“It is the science and engineering of making intelligent machines, especially intelligent computer programs.”
John McCarthy’s definition of artificial intelligence is direct: “It is the science and engineering of making intelligent machines, especially intelligent computer programs.” He had helped establish the field decades earlier, including proposing the term “artificial intelligence” for the 1956 Dartmouth workshop. The wording matters because it gives AI two linked identities. It is a science that investigates the principles behind reasoning, learning, perception, and action. It is also engineering that turns those principles into working artifacts. An elegant theory is incomplete if it cannot guide construction, while a useful program can raise scientific questions that its builders did not anticipate.
The definition also avoids tying intelligence to a single technique. AI has included symbolic logic, search, planning, expert systems, probabilistic models, neural networks, robotics, and combinations of all of them. Each method captures different regularities and fails in different ways. McCarthy’s emphasis on “making” encourages comparison by behavior and purpose rather than fashion. A system should be examined according to the task it performs, the knowledge or data it uses, the conditions in which it succeeds, and the kinds of mistakes it produces. Those questions are more informative than treating every new product as a mysterious general intelligence.
Calling AI engineering carries a further implication: design choices are human choices. Objectives, training material, interfaces, safety constraints, and deployment settings do not arrive naturally from mathematics. They reflect judgments about what counts as success and whose needs matter. A hiring model, medical assistant, or tutoring system may share underlying techniques, yet each belongs to a different institutional environment with different stakes. Responsible engineering therefore requires more than increasing benchmark scores. It demands testing with affected communities, monitoring after release, clear routes for appeal, and limits on uses where evidence is weak or harm would be difficult to reverse.
McCarthy’s definition remains useful because it keeps ambition and accountability in the same frame. AI can be a profound investigation into intelligence while also being a collection of tools built for particular purposes. Scientific curiosity explains why researchers explore unfamiliar capabilities; engineering discipline explains why systems need specifications, measurements, and failure plans. When public debate swings between magical promises and total dismissal, the quote offers a steadier view. Ask what kind of intelligence is being modeled, what has actually been built, how well it works outside a demonstration, and whether the people responsible have designed for the world in which the program will operate. Seen this way, every AI claim has two parts: a claim about the nature of intelligence and a claim about the performance of an engineered system. Keeping both visible helps society appreciate genuine invention while still asking for specifications, evidence, maintenance, and responsibility.
Science asks what intelligence requires and how it can be represented, learned, or evaluated.
Engineering converts those ideas into systems with defined goals, constraints, interfaces, and measurable performance.
McCarthy’s wording does not privilege rules, statistics, neural networks, or any particular architecture.
That breadth makes the definition durable across generations of AI research and changing technical fashions.
Intelligent programs inherit choices about objectives, data, tests, and acceptable failure from the people and institutions that build them.
Good engineering makes those choices visible and matches safeguards to the consequences of the system’s use.
Explore more "Quotes of The Day"
Continue exploring how writers, thinkers, artists, and advocates have described the choices that shape an everyday life.