AI, ML, and DL Have Boundaries

A map before the journey

Analogy: Artificial intelligence (AI) is the whole city: any engineered system that performs a task associated with human intelligence. Machine learning (ML) is one neighborhood where systems learn patterns from examples. Deep learning (DL) is one part of that neighborhood, built from neural networks with many successive layers.

AI is the broadest term. A hand-written chess search, a rules engine that approves expenses, and a model that recognizes speech can all be called AI, even though only the last system necessarily learned from data. ML replaces at least some hand-written decision rules with a model learned from examples. DL is ML performed with multi-layer neural networks; large language models (LLMs) are a deep-learning application specialized for sequences of tokens. These sets nest as LLM ⊂ deep learning ⊂ machine learning ⊂ AI, but marketing often blurs them.

Artificial intelligence
└── Machine learning
    ├── Classical ML: linear models, trees, clustering
    └── Deep learning: CNNs, RNNs, Transformers
        └── Large language models

A useful test is to ask where behavior came from. If a programmer explicitly wrote every decision rule, it may be AI but it is not learned ML. If parameters were adjusted from examples, it is ML. If those parameters live in a many-layer neural network, it is DL.

Warning: None of these labels guarantees intelligence, truth, autonomy, or suitability. They describe how a system is built, not whether it solves the user's problem safely or well.