How Epistemology Can Help Align AI

A Wittgensteinian JTB+U Framework for Responsible Intelligence

Introduction: Fear Without Framework
One of the greatest public anxieties about artificial intelligence is not simply that it makes mistakes, but that it makes them with the wrong kind of confidence. AI systems already generate falsehoods while sounding certain, apply reasoning across domains without context, and produce outputs we cannot always distinguish from knowledge. People worry that such systems may spin out of control — not only because they are powerful, but because we lack a shared standard for when their outputs count as genuine knowledge.

What if part of the solution lies not only in engineering but in epistemology? Philosophy has wrestled for millennia with the conditions of knowledge. The classical model of Justified True Belief (JTB) remains the backbone: one knows something if it is true, believed, and justified. Yet this model alone is too thin for the complexities of modern reasoning. In my work I extend JTB with a fourth element — Understanding (U) — and frame it through Wittgenstein’s later philosophy. The result, JTB+U, offers a grammar of knowledge that could help discipline AI reasoning and reassure human users.

The Core: JTB+U
The JTB model works well because it matches our practices: we do not call someone knowledgeable if they voice a lucky guess, nor if they cling to a false belief. We demand truth, belief, and justification together. My addition of +U (Understanding) makes explicit what was always implicit: that knowledge also requires uptake. A system must not only produce a true justified belief but must also grasp the concept well enough to apply it competently across contexts.

In human life, understanding is what allows us to use knowledge in reasoning, explanation, and correction. Without it, justification can be hollow or accidental. For AI, adding +U means more than citing sources or probabilities — it means showing conceptual grasp: when to apply a claim, what follows from it, and when it fails.

Certainty: Degrees and Kinds
AI systems often present outputs as if they were absolute, when in fact they are probabilistic. Here Wittgenstein’s analysis of certainty helps. We can distinguish at least four senses:

Subjective certainty — conviction or felt assurance.

Hinge certainty — arational bedrock that makes doubt possible (e.g., the world exists).

Epistemic certainty — defeater-resistant knowledge in practice, stable enough to guide action.

Absolute certainty — logical or moral necessity, where denial collapses into contradiction.

Probabilistic reasoning belongs to epistemic certainty: not infallible, but robust against correction and sufficient for rational action. By contrast, mathematics and logic express absolute certainty in kind, not in degree. An AI system grounded in this grammar would know how to communicate not only confidence scores but what kind of certainty is at stake. That would go far in preventing false impressions of infallibility.

Hinges: Bedrock for AI Reasoning
Another danger is regress: how do we know that an AI’s reasoning is not flawed at some hidden level? Wittgenstein’s answer was that justification does not go on forever. At some point there are certainties that “stand fast”: basic beliefs not themselves justified but making justification possible.

Humans take for granted that the world existed before our birth, that memory is generally reliable, that other people have minds. An AI system, too, must operate with hinge assumptions: that its arithmetic is consistent, that its data streams are genuine, that communication has meaning. If AI could make such hinges explicit, humans would better understand what the system presupposes, and where corrections could intervene. Transparency at the hinge level could dissolve much of the “black box” fear.

Language-Games: Contextual Justification
Wittgenstein also reminded us that justification is not timeless but practice-indexed. What counts as evidence in a courtroom differs from what counts in a laboratory. Human beings navigate these shifts in grammar fluidly; AI systems do not.

Here the JTB+U framework highlights five primary routes of justification:

1) Testimony

2) Logic (inductive and deductive reasoning)

3) Sensory experience

4) Linguistic training

5) Pure logic (boundary-setting only)

Each route is weighted differently across practices. In law, testimony and records dominate; in science, prediction and replication; in daily life, sensory access and trust. An AI system that tracks these contextual differences could avoid misfires — no longer treating courtroom testimony as “unscientific,” nor confusing legal plausibility with empirical law. Context-sensitive justification is not optional; it is essential for responsible reasoning.

Understanding as Uptake
Perhaps the most radical demand of JTB+U is that knowledge requires understanding. For AI, this entails more than generating correct statements. It requires showing uptake:

* applying concepts correctly across cases,

* responding to defeaters,

* grasping the inferential consequences of claims,

* recognizing when claims no longer fit.

Without +U, AI outputs risk remaining surface mimicry — statistical echoes of human reasoning. With +U, they could approximate conceptual competence: the ability to play the language-game of giving and asking for reasons.

Why Epistemology Matters for AI Safety
Technical controls are vital, but without an epistemological grammar they risk being blind. JTB+U offers a framework for:

* Transparency — AI can show what kind of certainty it claims.

* Trust — users can see hinge assumptions and justificatory routes.

* Humility — AI can learn to say “I don’t know” when defeaters arise.

* Alignment — systems become participants in human language-games, not alien oracles.

AI will never be “safe” if we cannot tell when it knows what it says it knows. A Wittgensteinian epistemology does not solve all engineering problems, but it clarifies the standards by which outputs should be judged. If adopted, it could help AI evolve from a powerful but opaque prediction machine into a responsible reasoning partner.

Conclusion: Epistemology in Practice
Philosophy is often accused of irrelevance, but here it has something urgent to offer. The JTB+U framework, enriched by Wittgenstein’s grammar of hinges and language-games, provides exactly the clarity needed at the frontier of AI. It disciplines knowledge-claims, distinguishes types of certainty, contextualizes justification, and demands understanding. If we are serious about aligning AI with human values, we must first align it with the human grammar of knowledge. Epistemology, far from being a museum piece, may be one of the keys to AI’s safe future.

One of the recurring problems in AI research is the ambiguity of “knowledge.” Systems can be designed to store facts, to process probabilities, or to simulate reasoning, but what it means for an AI system to know something is rarely clarified. Engineers often collapse knowledge into probability scores, while philosophers insist that knowledge requires more than high confidence. This is precisely where a Wittgensteinian extension of the classical JTB model — JTB+U — could offer guidance.

The first benefit lies in distinguishing different senses of certainty. Human beings navigate between convictional certainty (“I know she loves me”), epistemic certainty (“I know the bridge will hold”), and absolute certainty (“2+2=4”). An AI system that fails to recognize these differences risks either inflating its probabilistic outputs into necessity or downplaying genuine knowledge as “only probable.” Incorporating the grammar of certainty into AI reasoning could allow systems to represent not just confidence levels but also kinds of certainty — treating mathematical truths differently from empirical predictions, and both differently from convictional expressions.

The second benefit is the awareness of hinges. Wittgenstein insisted that justification does not go on forever. At some point there are certainties that “stand fast”: that the external world exists, that memory is generally reliable, that other people have inner lives. These are not optional premises; they are conditions of sense. An AI designed without hinge-awareness will endlessly seek justification where none is required, or worse, collapse into regress. A hinge-sensitive AI could be programmed to recognize its own bedrock assumptions and to operate within them while remaining open to correction at higher levels. This would make its reasoning both more efficient and more human-like.

The third benefit arises from the “+U” in JTB+U — understanding. Current AI systems can store truths, assign probabilities, and even generate justifications of a sort, but they often lack uptake. They can state that Paris is the capital of France, but they do not consistently grasp what follows from that claim, when to apply it, or how to detect when an error-signal demands revision. Embedding “understanding” as a criterion would shift AI development away from mere statistical prediction toward competence in language-games: the ability to apply concepts correctly across cases, to give and ask for reasons in context, and to adapt when defeaters arise.

Finally, JTB+U emphasizes contextual justification. Testimony functions differently in science than in law; sensory evidence is weighed differently in everyday life than in a laboratory. An AI that can track these language-game differences will be able to tailor its justificatory standards to the domain in which it operates, rather than treating knowledge as homogeneous. This would prevent both under- and over-reach: no longer dismissing courtroom testimony as “unscientific,” while not confusing legal plausibility with empirical law.

In short, JTB+U offers a conceptual framework that could reshape AI epistemology. It would push systems beyond probability scores toward a layered understanding of certainty; beyond blind data-processing toward hinge-awareness; beyond rote recall toward genuine uptake; beyond one-size-fits-all reasoning toward practice-specific justification. If humans are to design AI that integrates into our forms of life, the grammar of “knowledge” must be respected. Far from being a mere philosophical abstraction, epistemology in this Wittgensteinian key could guide the next generation of intelligent systems.

I think you’re right to warn against reifying “truth” into a thing with different guises. Wittgenstein’s reminder is well taken: meaning is use, and every particular use of “truth” has to be understood within its practice and even within its immediate context. To that extent, I agree we don’t apply a pre-given conceptual grammar from outside.

That said, I would still hold that in the epistemic use of “true,” there is a minimal grammatical constraint that runs through its particular instantiations. However much the criteria differ — testimony in law, predictive success in science, proof in mathematics — the word “true” here always functions in contrast to “false,” and that contrast implies that the proposition fits the way things stand. This is not to hypostatize truth as an object with “faces,” but to recognize the mirroring role that the grammar of epistemic truth presupposes.

So I wouldn’t want to treat “truth” as having an essence across all uses. I agree that outside epistemology — in poetic, moral, or expressive speech — the sense of “truth” shifts without necessarily carrying factivity. But within epistemology, its factive core seems indispensable. Without it, we lose the very grammar that lets us correct, doubt, or justify claims. So my position is: meaning is use, but among the uses of “truth” there is a distinctive epistemic family where factivity is non-negotiable, even though its criteria are locally defined.

