Issue #53

Abstract: This article is devoted to a philosophical and methodological analysis of the phenomenon of artificial intelligence (AI), which the author views not simply as another computing technology, but as a civilizational factor comparable in significance to the invention of writing or printing. The starting point for this discussion is Christopher Summerfield’s book, which captures a fundamental debate about the nature of large-scale language models (LLMs): whether they are simply complex computational devices or represent a new, different type of intelligence. The author demonstrates that two camps are engaged in a dialogue in modern science: computer scientists, who tend to interpret AI as computation, and philosophers and sociologists, who see this approach as naive reductionism. The author seeks to overcome this opposition by proposing to consider AI in two dimensions: first, as a neural network substrate providing computation and flexibility; second, as a symbolic activity modeling human intellectual operations — the construction of schemas, linguistic expressions, knowledge, and problem solving. Particular attention is paid to how neural networks, lacking formal logical rules, acquire the patterns of language and thought through statistical learning, forming “organizational rules” fixed in a multidimensional space of weights. Parallels are drawn with the functioning of the human brain as the substrate of consciousness, emphasizing that attempts to reduce the phenomenology of AI solely to neural network computations are as illegitimate as reducing human thinking to electrochemical processes in the cortex. The article analyzes the mechanisms by which prompts are transformed into meaningful responses, the role of embeddings and attention mechanisms, and also raises the question of the possible future acquisition of subjectivity by AI.