.)

Interesting juxtaposition. Unrelated essentially, but there are other kinds of “knowledge” than propositions with justification, like “I know how to put together a gun”, or, at least, the justification is putting together the gun. Or, as in On Certainty, “I know you are in pain”, as I walk away, or, alternatively, sit by your hospital bedside to console you. Again, maybe tangential, but just to mention cases outside the dichotomy of, let’s call it emotional vs rational, or to note these are legitimate (non-“private”) while still individual (vs general) or even personal.

Gettier cases lose much of their force because many depend on a mismatch between seeming justified and having justificatory standing. — Sam26

All I know about this to comment on is that the confusion attempting to be addressed is because people imagine knowing or understanding as objects, which we “have” or which are an internal state, instead of a judgment we make (based on criteria), i.e., they are not “physical”, but logical. “You know, you don’t.” If knowing or understanding is like an object that I possess, and then it turns out that I didn’t know, they ask: where/when was the understanding? instead of seeing the whole process and, as Sam says, the place of poor judgment, errors, misunderstandings, rectification, etc.

In some cases justifiable knowledge might not have, or at least require, justification, say, if it is uncontested. One reason being that justifications are not necessary beforehand, showing that knowledge is claimed, and so is something in time, an event, but also that it is claimed by someone. As with reasons (a different matter), we provide justification after the fact—though we may gather it beforehand—because we justify it to others, though the (philosophical) ideal is that it is to/for everyone, and even imagined as apart from anyone (wishing it were not a matter of judgment, but undeniable, thus forceful—able to change behavior or set “norms”).

Now a claim of knowledge may legitimately rest on authority. If you are an expert, that can be all we need as justification that you know. The proof is still in the pudding, but I bring this up because a claim to know is a stance that I take analogous to a promise to another [from Austen], that puts me—my authority and reputation—on the line, different than just saying (unsure, shrugging) “I believe(?) [think…] it’s…”. If I claim to know, I’m asking to be trusted, subject not just to being wrong, but being delegitimized, losing standing. Relatedly, a belief is not always a claim that is not justified, but can also just be a guess that is afterwards verified. So we don’t hold a person’s feet to the fire in the same way with belief as when they claim to know. An aside, the authority of conviction is not a claim of knowledge, nor even a strong belief (feeling) that a claim is true, but a willingness to be true to something (a claim of who I am, what I stand for).

Where would you draw the boundary between justification and understanding, if you think there is one. — Sam26

We may need to differentiate justifying knowledge from understanding. I would take “understanding” as, with the gun example, a claim that is demonstrated. “Do you understand?” “[Claim] I understand.” “Then show me… how to long divide, build a cabinet, critique a poem, justify a claim to know, etc.” When I claim that I understand, I subject myself to judgment, rather than, or more than, with a claim to knowledge, which, apart from, say, correctness, is subject to ongoing responsibility (to its promise).

But this also leads to whether judging what is knowledge (or whether I know) sometimes relies on different types of criteria depending on the case. It seems the variety of criteria is sometimes a matter of interest in the question. If I say I know a tune that is playing, I may be asked to give the title of the song (right/wrong), or to hum the rest of it. I may hum poorly but enough to verify it is the same song I claimed to know, but one could say I am not justified to claim I know it (well). Also, I could claim I know Bigfoot exists. When asked for a justification, I might say that I saw it. And in some cases, witnessing first-hand is considered evidence of knowledge, as in court, but then the legitimacy of the witness comes under judgment (with particular criteria), but differently then a claim to authority.