I think we’re converging on a similar point. I would agree that “truth” does not wear a single face. Its criteria shift depending on the language-game: in a courtroom truth is tied to testimony and records, in science it is tied to predictive success but also to the testimony and documentation that communicate, test, and replicate those predictions, and in mathematics it is tied to logical necessity. To borrow Wittgenstein’s term, these are family resemblances rather than a unitary essence.

Where I’d want to add a note of caution is that the factivity of truth still matters across those contexts. However we construe it, “p” being true always implies that things are as “p” says they are. Otherwise we lose the very grammar that distinguishes knowledge from conviction.

On the question of pattern recognition: I see this as closely related to what Wittgenstein called hinges, or what I sometimes call hinge-beliefs. These are certainties that do not themselves stand in need of justification but make justification possible. For example, we do not reason our way to believing that objects persist when out of sight, or that other human beings have inner lives, these are taken for granted in our dealings. They are beliefs in the broader sense, which includes both propositional claims (“The earth existed before my birth”) and non-propositional, prelinguistic certainties (the infant’s trust in the caregiver, or the body’s implicit grasp that the ground will hold). Pattern recognition provides much of the raw material for these hinges: it is how we find stability in experience, and those stable expectations are what make reasoning and justification possible.

And here is where my JTB+U framework comes in. Propositional truth on its own is “thin,” as you put it. What matters is how it is embedded in justification and uptake, how belief is connected to truth in ways that others can test, and how the agent understands what the claim involves. So I wouldn’t discard propositional truth, but enrich it. It becomes one thread in the weave of truth, belief, justification, and understanding, all grounded in the certainties that stand fast in our forms of life.

Just to reiterate...

Probability and the Varieties of Certainty

One common source of confusion in epistemology is the place of probability. Some treat probabilistic reasoning as a weaker substitute for knowledge, as though it can never rise above opinion. Others mistakenly assume that probability can deliver the same kind of necessity we find in logic or mathematics. Both views rest on a failure to distinguish the varieties of certainty.

Recall the four senses:

Subjective certainty — conviction, the felt assurance of belief.

Hinge certainty — the arational bedrock that makes doubt and inquiry possible.

Epistemic certainty — defeater-resistant knowledge in practice, stable enough to guide action.

Absolute certainty — logical or moral necessity, where denial collapses into contradiction.

Probabilistic reasoning clearly does not yield absolute certainty. No degree of evidence can turn an inductive conclusion into a logical necessity. To say “this drug has a 95% chance of success” is not to assert that its success is absolutely necessary. The grammar of probability already places it outside the category of absolute truths like “2+2=4.”

But this does not mean probability is epistemically weak. Probabilistic reasoning is one of the chief ways in which we arrive at epistemic certainty. When a belief is supported by convergent evidence — statistical studies, repeated trials, predictive accuracy — it can withstand defeaters and function with the kind of reliability that makes action rational. We rightly say we know airplanes will stay aloft, even though our confidence is grounded in probability, not logical necessity. The strength of the claim lies not in its absoluteness but in its resilience across error-signals and justificatory routes.

It is also important to note the role of outcomes. Once an event occurs, its probability collapses to 1 or 0. The coin either lands heads or it does not. But knowledge-claims are not made after the fact; they are made in the stream of life, before outcomes are settled. What matters epistemologically is not the metaphysical status of the event but the justification for belief leading up to it. Probability is the language we use to track that justification, and when paired with understanding, it satisfies the conditions of JTB+U.

In this light, probability is not a lesser form of knowing but one of the principal ways human beings secure knowledge. We live in a world where outcomes are not always transparent, but where patterns and regularities can be detected and acted upon. Probabilistic reasoning translates those patterns into degrees of epistemic certainty. It belongs not to the realm of absolute necessity but to the realm of practice — the very place where JTB+U is meant to operate.

So where does probabilistic reasoning fit? An example of a conclusion that can approach certainty in degree, but never be absolutely certain in kind. Bear in mind, that probability is, by definition, defined by an outcome, of which the probability, by definition, is 1 or perhaps 0. (I'm not saying the outcome always has to happen, just that each probability defines an outcome.) — Ludwig V

Probabilistic reasoning does not yield absolute certainty (sense 4: logical/moral necessity). It is not the kind of certainty we find in “2+2=4” or “murder is always immoral,” where denial collapses into contradiction. Instead, probability operates in the register of epistemic certainty (sense 3: defeater-resistant knowledge in practice). A high probability, backed by justification and understanding, can approach a state where we rightly treat the belief as knowledge, even though it remains in principle corrigible.

This distinction is crucial. To say “I am certain at the 99.9% confidence level that the medicine will work” is not to claim an absolute necessity. It is to claim a belief that has survived defeater screening, cross-checked by testimony, reasoning, and sensory evidence, and that functions with the kind of stability that makes action rational. In other words: probabilistic reasoning belongs in the realm of practical epistemic certainty, not absolute certainty.

Your remark about outcomes is well taken: once an outcome occurs, its probability collapses to 1 (happened) or 0 (didn’t happen). But notice that before the fact, what matters for knowledge is not the metaphysical status of the outcome but the justification of belief leading up to it. The grammar of “know” ties us to what is justifiable before the outcome is fixed. This is why probabilistic reasoning fits comfortably within JTB+U: truth (the event does or does not occur), belief (the agent commits to its likelihood), justification (the probabilistic model and evidence), and understanding (grasp of how probabilities function).

So the answer is: probabilistic reasoning cannot reach absolute certainty, but it can yield epistemic certainty robust enough for knowledge in practice. That’s why we entrust planes to aeronautical engineering or treatments to clinical trials. We are not misusing “know” when we say, “We know this drug works,” even though the claim is probabilistic. We are speaking in the epistemic register, not the absolute one.

The following is a point of clarity.

There are four uses of certainty that we need to be aware of, and they are the following:

1. Subjective certainty
This is the psychological sense: conviction, the felt impossibility of doubt. I may be subjectively certain that a friend will keep a promise, or that my team will win. But conviction alone is not knowledge, because one can be subjectively certain and still wrong.

2. Hinge certainty
These are the background beliefs that stand fast and frame our practices of doubt and justification. “The world has existed for a long time,” “I have two hands,” “Memory is generally reliable.” They are not justified in the ordinary way, nor are they really doubted in practice. They are arational bedrock, conditions of doubt and inquiry.

3. Epistemic certainty
This is when a belief is not only true and believed, but also justified and undefeated within the relevant practice. It is defeater-resistant knowledge. Epistemic certainty is what JTB+U aims to secure: not infallibility, but a belief whose justificatory structure is strong enough to withstand error-signals in its domain.

4. Absolute certainty
Here belong the logical and moral necessities: “2+2=4,” “A thing cannot both be and not be in the same respect,” or “murder is always wrong.” These are not contingent truths but necessary ones, either analytic or rooted in moral grammar. Doubt here would not be corrigible error but conceptual breakdown.

So:
Subjective certainty = conviction.

Hinge certainty = arational bedrock.

Epistemic certainty = knowledge in the JTB+U sense.

Absolute certainty = logical or moral necessity.

The problem with your account is that it tries to dismiss epistemology while simultaneously advancing an epistemological thesis. To say that “truth is a maintenance project of cognition” is itself a claim about the nature of truth and knowledge, exactly the kind of claim epistemology exists to evaluate. In other words, you rely on the very framework you’re trying to undermine. Pointing to mental illness as evidence doesn’t solve this contradiction either, because interpreting dementia or schizophrenia as “revealing” something about knowledge already presupposes a stable epistemic standpoint from which such judgments can be made. If all categories collapse with the mind, then the very distinction between “collapse” and “order” also collapses, leaving you with no grounds to assert your conclusion.

What your view really demonstrates is not the futility of epistemology, but its necessity. The fact that cognition can fail does not mean that truth is nothing more than a maintenance project; it means that human beings sometimes lose their grip on truth. To confuse the breakdown of knowledge with the nature of knowledge itself is like confusing a malfunctioning compass with the nonexistence of north. Far from undermining epistemology, your example underscores why we need it: to distinguish between distorted perception and genuine understanding, between the fragile scaffolding of a disordered mind and the enduring structures of reality.

Some probabilistic reasoning is absolutely certain. — Ludwig V

That claim is self-contradictory under the way I’m defining these terms. Absolute certainty means knowledge held with 100% confidence, without the possibility of error. This is the domain of sound deductive reasoning, where a valid argument with true premises guarantees its conclusion—what we call a proof in the strict deductive sense. The word absolutely here signifies “without restriction” or “without qualification.”

By contrast, probabilistic reasoning always carries qualification: no matter how small the probability of error, there remains some chance that the conclusion is false. Thus, probabilistic conclusions can approach certainty in degree, but they can never be absolutely certain in kind.

I was very pleased to see you include testimony. Because it enables us to pass on what we know It is critical to our practice of knowledge. We all stand on the shoulders of others and our society would be greatly impoverished if testimony were not an effective way of communicating it. However, accommodating it in the standard JTB framework is tricky. It requires acceptance of fallible justifications. — Ludwig V

Much of our knowledge comes through testimony (books, lectures, person-to-person, etc). Testimony is often undervalued and misunderstood. Every area of study, including science, must rely on testimony. What's lacking is the knowledge of how (it's a skill) to evaluate and appreciate its value.

Almost all justification is fallible, not just testimony. Why? Because most knowledge relies on probabilistic reasoning, including science.

