Note: This article is part of AI Technology for Lawyers, a series explaining the technical foundations of AI for legal professionals. Start with the series introduction.
The term "artificial intelligence" encompasses a broad range of technologies, and imprecise usage creates confusion. A chess program from 1997, a spam filter, a Netflix recommendation engine, and ChatGPT are all "AI" in some sense, but they work in fundamentally different ways and raise different legal issues. Getting the hierarchy right prevents confusion in everything that follows.
Four Levels of Specificity
There are four levels in the AI hierarchy, each a subset of the one above:
- Artificial Intelligence is the broadest category. It includes any system designed to simulate aspects of human intelligence. This includes rule-based systems where programmers explicitly coded decision logic, as well as systems that learn from data. The spam filters of the early 2000s were AI. So were the chess programs that preceded them.
- Machine Learning is a subset of AI where systems learn patterns from data rather than following explicitly programmed rules. Instead of a programmer writing "if the email contains 'Nigerian prince,' mark as spam," a machine learning system examines millions of emails and learns patterns that distinguish spam from legitimate messages. The system is not told the rules; it derives them from examples.
- Deep Learning uses neural networks with multiple layers. The "deep" refers to the number of layers between input and output, not to any quality of understanding. Think of it as machine learning with a specific architectural approach that has proven remarkably effective for certain tasks. Deep learning powers image recognition, speech processing, and language models.
- Large Language Models (LLMs) are deep learning models trained on enormous amounts of text to predict the next token (a chunk of text, not necessarily a whole word) given prior context. This sounds simple, but capabilities can appear abruptly with scale and training, sometimes unexpectedly. GPT-5, Claude, Gemini, and Llama are all LLMs.
Why the Hierarchy Matters
When someone says "AI," the appropriate legal analysis depends entirely on which level they mean. A rule-based system that follows explicit logic raises different liability questions than a probabilistic model that generates novel outputs. A narrow classifier that labels documents as "relevant" or "not relevant" has a different risk profile than a generative model that drafts contracts.
The hierarchy also clarifies marketing claims. When a vendor says their product uses "AI," ask which kind. A product built on simple decision trees has different capabilities and different failure modes than one built on a large language model.
Continue here to the next article in the series: How Models Learn
