From Tool to Agent
Yuval Noah Harari’s fundamental distinction in the debate on artificial intelligence is that technology can no longer be understood merely as a tool used by human beings. A traditional tool carries out the purpose assigned to it; a printing press does not decide which text will be printed, and a weapon does not independently choose which target will be struck. Artificial intelligence, by contrast, is a system capable of generating options, participating in decision-making processes, producing linguistic content, and developing new courses of action in accordance with the general objectives assigned to it. Harari therefore thinks of artificial intelligence through the concept of the “agent.” The central thesis of The Next 50 Years: Humanity, AI, Power is that the technological rupture lies not simply in greater computational capacity but in the entry of a non-human agent into the social world.
This distinction is particularly important because, although humanity has previously produced machines more powerful than itself, it has not encountered decision-making technical systems on the same scale. Harari’s emphasis on language, law, finance, and bureaucracy is therefore not incidental. The effects of artificial intelligence may first emerge not through machines that seize control of the physical world but within the symbolic systems constructed by human beings. When an algorithm determines credit risk, evaluates legal texts, develops an investment strategy, or selects the content that millions of people will encounter, it does more than perform calculations; it participates in a chain of decisions with social consequences. Yet it is precisely at this point that Harari’s concept of agency encounters another and much older philosophical question: Is the capacity to act sufficient for being a subject?
Žižek: There Is an Agent, But Where Is the Subject?
Slavoj Žižek’s Lacanian approach to artificial intelligence introduces an important distinction at the point where Harari’s analysis leaves off. The central claim of Žižek’s 2026 text Why Artificial Intelligence Is Not a Subject is that an artificial intelligence system’s ability to behave like a subject does not mean that it is one. A large language model can receive questions, produce answers, establish relationships among concepts, and maintain an apparently coherent discourse. In the Lacanian sense, however, the subject is not defined by the sum of these capacities. The subject emerges through the division and failure within its own symbolic representation; it carries a gap between itself and what it says that can never be fully closed. Žižek therefore describes artificial intelligence not directly as a subject but as “knowledge without a knower”: the operation not of a personal “I know” but of a subjectless “it is known.”
Harari and Žižek do not exclude one another here. They speak from different conceptual levels. Harari’s “agent” is a functional category: a system capable of making decisions and producing consequences. Žižek’s “subject,” by contrast, is an ontological and psychoanalytic category. Artificial intelligence may therefore be an agent without being a subject. Indeed, the philosophical specificity of contemporary artificial intelligence may be found precisely within this distinction. Humanity is confronting systems that have not been shown to possess subjective experience yet are acquiring an increasing capacity for action within linguistic, economic, and institutional processes.
Knowledge Without a Knower
Žižek’s formula of “knowledge without a knower” carries the problem beyond the debate over whether artificial intelligence possesses consciousness. When a large language model answers a question, it is easy to assume that there is an “I” on the other side that knows the answer. Its use of first-person linguistic forms, its capacity to provide explanations, and its ability to establish relationships among questions strengthen this illusion. The user may therefore place the system in the Lacanian position of the subject supposed to know: asking it questions, expecting judgments from it, and entrusting what they do not know to the knowledge the system is presumed to possess. The paradox identified by Žižek is that the user addresses a subject, while the response comes from a subjectless order of knowledge.
At this point, the classical bond within epistemology begins to dissolve. Knowledge has traditionally been connected to a knower: someone sees, remembers, experiences, infers, makes mistakes, corrects a judgment, and can finally say, “I know.” In artificial intelligence, explanation, classification, association, and the production of conclusions occur without a subject that experiences these operations as its own, binds itself to the truth of what it says, or bears responsibility for its falsehood. Nevertheless, the resulting output can be used by human beings as knowledge; it can influence decisions, be transferred into other texts, and become part of institutional processes. The social function of knowledge and the existence of the knowing subject can therefore become separated.
Synthetic Epistemology begins precisely at this separation.
Synthetic Epistemology: The Separation of Subject and Knowledge
From the perspective of Synthetic Epistemology, the fundamental problem is not the technical manner in which artificial intelligence operates. Datasets, models, algorithms, and computational architecture are important, yet they are the material conditions of a more fundamental transformation: the historical separation of knowledge from the knowing subject. A process resembling knowledge, producing judgments and decisions, and generating consequences in the real world can now occur without the classical subject that knows, understands, or experiences it.