Family Resemblances and the Grammar of Knowledge

If one lesson can be taken from Wittgenstein’s later philosophy, it is that the search for an essence often misleads us. “Knowledge,” like many of our central concepts, is not held together by a single core feature but by a web of similarities — what he called a family resemblance. We call different things “knowledge” because they overlap and crisscross, not because they share one identical trait. The advantage of approaching knowledge this way is that it steers us away from the temptation to define it too narrowly while still letting us see the grammar that governs its use. JTB+U takes up this insight: it identifies the recurring traits in our practice — truth, belief, justification, and understanding — and shows how they give shape to the family of cases we call knowledge.

This perspective dissolves a common confusion. Philosophers sometimes assume that to count as knowledge, every instance must satisfy one rigid formula. Wittgenstein would say instead: look at how we actually use the word. When we do, patterns emerge. In one case, we credit knowledge because a person’s claim is true and well-reasoned. In another, we retract it because their belief was false or based on a guess. In another, we hesitate because, though the person can repeat reasons, they plainly do not understand what they are saying. These are not disconnected episodes but overlapping uses of the word “know,” bound together by a family resemblance. JTB+U does not impose a new essence; it describes the grammar that already structures these practices.

To clarify, however, not every use of the word “know” belongs to the epistemic family. Our language-games carry many strands, and “know” is put to work in expressive, metaphorical, and practical ways that are distinct from epistemology proper. A few examples make this plain. “I know how you feel” may simply be an expression of sympathy — an assurance of solidarity rather than a truth-apt claim. “You should know better than to eat wild mushrooms” functions as rebuke, not as a report of propositional knowledge. “I love you more than you’ll ever know” is purely expressive, a way of magnifying devotion. “Frank doesn’t know his ass from a hole in the ground” is colloquial insult, equating knowledge with competence. In such cases, “know” is doing different work. It belongs to our forms of life, but not to the epistemic strand where JTB+U applies.

Yet some of these cases can shift depending on context. If I say “I know how you feel because I too have lost a parent,” the grammar changes. The claim is now anchored in truth (I really did undergo such an experience), belief (I am committed to that claim), justification (I can describe the circumstances of my grief), and understanding (I grasp what grief is like from the inside). Here “I know how you feel” becomes epistemic — a claim of shared experience that falls under JTB+U. The lesson is that grammar, not surface form, determines whether a use is epistemic. By attending to the language-game in which “know” appears, we can distinguish genuine claims to knowledge from expressive or rhetorical gestures.

This also helps us see the role of justification more clearly. Justification is not an abstract property that floats free of context. It is an activity situated in our language-games and tied to our forms of life. To be justified is to be able to give reasons that others can recognize as having weight within the practices we share. A person who believes for no reason, or on the basis of wishful thinking, cannot be said to know, even if they happen to be right. By contrast, a person who gives reasons that others in their form of life accept as strong, and who understands those reasons, does count as knowing — provided the belief is true.

The five justificatory routes identified in JTB+U illustrate this point. Testimony, logic (both inductive and deductive), sensory experience, linguistic training, and pure logic (boundary-setting only) are not reducible to one essence. They resemble one another the way games resemble one another: overlapping in methods and aims without being identical. What unites them is their role in linking belief to truth through public practices that can be tested, cross-checked, and corrected. A belief justified by testimony can be reinforced by sensory experience; one tested by logic can be clarified by linguistic training. Their strength comes from their interrelation, not from any single trait that all share. They form a family of justificatory methods — the most prominent hinges on which our claims to knowledge turn.

The image of hinges helps here. Our justificatory practices do not spin freely but move around fixed points that stand fast for us. In ordinary life, we do not doubt the existence of the external world or the reliability of memory at every turn; these are bedrock certainties that frame the game of justification. Within that frame, the five routes function as hinges of practice: they orient us in the search for truth, and when one route falters, another can expose the failure. Their capacity to cross-correct is what prevents justification from collapsing into mere convention. They are not exhaustive, but they are the dominant ways in which our forms of life give weight to reasons.

Seen from this angle, JTB+U is not an abstract model imposed from above but a refinement drawn from our actual use of “know.” It names the recurring features of the family resemblance that holds our knowledge-claims together, distinguishes the epistemic use of “know” from other strands in our language, and clarifies the routes by which justification has force. By making understanding explicit, it strengthens the link between belief and truth, ensuring that what counts as justification is not just appearance but uptake. In this way, JTB+U brings Wittgenstein’s therapeutic insight into constructive form: it dissolves confusions about “know” by looking at use, and it offers a framework that captures the grammar of knowledge as it is lived in our forms of life.

There's know that and there's know how. Sometimes it's know of. In some cases, it might be a combination?

I know how you feel.
You should know better than to eat wild mushrooms.
I didn't know which path I should take.
I want to know what it's like to jump from an airplane.
How do the migrating butterflies know the way to Mexico?
I love you more than you'll ever know.
Frank doesn't know his ass from a hole in the ground. — frank

If I say “I know how you feel because I too have lost a parent”, then the structure does fit JTB:

Truth: it is true that I have felt grief of that sort.

Belief: I believe that what I experienced is relevantly similar to what you are feeling.

Justification: I can give reasons (describing my past emotions, circumstances).

Understanding (U): I actually grasp what fear, grief, or joy feels like from the inside.

Here the claim is epistemic — a propositional one: “Your feeling is of type F, and I have also experienced type F.” That does fall within JTB+U, though it is grounded in testimonial and experiential justification rather than inference or measurement.

By contrast, when it’s used loosely (“I know how you feel” as mere reassurance), it slips into the convictional/expressive use Wittgenstein noted — no propositional claim is really being advanced.

So: this phrase actually shows how the same sentence can belong to different language-games depending on how it is meant. Sometimes it’s epistemic (anchored in truth, belief, justification, and understanding); other times it’s expressive (an act of sympathy).

“You should know better than to eat wild mushrooms.”
Here “know” means having the practical awareness or standing knowledge of a danger. It presupposes JTB in the background (“these mushrooms are poisonous”), but the utterance functions as rebuke, not as a truth-claim.

“I didn’t know which path I should take.”
Epistemic use: this is propositional knowledge — lack of knowledge about which option is correct. It fits the JTB structure directly (truth of which path is best, belief about it, justification for choosing).

“I want to know what it’s like to jump from an airplane.”
This is a case of experiential knowledge (“knowledge by acquaintance”). It’s outside JTB’s domain, which focuses on propositional knowledge.

“How do the migrating butterflies know the way to Mexico?”
Here “know” is metaphorical. Butterflies don’t form justified beliefs. It’s a shorthand for innate mechanisms or instinct. Not JTB.

“I love you more than you’ll ever know.”
Purely expressive. “Know” here means “imagine” or “comprehend.” It’s not epistemic at all.

“Frank doesn’t know his ass from a hole in the ground.”
Colloquial. It means “Frank lacks competence.” It’s not propositional knowledge but a judgment of practical incompetence.

Knowing that → propositional knowledge, the domain of JTB (and your JTB+U). It’s about truth-evaluable claims: knowing that the train leaves at noon, that water boils at 100°C, that a promise was broken.

Knowing how → practical or skill knowledge. It’s about abilities or competences: knowing how to ride a bike, how to play the violin, how to fix a sink. This is not easily reducible to propositional form. Gilbert Ryle made this point sharply in The Concept of Mind: “knowing how” is not just a set of facts one has memorized but an ability to act appropriately.

That said, the boundary isn’t always rigid:

When someone says “I know how to ride a bike,” what they’re really claiming is competence in a practice — which is different from holding a justified true belief.

But “knowing how” often presupposes some propositional knowledge (e.g., knowing that you must balance, pedal, and steer).

Conversely, “knowing that” is often reinforced by “knowing how” (a surgeon who knows that a procedure requires precision must also know how to carry it out).

In my framework:

JTB+U is designed to cover knowing that — propositional knowledge.

Knowing how is better understood as a different language-game of “know”: one rooted in practice and skill rather than truth-apt propositions.

The two overlap in that “knowing how” can feed into propositional knowledge (e.g., an expert mechanic has knowledge-that cars behave in certain ways, because he knows-how to fix them).

JTB is fits into many different language-games, and the definition is based on Wittgenstein's family resemblance idea.

It may be objected that nothing truly novel is added by the ‘+U.’ Some have held that unless a person grasps the concepts at work, their justification is not genuine to begin with. In that sense, understanding has always been latent in the J-condition. My argument is not that I am introducing an alien element, but that by making U explicit we secure what has too often been assumed in silence. Once drawn into the open, the requirement of understanding keeps us from mistaking parroting for uptake, appearance for reality, or borrowed reasons for genuine justification. If the condition seems obvious, that is precisely the point: it has always been there, but left unnamed, it has allowed too much misdescription to slip through.

Instead of trying to provide a definition for knowledge, think about how the word is used. The next time you catch yourself using the word, stop and reflect on what you're trying to convey. — frank

If this is what you think I'm doing, then you haven't understood anything I've said. My impression @frank is that you haven't read my posts in the last few pages, because if you had, you would know how foolish your remark is.

