The TUDOK Operator as a Proposed Operational Measurement Layer for the Rome Call for AI Ethics

Abstract

In 2020, the Rome Call for AI Ethics articulated a shared, human-centred ethical commitment that organizes the development and use of artificial intelligence around three areas of impact-ethics, education, and rights-and six principles: transparency, inclusion, responsibility, impartiality, reliability, and security and privacy. The normative strength of the document lies in bringing human dignity, the common good, the protection of vulnerable persons, and technological responsibility into a shared ethical language. A recurring problem in the literature on AI ethics, however, is that a practical implementation gap remains between statements of principle and operational, documentable practice. This study examines how the previously published TUDOK Operator may help bridge a narrowly defined part of that gap. TUDOK is a model-agnostic, output-level calibration measure that evaluates AI outputs according to four components: semantic coherence, internal consistency, contextual truth- and reality-alignment, and distortion pressure. TUDOK is not a fact-checking engine, a moral arbiter, a comprehensive AI-safety system, or a compliance-certification mechanism. The study analyses the six principles of the Rome Call through a conceptual crosswalk method. For each principle, it separately identifies the normative content, the observable output-level risk, the relevant TUDOK components, the diagnostic signals, the required evidence trail, the point at which human review is needed, and the limits of the method. The analysis suggests that TUDOK may provide its strongest support to certain output-level aspects of transparency, responsibility, and impartiality; may offer a supplementary calibration signal for reliability; and may play only a partial role in relation to inclusion, security, and privacy. It does not replace fairness, accessibility, data-protection, or cybersecurity assessments. Finally, the study proposes an audit and review process aligned with the principles of the Rome Call while remaining institutionally independent from it. It treats the existing demonstration and the analysis of differences observed during a second evaluation pass as the starting point, and proposes a confirmatory study involving 30–50 task prompts, multiple models, and at least two genuinely independent evaluators. The programme would separately assess inter-evaluator agreement, paraphrase and cross-lingual stability, the relationship with external fact-checking, and the reproducibility of the Rome Call crosswalk. The aim is not to create a new certification system, but to offer an openly contestable research instrument for making truth-fidelity, uncertainty handling, and the need for human review more visible.