Multiple Sequence Alignment (MSA) presents a significant difficulty in computational biology, with essential consequences for evolutionary inference, protein structure prediction, and functional annotation. Conventional MSA methodologies frequently depend on heuristics or dynamic programming, which can become computationally intensive and less efficient when addressing complex or large scale sequence sets. The Augmented Random Search (ARS) algorithm is adopted as an alternate optimization approach to tackle these difficulties. This study proposes the utilization of ARS, a gradient-free reinforcement learning algorithm, to tackle the problem of MSA. MSA is treated as a black-box optimization issue, where ARS iteratively enhances sequence configurations by random perturbations of a linear policy, guided by alignment quality measurements like the Sum-of-Pairs (SP) score. The method has proven effective in exploring the solution space and generating high-quality alignments, suggesting its promise as a viable and scalable algorithm for solving the MSA problem.