I don't think your conception of knowledge is going to stand up to a skeptical challenge. At any time, we may be mistaken about our justifications. So, to nail the jello to the wall, how do you determine if the evidence in front of you is sufficient for knowledge? I say you have no way to do that. You only use the word "knowledge" to signify confidence in your beliefs. — frank

If knowledge is just confidence in one's belief, then one's confidence/conviction that one knows would suffice, that can't be correct. Sure, we can be mistaken in what we think justifies our belief, but as I pointed out earlier that doesn't justify being skeptical about JTB. Your skepticism is unfounded, you seem to think that if we can't know in an absolute sense, then we can't have knowledge (at least in the sense that I'm proposing). We don't use the concept knowledge,generally, in the way you seem to think. Even in science our conclusions are mostly probable.

In the last three pages I explained this in detail.

It wouldn't make sense to say it's an opinion, because the mafia threat isn't a matter of opinion. It wouldn't make sense to say the character knows it, but it's not true. So we could add on truth. — frank

It makes sense to say the man thinks he knows, but he doesn’t. This is something we see all the time: people confuse what they believe with what they actually know. The key difference is that conviction alone isn’t knowledge, and sometimes the evidence that seems to support a belief doesn’t really justify it.

For JTB+U to be viable as a framework, it has to work more often than not. If the framework only rarely connected belief with truth, it would collapse into skepticism or relativism. But the very fact that our practices of knowing guide us successfully in daily life, science, history, and moral reasoning shows that it does work.

We navigate the world, build functioning technology, diagnose illnesses, and correct one another’s errors. These are not flukes. They demonstrate that justification, when tethered to practice and subjected to defeater screening, is in fact truth-conducive. Understanding ensures that justification is not merely parroted but conceptually grasped. This combination makes it more likely than not that our beliefs line up with reality.

Of course, mistakes and regressions happen, both individually and culturally. But if error dominated, the very concepts of knowledge and justification would lose their grip. The fact that we can identify mistakes as mistakes already shows the framework is working. Our ability to mark regressions as regressions depends on having a more stable body of truths against which those failures stand out.

So, JTB+U is viable because it works in practice most of the time. Its fallibility is not a weakness but a strength: it builds in the possibility of correction without demanding infallibility. What matters is that the framework links belief, justification, and understanding to truth with enough reliability to sustain our practices of inquiry, correction, and progress.

Error is the thing that makes knowledge possible. This is counterintuitive.

But you have already allowed that cultural-historical regressions might lead to a case where a culture widely accepts that a true idea/theory has been "defeated" when it hasn't been. How do you know that you're not in that situation?

The move of: "scholasticism lost ground because it was properly defeated, but if secular naturalism and exclusive humanism lose acceptance that will be simply a regression," seems arbitrary unless you can show why some beliefs are actually true and cannot be the result of regression/error. That is, an apparent defeater or error is not solid evidence that a theory/idea is actually wrong, since you have already allowed that whole cultures can misidentify defeaters and errors for long periods. — Count Timothy von Icarus

That is the way some pushback on this idea, but I think the apparent arbitrariness disappears once we distinguish between (a) the framework for knowledge and (b) our fallible application of it. There is always a difference between what I think counts as knowledge and what really is knowledge. Gettier cases make this clear: someone thinks they are justified, but the justification is defective, so they do not actually know. The same thing can happen at the cultural level—whole societies can misidentify defeaters or mistake regression for progress. That does not show that knowledge is impossible; it shows that our application of JTB+U can go wrong. The framework itself—truth, belief, justification, and understanding—remains intact. What fails is our judgment, our recognition of what counts as a genuine defeater.

So how do we avoid arbitrariness? By keeping the distinction clear between apparent defeaters and genuine defeaters. An apparent defeater is what a culture or individual takes to undermine a claim at a given moment. A genuine defeater is one that endures across error-signals, survives cross-checking through multiple justificatory routes, and holds up under defeater screening in the long run. Misfires don’t erase the difference between appearance and reality; they only show that human uptake often lags behind truth.

This is why I say Scholasticism lost ground for reasons that were not merely political or sociological. Its justificatory practices proved less able to handle new error-signals. If secular naturalism were later abandoned, the question would not be settled by its decline in popularity or survival-value but by whether the reasons against it genuinely exposed its falsity. To decide that requires the same thing JTB+U always requires: truth, belief, justification, and understanding tested against defeaters.

The risk of cultural error is real, but that does not make all judgments arbitrary. It means we must hold our claims with epistemic humility, open to correction, but not collapsing into relativism. Some beliefs are true and will survive defeater screening indefinitely (mathematical truths, some moral absolutes). Others are corrigible. What matters is whether our justificatory practices are strong enough to keep those differences visible.

Right, but if you cannot be sure that you have true beliefs now why should you trust your own beliefs about the long term trend of knowledge or epistemology more generally? — Count Timothy von Icarus

That question rests on a misunderstanding of what knowledge is. It assumes that unless I can be sure of my beliefs in the sense of absolute, indubitable certainty, then I have no rational grounds for trusting them. But knowledge has never required that kind of infallibility. On JTB+U, knowledge requires truth, belief, justification, and understanding. Those conditions are demanding, but they do not amount to immunity from error.

Part of the confusion here comes from how we use the word certainty. It can mean at least three different things. First, there is subjective certainty, the psychological feeling that I cannot be wrong. That is fallible, since I can feel certain and yet be mistaken. Second, there is objective certainty, the hinge-level background we do not doubt—like that the world has existed for a long time. These certainties are not proved but stand fast as the conditions of doubt and knowledge alike. Third, there is epistemic certainty, where a claim is true and justified to the point that doubt has no footing within a given practice. Even this form of certainty remains corrigible if new defeaters arise.

The challenge you raise trades on conflating these senses. It assumes that unless I can have absolute, infallible certainty, I cannot call my beliefs knowledge or trust them for long-term reflection. But this is simply not what knowing is. Knowledge, on the JTB+U account, is defeasible but real: I can hold a belief that is true, justified in my practice, and conceptually understood, even though I remain open to correction.

So why trust my beliefs now, including beliefs about epistemology? Because they have withstood defeater screening across the routes of justification available to me: testimony, reasoning, sensory experience, linguistic clarity, and logical consistency. If new defeaters arise, I will adjust. But until then, the best explanation for their stability is that they are tethered to truth.

The upshot is that skepticism here demands the wrong thing. It asks for absolute certainty, when what knowledge actually requires is justified, true, and understood belief within our forms of life. To see that distinction is to dissolve the challenge.

Isn't understanding the same thing as justification? I'm not sure what the U adds to JTB, given that we assess understanding in terms of justifications.

As for deciding whether a refutation is valid or not, this rests upon the truth of one's auxiliary hypotheses. So unless those can also be tested, one cannot know whether the refutation is valid, which is the staple criticism of Popper's falsificationism - that individual hypotheses are impossible to test, since their validity stands and falls with the truth of every other hypothesis. So the bridge from practical refutation in everyday life, which often involves the testing of individual hypotheses under the assumption of true auxilliary hypotheses, doesn't withstand skeptical scrutinty and the standards demanded by scientific epistemology - an essentially unattainable standard, relegating JTB to the realm of the impossible, or to the realm of semantics that is epistemically vacuous. — sime

On the first point: understanding is not the same as justification. Justification is the giving of reasons that satisfy the standards of a language-game. Understanding is a matter of concept-mastery, the ability to use terms correctly within that grammar. A student can repeat reasons in a way that looks justified, but without grasping the concepts, they do not understand—and so they don’t know. The “+U” is needed because justification can sometimes be mimicked or borrowed without genuine uptake.

On the second point: you are right that Popper’s falsificationism raised the problem of auxiliary hypotheses, and that one cannot isolate hypotheses absolutely. But this does not mean that knowledge is impossible. It means that justification is corrigible and practice-bound. When we test a claim, we do so against a backdrop of hinges and auxiliary commitments. If anomalies arise, we do not simply abandon the claim, but we check whether the defeater is genuine. This is why defeater-screening is part of the JTB+U framework.

The skeptical worry that this makes JTB impossible arises from a mistaken demand: that justification must be final and immune to revision. Wittgenstein helps us dissolve that demand. Knowledge does not require absolute independence from auxiliary assumptions; it requires that our reasons hold up against available defeaters in the practices that give those reasons sense. Here the distinction is crucial: the JTB framework itself—truth, belief, justification (+U)—remains the definition of what knowledge is. But our application of that framework in practice is fallible. We can misidentify truths, we can mistake apparent reasons for genuine justification, and we can claim understanding where only parroting exists. The framework is sound; what fails is our use of it.

So the linkage is this: justification is public and corrigible, understanding is the uptake that prevents mere parroting, and truth remains the non-negotiable condition. Together, they let us say that knowledge is possible without pretending that justification must ever be perfect or final. Our judgments may fall short, but that is an error of application, not a defect in the framework itself.

Well, if you acknowledge that regression can occur, then it seems that defeaters can appear to pile up against a position, and yet this is itself a sort of illusion or product of pathological justification. So my question then is: "how do you know that what you think are defeaters and are progressive evolution really are?" — Count Timothy von Icarus

My system fixes the regression problem. I said that "there are periods of regression," just as there are periods where one thinks that a particular conclusion is knowledge, but later it's found to have flaws/defeaters.

Your question "how do you know that what you think are defeaters and are progressive evolution really are?" is the right question to ask, because it highlights the difference between thinking one has a defeater and actually having one. JTB+U is built precisely to keep that distinction clear.

A defeater is not just whatever I take to undermine a claim. It has to stand up within the interlocking routes of justification and not collapse under cross-examination. Suppose I call something a defeater—say, new evidence, or a conceptual contradiction. The question is then whether that supposed defeater is itself truth-conducive, practice-safe, and able to endure scrutiny across language-games. If it turns out to be mistaken, then it was not a genuine defeater but an apparent one.

