This paper explores the potential of Kolmogorov-Arnold Networks (KAN) as an alternative to traditional Multi-Layer Perceptrons (MLPs) for hand gesture classification. While MLPs are integral to many AI applications, they have limitations such as inefficiency in parameter usage and a lack of interpretability. KANs address these issues by introducing adaptive, univariate functions that serve as both weights and activation functions, enhancing both the efficiency and flexibility of the network. Experiments were conducted using surface electromyography (sEMG) data from 30 participants performing seven hand gestures. The data were processed through a series of steps, including signal filtering, windowing, and feature extraction. The results showed that KANs performed comparably to, or better than, traditional classifiers such as Neural Networks, Random Forest, and Extreme Gradient Boosting, particularly when larger window sizes were used. However, the training time and risk of overfitting remain challenges. Overall, KANs demonstrate promise for low-resource systems like prosthetics, warranting further exploration in real-world applications and with larger datasets.