Human intelligence is distinguished not merely by the capacity to solve problems, but by the ability to generate new explanatory structures — to create, test, and revise ideas about the world. This paper introduces Creative Scientific Intelligence (CSI) as a theoretical and computational model of this uniquely human capability. Building on insights from cognitive psychology, epistemology, and artificial intelligence, CSI formalizes creativity as a recursive process of epistemic calibration: agents generate hypotheses, perform interventions, and revise internal models to resolve tension between expectation and observation. Structurally, CSI integrates three mechanisms central to human creative reasoning — recursive abstraction, analogical transfer, and epistemic tension regulation — implemented through multi-timescale calibration loops that parallel human learning, development, and insight formation.

Empirical demonstrations in symbolic physics and ecosystem-simulation environments show how a CSI agent can autonomously discover hidden causal laws and generate novel structural models through self-directed exploration. These results suggest that scientific creativity and cognitive intelligence share a common recursive architecture grounded in explanation-driven learning. By modeling intelligence as the dynamic capacity to construct, test, and refine internal theories, CSI provides a formal account of how understanding — and not just performance — emerges from curiosity, surprise, and self-correction.

The framework offers a unified view of creative and scientific cognition, with implications for measuring and enhancing human intelligence across educational, artistic, and scientific domains.