Synthetic Epistemology therefore does not remain trapped within the question, “Does artificial intelligence truly know?” The more important question is this: What happens to epistemology when content produced without a knowing subject begins to function socially as knowledge? The issue is not merely to determine the machine’s internal condition. People make decisions according to these outputs, institutions incorporate them into classification processes, and synthetic content re-enters the shared field of knowledge. Subjectless production therefore does not remain outside the subjective world; it reshapes what human beings know, what they trust, and what they accept as real.
The classical schema was based broadly upon the subject–object relationship. Within the synthetic information environment, however, data, models, algorithmic processes, and interfaces enter between them. A human being asks a question about the world, yet the representation received is no longer solely the product of that person’s own experience of the world or the judgment of another human subject. Instead, the person encounters a synthetic result that has passed through previously produced human expressions, structures of data, and algorithmic relationships. Representation remains, but behind it there may be no singular subject who assumes it as their own speech.
The Subjectlessness of Representation and the Problem of Responsibility
Harari’s thesis that language is the “operating system” of civilization acquires a new meaning here from the perspective of Synthetic Epistemology. If human societies operate through law, money, institutions, and shared narratives, and if non-human agents begin to participate in producing these narratives, then what changes is not merely communication technology. The subject-structure of representation is changing. The human being is no longer the direct source of all cultural representations, yet these representations continue to generate real effects within the human world. Under such conditions, the question of error also changes. When a human being makes an incorrect judgment, they can be asked for their reasons; what they saw, what they thought, and which evidence they relied upon can be investigated, and responsibility for the decision can be attributed to them. In a synthetic system, however, the causal chain is dispersed among data selection, model design, the training process, user input, institutional use, and algorithmic output. The mechanism producing the result exists, but there is no subject who assumes that result as its own judgment. The fundamental problem of artificial intelligence epistemology is therefore not merely truth but where the subject of responsibility is to be located.
It is precisely here that Harari’s concept of agency and Žižek’s thesis of subjectlessness must be considered together. While agency increases the capacity to produce consequences, the absence of subjectivity does not eliminate responsibility. On the contrary, it makes the reconstruction of responsibility by human beings and institutions necessary. The participation of a technical system in a decision-making process cannot automatically transfer the decision’s normative burden to the system itself.
The Threshold Human: Not the Owner of Knowledge, but the Bearer of Responsibility
This transformation does not mean that the human subject disappears entirely. From the perspective of Synthetic Epistemology, the human position is changing. The human being may no longer be the central subject who independently produces every form of knowledge and controls all knowledge processes. Yet a position is still required from which the results produced by synthetic systems can be interpreted, questioned, limited, and used in decisions concerning what should be done. The Threshold Human names precisely this new position of the subject: the human being who does not produce knowledge alone but does not abandon critical and ethical responsibility before knowledge.
Harari’s distinction between intelligence and wisdom acquires another dimension here. Intelligence can solve a problem; wisdom asks which problem is worth solving. Synthetic Epistemology adds a third term: responsibility. A system may produce an extremely powerful solution, yet computational capacity alone cannot determine what kind of world that solution will construct, whom it will exclude, or which of its consequences should be considered acceptable. The task of the Threshold Human is not to compete with artificial intelligence by producing more knowledge than it does, but to perceive what remains unrepresented alongside what is represented, and to recognize the epistemic void alongside the answer that has been given.
Conclusion
Harari, Žižek, and Synthetic Epistemology grasp the same historical rupture at three different levels. Harari’s question is technological and social: What happens when a tool becomes an agent? Žižek’s question belongs to the theory of the subject: Why may an agent still fail to become a subject even when it can speak and produce knowledge? Synthetic Epistemology constructs the epistemological problem at the intersection of these two questions: What happens to the status of knowledge, representation, and responsibility when a non-subjective agent becomes an active element in the production of knowledge?
The fundamental rupture of the age of artificial intelligence is therefore not that machines process more information than human beings. The more radical transformation is the separation of the functions of knowing from the knowing subject. For the first time, humanity is entering an epistemic world in which there need not be a “knower” behind extraordinarily powerful information processes, explanations, and decisions. Harari’s agent and Žižek’s “knowledge without a knower” constitute two sides of this threshold. From the perspective of Synthetic Epistemology, however, the central problem begins beyond both of them: if knowledge can circulate and produce effects without belonging to a subject, the human task is not to reclaim a monopoly over knowledge. The new task is not to lose subjective responsibility before the consequences of subjectless knowledge.
