Data-Centric Governance Models Using Trustworthy AI: Strengthening Transparency, Bias Control, and Policy Compliance in Welfare Management

Abstract

The increased applications of AI-based decision making in the welfare area of the government have heightened the issues associated with the lack of transparency, the bias of algorithms, and the uneven compliance with the provisions of the policy. Current welfare systems often have disjointed data streams and black box models, which creates quantifiable differences in benefit eligibility determinations across demographic categories and opening rates in automated decision libraries of more than 20 percent. To overcome these obstacles, this article proposes an integrated model of data-centric governance that implements reliable principles of AI, combining the promotion of transparency, the reduction of the effect of bias, and the possibility of automatic verification of policy adherence. The structure takes into consideration organized data administration, impartiality-conscious modeling, decipherable choices and a guideline-driven conformity execution to guarantee uniform, auditable welfare results. Empirical experiments done on welfare-analogous datasets indicate that the proposed model narrows demographic gaps by 31-38% and leads to greater compliance accuracy of policies (78 vs. 96) and higher transparency scores (42 vs. baseline machine learning systems). The governance layer is also computationally efficient and has a mean runtime overhead of 69-9%. These findings indicate that data-fiduciary trust AI: This finding shows that sound, trustworthy, and regulatory consistent welfare decision-making through data-centric AI provides a promising opportunity to establish fairness, reliability, and regulatory consistency in the application of an AI to the population.

Ähnliche Arbeiten