So yes, to make a claim to knowledge (of the justified kind), we should be familiar with (understand) the standards for justification. This may be part of Sam’s animus here, to say that: not only can a claim to knowledge be wrong, but illegitimate, because of a misunderstanding of the criteria for justification. I might know the answer about the elements of water, but I don’t need to understand chemistry. What I do understand (demonstrate) though is the way truth and falsity are judged, through verification. If we make a claim to knowledge without understanding the workings and criteria for justification, I would think we break them, or ignore them. Witnesses can be impeached, evidence can be illegitimate, and experts can be discredited, but are there consequences of not understanding/demonstrating the standards of justification? There is the abuse of ignoring them and simply relying on authority (say, power), but there are also abuses in not accepting justification, say, in wrongly delegitimizing a claimant, perhaps even because of or through the use of criteria of justification.

No, the hinges are neither habitual nor pragmatic/convenient. But . . . "By which we take ourselves to be in contact" -- that's the rub. What is the practice being described which can result in contact with reality, aka, that which ontology studies? This question isn't limited to Witt-related thought, of course, but nor do the Wittgensteinian moves render it unproblematic.

Another way of saying it: To carry ontological commitments -- which I agree that hinges do -- is not to be part of what ontology studies or describes. There remains the question of the status of our epistemic practices as they relate to what we're pleased to call "reality." That is an Ur-metaphysical question, so possibly out of bounds for the Witt line of thought? — J

I think that’s the right pressure point, and I don’t want to dodge it with Wittgenstein slogans.
When I say by which we take ourselves to be in contact with reality, I’m not trying to smuggle in a practice that gives us access to Being. I mean something more minimal, i.e., that our ordinary epistemic practices already operate with a contrast between getting it right and getting it wrong, between correction and mistake, between appearance and reality. Hinges are part of what makes that contrast intelligible. They don’t secure contact with reality, they are the background commitments that make the very idea of contact, miscontact, error, and correction usable.

So, I agree that carrying ontological commitments is not the same as doing ontology. Hinge talk isn’t an ontology and it doesn’t settle ontology. What it does is clarify the foundational/bedrock commitments under which ontological discourse, or any discourse can even get off the ground. In that sense it’s neither merely a pragmatic convenience nor a metaphysical proof. It’s a claim about the grammar (Wittgensteinian grammar of course).

On the fundamental metaphysical question about the status of our practices in relation to reality: yes, that question remains. Wittgenstein doesn’t abolish it. What his line of thought does is block a certain way of posing it, the way that tries to demand a justification for the whole framework while still using the framework’s notions of justification, evidence, and correction. If you ask the Ur-question as a metaphysical project, fine. My point is just that it’s no longer an ordinary epistemic demand, and it can’t be answered by ordinary justificatory moves, because those moves presuppose the very standing-fast background under discussion.

I’d say it this way. Hinges have ontological commitments, and in that sense they touch ontology. But they don’t deliver ontology. They set the stage on which ontological arguments can be intelligible, and they explain why some global demands for justification misfire: not because reality is off limits, but because the demand is asking for a kind of validation that cannot be supplied without circularity or infinite regress.

I don't want to make the thread all about hinges, but I also don't want to not take questions about hinges.

Continuing with paper...
Post #10

6.3 Defeater Screening
The third guardrail concerns challenge. Justification is not only a matter of placing a belief within support. It is also a matter of whether that standing survives relevant disruption. Defeater Screening names the fact that, within our practices, certain challenges count as undermining, and when they are present and undefeated the belief no longer has the standing required for knowledge.

This does not mean that one must answer every skeptical possibility. Our language-games do not treat every imaginable doubt as relevant. What matters is the kind of defeater that the practice itself recognizes as bearing on the standing of the claim. Sometimes a defeater shows that the grounds were false. Sometimes it shows that the route was misapplied, that the inference did not hold, that the perceptual situation was misleading, that the testimony lacked standing, or that a concept was applied outside its proper range. In each case the issue is not private reflection but public standing. A belief is not counted as knowledge when it stands under an undefeated challenge that the practice takes to be disqualifying.

Defeater Screening therefore belongs to the grammar of justification. It is part of what it is for a belief to stand properly within a practice, rather than to be held in a way that collapses under the first relevant counter-consideration.

