By Geoffrey Chen

In traditional epistemology, the definition of knowledge has always been a central issue. The most classic formulation is JTB, or justified true belief. A person believes something, that thing is true, and the person has sufficient reason to believe it. For a long time, this definition appeared quite stable. If a person does not believe something, then obviously he does not know it. If what he believes is false, then it cannot count as knowledge. If he happens to say something true but only by guessing, that also does not seem to be genuine knowing. So truth, belief, and justification seemed to form the basic framework of knowledge.

Later, the Gettier problem raised a famous challenge to this definition. Its basic point is that even when all three conditions are satisfied, that may still not be enough for knowledge. A person may form a belief that is justified and true, and yet we still do not want to say that he genuinely knows. In other words, knowledge requires truth, but it also requires that the truth not be reached merely by luck. That is the importance of the Gettier problem. It reminds us that knowledge is not simply a belief that looks qualified on the surface. Knowledge must also exclude a certain kind of epistemic accident.

There is no real problem with that point itself. I do not deny the role of the Gettier problem in the history of epistemology. However, since this book does not directly discuss the definition of knowledge, I only want to say a little here about how I see Gettier.

My view is that the examples Gettier originally used were mainly academic counterexamples aimed at professional philosophical readers, not ordinary explanations meant for general readers. Their purpose was to find the shortest argumentative path for breaking a definition. For that reason, they often carry a strong formal character. They do not feel natural, and they do not resemble ordinary experience. For professional readers, this is not strange, because what matters to them is whether the counterexample works structurally, not whether it is elegant or close to everyday life. But for ordinary readers, these examples easily produce misunderstanding. People may take the issue to be a logical mistake in the process of justification, or merely a strained language game.

In fact, what Gettier wanted to challenge was not logic itself, but the sufficiency of the traditional definition of knowledge. Logic was only the tool used in the argument. He was not saying that some step of inference broke the rules. He was saying that even if a belief formally satisfies the conditions of justified true belief, it may still fail to be knowledge because too much accident has entered into it. That point is clear enough in a professional context. But in a broader reading environment, because the original examples are so artificial, readers are often distracted by the unnaturalness of the examples themselves and miss the fact that the real target is the traditional definition of knowledge.

I also think that unnatural examples do not merely weaken persuasiveness. At the level of the general public, they may even produce the opposite effect. JTB has remained influential not only because it is philosophically neat, but because it also matches the basic intuition ordinary language has about the word “know.” In everyday language, when people say “I know,” what they usually mean is simply this. The thing is true, I believe it, and I have reason to say so. That stability is not accidental.

By contrast, an example that is too artificial and too dependent on formal construction can easily make people feel that the problem lies not in the definition of knowledge, but in the strained nature of the example itself. More importantly, natural language is not a mechanical shell for formal logic. In real understanding and everyday judgment, people constantly perform semantic filtering, relevance judgment, and contextual correction. They naturally exclude situations that are too disconnected, too stitched together, or too unnatural. For that reason, some counterexamples that can be constructed in formal terms may not appear very often in real cognition, and may not appear easily at all. What is formally constructible is not automatically cognitively representative.

In that sense, the force of Gettier’s original examples is not exactly the same for professional readers and for the general public. For professional discussion, they may be sufficient, because a philosophical paper only needs one structurally valid counterexample in order to shake a definition that claims universal validity. But at the public level, these examples may not truly dislodge JTB. On the contrary, they may lead people to think that the author is creating trouble through unnatural cases. This does not mean that the Gettier problem is entirely invalid. It only means that its original mode of presentation had clear limits tied to an academic context.

That is also why After the Knowing Subject does not spend its main energy on this line of discussion. It is not because these debates are worthless. It is because this book is concerned with a different level of the problem. It does not begin by asking how knowledge should be defined. It begins more directly with a reality. Today, a large amount of what we call knowledge can already be processed, organized, called upon, and output without depending on a traditional knowing subject. This is not a marginal phenomenon. It is a real shift that has already happened and is still expanding.

Whether in search, translation, code generation, text analysis, pattern recognition, medical assistance, legal summarization, scientific research, or everyday decision-making, artificial intelligence systems are dealing with the very contents that used to belong to knowledge activity. They process facts, relations, rules, chains of reasoning, semantic structures, and judgments under complex conditions. More importantly, these outputs have already been incorporated into social practice. They have become something people rely on.

Once things reach this point, the center of the problem changes. The real question is no longer only which condition the definition of knowledge is still missing. The real question is whether, once the actual operation of knowledge has partially detached itself from the traditional subject, we should still continue to understand knowledge entirely as an inner state of a subject. Traditional epistemology begins from the assumption that knowledge is something possessed by a subject. So the problem of knowledge is usually written in the following form. Who knows, under what conditions does that person know, and why does that person have the right to say that he knows. This framework had its reason in a human-centered age, but today it is clearly no longer enough.

