The term encompasses a range of mathematical, statistical, and computational methods aimed at developing autonomous algorithms (not created by humans) capable of solving tasks by identifying patterns in various input data. These patterns may be imperceptible to humans. A vivid example is facial recognition: AI currently outperforms humans in this task by a considerable margin. Humans might mistake twins for the same person or consider an individual to be different people if captured with varying facial expressions, from different angles, and under different lighting conditions.
Before the advent of machine learning, programs were deterministic, operating based on clear-cut algorithms. Now, programs employing ML autonomously select available methods for solving given tasks. They also have the ability to learn from their mistakes and use rewards for correct responses to improve performance.