The Centaur Situation: The Grounds and Limits of the Cognitive Alliance

Abstract

The Epistemic Centaur Manifesto, published in this issue, proposes a strategy of alliance between the human researcher and artificial intelligence in the human and social sciences. Its proposi- tions are stated as theses, without argument or references, and this article sets out the grounds on which they rest and clarifies the points that invite disagreement. Drawing on recent research literature, the authors show that generative models are already built into every stage of the research cycle, from hy- pothesis formation and literature reviews to synthetic respondents, the recovery of missing survey data, machine interviewers, text annotation and statistical coding. The authors call this state of affairs the cen- taur situation, one in which researchers find themselves whether or not they have chosen the alliance. To analyse it, the article develops the distinction between the functional agency of artificial systems and human subjectivity that proposition 2.2 of the Manifesto introduces, and shows that the irreducibility of a result to the contributions of its participants does not rule out disclosing what the machine did. The article then takes up the objection that responsibility for a conclusion the researcher does not fully understand becomes a ritual admission of guilt, and argues that such responsibility is meaningful as long as the researcher retains the right to reject a conclusion and verifies it in proportion to its consequences. Finally, experimental evidence on human-machine collaboration gives no support to the idea that the alliance is productive in itself, which is precisely why the demands of personal discipline, of a community of practice and of rethinking the aims of university education carry the weight they do.