Artificial Curiosity, also known as intrinsic motivation in AI, is a research area that aims to create systems with an inherent desire to acquire knowledge independently, much akin to human curiosity. The concept is derived from the theories of developmental psychology and cognitive neuroscience. It allows AI systems to go beyond the limitations of pre-programmed knowledge and discover information autonomously by interacting with their environments. The model is often used in reinforcement learning, in which the AI, driven by artificial curiosity, explores new states and improves its policy over time. Notably, this ability for self-directed learning enables AI to tackle problems where supervision or rewards may be sparse or delayed, facilitating its versatility and adaptability.
The concept of artificial curiosity emerged in the 1990s' cognitive sciences, but it was first applied to AI and robotic systems by researchers such as Juergen Schmidhuber and Pierre-Yves Oudeyer in the early 2000s. The term gained popularity over the following decade as AI systems started showing significant learning enhancements when imbued with this attribute.
Researchers like Juergen Schmidhuber, known for his work on neural networks and AI, and Pierre-Yves Oudeyer, a leader in developmental robotics, significantly contributed to the development and application of artificial curiosity in AI systems. Their pioneering work established the foundations of artificial curiosity, influencing its subsequent evolution and application across various AI domains.