Anthropopractic for neural networks: training and learning to understand

The article examines the phenomenon of artificial intelligence through the lenses of anthropology, semiotics, and the philosophy of technology. Starting from the observation that reflection does not keep pace with the rapid development of technologies, the author draws parallels between the processes of understanding and "training" neural networks. Using dialogues with the neural network deepseek and analyzing the works of Maria Kuvshinova, the nature of the neural network's learning and the "self-organization" of AI is explored. Specialists train AI using the same types of connections (subject-related, linguistic, logical, task-based) as in modern human education. The analysis of AI's functioning shows that, in addition to the subject-related connections established through schemes, the neural network also operates with three other types of connections: linguistic (using grammar, syntax, and other language norms), logical (based on rules and patterns for constructing reasoning, proofs, research, etc.), and conditionally "task-based" (within the framework of problem-solving; these are studied in heuristics, TRIZ, cognitive psychology). These types of connections are established in the process of training neural models; additionally, the neural network finds them on the Internet. It also discovers various forms of reflection on the Internet (in the form of samples and methodological awareness) regarding the relationships that connect subject-related, linguistic, logical, and task-based modes of thinking. This allows the author to characterize contemporary AI as a "faceless subject." Furthermore, the article distinguishes between "the formation of a specific type of neural network," during which a "generative organization" of the neural network is developed (i.e., a neural network with changes that allow it to understand the user and solve the problems posed), and "the operation of the neural network," where through the actualization of the generative organization, the neural network adequately understands and solves tasks. In its non-operational state, the neural network is a "turned-off machine"; in its operational state, on one hand, it is a "working machine" and "technology," while on the other, it is a "human-controlled subject" capable of understanding us and solving the tasks we set. The article offers a cultural-anthropological perspective on AI as a new form of "anthro-techno-social" reason, generated by humans.