AI is a very diverse field, with many subfields, including:
machine learning (ML) is the study of algorithms that learn patterns from data to make predictions or decisions without explicit programming. As of this writing (2025), it is widely agreed that ML is the central subfield within AI, largely due to the deep learning revolution. ML includes the following subareas/topics:
supervised learning: the most common setting of ML, where the training data contains an annotated label for each training example; e.g., each house has a price so that you can train a housing price predictor, or each email has a spam/not-spam label so that you can train a spam filter. We will discuss supervised learning in Unit 3 of our course, which includes two subsettings:
unsupervised learning: training data does not include labels. Here we aim to find the inherent structures within the data, e.g., clustering and dimensionality reduction.
reinforcement learning: the training data (in this case, the optimal sequence of actions to perform a task) is not labeled, but you can get rewards from the environment. Think about playing chess: there is no annotation to tell you whether each move is good or bad, but at the end of each game, you get a reward signal of win (+1), lose (-1), or draw (0). The goal is to learn from these (often delayed) reward signals and attribute them to individual actions, so that good actions (which eventually lead to positive rewards) will be preferred.
deep learning (DL): a branch of ML using multilayer artificial neural networks to model complex, high-dimensional data such as images, speech, and text. We will discuss DL in Unit 4.
natural language processing (NLP) and speech: NLP focuses on enabling computers to understand, interpret, and generate human language in the text format, powering tools like chatbots, translation systems, and writing assistants. Speech processing includes speech recognition and speech synthesis, enabling computers to understand and generate natural speech. NLP and speech technologies are often combined in systems such as speech translation tools. We will discuss these topics in Unit 4.
computer vision (CV) gives machines the ability to see and interpret visual information, such as recognizing objects, faces, gestures, or scenes from images and videos. More recently, CV also expands to include image generation and video generation. There is also interaction between NLP and CV, such as image captioning and text-to-image generation. We will discuss these topics in Unit 4.
robotics combines AI with mechanical engineering to create autonomous systems capable of sensing, planning, and acting in the physical world. As of this writing (2025), robotics still lags far behind NLP: general-purpose chatbots can win Gold Medals in International Olympiads in both Mathematics and Informatics (programming) which represent the highest level of human intelligence, but robots still lack the abilities to perform very basic tasks.
knowledge representation and reasoning (KRR): Involves structuring information so that AI systems can reason logically, make inferences, and answer questions about the world. This subfield was at the center of AI in the 1980s but lost its center stage to ML/DL since 1990s.
AI planning: a subfield of AI deciding what sequence of actions an intelligent agent should take to achieve a specific goal, given a model of the world. The system must reason about the future, predicting how its actions will change the world. Traditionally, it often uses search and logic-based methods to explore possible action sequences, but recently, it is more and more associated with reinforcement learning.
AI search is the process of systematically exploring possible actions or states to find a path from a starting point to a goal. It is similar to but simpler than AI planning in that it does not need to explicitly model the world. The central topic here is “heuristic search”. Historically, this was the first AI subfield to boom (in the 1960s). Applications include reasoning, theorem proving, and game playing (Chess/Go). For example, the landmark AlphaGo (2016) project combines AI search (Monte-Calo Tree Search) with deep reinforcement learning.
The last three (more symbolic) subfields will be discussed in Unit 2.