I’m not addressing knowledge as a skill here, knowing how to build a cabinet, assemble a gun, play a tune, etc. That’s important, but it’s a different topic. My focus is propositional knowledge, claims of the form “I know that p,” where questions about justification, error, correction, defeaters, and Gettier pressures actually arise.

On that score, I agree that knowing and understanding aren’t inner objects we possess. They’re statuses we attribute within practices, and practices have standards for what counts as getting it right and what counts as correction. That’s why I talk about justificatory standing rather than private confidence or conviction.

Where the “+U” bites is this: in the propositional cases I’m focused on, understanding is competence with mistake-conditions. It’s being able to say what would count against the claim, what would defeat it, and what would correct it. That’s what keeps justification from collapsing into citation, authority, or lucky success.

And on authority: yes, propositional knowledge often rests on testimony and expertise, but authority isn’t self-certifying. It’s answerable to provenance, track record, and defeaters. If someone can’t track those, then there's an obvious problem, and that is exactly what many Gettier style worries are seeming to expose.

A lot of recent hinge work has been shifting away from hinges as a magic bullet against skepticism (I don't claim it as a magic bullet, but I do think it shows how global skepticism misfires) and toward practice facing issues: trust, testimony, deep disagreement, and the way background commitments shape what even counts as evidence and correction. Coliva’s work on “hinge trust,” and the newer “social hinge epistemology” literature, are good examples.

Where I separate myself. Contemporary hinge epistemology often gets stuck arguing over one global theory of what hinges are, or over whether hinge frameworks really deliver the anti-skeptical payoff people want. You can see that pressure in recent work that argues the prominent “framework” formulations don’t succeed on their own terms. My paper isn’t mainly trying to win that internal hinge debate. I use hinges to mark the structural background of epistemic assessment, and then I focus on what actually disciplines claims to propositional knowledge inside language-games and forms of life.

My distinctive move: I start with JTB as a familiar grammar for propositional knowing, then I tighten the “J” condition with explicit guardrails, and I treat “+U” as a clarification of justificatory standing, not a fourth ingredient. The separate the person who can track mistake conditions, defeaters, and correction, from the person who can only recite the right conclusion. That’s also why testimony matters so much in my account, it’s not a lesser substitute for real knowledge, it’s a primary route that has to be disciplined by the same practice governed rules.

One more divergence, and I’m happy to own it. I treat hinges as arational and foundational, and I also think they carry ontological commitments. That’s stronger than a purely “methodological” reading, but it explains why global skepticism misfires, viz., it tries to call into question the very background that gives doubt, checking, and correction their sense. (Whether you want to call that a “solution” or a “dissolution” is secondary.)

On the fundamental metaphysical question about the status of our practices in relation to reality: yes, that question remains. Wittgenstein doesn’t abolish it. What his line of thought does is block a certain way of posing it, the way that tries to demand a justification for the whole framework while still using the framework’s notions of justification, evidence, and correction. — Sam26

OK, good enough.

I’m not trying to smuggle in a practice that gives us access to Being. I mean something more minimal, i.e., that our ordinary epistemic practices already operate with a contrast between getting it right and getting it wrong, between correction and mistake, between appearance and reality. — Sam26

Also OK, as long as we don't get too excited about "appearance and reality." That, as you say, is a different metaphysical animal in most philosophical approaches.

I don't want to make the thread all about hinges, but I also don't want to not take questions about hinges. — Sam26

Fair enough. I don't want to pull you away from the focus of the thread, though I think we both agree that JTB+U is an attempt to shore up a strong notion of what justification actually means, which requires the hinge idea to carry a lot of weight, as a limit point for ordinary epistemic justification.

If I say I know a tune that is playing, I may be asked to give the title of the song (right/wrong), or to hum the rest of it. — Antony Nickles

This example raises a possibly disconcerting question. Suppose what I say is, "I know I've heard that tune before." Is there any justification I can be asked for, or can offer? This seems different from the earlier "purple cow" example, where we can separate out "purple," "cow", etc., which are public, from the private data of "I am thinking of . . ." What would be the equivalent public criteria for "to know one has heard before"?

