This paper proposes a learning framework that empowers AI agents to develop intrinsic exploration motivation based on cognitive and achievement emotions (surprise and pride) generated during data observation. This approach aims to empower AI models with the ability to explore the physical environment, essential for information acquisition and knowledge integration in living organisms. The proposed dual-module reinforcement learning framework induces surprise or pride based on data analysis scores and optimizes the correlation between these emotional states and exploration behaviors to enable agents to achieve learning objectives. Experimental results demonstrate a causal relationship between emotional states and exploration behaviors in most agents, with surprise increasing by an average of 15.4% and pride decreasing by an average of 2.8%. The correlation coefficients for surprise and pride, ρ surprise = 0.461 and ρ pride = -0.237, respectively, are consistent with previous research on human behavior. In conclusion, this paper demonstrates that biologically inspired AI development can impart life-like advantages, such as autonomy, to AI, and empirically demonstrates that AI methodologies can support findings from human behavior research.