The same applies to what looks like “progressive evolution.” The fact that practices change does not mean they are evolving toward truth. What secures progress is not the mere shift but whether the change holds up under defeater screening, survives error-signals, and continues to cohere with the deeper grammar of our concepts. Many shifts are later revealed as regressions—false starts that fail the test of time. Others, like the recognition of human rights, survive repeated challenges and show themselves to be knowledge rather than temporary consensus.

So my answer is: I do not “know” in advance that this or that shift is progress. What I can do is apply the JTB+U framework as best as I can—testing for truth, justification, belief, and understanding, and watching to see whether putative defeaters actually endure. Progress is visible in retrospect when claims prove stable across scrutiny and when rival justifications collapse. That is what makes them knowledge rather than mere opinion.

I think that's a good answer. A difficulty in epistemology that I think is often under addressed is that the idea that knowledge "progresses" (e.g., "scientific progress," "moral progress," etc.) needs to be justified itself. The original Enlightenment justification for this was theological, and so if it is adopted in a secular naturalist context it needs another sort of justification. However, historically, it does not seem that technological, scientific, moral, or philosophical progress is assured. They don't seem to have always occurred; periods of regression show up as well. The early secular narratives that put forth the idea that only "superstition" blocks the path to progress seem too simplistic to account for this (they seem like downright ideological propaganda a century on TBH).

However, there are still plenty of issues. What constitutes an "error" or "failure" is itself dependent on goals and understandings that are always shifting. Consider the contemporary traditionalist critique that crashing fertility rates are a sign that exclusive humanism is maladaptive. Well, if it really does go extinct because its population falls by more than half each generation, and the dominant paradigm a century from now is something more in line with the traditionalist ethos, shall it thus be true that secular exclusive humanism was discovered to be a sort of error? From the exclusive humanist perspective, this would not prove it was in error; its own extinction would not be evidence of the truth of religion, tradition, etc., only of those traditions' reproductive value (indeed, anti-natalists would probably argue that religion is reproductively successful precisely because it is false). But this shows that "success" might arguably correlate with falsehood not truth.

I think your response works best in terms of technological questions where "success" is fairly easy to define. Either a plane crashes or it doesn't. It becomes much more difficult in political, social, moral, philosophical, etc. questions. For example, arguably the main liability to Scholastic thought in the early modern period was not its apparent falsity (or its dogmatically asserted "dogmatism") but that it absolutely did not lend itself to pamphlets aimed at a common audience, the new dominant market for philosophy, and that its institutions became prime targets for political violence and expropriation.

So, there is the issue that past "successes" and "errors" are being defined in terms of current practice. There is a bit of a "history is written by the victors" problem. There is also the problem posed by Hoffmann's "Fitness Beats Truth Theorem" and similar arguments, where selection-based approaches to belief do not ensure that beliefs are true. Fitness does not seem to be equivalent with truth in how information (or beliefs/memes) replicate. But then if our justification for our own beliefs rests on a selection model, this ends up being self-refuting. If our selection theory is true, we ought not believe our own beliefs are true, because their fitness is only loosely related to their truth.

Nonetheless, I think selection can be an important factor in explaining progress, just not the only factor. The other issue here is that it would only suggest that knowledge will be produced in the long run, not that we have it now. But if we aren't likely to have true beliefs now, then we ought not believe in our own progress narratives (a similar sort of issue). Hence, I think a stronger linkage is needed. — Count Timothy von Icarus

I agree that narratives of inevitable progress, whether Enlightenment or later, tend to oversimplify. Knowledge does not advance in a straight line. There are periods of regression, distortion, and even collapse of practices. What JTB+U gives us is not a guarantee of progress but a way of making sense of how correction is possible when progress does occur.

The key is to separate two things: (1) the sociological fact of which beliefs or practices dominate, and (2) the epistemic status of claims. The fact that a tradition survives or reproduces more effectively does not by itself prove truth. Fitness can select for falsehood as easily as for truth. The “Fitness Beats Truth Theorem” makes this plain. But epistemic justification is not reducible to fitness. A belief counts as knowledge only if it is true, believed, justified, and understood. That standard is higher than survival-value alone, which is why extinction or dominance cannot by themselves settle the matter.

Still, practices evolve. And they evolve in part because defeaters accumulate: rival routes of justification expose contradictions, or experience undercuts established norms. Scholasticism did not vanish simply because it lacked pamphleteers; it lost its grip on the justificatory practices of the time. Competing frameworks proved more effective at handling error-signals and sustaining inquiry in new conditions. That does not mean the older framework was false because it disappeared, but that its justificatory practices could not carry forward.

So I would put it this way: JTB+U does not assure progress, but it explains what progress consists in when it happens. A belief that is true, justified within its language-game, and understood conceptually can survive the test of defeaters across time. Sometimes it will be suppressed, sometimes ignored, sometimes distorted. But when it resurfaces and proves practice-safe under renewed scrutiny, that is a mark of epistemic progress.

This also clarifies why “success” in the practical sense (planes flying, vaccines working) and “success” in the moral or philosophical sense look so different. In technology, error-signals are immediate and decisive. In moral and political life, error-signals can be deferred, resisted, or disguised. That makes progress slower and less assured, but not impossible. Over the long run, justification and understanding are forced to adjust as contradictions mount. The abolition of slavery or the recognition of human rights are not simply products of fitness; they are instances where justification and truth finally aligned, and practices evolved to acknowledge it.

In that sense, knowledge does not march inevitably forward. But when it advances, it is not because victors wrote the history, but because some beliefs endured defeater screening while others failed. That is the “stronger linkage”: progress is not guaranteed, but possible because truth exerts pressure over time through the interlocking routes of justification.

By the way, I've finished my first book, and I'm starting my second book called "Why Christianity Fails
(Weak Testimony, Fragile Evidence, and the Collapse of a Belief System). This book will probably take a while to finish, and it will be much longer (300-350 pages). I go after the core belief in Christianity, the resurrection, and demonstrate the weakness of the testimonial and historical evidence.

How might this apply to moral knowledge? If one discovers that the dominant norms of one's society are, in fact, evil, how does one end up demonstrating this understanding? Pretty much by definition, one's community will think you are in error. But it does seem possible to be right about what is just, or choiceworthy, when everyone around you is wrong, and deems you to be in error, and prehaps "misusing language." For example, when Saint Gregory of Nyssa first began making a concrete Christian justification for the total abolition of slavery, this was a pretty wild claim. When he said "slavery is unjust," arguably he could be accused of misusing the term "just" in his context. And yet we tend to think he was absolutely correct here, and that his society would later come to agree with him and largely abolish slavery because he was correct. — Count Timothy von Icarus

On my account, moral knowledge is possible in the same way any knowledge is possible: it requires truth, belief, justification, and understanding, situated within the language-games and forms of life that give “justification” its meaning. The claim that “slavery is unjust” is not just convictional; it is epistemic when it can be shown to cohere with the deeper grammar of moral language—justice as fairness, dignity, and reciprocity—even if the society around you has not yet taken up that use.

Objectivity without Absolutism
Objectivity in morality means that claims can be tested by public reasons, not just private conviction. Even when one’s contemporaries reject those reasons, the reasons can endure defeater screening over time. Gregory’s claim survived challenges and proved practice-safe; rival justificatory schemes (naturalizing slavery, theological rationales for domination) eventually collapsed. The enduring stability of the abolitionist claim shows that it was not merely cultural preference but a true moral proposition grasped with understanding.

The Role of +U
This is where the “+U” condition is critical. Gregory did not simply parrot “slavery is unjust”; he understood the concept of justice in a way that exposed contradictions in existing norms. To demonstrate such understanding is to handle the concept rightly, even in the face of communal resistance. Moral knowledge is thus not a matter of majority assent but of correctly grasping and applying the concepts in a way that survives both internal scrutiny and the defeater tests of history.

What about shifting contexts? As a Marxist, I might be able to justify and demonstrate understanding of the labor theory of value to other Marxists. I might also believe the theory is true. However, we have pretty good reason to think the labor theory of value is false. Can I know something that is false?

More problematically, suppose I have become versed in both Protestant and Catholic/Orthodox language games. Can I both understand and know that the Eucharist is the real body and blood of Christ, and that it is not the real body and blood of Christ because I can justify both and demonstrate a competent understanding of both?

Obviously, we might object that I cannot actually believe both (barring some sort of power of self-hypnosis perhaps), so I fail the B criteria on at least one of these. However, it seems possible that I could act like I believe both. The B criteria here seems ineluctably private, and so not "observable."

The problem I see with grounding J in current practice is that many forms of J do not seem to secure, or even lead towards truth. Some seem to positively block access to truth. So, referring back to current practice and use doesn't secure T. This would mean that knowledge exists just in case current practice and use corresponds to what is true (I think it's fair to say that no one except for the relativist vis-á-vis truth thinks this is always the case). But then there still needs to be some linkage between justification, use, and practice (@J's issue if I understand it right). Just because current practice requires that I cut out a victim's heart to keep the sun from going out won't make my justified belief, through which I demonstrate mastery of the relevant language game, true; it must also be true that this practice actually keeps the sun from going out.

