This working paper introduces a Decision Engineering Science (DES) perspective on recent developments in predictive cognitive AI, particularly approaches based on filter equivariance and structural stability as explored by MindCast AI and related research directions associated with Google DeepMind. While predictive cognitive AI systems increasingly ensure coherence and robustness at the level of representation, they do not inherently guarantee the quality, alignment, or robustness of decisions derived from those representations. This paper identifies a structural gap between predictive performance and decision performance. To address this gap, the paper introduces a decision layer as a distinct system component and extends the concept of equivariance into the decision domain through the Decision Consistency Principle. In addition, it presents the Decision Quality Index (DQI) as a multi-dimensional framework for evaluating how effectively systems translate signals into decisions under uncertainty, constraints, and risk. The paper positions predictive cognitive AI and Decision Engineering Science as complementary layers within a unified architecture for decision systems. In this architecture, predictive models provide signal processing and representation capabilities, while DES provides decision architecture, evaluation mechanisms, and decision quality optimization. This work contributes to the emerging shift in artificial intelligence from prediction-centric systems toward decision-centric systems. It is particularly relevant for applications in high-stakes domains, including finance, healthcare, governance, and legal systems, where the quality of decisions is critical. The paper is intended as a foundational note within the broader development of Decision Engineering Science and serves as a bridge between predictive AI research and the engineering of decision systems.
Aleksandra Pinar (Thu,) studied this question.