“Intelligence is that quality that enables an entity to function appropriately and with foresight in its environment.”
Nils J. Nilsson spent his career helping define artificial intelligence through robotics, search, planning, and the history of the field. In “The Quest for Artificial Intelligence,” he offered a functional definition: “Intelligence is that quality that enables an entity to function appropriately and with foresight in its environment.” The wording avoids requiring an entity to think exactly like a human. Instead, it asks whether behavior fits a situation and anticipates what may happen next. Intelligence becomes a relationship among an agent, its goals, and its environment rather than a trophy awarded for performing one impressive task.
“Appropriately” is the demanding word. A chess program may play brilliantly inside a formal board game while being unable to recognize a broken piece or explain why anyone plays. A language model may produce an excellent plan but fail when facts are missing, instructions conflict, or the social setting changes. Performance cannot be judged apart from the environment in which a system will operate. The criteria for a warehouse robot differ from those for a medical assistant or a classroom tutor. Each system needs evidence that covers ordinary conditions, edge cases, changing circumstances, and the consequences of an error.
Foresight adds another dimension. Intelligent action is not only a reaction to the present; it considers downstream effects. For AI, that may mean planning several steps ahead, estimating uncertainty, or avoiding choices that create irreversible harm. For the people deploying AI, foresight means asking how incentives will change once a system enters an institution. A scoring tool can alter the behavior of applicants, workers, or officials. An assistant that makes work faster can also reduce opportunities for learning. Anticipating those feedback effects requires social knowledge as well as technical evaluation.
Nilsson’s definition offers a practical antidote to debates based solely on scale or spectacle. A system is not intelligent in the abstract; it is more or less capable relative to a setting, objective, and time horizon. That perspective encourages precise claims and safer deployment. Describe the environment, define what appropriate behavior means, test whether the system recognizes uncertainty, and examine how far its planning reaches. Most importantly, keep human institutions responsible for choosing the goal. Foresight is valuable only when the future being pursued is one that affected people have reason to accept. This approach also rewards modest systems that work reliably in a well-defined environment over grand claims unsupported by deployment evidence. Intelligence becomes something demonstrated through appropriate, anticipatory conduct, not a label granted by impressive language or the size of a model.
Nilsson defined intelligence by successful, anticipatory behavior rather than by resemblance to a particular human mental process.
That makes the definition applicable to animals, people, robots, and software while still requiring context-specific evidence.
A capability demonstrated in a benchmark may disappear when inputs, goals, or consequences differ from the test conditions.
Responsible evaluation names the environment and includes the uncertainty, edge cases, and feedback found there.
An AI system may plan ahead, but deployers must also anticipate how its presence changes behavior, incentives, and access.
Longer-term effects belong in the definition of appropriate use, especially when decisions are difficult to reverse.
Explore more "Quotes of The Day"
Continue exploring how writers, thinkers, artists, and advocates have described the choices that shape an everyday life.