Artificial intelligence systems do not possess subjectivity in the traditional philosophical sense. They do not have the inner self-experience of human beings. They do not have a unified stream of consciousness. They do not have the kind of structure that gathers belief, reflection, memory, and self-understanding into one center of existence. If we insist on the old standard, then they would seem not to belong within the discussion of knowledge. But reality does not look like that. The reality is that they are already performing more and more tasks that previously could only be performed by a “knower,” and the results of these tasks are already being treated as knowledge. That means the problem is not whether AI knows in the same way humans do. The problem is whether our inherited understanding of knowledge relies too heavily on a specific model of the human subject, so heavily that it can no longer explain the new situation now before us.

This is the starting point of After the Knowing Subject. It does not first accept a fixed definition of knowledge and then check whether AI fits it. It proceeds more from the actual operation of knowledge in the world and then asks whether the old concept of knowledge is still adequate. In other words, this book does not stand at the gate of traditional epistemology holding up an old standard to inspect AI. It stands before a changed reality of knowledge and reexamines traditional epistemology itself.

From this perspective, the Gettier problem remains important, but its importance is limited. What it strikes is still an issue internal to traditional subject-centered epistemology. It tells us that even if a subject has a true belief supported by reasons, that may still not be genuine knowledge. This is indeed a deep insight. But it does not touch a larger question. If the production and operation of knowledge are increasingly taking the form of a systematic, distributed, and de-subjectified process, then why should we continue to understand knowledge only as some belief-state inside a subject. The Gettier problem forces us to rethink what kind of subject-state counts as knowing. After the Knowing Subject is dealing with a more prior question, namely whether knowledge must still take subject-states as its basic unit at all.

That is why this book does not dwell on JTB or on its many revised forms. This is not because those discussions are irrelevant. It is because this book does not want to place the starting point of the problem inside traditional definitions. No matter how refined, complex, or careful a definition becomes, if it still assumes that knowledge first belongs to an inner subject, then it still has not faced the actual transformation of the present. What needs explanation today is not simply how one person satisfies the conditions of knowledge. What needs explanation is why the contents and functions previously treated as knowledge can now stably exist and operate within nontraditional systems.

The key point here is simple. However we define knowledge, AI is still processing the very kind of things we have always called knowledge. It processes historical information, linguistic rules, mathematical relations, legal structures, medical materials, logical arguments, and cross-textual connections. These contents do not suddenly lose their knowledge-character merely because the processor is no longer an individual with human consciousness. If when humans process these things we call them knowledge, but when AI processes them we suddenly say they are not knowledge, then the problem usually lies not in reality, but in the concept. The concept has fallen behind reality. That is what really needs to be pointed out.

For that reason, the focus of After the Knowing Subject is not to patch the boundary of the word “knowledge,” but to reassess the mode of existence of knowledge itself. Traditional epistemology is used to thinking of knowledge as something attached to the interior of a subject, as if without a knowing subject there could be no knowledge. But what is becoming increasingly obvious today is that knowledge exists in other forms as well. It can exist in databases, in model weights, in algorithmic procedures, in retrieval networks, in organizational structures, and in human-machine systems of cooperation. It no longer always appears as “a justified true belief inside someone’s mind.” It increasingly appears as an external structure that can be generated, called upon, integrated, and validated.

Of course, this does not mean that the human subject has become unimportant. Human beings still participate in judgment, selection, interpretation, goal-setting, and responsibility, and they still play a decisive role in many critical places. But this is no longer the same as the subject-centered model of traditional epistemology. The old picture was that all knowledge must finally be traced back to one complete, self-transparent, unified cognitive subject. The more realistic picture today is that knowledge increasingly flows through systems, networks, procedures, and relations of cooperation, and that the human subject is only one node among them, not the sole center of carriage.

So After the Knowing Subject does not focus on defining knowledge, not because it neglects the philosophical tradition, but because it thinks philosophy now has to face something larger than internal repairs to old definitions. JTB and the Gettier problem represent an important historical stage. They help us understand tensions within the traditional concept of knowledge, and they help us see the complexity of luck, reason, and truth. But if discussion remains there, it is still circling on the old foundation. The larger question today is whether, once knowledge can be processed and made effective without the traditional knowing subject, we can still continue to use a subject-centered concept of knowledge without radical revision.

This is where the position of this book begins. It is not a book that tries to make the definition of knowledge more precise. It is a book that tries to explain why knowledge can no longer be fully understood through the traditional knowing subject. Its concern is not the final patch on an old definition. Its concern is whether the entire picture of knowledge has undergone a structural change. For that reason, it does not lock itself into disputes about JTB, Gettier, or any other traditional definition. It starts instead from the actual practices of knowledge in the world and rethinks knowledge itself.