This paper presents a conceptual and theoretical proposal for understanding cognitive development in artificial agents beyond performance-driven optimization. While contemporary AI systems—particularly large language models—exhibit impressive linguistic and problem-solving capabilities, they remain largely static, lacking a persistent cognitive history that shapes how they think over time. The work introduces the notion of an Evolving Cognitive Agent (ECA), a bounded artificial architecture designed to accumulate experience, preserve refuted and contradictory possibilities, and modify internal cognitive strategies through self-reflection rather than parameter updates or retraining. Imagination is treated not as a correctness-oriented mechanism, but as an unconstrained space in which logical, illogical, possible, and impossible ideas may coexist. The paper does not claim consciousness, subjective experience, or human-like imagination. Instead, it offers a clear theoretical framework intended to stimulate discussion in developmental AI, cognitive architecture research, and the philosophy of artificial intelligence. The proposed model aims to shift the focus from optimizing outputs to understanding how artificial systems might undergo meaningful cognitive change as a result of their own histories.
Najm abe housh (2025) studied this question.