THE SHORT ANSWER
Artificial intelligence is a broad family of machine-based systems that infer how to produce outputs—such as predictions, recommendations or content—from inputs. An AI system can be useful at a task without thinking or understanding in the human sense.
A category of systems, not one kind of mind
The OECD definition focuses on inference, objectives and outputs. It includes approaches based on machine learning as well as knowledge-based approaches. AI therefore does not mean only chatbots, and it does not require a claim that a machine is conscious.
Three questions make the label more useful: What input does this system receive? What output does it produce? Who decides what happens next? A system that ranks documents and a system that controls a robot may both use AI, but their consequences differ considerably.
Evidence & context: OECD
Learning a pattern versus following a fixed rule
Consider an illustrative school-library task: sorting book enquiries. A fixed rule might route every message containing ‘renew’ to a renewal queue. A machine-learning approach could learn patterns from labelled examples, including different ways people ask for more time.
The learned approach may handle variation the rule misses. It can also make mistakes that are harder to explain. The choice is not automatically ‘AI is better’. If a simple rule solves the actual problem reliably, the extra flexibility may not be worth the uncertainty.
Separate the model from the product
A model is one component. A finished application can add a user interface, a database of documents, search tools, permissions and human review. A product's ability to find current information may come from its search connection rather than the model alone.
The distinction matters when comparing generative AI with an AI agent. Generating text describes an output; choosing and carrying out steps describes how a larger system operates.
Judge the task, not the impressive demonstration
Imagine the library tool routes a request to the wrong shelf. That is inconvenient. If a similar classification system affects access to a service, an error may have much greater consequences. Ask what kinds of mistakes matter, whether affected people can challenge them and what happens when the system is unsure.
- Name one concrete task and the person it is meant to help.
- Describe a useful output and a harmful mistake.
- Identify who checks the output and can correct the decision.
- Compare the AI option with a simpler method.
These questions do not settle every debate about AI. They turn a vague claim about intelligence into a problem you can investigate.
Sources & further reading
- What is AI? The updated OECD definition
OECD. A policy definition and explanation of the boundaries of an AI system; not a test of consciousness.
Examples and exercises are illustrative unless attributed to a source. No independent expert review is claimed.
A correction, a counterexample or an experience worth sharing?
Join the conversation ↗