The paper outlines an epistemological self-positioning of the scientific community towards a new type of cognitive agent, artificial neural networks. The inclusion of neural-network technologies in research practice raises questions that go beyond the instrumental use of new tools. If the results obtained with neural networks claim the status of scientific knowledge, the scientific community is required not only to verify them methodologically, but also to rethink the very structure of the cognitive situation. Drawing on V.I. Arshinov’s idea of the neural network as an “observer of complexity” and on G. Simondon’s philosophy of technology, the author analyses two recent cases, the gluon scattering problem solved by OpenAI and A. Strominger, and Google DeepMind’s autonomous mathematical agent Aletheia. The paper discusses neural-network hallucinations and sycophancy, limits of interpretability, and recent Anthropic findings on internal computational organization and emergent introspective awareness. Heidegger’s notion of aletheia is invoked to argue that the central issue is not the truthfulness of neural-network knowledge but the readiness of the scientific community to encounter the radically new. In conclusion, the author suggests treating the neural network not as a “black box” or a replacement for the researcher, but as a partner in joint thinking, which presupposes the development of corresponding institutional, educational, and communicative forms.
artificial intelligence; neural networks; philosophy of science; observer of complexity; V.I. Arshinov; interpretability; epistemology of scientific knowledge.