This article proposes a redefinition of scientific authorship under conditions of algorithmic mediation. We shift the discussion from the ontological dichotomy of “tool versus author” to an operationalizable epistemology of contribution. Building on the philosophical triad of instrumentality—intervention, representation, and hermeneutics—we argue that contemporary AI systems (notably large language models, LLMs) exceed the role of a merely “mute” accelerator of procedures. They now participate in the generation of explanatory structures, the reframing of research problems, and the semantic reconfiguration of the knowledge corpus. In response, we formulate the AI-AUTHorship framework, which remains compatible with an anthropocentric legal order while recognizing and measuring AI’s cognitive participation. We introduce TraceAuth, a protocol for tracing cognitive chains of reasoning, and AIEIS (AI epistemic impact score), a metric that stratifies contributions along the axes of procedural (P), semantic (S), and generative (G) participation. The threshold between “support” and “creation” is refined through a battery of operational tests (alteration of the problem space; causal/counterfactual load; independent reproducibility without AI; interpretability and traceability). We describe authorship as distributed epistemic authorship (DEA): a network of people, artifacts, algorithms, and institutions in which AI functions as a nonsubjective node whose contribution is nonetheless auditable. The framework closes the gap between the de facto involvement of AI and de jure norms by institutionalizing a regime of “recognized participation,” wherein transparency, interpretability, and reproducibility of cognitive trajectories become conditions for acknowledging contribution, whereas human responsibility remains nonnegotiable.
Adilbek K. Bisenbaev (Tue,) studied this question.