Artificial Creativity: from predictive AI to Generative System 3.

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Computational Creativity

Background:

• Large language models (LLMs) excel at text generation but lack sustained novelty, task relevance, and diversity.
• Human creativity involves a tri-process loop: spontaneous ideation (System 1), goal-directed evaluation (System 2), and metacognitive control (System 3).
• Current LLMs implement only fragments of this loop, hindering their creative potential.

Purpose of the Study:

• Introduce Generative System 3 (GS-3), an architecture-agnostic design pattern to foster artificial creativity.
• Address the limitations of current LLMs in generating novel, relevant, and diverse content over extended contexts.
• Provide a formal framework and testable roadmap for developing genuinely creative generative systems.

Main Methods:

• Conceptual analysis formalizing novelty, usefulness, and diversity with operational definitions.
• Development of multiple gain-update policies (exponential, linear, logistic) with stability constraints.
• Derivation of falsifiable behavioral indices: associative-distance density, analytic-verification ratio, and convergence latency.
• Proposal of a proof-of-concept blueprint and evaluation protocol for GS-3.

Main Results:

• GS-3 integrates a high-entropy generator, a learned critic, and an adaptive gain controller.
• Identifies key missing components for artificial creativity: internal evaluation, endogenous control over sampling entropy, and adaptive priors.
• Defines formal metrics and pass-fail criteria for evaluating creative AI systems.
• Outlines ethical considerations, including bias mitigation and reward gaming prevention.

Conclusions:

• GS-3 offers a testable roadmap for advancing generative AI beyond mere prediction towards genuine creativity.
• The proposed framework enables the development of AI systems capable of sustained novelty, relevance, and diversity.
• GS-3 provides a foundation for more sophisticated co-creative systems and a deeper understanding of artificial creativity.