Algorithm stability refers to the resilience of an algorithm's output to minor variations in input data, essentially determining its consistency and reliability. It becomes particularly crucial in machine learning, where stable algorithms tend to generalize better to unseen data, making them more robust and dependable in practical applications.