Returning to your earlier post from yesterday:

So when Witt indicates a stopping point, the point isn’t “here is where I recommend you stop.” It’s “past this point your demands no longer operate as epistemic demands.” — Sam26

So the question is, Is this stopping point as clear as the one that ends the 500-yd race? The right answer might be, Some are and some aren't. In fact, it might be helpful to take a pair of alleged hinges that may be central to epistemic justification and really work through a demonstration of why one is, while the other may not be. How exactly does skepticism about, say, "There is an external world" necessarily and unambiguously undercut the concept/grammar of doubting? (I think your familiarity with all this may make it appear more obvious than it is to others.) I'm not suggesting this is an impossible challenge, quite the contrary. I think the more we understand about why the limits of justification are what they are, the better we'll be able to circle back and ask what knowledge really is.

Which is all merely to affirm what you say here, especially about the hierarchy:

The disagreement isn’t whether practices have hinges, they do. The disagreement is where to locate them, how to describe them (I think there's a hierarchy of hinges), and whether someone’s philosophical demand has genuinely left the space of epistemic assessment or is still a legitimate request for further justification within it. — Sam26

Lastly:

“this is where we stop” isn’t a gesture, it’s a disciplined account of how our epistemic life actually works. — Sam26

A tangential question: Would you allow any of this to be culturally conditioned? That is, might Chinese philosophy's epistemic life "actually work" differently? Or is that move the same as the psychological move concerning an individual, where justification becomes an empirical study about the human consciousness rather than an analysis of concepts/grammar?

And now . . . I'll read your paper. Are you looking for more or less the same kinds of feedback as the earlier questions for the original posts?

It all hinges upon whether the cartesian notion of belief is admissible. According to naturalized epistemology, beliefs are stimulus-response dispositions that are conditioned by a community to approximately reproduce some aspect of a shared semantic network, as when training an LLM to respond "correctly".

Ask ChatGPT what its hinges are. Even if you agree with its answer, is it in a position to know what it says, given that its output is a deterministic consequence a sequence of transformer blocks applied to a query? Where precisely do hinges fit in the machine learning pipeline?

I mean justificatory standing, the sort of standing a belief has when it is supported by the standards that govern a practice, standards for what counts as evidence, what counts as error, and what counts as correction. — Sam26

When I think of a justified claim, I imagine judging the claim. When I think of a practice, I picture the person conducting the practice. Is this a separate concern of the legitimacy of the claim at all? as “That belief is not justified.” is different than “That is not justifying!”

understanding is competence with mistake-conditions. It’s being able to say what would count against the claim, what would defeat it, and what would correct it. — Sam26

I agree that understanding justification is being competent in the practice. But we do not “understand”, or “have” competence, in the same way we don’t have an internally-pictured object; we are only judged to have understood, i.e., we demonstrate understanding. So understanding is not a matter of saying (or “knowing”) something (beforehand), because, as I mentioned, it’s a matter of timing. Only in our adherence to the practices of justification during the process, do we demonstrate our understanding of justification, and so it is impossible to judge competence before an attempt at justification is made. Otherwise, we would never have the possibility to judge between a mistake in justification and a lack of understanding of justifying. And thinking we can judge beforehand ignores the same error and correction in the practice (afterwards) as with the claim.