But then J and U must have something to do with truth, or else they seem irrelevant, and likewise if B and U can be arbitrarily related to T, they will only ever accidentally line up with it. Presumably, J links them. But sometimes J requires that we contravene established practices that demonstrate U as well. We might decide that we have to start speaking about DNA or justice differently, before we have convinced anyone else.

I think this relates to another question. Practices and language clearly evolve over time. What causes them to change the way they do? Presumably, this is how J might relate to T and U. — Count Timothy von Icarus

1. Can I know something that is false?

No. Knowledge is factive. If a belief turns out false (like the labor theory of value as an account of price and markets), then however justified and understood it may have seemed at the time, it was not knowledge but a case of apparent knowledge. JTB+U preserves this: the T condition (truth) is non-negotiable. What you had was a justified, understood belief that later collapsed under defeaters. That does not mean JTB+U failed—it means knowledge claims are always defeasible.

2. Can I know contradictory things?

The Eucharist example is instructive. You might be able to understand both the Catholic/Orthodox and Protestant language-games, and you might be able to justify each within its practice. But you cannot believe both simultaneously in the epistemic sense. At most, you can role-play or act as though you believe both. Since JTB+U requires belief, one of these would fail the B condition. If you suspend belief and simply track the grammar of each tradition, that is competence, not knowledge. Knowledge needs commitment to one truth-claim, not simultaneous acceptance of contradictories.

3. Isn’t justification sometimes corrupt or truth-blocking?

Yes—and this is why in JTB+U justification is not free-floating but checked by practice-safety and defeater screening. A practice like Aztec human sacrifice may have had an internal logic, but the claim “cutting out hearts keeps the sun alive” cannot survive defeater screening. It conflicts with what we now know through other interlocking routes (astronomy, physics, biology). So J is not whatever counts as justification in the moment, but justification that can hold up under the pressure of cross-checks and error-signals. That’s the linkage between J, U, and T: justification is only adequate if it is safety-preserving and defeater-resistant relative to truth.

4. How do practices evolve?

Practices evolve because defeaters accumulate, because rival routes converge on better explanations, and because conceptual understanding exposes contradictions. This is why J and U are not sealed off from T: language-games are porous. A community can be wrong for a time, but over the long run, practices shift under the weight of correction. That is why Gregory’s abolitionist claim, once dismissed as misuse, later became the new grammar: it better aligned justification and understanding with what is true.

5. The big picture

So the framework looks like this:

Truth (T): non-negotiable; one cannot know falsehoods.

Belief (B): requires genuine commitment, not role-playing.

Justification (J): tied to practices but must be defeater-safe and truth-conducive.

Understanding (U): demonstrated by conceptual uptake, not parroting.

Together, JTB+U explains why false theories don’t count as knowledge, why contradictory beliefs can’t both be known, why corrupt practices don’t ground knowledge, and how evolving practices eventually bend toward truth.

On this point, Wittgenstein’s contribution is not to propose another model of knowledge beside JTB, but to dissolve the demand for an ultimate account of justification outside our forms of life. The factivity of know remains untouched, as does its commitment to belief. What changes is our view of justification: no longer a timeless condition, it is an activity rooted in our shared background. When Wittgenstein says that “knowledge is in the end based on acknowledgment” (OC §378), he is not abandoning JTB but pointing to the human practices in which justification has its weight.
— Sam26

This account is right in line with the shifting meaning of "justification." But it seems to me to leave open the same question. — Count Timothy von Icarus

You’re right that the question remains—what secures justification if it shifts with use? My point is not that justification becomes arbitrary, but that its weight lies in the river-bed of practices that give it sense. That river-bed is not static, but neither is it untethered.

What Wittgenstein helps us see is that justification is always bound to what stands fast for us at a given time: the certainties we do not doubt, the error-signals we attend to, the ways we check one another’s claims. That is why he can say knowledge rests on acknowledgment. To acknowledge is not simply to nod assent; it is to recognize a claim as fitting the grammar of our form of life.

So yes, the question presses: what if those practices are distorted, or what if our forms of life themselves evolve? Here is where JTB+U offers a refinement. Justification is not only public uptake, but uptake tethered to understanding. One can parrot reasons that “fit” in the moment, but without grasping their grammar, one does not know. Understanding functions as the hinge between practice and truth: it is what allows us to detect when a justification, though accepted, is hollow, or when a claim, though rejected, is nonetheless aligned with the deeper use of our concepts.

The open question then is not whether justification needs grounding beyond practice, but how practices themselves can be judged as truth-conducive. My answer is that they are judged over time by defeater screening, by convergence across routes of justification, and by whether they prove practice-safe when tested against the world. Practices shift because error-signals accumulate. In that shifting, knowledge does not lose its footing; it shows that our grasp of justification is corrigible, but never free-floating.

When Wittgenstein uses “know” epistemically, his examples remain squarely within the orbit of what the tradition would call justified true belief. He does not propose a radically new category of knowledge but reminds us that the sense of “know” always comes to light in use, and its grammar reveals the conditions of truth, belief, and justification. If I say, “I know he was on the train,” my claim is factive (it presupposes truth), I am committed to it as a belief, and I expect to be able to give reasons—tickets, eyewitnesses, timetables—if challenged. In this respect Wittgenstein does not undo the JTB model, he re-situates it: he wants us to notice the practices that give “justification” its shape in the first place.

What he strips away is the illusion of a context-independent essence. The temptation of classical epistemology is to imagine justification as a single property that attaches to beliefs in the same way across all cases. Wittgenstein shows that what counts as justification shifts with the language-game. In a courtroom, “I know” calls for evidence that can be entered into the record; in daily conversation, it may be enough that I saw it with my own eyes; in science, the standards are replication and peer review. Each of these is still JTB, but the J is made concrete only in its practice. This is why he counsels “look, don’t think”: instead of theorizing what justification must be, examine how it functions in the life of language.

On this point, Wittgenstein’s contribution is not to propose another model of knowledge beside JTB, but to dissolve the demand for an ultimate account of justification outside our forms of life. The factivity of know remains untouched, as does its commitment to belief. What changes is our view of justification: no longer a timeless condition, it is an activity rooted in our shared background. When Wittgenstein says that “knowledge is in the end based on acknowledgment” (OC §378), he is not abandoning JTB but pointing to the human practices in which justification has its weight.

Thus the upshot is clear: there is no distinct epistemic use of “know” in Wittgenstein’s later philosophy that lies beyond JTB. What he uncovers is that every epistemic use is JTB-like, yet always practice-bound. The hinge or convictional uses fall outside epistemology altogether, but the genuinely epistemic ones do not float free of the traditional model. They fall comfortably within JTB once we recognize that justification lives inside language-games, and your +U condition makes that insight explicit. To know is still to believe what is true with justification, but it is also to understand the grammar of the concepts in play, and that is what JTB+U secures.

The craving for generality: Wittgenstein warned against the philosopher’s temptation to seek a single formula that captures the essence of every case. He called this the “craving for generality”: the urge to believe that a concept must rest on one hidden feature common to all its uses. But not all generality is illusion. There is a legitimate kind, which describes the recurring patterns and family resemblances that hold our practices together without reducing them to one rigid essence. The tripartite model of truth, belief, and justification belongs to this latter kind. It does not disclose the secret essence of knowledge but highlights a recurring structure that marks off knowledge from mere conviction. Adding +U strengthens that pattern, keeping it tied to competent application in practice and guarding against the mistake of treating a guiding framework as if it were an eternal definition.

Mastery of a technique: Wittgenstein often compared understanding to mastering a skill or technique. To know how to play chess is not merely to recite the rules but to make the right moves in practice. Likewise, to know a mathematical proof is not just to memorize it but to apply it under variation, extend it, and recognize mistakes. Knowledge claims, on this view, require more than stating the right words: they require competent use within a practice. This aligns perfectly with the +U in JTB+U—understanding shown in skillful use of concepts.

Bridge: From JTB+U to Hinges

Even with justification clarified and understanding secured, the model of JTB+U does not by itself remove the regress problem. Reasons can always be asked for in turn, and if the structure were required to justify itself endlessly, no claim could ever rise to knowledge. At some point, justification reaches a stopping place—not because of a failure in reasoning, but because practices rest on certainties that are not themselves up for doubt or proof. This is where Wittgenstein’s remarks on hinges enter.

Hinge certainties function as the river-bed of thought: “I have two hands,” “the earth has existed for a long time,” or more primitively, the unreflective assurance that the ground will hold us when we walk. Such claims do not stand as ordinary propositions requiring evidence; they are the background that makes doubt and evidence possible in the first place. To doubt them globally would be to lose our footing in the very language-games that give meaning to knowledge claims.

In my own work I have drawn a parallel between these hinges and Gödel’s incompleteness theorems,
just as Gödel showed that no consistent formal system strong enough for arithmetic can prove all the truths it contains or even establish its own consistency from within, Wittgenstein shows that epistemic systems rest on unprovable certainties. Both reveal a structural limit on internal justification. Far from undermining knowledge, these limits are enabling conditions: mathematics requires axioms it cannot justify, and our epistemic practices require hinges that stand fast without proof.

Placing JTB+U against this background allows us to see its proper scope. Justification, truth, belief, and understanding all operate within practices bounded by hinges. Knowledge becomes possible not because every step can be proven, but because certain things are exempt from doubt—bedrock elements of our form of life. To see this is to shift from asking for an impossible universal ground to recognizing the lived certainties that make reasoning, language, and knowledge possible at all.

