Decision Dynamics: A Formal Theory of Decision Systems Across Biological and Artificial Agents Toward a Unified Framework for Cognitive Alignment and Action Under Constraints

Contemporary artificial intelligence systems have achieved significant advances in prediction, inference, and representation learning. However, a fundamental gap persists between predictive capability and effective action. Systems optimized for accuracy do not necessarily produce coherent, admissible, or aligned decisions in real-world environments. This work introduces Decision Dynamics as a formal framework for modeling decision-making as a dynamical, constraint-driven process, rather than as a byproduct of predictive inference. Within this perspective, decisions are defined as structured selections over feasible action spaces under evolving constraints, embedded within temporal trajectories and feedback mechanisms. The paper establishes a unified architecture for decision systems, integrating: state-dependent action formation, constraint-based admissibility, dynamic interaction between objectives and limitations, feedback-driven adaptation, and governance as a meta-structural layer. A key contribution is the introduction of Decision Quality Index (DQI) as a structural metric for evaluating decision systems beyond traditional performance measures. The framework further formalizes signal sensitivity as a critical property governing how systems detect, interpret, and respond to relevant changes in their environment. Importantly, the work connects formal decision system architecture to neurobiological mechanisms, interpreting cortico–basal ganglia circuits as action selection systems, prefrontal cortex as a constraint integration layer, and dopaminergic processes as dynamic valuation and learning signals. This establishes a bridge between biological and artificial decision systems within a common theoretical framework. The paper argues that contemporary AI architectures lack an explicit decision layer, leading to systemic failures such as decision drift, constraint violations, feedback degradation, and temporal misalignment. In response, it proposes a shift toward decision-centric system design, where normative, predictive, and decision layers are explicitly separated and integrated. This work positions Decision Engineering Science™ as a foundational discipline for the design and evaluation of decision systems across domains, including enterprise systems, healthcare, autonomous systems, and human–AI collaboration. By reframing intelligence as the capacity to generate coherent, constrained, and adaptive action over time, Decision Dynamics provides a theoretical and engineering foundation for the next generation of decision systems.