justification is a normative standing within a practice — Sam26

Is “justified” the standing? I may grant that your belief is justified, but why must that be a “status”? Because justification is the practice. Like all practices, it is the practice that is “normative”. If you don’t follow the practice of apologizing, the apology is not wrong, you just aren’t apologizing (didn’t). And the statement of a proposition as true or false does not require justification, just verification. So the correct answer to a math problem is true whether it was a “lucky guess” or not. However, if you apologize correctly, there is still no assurance it will be accepted; that’s the nature of that business. A claim to knowledge also does not start out requiring justification (presumably all done beforehand). We may request it, and the claim must be open to repudiation (not rest on authority). If you follow the practice of justification, the claim now has justification. But is it necessarily justified (reached a standing or status in and of itself)? say, as something is necessarily true if it is verified? There still appears to be the process of judging whether the justification of the claim is sufficient, and thus the question of where we stop. But perhaps these requests for justification are part of the practice, subject to being appropriate, or correctly requested.

understanding is the competence by which a person can genuinely participate in [the practice of justification], not merely mimic its conclusions. — Sam26

I’m not sure why “genuineness” comes into it. Why is that even introduced into the conversation? Maybe we are thinking of participating completely? “operatively”? being able to respond appropriately to all requests for justification? Are we making a comment that justification is a process, a conversation, rather than just conclusions, only correct answers?

belief can look justified from the outside, and even to the person themselves, but the justificatory standing is fragile, because the person does not reliably track the mistake-conditions that the practice treats as decisive. — Sam26

The bolder text is problematic for me. I take “from the outside” as not a measure opposed to something I may have or not, but referring to superficiality, without depth. As you say, “fragile”. And so “decisive” is not the correct end, as it would be (could be) with a matter like truth, rather than a process of, say, reaching or creating trust.

And we appear to again be putting the cart before the procedural horse in judging a superficial justification “does not reliably track” (that is a critique after a request that might led to ignoring a condition for a mistake—only able to be judged ongoingly). Stopping superficially may only be a reflection of the low stakes.

Understanding is what makes the support more than a recitation, it is the ability to locate the claim within the space of reasons, objections, defeaters, and revisions that the practice recognizes. That is not infallibility, and it is not an impossible demand. It is simply the difference between having a standing and merely borrowing one. — Sam26

This mirrors my point of understanding justification as a demonstration of a process in time, and so, yes, not a matter of “infallibility”, but then also not a matter of ending in “decisiveness”. If this is a process, do we “have” a standing (when?), rather than stand ready to justify (more)? From what I can tell so far, the criteria here are (or wish to be) a matter of just exclusion (from participation, from having standing to participate), but I will continue. Perhaps it is a matter of holding up (appropriately) under the demands for justification.

Can you think of a counterexample, a case where someone lacks this competence but still seems to have genuine justificatory standing. — Sam26

And here we come up against it. Are we entitled to exclude someone from the process due to incompetence? Yes. Must we? No, as it is not the description of a status (knowing or understanding), but the outcome of a process; a process that may be set on the tracks again. As Wittgenstein qualifies, we are only “inclined” to throw up our hands (PI #217)—it is up to us whether we start again. Also, as I said, how are we to (outside of going forward) judge between a lack of competence and a mistake (or even defiance)? Is the questioner the only one with the power to call foul?

No False Grounds (NFG). A belief cannot have genuine justificatory standing if the support it depends on is false, or if it is being carried by a false presupposition that is doing the real work. The point is simple: if the ground is false, whatever looks like support is a counterfeit support. — Sam26

This of course makes perfect sense within the framework of our present knowledge. So not sure this matters, but there is the case where a belief turns out (in the future) to be based on false support that was true/legitimate at the time. Now, does this mean we didn’t know? Sure, but that might just be to say that we didn’t know the support was false. The claim was justified. In the case of WMDs, we could say we were simply never given the time to adequately justify the claim, refute the false support. In the case of relativity, we didn’t make a mistake, but, say, overlooked something. I take this to show there are different cases, and so not just that “seeming” to be justified is not a thing, but being justified is also not a matter of the opposite being the case (being true, genuine, etc.).

Defeater Screening. Even when the grounds look good, justificatory standing is undermined when there are live defeaters that have not been faced. — Sam26

But of course we cannot face all live defeaters before we would say a claim is justified. I would suggest this again demonstrates the ongoing nature of justification, our responsibility for it (to respond), so we do not reach a “decisive” end.