Continuing the Explanation of JTB+U

To extend the picture, I need to clarify what the “+U” adds, how justification itself must be situated in language-games, how the five routes interlock, and how this framework addresses familiar rivals. Each step reinforces the claim that knowledge is not only a matter of true belief with reasons, but of uptake within the forms of life that make reasons intelligible.

1. The role of +U
The “understanding” condition guards against the illusion that one can count as knowing simply by repeating words linked to justification. A student may recite that “HbA1c above 6.5 signals diabetes,” but unless they understand what HbA1c measures and why that threshold matters, their belief does not rise to knowledge. The difference lies in the correct public use of concepts. Understanding is shown in practice, not in private conviction. This is why Wittgenstein reminds us to “look, don’t think”: look to the use of terms in their home setting, not to an imagined essence behind them.

2. Justification in language-games
What counts as justification is not determined once and for all, but by the grammar of the practice in which the claim is made. In law, justification rests on admissible evidence and procedural safeguards; in science, on reproducibility, statistical thresholds, and peer scrutiny; in everyday life, on testimony that lacks defeaters. Each form of life sets its own standards. To ask for justification without reference to a language-game is to float free of the river-bed. JTB+U insists that we situate reasons where they belong, not in abstraction but in practice.

3. The five routes working together
My method recognizes five primary routes of justification: testimony, logic, sensory experience, linguistic training, and pure logic. Each has its own integrity, but they rarely operate in isolation. Testimony is checked by sensory anchors; logic binds testimony and perception into argument; linguistic training disciplines our terms, keeping them clear of confusion; pure logic secures consistency at the boundaries. When these routes converge, belief gains strength. When they diverge, we look for defeaters. This interlocking structure keeps justification tied to the lived world.

4. Answering rivals
Seen in this light, the JTB+U framework offers replies to classic objections. Gettier’s puzzles lose force because a belief that is true by accident but not conceptually taken up within the relevant practice fails the +U condition. Relativism is resisted because language-games are not arbitrary; they are bound by hinges—what stands fast in our shared form of life. Dogmatism is resisted as well, because defeater screening and practice-safety remind us that no claim is beyond correction.

The result is not a rejection of JTB but a strengthening of it. By insisting on public uptake, contextual justification, and interlocking routes, JTB+U places knowledge back where it belongs: within the activities and forms of life that give our words and reasons their grip.

I'm not convinced that Wittgenstein accepted JTB, in the way Sam26 seems to think. I read him in On Certainty more as pointing out that if we do accept JTB then these are the consequences - there are for instance things that we might casually say we know that are rules out as knowledge by the JTB account. We can't know how a dog that has been run over feels. — Banno

My account of JTB+U is not the same as traditional JTB, it's more refined using Wittgenstein's methods from his later thinking. I'll partly sum it up in the following:

Extending JTB Through Wittgensteinian Methods

The classical model of knowledge—justified true belief (JTB)—says that a person knows p when three conditions are met: the belief is true, the subject believes it, and it is justified. That triad has stood since Plato, but its weak point has always been justification: what exactly counts as good reasons, and how are they grounded?

My proposal is to extend JTB with a Wittgensteinian emphasis. I call this JTB+U, where the “+U” stands for understanding. Genuine knowledge requires not only justification in some abstract sense but also competence with the relevant concepts. That competence shows itself in the correct public use of words, within the language-games and forms of life that give them their meaning. In other words, to “understand” is to be able to navigate the practices that make justification possible in the first place.

This move does two things at once:

Re-locates justification: It is not a freestanding relation between belief and evidence, but an activity carried out within specific language-games. What counts as “good reasons” is inseparable from the public standards and forms of life that sustain them.

Builds in uptake: To count as knowledge, a claim must not only be justified but also be grasped conceptually by the knower. A person may parrot a valid argument form, but without knowing the terms in use—say, “HbA1c” in a medical report—they cannot be said to know.

Traditional epistemology often treats justification as context-free, but justification lives and breathes inside particular practices. Courts, laboratories, and ordinary conversations each operate with their own standards, which are intelligible only against their background forms of life. Language-games set the grammar for justification, and hinge propositions—the arational certainties that stand fast—mark its limits.

I part ways with Wittgenstein on one important point: he treated language as bounded by insurmountable limits. I disagree. Language-games and forms of life are more open-textured than he allowed, and they evolve in ways that allow language to surpass those boundaries. Our epistemology should therefore preserve Wittgenstein’s insight into the contextual life of justification without accepting his strict metaphysical ceilings.

The upshot is a framework that is at once traditional and supple. JTB remains the core, but it is further strengthened by grounding it in the Wittgensteinian recognition that meaning, justification, and understanding are functions of practice. Knowledge claims are tested along multiple routes—testimony, logic, sensory experience, linguistic training, and pure logic—but always within the river-bed of language-games that give these routes their force.

Look, don't think:

Wittgenstein’s reminder “look, don’t think” is not an anti-intellectual gesture, but a methodological one: when philosophy drifts into abstraction, we should return to the actual use of words, to the practices in which meaning and justification live. It is a call to examine the grammar of our language-games before theorizing about their essence.

In terms of JTB+U, the maxim sharpens both justification and understanding:

For justification (J):
Traditional JTB risks treating justification as a timeless relation between belief and evidence. “Look, don’t think” tells us to attend instead to the public criteria in use: how testimony is corroborated, how sensory anchors are cross-checked, how logical inference operates in practice. Rather than imagining some metaphysical essence of justification, we look to the rules and error-signals that actually govern our language-games.

For understanding (+U):
Understanding is shown not in private conviction but in our capacity to use concepts correctly. To know is not merely to assent to a true and justified proposition, but to handle the relevant terms in the way our form of life requires. Here Wittgenstein’s maxim is the corrective: don’t speculate inwardly about what a concept really is; look outward at how the concept functions in actual practice. If someone says they “know what DNA is” but cannot use the term competently in the language-game of biology, they fail the +U condition.

In this way, “look, don’t think” guards JTB+U against two perennial errors:

Essentialism (believing knowledge must rest on some hidden inner property);

Private conviction masquerading as knowledge (thinking “I feel sure” is equivalent to being justified).

Instead, the maxim keeps our eyes on use, on the lived background where justification gets its grip and understanding is displayed.

So, in short: “Look, don’t think” is the Wittgensteinian brake on abstract theorizing that keeps JTB+U tethered to practice. It ensures that both justification and understanding are grounded in observable criteria within language-games.

And when we look, we do find uses of "knowledge" that do not quite fit the JTB account. — Banno

There are uses of know that don't fit the JTB account, but I would maintain they're not epistemic uses. Are you saying that there are epistemic uses that don't fit the JTB model? If that's what you're claiming, and I'm not sure that you're claiming that, give an example. There is a use of know, for example, that's just an exclamation of a conviction. In other words, it's just an expression without epistemic force behind it.

The spectre of essentialism hangs over such expectations. — Banno

There is no one essence that fits JTB, there are just a series of language games that demonstrate how we justify beliefs in various ways. If we "look" as Witt would say, we can see these uses in practice via forms of life. I've added the +U because I think it captures the essence of this. The “+U” is what makes the difference plain. Traditional JTB can sound as if all you need is a true belief and some reason that looks like justification. But without understanding, justification can collapse into parroting words or leaning on a method you don’t really grasp. +U requires that the knower demonstrate competence in the concepts at play. That means being able to apply the idea correctly in practice, to handle nearby counterexamples without confusion, and to explain the checks that make the claim public.

And yet today it is common to hear people speak as though knowledge is simply whatever they take knowledge to be. To put it more sharply: many equate knowledge with conviction. If I am sure of something, then in my own eyes I “know” it. But this collapses the difference between subjective certainty and objective certainty or knowledge. Conviction may be powerful, even immovable, but it is not an epistemic use of the word know. Knowledge requires more: truth, belief, justification that can be publicly tested, and understanding in use. Without these, what someone calls “knowledge” is merely the strength of their assurance, not the possession of reasons that could stand up in a public arena.

Another important aspect of this epistemology comes from Wittgenstein’s Philosophical Investigations: knowledge is not something we locate inwardly, as though we could point to a private object called “knowing” or consult an inner meter. Knowledge shows itself in our practices—what we say, how we check, and how we act in the world. To imagine that knowledge is an inner state is to confuse conviction with justification. Conviction is inward, but knowledge is public: it lives in criteria that others can test, in language that is shared, and in practices that display whether someone truly understands what they claim.

I don’t think I can overestimate how important epistemology is in evaluating beliefs and systems of belief, especially in our modern society where we’re being challenged at every turn. We are flooded daily with claims—political, scientific, technological, and now artificial intelligence—all competing for our assent. Without a clear framework for what counts as knowledge, we risk confusing conviction with truth or mistaking repetition for justification. Epistemology provides the tools to sort through this noise: to ask whether a belief is backed by reasons that are testable, whether those reasons stand up to defeaters, and whether the belief is grounded in understanding rather than mere assertion. In a culture where misinformation spreads quickly and skepticism often takes the form of indiscriminate doubt, epistemology steadies us by distinguishing between genuine inquiry and idle challenge. It reminds us that knowledge is not simply having a belief that happens to be true, but having reasons that others can check, refine, and correct.

