Purpose : This study aims to develop a game-based exercise content that integrates a convolutional neural network (CNN) algorithm with artificial intelligence (AI), and evaluate its performance. Methods : Thirty healthy adults were assigned to either the experimental group (AI convergence exercise; EG) or the control group (CG). The AI convergence exercise, based on the CNN algorithm, not only analyzes the movements of a user in real time, but also enables immediate integration with the game program. The participants are tasked with controlling an avatar, whose movements are instantly synchronized with their own, to perform activities such as evading by moving their entire body in all directions (forward, backward, left, and right). The outcome of the intervention was assessed using proprioceptive sensory measure and the limits of stability in left, right, forward, backward, and total directions. Results : The results showed significant improvements on the proprioceptive sensory measure and the limits of stability in the left, right, forward, backward, and total directions in the EG post-intervention. Specifically, there were notable changes in the limit of stability in all directions. In the CG, no significant change was observed in proprioceptive sensory of dorsi, plantar, and knee flexions; however, significant changes were observed in limits of stability in the left, right, forward, backward, and total directions. There was no significant difference between the groups in proprioceptive sensory measures; however, significant differences were observed between the groups in limits of stability in the left, right, forward, backward, and total directions. Conclusion : This study demonstrates that AI convergence exercise positively affects the proprioceptive sensory and balance ability of healthy adults. These results suggest that the AI convergence exercise is beneficial for balance and proprioceptive sensory improvements.