I agree with your “some are and some aren’t.” The stopping point isn’t always as clearly delineated as a finish line. Take “there is an external world.” The point isn’t that this is a proposition we happen to feel very confident about. It’s that the skeptical demand to doubt it globally can’t be carried out without using the very grammar it is trying to suspend. To be able to doubt in a way that counts as doubt, you need criteria for checking, what would count as error, what would count as correction, and what would count as a defeater. But those notions are already world involving, they presuppose stable objects, re identification across time, public conditions of correction, and a contrast between appearance and reality. The global doubt doesn’t undercut a particular belief, it undercuts the practice conditions that give doubt its meaning. That’s why it’s hinge territory.

As for the cultural question, I think cultural conditioning in a limited sense is only natural (how much depends on the context). Different cultures can weight different routes, emphasize different norms of authority, etc. But that’s not the same as saying the grammar of justification is an empirical study of how this group thinks. The hinge point is structural, any community that has a practice of giving and asking for reasons, correcting mistakes, and distinguishing seeming from being will have some things that stand fast in order for the practices to function. So yes, there can be variation in which commitments play hinge roles and how they’re expressed, but the need for a bedrock background isn’t optional, and it’s not reducible to psychology. Just as the hinge background of chess (board and pieces) aren't optional if you're playing traditional chess.

Pure logic (boundary setting only). There are limits that are not empirical discoveries but logical constraints, what is possible, what is coherent, what follows from definitions, what collapses into contradiction. This route does not supply new facts about the world. It sets boundaries, clarifies entailments, and exposes category mistakes. — Sam26

Of course there is formal logic, which has it uses, but I would think more important is the internal logic of a practice, some of which are particular versions of ones you mention. As I said above, there are criteria for what we consider to be an apology, a sufficient one to categorize it as an apology. There are constraints for correctness, boundaries for appropriateness. The implications of certain acts within a practice have specific implications. All this is to say that formal logic is but one practice, no better, more important, or more necessary/powerful than others. Of course, its independence and certainty make it more desirable.

Do you think “linguistic training” deserves to be a distinct route, or is it better treated as part of the background of the other routes. — Sam26

I think yes, it is like an underlying condition. Perhaps tangential, but it would be enough to say we are trained in practices (indoctrinated, pick up by osmosis). We are sometimes taught with language; sometimes taught how and when to use language, but to label it “linguistic” is to trivialize that we are really learning how the world works.

Of course there is formal logic, which has it uses, but I would think more important is the internal logic of a practice, some of which are particular versions of ones you mention. As I said above, there are criteria for what we consider to be an apology, a sufficient one to categorize it as an apology. There are constraints for correctness, boundaries for appropriateness. The implications of certain acts within a practice have specific implications. All this is to say that formal logic is but one practice, no better, more important, or more necessary/powerful than others. Of course, its independence and certainty make it more desirable. — Antony Nickles

Formal logic is one practice among others, and a lot of what governs our epistemic life is the internal logic of practices, what counts as an apology, what counts as correction, what counts as evidence, and so on. That’s why I talk about objective justification as part of the five ways we justify our beliefs, rather than treating justification as a purely formal relation between propositions.

Formal logic isn’t more important in the sense that it's some master key to knowledge, but it has a particular role. It doesn't supply the whole grammar of justification on its own, because most practices involve standards of relevance, error, correction, and defeaters that aren’t reducible to formal entailment. I’m happy to say, logic is indispensable for certain jobs, but it’s not the whole story, and it shouldn’t be treated as the model for every kind of justification. Finally, I make a distinction between pure logic and the logic of deductive and inductive reasoning.