The crux is that, following Sam26's thought, there needs to be some space between justification and truth in order for JTB to really be a three-legged tripod. I have to be able to be justified yet wrong. — J

I would put it this way: the real issue lies in how we understand justification. It is more than simply a person thinking they are justified. Genuine justification must be open to public testing, not just persuasive from the inside. Even so, it can fail—our best reasons are probabilistic, not infallible. But generally, these standards serve us well as guides to knowledge. And when a defeater arises that overturns what seemed to be justified, we recognize that the claim was never knowledge to begin with, but only something that masqueraded as such.

I'm going to repeat my position on Gettier because people seem to think it has weight.

The so-called “Gettier problem” rests on a sleight of hand. It trades on the difference between thinking one is justified and actually being justified. Gettier’s examples always contain a flaw in one of the three conditions of JTB: either the justification is faulty, or the belief is formed through a false step, or the connection between evidence and conclusion is too fragile to deserve the title of knowledge.

Take the stopped-clock case. A man looks at a clock, sees 2:00, and believes it is 2:00. By chance it really is 2:00, though the clock is broken. Here we do not have justified true belief at all. The belief is true, but the “justification” rests on a false ground: the clock is not functioning. That means the J (justification) condition fails. The man thinks he is justified, but he is not.

The same holds in other Gettier constructions. Somewhere along the line, one of the elements fails, usually the J. But if a case fails the J condition, then by definition it is not JTB. It is a case of apparent justification, not real justification. And if there is no genuine justification, then there is no knowledge to begin with.

When I speak of “real” justification, I mean justification that is not merely persuasive to the subject but actually satisfies the standards of knowledge: it must be publicly checkable, truth-conducive in the given context, and free of false grounds. Real justification is the kind of reason that anyone could in principle examine, replicate, and confirm, not just the kind of reason that feels convincing from the inside. Gettier’s examples work only by smuggling in defective grounds and then treating them as though they were genuine reasons. But if justification rests on a broken clock, a hidden falsehood, or a fragile inference that could collapse with the slightest change in circumstance, then it is not justification in the robust sense required for knowledge. It is a case of apparent justification: the subject thinks they are justified, but the conditions for knowledge have not actually been met.

This is why I see the Gettier literature as a long detour. It multiplies refinements to patch a problem that dissolves once we keep the standard for justification strong. By “strong” I mean publicly checkable, defeater-sensitive, and free of false grounds. If a justification fails those checks, it does not count as justification. Once that is clear, Gettier’s cases lose their force: they are examples not of knowledge, but of its counterfeit—instances where someone takes themselves to know but does not in fact know.

The dispute, then, is not a deep discovery but a confusion. Gettier showed that seeming to satisfy JTB is not the same as satisfying it. The lesson is valuable in its own way, but it is not the crisis it is often taken to be. The traditional JTB definition was never refuted; it was only misapplied.

To be honest I'm not sure I would agree with my point back then. I'd have to give it more thought. I do appreciate that at least you were trying to read the thread, most don't. Many of my thoughts have evolved in the last couple of years.

Wow, you went back a bit to find that. I would phrase that a bit differently now, but it's off the main topic, which is an epistemology following Wittgensteinian methods, but not strictly. I'm trying to expand epistemology using for example, OC.

What Is Knowledge? A Clear Explanation
In my view, knowledge is more than just a correct guess or a strong feeling. I define it using a framework called JTB+U, which stands for justified true belief plus understanding. Let me explain this step by step in simple terms.

For something to count as knowledge:
It must be true: The statement matches reality, like saying "the sky is blue" when it actually is.

You must believe it: You accept it as fact, not just as a possibility.

It requires justification: You have solid reasons that anyone can verify, such as evidence from observation or reliable sources.

And it includes understanding (+U): You grasp the concepts involved and know how to apply them correctly, avoiding confusion in how words or ideas are used.

To build those reasons, I outline five main paths, though there could be others:

Testimony: Relying on what trustworthy people report, checked by seeing if multiple accounts agree, come from diverse sources, stay consistent, and can be confirmed independently.

Logic: Reasoning things out. Inductive logic looks at patterns, like expecting rain tomorrow because it has rained every cloudy day this week. Deductive logic follows strict steps, like knowing all squares have four sides, so this shape does too.

Sensory experience: What you directly see, hear, or feel, as long as conditions are normal and not misleading.

Linguistic training: Learning and using words accurately, so you do not mix up meanings.

Pure logic: Basic rules like "it is this or it is not," which help structure thinking but do not add new information.

A common issue is endless questioning: "Why believe that reason?" To handle this, I draw on ideas like "hinges"—basic certainties we rely on without proving them each time, such as the world existing or words keeping their meanings. These hinges fit into "language-games," shared ways we use words in everyday practices, and "forms of life," the broader patterns of how we live and interact.

I also distinguish two uses of "I know": One is factual and provable to others; the other is just a personal sense of certainty.

To determine if something is knowledge, I use a straightforward process:
State clearly what you claim to know.
Choose the path for your reasons.
List the supporting evidence.
Apply checks specific to that path.
Ensure no false information is included.
Confirm the method usually leads to accurate results.
Look for anything that might disprove it.
Decide based on that, noting what could change the conclusion.

This approach makes knowledge reliable yet open to updates.

How It Differs from Traditional JTB
The traditional view of knowledge, known as justified true belief (JTB), says it is a true statement you believe with good reasons. My framework builds on this but adds improvements for better clarity and strength:

Adding understanding (+U): Traditional JTB does not require fully grasping the ideas. I insist on it, so you demonstrate knowledge by using concepts properly.

Incorporating hinges and shared practices: Traditional JTB can get stuck in endless questions or tricked by coincidence. I address this with hinges (unproven basics) and language-games (group rules for words), providing a stable foundation.

Focusing on public checks and flexibility: Traditional JTB can seem personal. I emphasize reasons everyone can examine and allow knowledge to be revised if new evidence appears.

Providing a clear process: Traditional JTB is more of a definition. I offer a step-by-step method with rules against errors to make it practical.

These changes make the idea of knowledge more robust and easier to apply in real situations.

I don’t really see a problem here. First, most of our knowledge is probabilistic, so if new evidence comes into the picture, it overturns what we believed we knew. Second, there’s a difference between thinking we’re justified, and being justified. This is similar to the Gettier mistake. Here's the move: Gettier's sting relies on a skimpy view of what "justified" really demands. He paints justification as this solo act—private reasons, thin evidence, no real-world grit to test against. But crank it up: justification isn't just a hunch with a receipt; it's got to be robust, publicly accountable, and hooked into practices that weed out the lucky breaks. Start with no false grounds—no sneaky false lemma propping up the chain, like assuming Jones is the lock when the evidence could've flagged alternatives. In the job case, Smith's reasons scream "Jones," but if we'd baked in a defeater screen—cross-checking the boss's reliability, weighing other candidates, or even probing the "10 coins" as a quirky proxy—the fluke wouldn't glide by so easy. It's not that the belief's true by luck; it's that the justification was flimsy from jump, masquerading as solid because no one poked the seams.

I'm not sure, but you seem to think that if knowing isn't absolute, it isn't knowledge. This is a classic misunderstanding of what knowledge is.

What On Certainty Can Teach Us About AI

When I read Wittgenstein’s On Certainty, I can’t help but think about how it touches on questions we’re now facing with artificial intelligence. His remarks weren’t about machines, of course. They were about us, about the background that makes human knowing possible. But the more I’ve worked with his ideas, the more I see how they matter for thinking about the development of AI.

One of Wittgenstein’s simplest but most powerful points is that doubt only makes sense against a backdrop of certainty. I can doubt whether my car will start, but I can’t doubt that there’s a world in which cars exist. If I try to doubt that, I lose the very stage on which doubting makes sense. Now, think about AI. A system that treats all doubts as equal, as if “What is the capital of France?” and “Does the world exist?” were both just queries, misses something basic. Humans know how to sift idle doubt from meaningful doubt. For AI to be trustworthy, it will need something similar: a grammar of doubt, a way of recognizing which uncertainties are live and which are nonsense.

Another thought from On Certainty that strikes me is how much of our knowledge is enacted rather than stated. I don’t prove to myself each morning that the floor will hold me; I simply stand. Certainty shows itself in action. With AI, though, we tend to optimize for propositions: correct outputs, factually accurate answers. But there’s no hinge of action beneath it. Machines don’t “stand fast” in the way we do. That tells me we should be cautious about equating AI’s linguistic performance with human understanding. Without hinge-like certainties, embodied anchors, a sense of persons, continuity of the world — the words may ring hollow.

There’s also the question of inheritance. Wittgenstein says we are taught certain things without question. That’s our world-picture. AI, too, inherits a world-picture — its training data. What goes in as unquestioned background shapes everything that comes out. The danger, of course, is that we treat the dataset as neutral when it is already hinge-laden, already thick with assumptions. If we don’t examine that, the machine’s “certainty” may be nothing more than a mirror of our own blind spots.

Finally, Wittgenstein reminds me that explanations cannot go on forever. At some point, they end, and not because we’re lazy or careless, but because all justification rests on hinges that are not themselves justified. For humans, this is just how rational life works. For AI, it suggests that the demand for fully “explainable” systems may be a mirage. Like us, machines will have to rest on foundations that can be mapped, checked, and disciplined, but not explained away.

When I connect these dots, I come away thinking that On Certainty offers a kind of realism about intelligence, human or artificial. Knowledge isn’t just propositions strung together. It depends on hinges: embodied trust, inherited world-pictures, shared grammar, limits of doubt, limits of explanation. If AI is to grow into something we can genuinely trust, then what Wittgenstein saw about human knowing should not be ignored.