Neural Networks in Evidence-Based Governance: From Analytical Enhancement to Cognitive Dependence

  • O. O. Horobets
Keywords: evidence-based governance, neural networks, cognitive dependence, cognitive autonomy, algorithmic systems

Abstract

The article highlights the role of neural networks in evidence-based governance and identifies the conditions under which their use can enhance the analytical capacity of public administration without undermining the cognitive autonomy of public institutions. The rapid expansion of neural network technologies is transforming the traditional process of evidence formation, as algorithmic models are increasingly used not only for processing large volumes of structured and unstructured data, but also for forecasting, classification, risk assessment, pattern detection, and the generation of recommendations for managerial and policy decisions. It is emphasized that high predictive accuracy should not be equated with the reliability of evidence or the validity of a final governance decision. The use of neural networks is associated with methodological, institutional, and epistemic risks, including dependence on data quality and representativeness, limited model transparency and interpretability, algorithmic bias, automation bias, excessive reliance on automated recommendations, and the possible diffusion of institutional responsibility. Cognitive dependence is conceptualized as a specific risk of the algorithmization of public administration. It arises when a public institution gradually loses its capacity to independently assess data, verify the logic of an algorithmic model, critically interpret its outputs, and consider alternative courses of action. Particular attention is attached to the distinction between technological dependence and cognitive dependence: even when a state formally controls its data, it may remain dependent on external models, methodologies, digital platforms, and algorithms through which data are transformed into assessments, forecasts, and recommendations. The author argues that preserving cognitive autonomy requires five interrelated conditions: control over data, model transparency, interpretative capacity, human verification, and institutional accountability. On this basis, a model for the safe integration of neural networks into evidence-based governance is proposed. It combines the sequence “data – model – interpretation – verification – management decision” with corresponding institutional safeguards. Neural networks should, therefore, be regarded as instruments for augmenting institutional analytical capacity rather than as autonomous decision-makers. The preservation of meaningful human and institutional control over the formation, interpretation, verification, and use of evidence is identified as a key condition for preventing cognitive dependence in algorithmically supported governance.

Downloads

Download data is not yet available.

Abstract views: 0
PDF Downloads: 0
Published
2026-08-31
How to Cite
Horobets, O. O. (2026). Neural Networks in Evidence-Based Governance: From Analytical Enhancement to Cognitive Dependence. Scientific Bulletin of the National Academy of Statistics, Accounting and Audit, (3), 87-97. https://doi.org/10.31767/nasoa.3-2026.07