Continuing with paper...
Post #11

6.4 The Guardrails as Clarifications of Justification

Taken together, these guardrails clarify the shape of justification as it functions. No False Grounds excludes cases in which the support is defective at a crucial point. Practice Safety excludes cases in which truth is reached only by luck. Defeater Screening excludes cases in which the belief cannot retain standing under relevant challenge. None of this adds a new condition to JTB. It makes explicit the discipline that our justificatory practices already embody.

This is also where the role of understanding, as I use the term, becomes sharper. Understanding is not a separate achievement layered on top of justification. It is the conceptual competence through which justificatory support has its proper use within a language-game. The guardrails describe the constraints that this competence must respect if a belief is to have the standing required for knowledge. When a belief violates these constraints, it may still be true, and it may still feel compelling, but it does not stand as knowledge within our practices of justification.

With these guardrails in view, the classical model is not weakened but clarified. We can see why some beliefs that look well supported nevertheless fail to constitute knowledge, without treating Gettier-style cases as decisive objections. The next step is to apply this clarified grammar to contemporary pressures, including the temptation to describe artificial systems as knowers and the need to preserve the ordinary concept of knowledge in an information environment that often rewards persuasion over justification.

Rule-following brings the point into sharper focus. To use a concept correctly is not to consult a private rule, nor to match an inner image, but to have been trained into a practice where “correct,” “mistake,” “same,” and “different” have their use. The standards that govern justification are therefore not hidden in the mind. They are displayed in our shared procedures of correction and agreement, in what counts as getting it right and what counts as needing revision. This is why justification is public in its grammar even when it concerns matters that are privately experienced. If there were no practice in which correctness could be shown, there would be no sense to the claim that a belief is justified.

Continuing with paper...
Post #12

7. JTB+U and Artificial Intelligence: Why AI Does Not “Know”

The present interest in artificial intelligence has brought an old temptation back into view. We are inclined to treat fluent performance as if it were knowledge, and to treat the production of correct sounding answers as if it were understanding. This temptation is understandable. The outputs of large language models can resemble the surface of competent human speech. They can summarize, explain, and argue, and they can do so in a way that often passes casual scrutiny. Yet the resemblance is grammatical only at the level of appearance. When we look more closely, we see that the ordinary criteria for knowledge are not satisfied, not because the machine lacks a private mental state, but because it does not stand within the practices that give the concept of knowledge its use.

Truth remains the success condition for knowledge, and nothing in what follows weakens that point. An artificial system can produce a true statement, sometimes with striking reliability. But knowledge is not merely the arrival at truth. Knowledge is true belief that stands within a practice of justification, and the standing is not a decorative label. It depends on the routes by which the belief is supported, the guardrails that discipline that support, and the background of bedrock certainties that makes the whole practice possible. This is the first reason the language of knowledge becomes slippery when we apply it to machines. The system produces assertions, but it does not participate in the language-games in which assertion, challenge, withdrawal, and justification have their life.

This is also where the role of bedrock certainties becomes decisive. Human justification presupposes a background that stands fast for us. These certainties are not items we know. They are the inherited conditions under which doubting and knowing take place. They form a hierarchy in the sense that some stand deeper than others, and they are displayed in action rather than defended by argument. The point is not that a machine lacks a set of stored assumptions. It is that the machine is not trained into a form of life in which such certainties function as the background of justification. An AI system does not stand within the practices that define what counts as a mistake, what counts as correction, and what counts as the withdrawal of a claim. It can be updated, constrained, and fine-tuned, but this is not the same as occupying the human space in which bedrock certainties show themselves as what stands fast.

The five routes also clarify the difference. When a person justifies a belief through testimony, inference, sensory experience, or linguistic training, the support is situated within a practice in which the believer can be held responsible to standards. These standards are public and they include the possibility of being corrected in the relevant way. A language model can mimic the outward form of these routes. It can cite sources, draw inferences, and use perceptual language, but these are linguistic gestures, not placements within the practice itself. The model does not have testimony in the human sense, since it is not a participant in the practices that give testimony its standing. It does not infer in the human sense, since it does not operate with the conceptual competence that makes an inference a movement within a language-game rather than a pattern of token transitions. It does not perceive, and so it does not have sensory experience as a route of justification. It displays linguistic training in the limited sense that it has been trained on linguistic material, but this is not the kind of training by which a human learner comes to grasp the use of a concept within a form of life. It is closer to the acquisition of a statistical profile of usage than to the possession of a concept.

