AI and deep learning can be taught to recognize images through supervised learning using labeled training data. This allows systems to correctly identify objects like faces 95% of the time. Unsupervised learning allows AI to analyze data without labels, like YouTube videos, to learn patterns on its own. Applications of image recognition include Facebook's Deepface system which can identify 97% of faces, and medical diagnosis tools. While classical AI relied on extensive rule-based programming, modern AI using deep learning has vastly improved thanks to increased computing power and its ability to handle problems that can't be easily reduced to equations.