Non-Linear Computational Modeling of Emotional-Like States in Artificial Agents

Modelling of internally generated affective dynamics without explicit symbolic encoding remains a central challenge in the long-term quest for Artificial Consciousness. This paper proposes a unified computational framework for generating non rule-based emotional dynamics, moving beyond symbolic rule-based approaches. We define emotion as the emergent result of a weighted, non-linear integration of heterogeneous cognitive factors, including social pressure, internal metabolic states, and memory traces. Our architecture relies on continuous transfer functions (hyperbolic tangent, sigmoid) and a Softmax-based competitive mechanism to model resource allocation and emotional saturation. We further introduce an endogenous stochastic modulation term, characterized by 1/f noise, to ensure trajectory uniqueness and avoid deterministic redundancy. Simulation results demonstrate the model's capacity to generate distinct, stable psychological profiles (e.g., Extremely Sociable vs. Very Stressed) and to reproduce biological properties such as hysteresis and resilience. These findings suggest that "high-intensity affective functional state" can be rigorously defined as a functional state of high-intensity cognitive integration, providing a scalable foundation for autonomous ethical agents.