This is why the distinction between statistical competence and conceptual competence matters. A model can be statistically competent, in the sense that it produces language that fits patterns in its training data. It can do this at scale and with impressive fluency. But conceptual competence is not the possession of patterns. It is the ability to use a concept correctly within a practice, to respond to correction, to recognize when a challenge is relevant, and to withdraw a claim when the practice requires it. These are not private mental achievements. They are displayed in the way one stands within a language-game. The machine can be made to output a retraction. It can be prompted to list possible objections. Yet these are outputs, not the standing of a belief within a practice of justification.

The guardrails bring the point into sharper focus. No False Grounds matters because a model can generate support that looks acceptable and yet includes a false claim doing essential work. Practice Safety matters because a model’s correct output may be the result of a fortunate match rather than stable standing, especially in domains where the system has not been constrained by reliable sources. Defeater Screening matters because, while a model can generate lists of objections, it does not occupy the public discipline in which defeaters arise as challenges that change the standing of a belief. The model can simulate the discourse of justification, but it does not stand within a practice where its claims are owned, corrected, and withdrawn in the way that our language-games require.

None of this implies that AI is useless in epistemic life. The opposite is true. Artificial systems can be powerful instruments within human practices of justification, especially when their outputs are constrained by reliable data and when they are treated as aids rather than as knowers. They can help gather information, surface patterns, and organize arguments. But this usefulness does not collapse the grammatical distinction between producing true sentences and knowing. To treat the machine as a knower is to project the grammar of our concept of knowledge onto something that does not meet its criteria of use.

This is why JTB+U is especially valuable in the present environment. It gives us a disciplined way to distinguish persuasion from justification, fluency from conceptual competence, and the appearance of support from genuine standing within a practice. It also helps explain why the language of certainty is often misused in discussions of AI. A model can produce confident sounding claims, and this can resemble subjective certainty. But hinge certainty is not confidence, and epistemic certainty is not mere persistence under repetition. The kinds of certainty that matter to knowledge are rooted in practices and in what stands fast within them. The machine does not inhabit that structure, even when its outputs resemble the surface of a human epistemic performance.

For these reasons it is better to say that artificial systems can produce true statements, and can assist human beings in practices of justification, without saying that they know. The temptation to speak otherwise is understandable, but it blurs the grammar of knowledge at exactly the moment when we most need it to be clear.

This is a strong reply, and I agree with much of it, but I don't think it gets to the heart of @J's concern. My interpretation is that J is not questioning whether global doubt is incoherent, but is asking why grammar should be considered sufficient to settle the issue. In other words, what explains why grammar imposes the limits it does?

I would argue that Wittgenstein's anxieties over transcendental reification make it difficult for him to adequately address this question. I don’t think the deepest explanation can be grammatical. Grammar registers the limits, but it doesn’t generate them. These ultimately need to be grounded in the structure of our normative/epistemic acts themselves: to doubt, correct, or inquire is already to be oriented toward what is the case, toward conditions of fulfillment that distinguish seeming from being. An act of doubt misfires when it asks for fulfillment while cancelling the conditions of fulfillment.

So doubting is a form of judgment guided by reasons. Reasons presuppose the possibility of correcting mistakes by attending to data and testing insights. If you globally deny the existence of any constraint on the data of experience then you undercut the very idea of error, correction, learning and also doubt itself. Inquiry is intelligible only as a self-correcting process of answering questions about what is the case and is therefore rendered unintelligible under the assumption that there is nothing in principle that can settle such questions.