Algorithmic Decision-Making and the Legitimacy of Arbitral Justice: Toward a Multilevel Framework of Normative Governance

The integration of artificial intelligence into international commercial arbitration has engendered a structural fracture between legal rationality, grounded in normative justification, and the statistical logic of algorithms, based on opaque correlations. This epistemic tension compromises the arbitrator's agency and erodes the legitimacy of the arbitral award, particularly in the absence of regulatory frameworks capable of translating algorithmic outputs into legal reasons. The objective of this research is to design a multilevel normative model that enables such epistemic translation by articulating protocols for counterfactual explainability, substantive audits of normative biases, and interoperability standards to safeguard legal diversity against algorithmic homogenization. The findings enable the reconstruction of arbitral deliberation as an autonomous rational practice, even within AI-assisted contexts. It is concluded that human oversight is legally effective only when integrating normative mechanisms that guarantee the rational controllability of the award, thus configuring a pluralistic technological governance compliant with transnational procedural public policy.