The journal "Higher Education in Russia" has published the "Manifesto of the Epistemic Centaur" — a programmatic text on the state of human and social sciences in the context of the spread of generative artificial intelligence. The manifesto was developed by a research seminar at the Institute of Humanities and Social Technologies of HSE University; among the authors are HSE Vice Rector Dmitry Zemtsov and the journal's editor-in-chief Vladimir Nikolsky.
The authors pose the problem of competition of "simple answers": AI produces them faster than science manages to obtain substantiated knowledge, and social phenomena change even before research is completed. In their opinion, the humanities and social sciences risk losing the initiative in explaining humans and society, as well as control over the language of their description.
The authors call the answer to this challenge the "epistemic centaur" — a single functional subject of cognition, in whose common cognitive circuit human consciousness and artificial intelligence conduct cognitive work. The result obtained in such a circuit is irreducible to the individual contributions of the participants, and responsibility for it remains entirely with the researcher.
Knowledge created by the centaur is recognized as equal in status to knowledge created by a human — under three conditions:
- the nature of interaction with AI is disclosed,
- the results are verified,
- responsibility is accepted.
At the same time, the principles of the ethos of science — universalism, collectivism, disinterestedness, and organized skepticism — remain valid in centaur research. The researcher must correlate the speed of the machine's work with the depth of their own understanding and avoid oversimplification when studying complex phenomena. "The machine can make mistakes, but it cannot be surprised," the authors emphasize: noticing the unusual is a human duty, and an anomalous result can become a source of new knowledge only after rigorous verification.
Particular caution is required for a conclusion that sounds convincing and entails serious consequences, but is opaque in origin. Ethical limitations embedded in AI do not relieve the researcher of their own assessment: the conclusion must be weighed both by the measure of scientific reliability and by the measure of human ethics.
The authors also warn that generative AI is not neutral: it assimilates ways of explaining the world, omissions, and blind spots from training texts. The picture proposed by the machine cannot be taken as universal.
They propose to leave the last word to humans
We propose to perceive the manifesto as an occasion to rethink university policies in the field of AI use. It is important to determine how to assess the independence of the researcher, what to teach students, and how to ensure the reliability of knowledge created in cooperation with a machine
For the development of practice, the authors propose a horizontal community of researchers that will exchange methods, discuss difficulties, and verify results; in the long term, the manifesto states, this will require rethinking the goals of university education.